WO2017205836A1

WO2017205836A1 - Systems and methods for categorization

Info

Publication number: WO2017205836A1
Application number: PCT/US2017/034843
Authority: WO
Inventors: Shameer KHADER; Alla Mabrouk Salem OMAR; Joel T. DUDLEY; Partho P. Sengupta
Original assignee: Icahn School Of Medicine At Mount Sinai
Priority date: 2016-05-26
Filing date: 2017-05-26
Publication date: 2017-11-30

Abstract

Methods and apparatus for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications are provided in which a plurality of ultrasound gray-scale measurement images of the heart across a plurality of heartbeats is obtained. A plurality of cardiac parameters is determined from the measurement images. In this determination, two or more of the images contributes to each cardiac parameter. The parameters are subjected to a linear discriminant function thereby obtaining a first prediction of the classification. The parameters are also subjected to a weighted neighborhood scheme thereby obtaining a second prediction of the classification. The parameters are further subjected to an artificial neural network thereby obtaining a third prediction of the classification. The first, second, and third prediction are applied to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart.

Description

SYSTEMS AND METHODS FOR CATEGORIZATION

CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority to United States Provisional Patent Application

Number 62/341,813, entitled "Diagnosis and Classification of Left Ventricular Diastolic

Dysfunction Using a Computer," filed May 26, 2016, which is hereby incorporated by reference.

TECHNICAL FIELD

[0002] This following relates generally to apparatus and methods for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications.

BACKGROUND

[0003] At present, 500,000 new cases of congestive heart failure are diagnosed every year, and between 40% - 60% of those are instances of diastolic dysfunction. This condition is the most common discharge diagnosis for patients aged over 65 and is the most expensive disease for the Center for Medicare and Medicaid Systems. Early diagnosis and proper management of diastolic dysfunction can improve the prognosis of patients who have this disease. However, this requires not merely identification of the existence of diastolic

dysfunction, but classification of such dysfunction into established categories of severity.

[0004] Diastolic dysfunction is classified into three classifications: mild dysfunction

(Grade I), dysfunction (Grade II), and severe dysfunction (Grade III). In some instances, the classification of a particular patient's diastolic dysfunction is not clear or cannot be made at all. This is because patients are not selected for diagnosis using a single diagnostic methodology and some patients lacking diastolic dysfunction are selected for diagnosis for this condition.

Furthermore, such classification is carried out by physicians and subjectivity cannot be completely eliminated from the classification process. [0005] Medical imaging in the era of precision medicine can be directed for extracting large-scale information for accurate disease phenotyping and personalized therapies. Leveraging algorithms designed for big data analytics could help to develop automated, precision

phenotyping models using cardiac imaging datasets. In particular, the use of the new variables to develop automated predictive models for phenotyping different modalities of diastolic dysfunction.

[0006] Heart failure (HF) is a major public health problem in the US. It is estimated that by 2030, HF prevalence will grow by 25% and annual costs of care will increase from $21 to $53 billion. See, Heidenreich, et al., 2013, "Forecasting the impact of heart failure in the United States: a policy statement from the American Heart Association," Circulation Heart failure 6(3):606-619. Clinical, laboratory and cardiac imaging multi-scale data need integration for accurate phenotypic characterization in HF and designing cost-effective strategies for reliably identifying early stages and high-risk populations. See, Shah et al., 2015, "Phenomapping for novel classification of heart failure with preserved ejection fraction," Circulation 131(3):269- 279; and Francis et al, 2014, "The heterogeneity of heart failure: will enhanced phenotyping be necessary for future clinical trial success?", Journal of the American College of Cardiology 64(17): 1775-1776.

[0007] Amongst the cardiac imaging modalities, two-dimensional and Doppler echocardiography techniques are widely used in HF patients for assessing left ventricular (LV) structural and functional abnormalities. See Nagueh, 2009, "Echocardiographic assessment of left ventricular relaxation and cardiac filling pressures," Current heart failure reports. 6(3): 154- 159, which is hereby incorporated by reference. In addition newer approaches in tracking natural myocardial markers, or speckles in echocardiography images provide incremental

characterization of myocardial functional abnormalities beyond ejection fraction. See Lang et al., 2015, "Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging," J Am Soc Echocardiogr. 28(1): 1-39 el4, which is hereby incorporated by reference. Recent multicenter studies and global scientific consortiums have therefore enthusiastically endorsed standardizing and automating speckle tracking echocardiography (STE) approaches for routine clinical application. See, Yang et al., 2015, "Improvement in Strain Concordance between Two Major Vendors after the Strain Standardization Initiative," J Am Soc Echocardiogr. 28(6):642-648 e647; Voigt et al., 2015, "Definitions for a common standard for 2D speckle tracking echocardiography: consensus document of the

EACVI/ASE/Industry Task Force to standardize deformation imaging," J Am Soc Echocardiogr. 28(2): 183-193; and Knackstedt et al., 2015, "Fully Automated Versus Standard Tracking of Left Ventricular Ejection Fraction and Longitudinal Strain: The FAST-EFs Multicenter Study," Journal of the American College of Cardiology 66(13): 1456-1466.

[0008] It would therefore be advantageous to provide a computer based method for diagnosing the existence of diastolic dysfunction and for classifying the severity of such dysfunction into predetermined classifications, particularly mild dysfunction, dysfunction, and severe dysfunction (Grade I, Grade II, and Grade III dysfunction, respectively). However, this is difficult to accomplish because it is difficult to obtain data for a large population of subjects that have diastolic dysfunction without confounding factors such as systemic co-morbidities (e.g. malignancies, terminal hepatic failure, and end-stage chronic renal disease on dialysis), more than mild degree of valve disease, and pericardial diseases. As such, conventional classifiers trained using data from subjects with diastolic dysfunction are prone to error and typically do not have accuracies exceeding eighty percent.

SUMMARY

[0009] The present disclosure addresses the need in the prior art by providing machine learning techniques that diagnose whether diastolic dysfunction exists in a patient, and where dysfunction is found to be present, to classify it into a one of a plurality of predetermined classifications. More particularly, three different types of machine learning algorithms - a linear discriminant function, a weighted neighborhood scheme, and an artificial neural network program are trained in an initial training phase. A plurality of cohorts is identified during this phase. In one cohort, the individuals all lack diastolic dysfunction. And, for each of the predetermined classifications, a cohort of individuals having that classification of diastolic dysfunction is identified. From each individual in each cohort, and for at least one complete cardiac cycle in each individual, a set of image data representing motion of that individual's heart is acquired.

[0010] From each set of image data, a corresponding set of training data representing distortion of preselected regions of each individual's heart is acquired. Then, for each cohort, all sets of training data are input to the linear discriminant function, and the cohort-by cohort variation in linear discriminant function from the linear classifier program is noted. Similarly, for each cohort, all sets of training data are input to the weighted neighborhood scheme, and the cohort-by-cohort variation in the weighted neighborhood scheme is noted. Likewise, similarly, for each cohort, all sets of training data are input to the artificial neural network program, and the cohort-by-cohort variation in artificial neural network output from the artificial neural network is noted. The linear discriminant function, the weighted neighborhood scheme, and the artificial neural network are then output to a stratifier program, and cohort-by-cohort variation in stratifier output from the stratifier program is noted. The stratifier outputs are then associated with the existence and classifications of diastolic dysfunction in individuals, and once this association has been made, the preliminary training phase is concluded. By using this unique combination of classifiers, a fairly small patient population can be used to produce a trained ensemble classifier that has a high degree of accuracy. This is highly advantageous because a large training population is difficult to obtain given the highly invasive procedures necessary to acquire the training data and because data from a substantial number of potential training subjects must be discarded due to poor image quality, existence of systemic co-morbidities (e.g. malignancies, terminal hepatic failure, and end-stage chronic renal disease on dialysis), more than mild degree of valve disease, and existence of pericardial diseases.

[0011] After the preliminary training phase has been concluded, diagnosis of new untrained subjects is possible with the trained ensemble classifier. In some embodiments, the trained classifiers discussed above are incorporated into an apparatus for categorization of the diastolic dysfunction of a heart of a human patient into a classification in a set of classifications. The apparatus comprises one or more processors and memory storing one or more programs for execution by the one or more processors. The one or more programs singularly or collectively execute a method.

[0012] In the method, a plurality of ultrasound gray-scale measurement images of the heart of the subject is obtained across a plurality of heartbeats of the heart. A plurality of cardiac parameters is determined from the plurality of measurement images. In this determining process, two or more images in the plurality of images contributes to each respective cardiac parameter in the plurality of cardiac parameters. [0013] The plurality of cardiac parameters is subjected to a linear discriminant function thereby obtaining a first prediction of the classification in the set of classifications. The plurality of cardiac parameters is also subjected to a weighted neighborhood scheme thereby obtaining a second prediction of the classification in the set of classifications. The plurality of cardiac parameters is also subjected to an artificial neural network thereby obtaining a third prediction of the classification in the set of classifications.

[0014] The first prediction, the second prediction, and the third prediction are

collectively applied to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart of the patient.

[0015] In some embodiments, the linear discriminant function is a kernel function. In some embodiments, the kernel function is a support vector machine. In some embodiments, the weighted neighborhood scheme is a random forest or a k-nearest neighbor algorithm.

[0016] In some embodiments, each image in the plurality of images is two-dimensional.

In alternative embodiments, the plurality of images provides a three-dimensional image of the heart (e.g., either natively or collectively through a series of rotated two-dimensional images).

[0017] In some embodiments, the set of classifications consists of a first classification and a second classification. For instance, in one such embodiment, the first classification is grade I diastolic dysfunction or normal and the second classification is grade II diastolic dysfunction or grade III diastolic dysfunction. In other words, if a subject is scored "grade I diastolic dysfunction" or "normal" they are placed in the first classification, and if a subject is scored "grade II diastolic dysfunction" or "grade III diastolic dysfunction" they are placed in the second classification. In some such embodiments, the majority voting method deems the classification to be the classification in the set of classifications that matches two or more of the group consisting of the first prediction, the second prediction and the third prediction.

[0018] In alternative embodiments, the set of classifications comprises four or more classifications. For instance, in some such embodiments, the set of classifications consists of a first classification, a second classification, a third classification, and a fourth classification. For instance, in one such embodiment, the first classification is grade I diastolic dysfunction, the second classification is grade II diastolic dysfunction, the third classification is grade III diastolic dysfunction, and the forth classification is normal. In order to score four classifications, in some embodiments, the linear discriminant function is a plurality of support vector machines, implemented on a one-versus-all or a one-versus-one basis, where each respective support vector machine in the plurality of support vector machines distinguishes between a different pair of classifications in the set of classifications and where the plurality of support vector machines thereby represents each possible pair of classifications in the set of classifications. In such instances, the classification that wins the most in the plurality of support vector machines is deemed by the linear discriminant function to be the classification of a test subject. In some particular embodiments, the plurality of support vector machines is implemented on a one- versus-one basis.

[0019] In embodiments where the set of classifications comprises four or more classifications, the majority voting method deems the classification to be the classification in the set of classifications that matches two or more of the group consisting of the first prediction, the second prediction and the third prediction. If the majority voting method does not achieve two or more of the group consisting of the first prediction, the second prediction and the third prediction, meaning that each classifier voted for a different classification in the set of classifications, then the classification is not called. In such situations, conventional diagnostic methods are invoked or the subject is reimaged and the classification is repeated using the ensemble classifier.

[0020] In some embodiments, the set of classifications consists of three classifications.

In some embodiments, the set of classifications comprises five or more classifications.

[0021] In some embodiments, the plurality of ultrasound gray-scale measurement images are obtained from an echocardiographic quantification system. For instance, in some

embodiments, the echocardiographic quantification system is a speckle tracking

echocardiographic quantification system. In some embodiments, the echocardiographic quantification system is a tissue Doppler system.

[0022] In some embodiments, the plurality of ultrasound gray-scale measurement images are taken of the heart over at least a two minute period. In some embodiments, the plurality of ultrasound gray-scale measurement images are taken of the heart over at least a three minute period. In some embodiments, the plurality of ultrasound gray-scale measurement images are taken of the heart over a period of between two minutes and ten minutes. In some embodiments, the plurality of ultrasound gray-scale measurement images are taken of the heart over a period of time that is less than eight minutes. [0023] In some embodiments, the plurality of ultrasound gray-scale measurement images comprises a first subset of images and a second subset of images. In such embodiments, the first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2- chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable, and the second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. In some such embodiments, the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure in which a first left atrial strain curve is computed from the first subset of images, and a first left ventricular strain curve is also computed from the first subset of images. Further, a second left atrial strain curve is computed from the second subset of images, and a second left ventricular strain curve is computed from the second subset of images. The first left atrial strain curve and the second left atrial strain curve are averaged thereby forming an averaged left atrial strain curve. Also, the first left ventricular strain curve and the second left ventricular strain curve are averaged thereby forming an averaged left ventricular strain curve. Further still, a cardiac parameter in the plurality of cardiac parameters is atrio-ventricular strain of the heart (AV-S), where AV-S is computed as the average of simultaneous peak left ventricular systolic strain (LV-S) and peak left atrial strain during left ventricular systole (LA-S), wherein the LV-S and LA-S are derived from the averaged left atrial strain curve and the averaged left ventricular strain curve, in accordance with the formula:

AV-S = [(LA-S + LV-S)/2].

[0024] In some embodiments, the plurality of ultrasound gray-scale measurement images comprises a first subset of images and a second subset of images. In such embodiments, the first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2- chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable, and the second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. In some such embodiments, a cardiac parameter in the plurality of cardiac parameters is an averaged left atrio-ventricular end diastolic volume (TLH-diast) of the heart. Further, the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure in which a respective left atrio-ventricular end diastolic volume is computed from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the first subset of images thereby obtaining a first plurality of left atrio-ventricular end diastolic volumes. In this procedure a respective left atrioventricular end diastolic volume is also computed from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the second subset of images thereby obtaining a second plurality of left atrio-ventricular end diastolic volumes. The first plurality of left atrioventricular end diastolic volumes and the second plurality of left atrio-ventricular end diastolic volumes are averaged thereby forming the averaged left atrio-ventricular end diastolic volume (TLH-diast).

[0025] In some embodiments, the plurality of ultrasound gray-scale measurement images comprises a first subset of images and a second subset of images. In such embodiments, the first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2- chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. The second subset of images comprises a second collection of gray-scale 2- dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. In some such embodiments, a cardiac parameter in the plurality of cardiac parameters is an averaged left atrio-ventricular end systolic volume of the heart. In such embodiments, the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure in which a respective left atrio-ventricular end systolic volume is computed from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the first subset of images thereby obtaining a first plurality of left atrio-ventricular end systolic volumes. Further, a respective left atrio-ventricular end systolic volume is computed from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the second subset of images thereby obtaining a second plurality of left atrio-ventricular end systolic volumes. The first plurality of left atrio-ventricular end systolic volumes and the second plurality of left atrio-ventricular end systolic volumes are averaged thereby forming an averaged left atrio-ventricular end systolic volume (TLH-syst).

[0026] In some embodiments, the plurality of ultrasound gray-scale measurement images comprises a first subset of images and a second subset of images. In such embodiments, the first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2- chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable, and the second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. In some such embodiments, the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure in which a first curve comprising a volume rate plotted against time during a plurality of cardiac cycles for left atrial is computed from the first subset of images, and a second curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left ventricular is also computed from the first subset of images. Further, a third curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left atrial is computed from the second subset of images, and a fourth curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left ventricular is computed from the second subset of images. The first curve and the third curve are averaged thereby forming a first averaged curve comprising a volume rate plotted against time during the plurality of cardiac cycles for the left atrial of the heart. The second curve and the fourth curve are also averaged thereby forming a second averaged curve comprising a volume rate plotted against time during the plurality of cardiac cycles for the left ventricular of the heart. Then, an average simultaneous single beat atrio-ventricular peak volume rate during left ventricular systole (AVVR-S / AV s / AVVR.S) is computed, as a first cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve. Also, an average simultaneous single beat atrio-ventricular peak volume rate during early diastole (AVVR-E / AVVR.E / AVVRE) is computed, as a second cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve. Further, an average simultaneous single beat atrio-ventricular peak volume rate during left atrial contraction (AVVR-A / AVVR. A / AVVRA) is computed, as a third cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve.

[0027] In some embodiments, the plurality of ultrasound gray-scale measurement images comprises a first subset of images and a second subset of images. In such embodiments, the first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2- chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable, and the second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. In some such embodiments, the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure in which a first curve comprising a strain rate plotted against time during a plurality of cardiac cycles is computed for left atrial from the first subset of images. Further, a second curve comprising a strain rate plotted against time during the plurality of cardiac cycles is computed for left ventricular from the first subset of images. Further still, a third curve comprising a strain rate plotted against time during the plurality of cardiac cycles is computed for the left atrial from the second subset of images. Also, a fourth curve comprising a strain rate plotted against time during the plurality of cardiac cycles is computed for the left ventricular from the second subset of images. The first curve and the third curve are averaged thereby forming a first averaged curve comprising a strain rate plotted against time during the plurality of cardiac cycles for the left atrial of the heart. The second curve and the fourth curve are averaged thereby forming a second averaged curve comprising a strain rate plotted against time during the plurality of cardiac cycles for the left ventricular of the heart. An average simultaneous single beat atrio-ventricular peak strain rate during left ventricular systole (AVSRS / AVSR-S / AVSR. S) is computed, as a first cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve. An average simultaneous single beat atrio-ventricular peak strain rate during early diastole (AVSRE / AVSR-E / AVSR.E) is computed, as a second cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve. An average simultaneous single beat atrioventricular peak strain rate during left atrial contraction (AVSRA / AVSR-A / AVSR. A) is computed, as a third cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve.

[0028] Another aspect of the present disclosure provides a method for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications. The method comprises obtaining a plurality of ultrasound gray-scale measurement images of the heart of the subject across a plurality of heartbeats of the heart. A plurality of cardiac parameters is determined from the plurality of measurement images, wherein two or more images in the plurality of images contributes to each respective cardiac parameter in the plurality of cardiac parameters. The plurality of cardiac parameters is subjected to a linear discriminant function thereby obtaining a first prediction of the classification in the set of classifications. The plurality of cardiac parameters is subjected to a weighted neighborhood scheme thereby obtaining a second prediction of the classification in the set of classifications. The plurality of cardiac parameters is subjected to an artificial neural network thereby obtaining a third prediction of the classification in the set of classifications. The first prediction, the second prediction, and the third prediction are applied to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart.

[0029] Still another aspect of the present disclosure is a non-transitory computer readable storage medium storing one or more programs for execution by one or more processors in a computer system for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications. The one or more programs comprise instructions for obtaining a plurality of ultrasound gray-scale measurement images of the heart of the subject across a plurality of heartbeats of the heart. A plurality of cardiac parameters is determined from the plurality of measurement images. Two or more images in the plurality of images contributes to each respective cardiac parameter in the plurality of cardiac parameters. The plurality of cardiac parameters is subjected to a linear discriminant function thereby obtaining a first prediction of the classification in the set of classifications. The plurality of cardiac parameters is subjected to a weighted neighborhood scheme thereby obtaining a second prediction of the classification in the set of classifications. The plurality of cardiac parameters is subjected to an artificial neural network thereby obtaining a third prediction of the classification in the set of classifications. The first prediction, the second prediction, and the third prediction are applied to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart.

BRIEF DESCRIPTION OF THE DRAWINGS [0030] In the drawings, embodiments of the systems and method of the present disclosure are illustrated by way of example. It is to be expressly understood that the description and drawings are only for the purpose of illustration and as an aid to understanding, and are not intended as a definition of the limits of the systems and methods of the present disclosure.

[0031] FIG. 1 illustrates an apparatus for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications in accordance with some embodiments.

[0032] FIGS. 2A, 2B, and 2C collectively illustrate a method for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications in which elements in dashed boxes are optional, in accordance with some embodiments.

[0033] FIG. 3 illustrates an artificial neural network implemented in an embodiment of the present disclosure.

[0034] FIG. 4 illustrates a random forest implemented in an embodiment of the present disclosure.

[0035] FIG. 5 illustrates a support vector machine in accordance with an embodiment of the present disclosure.

[0036] FIG. 6 illustrates speckle tracking echocardiography (STE) of volume expansion of the left atrial (LA) of the heart in accordance with an embodiment of the present disclosure.

[0037] FIG. 7 illustrates speckle tracking echocardiography of wall deformation in accordance with an embodiment of the present disclosure.

[0038] FIG. 8 illustrates volumes plotted against time (volume curves) during a cardiac cycle for the left atrial (LA) (802), left ventricular (LV) (804), as well as total left heart volume (TLH) (806), and further in which the atrio-ventricular volume at left ventricular end-systole TLH-syst (TLV-s) and the atrio-ventricular volume at left ventricular end-diastole (TLH-diast (TLV-d) are illustrated in accordance with some embodiments.

[0039] FIG. 9 illustrates strain plotted against time during a cardiac cycle for left atrial

(LA) (blue) (902), left ventricular LV (red) (904), in which the magnitude of global left atrial strain (LAS) and the magnitude of global left ventricular strain (LVS) is depicted and the calculation of the atrio-ventricular strain (AV-S) as half the sum of the instantaneous maximal absolute values of LAS and LVS (906) is further depicted in accordance with some embodiments.

[0040] FIG. 10 illustrates volume rates plotted against time during a cardiac cycle for left atrial LA (blue) (1002), left ventricular LV (red) (1004) in which atrio-ventricular volume rate during left ventricular systole (AVws, black dotted arrow 1006), during early diastole (AVVRE, black dotted arrow 1008) and during late diastole (AVVRA, black dotted arrow 1010) were calculated as half the sum of the instantaneous maximal absolute corresponding values of the left atrial LA and left ventricular LV (LAVR-S and LVVR-S for AVws; LVVR-E and LAVR-E for AVWF; and LVVR-A and LAVR-A for 1010) in accordance with some embodiments.

[0041] FIG. 11 illustrates strain rate plotted against time during a cardiac cycle for left atrial LA (blue) (1102) and left ventricular LV (red) (1104) in which atrio-ventricular strain rate during systole (AVSRS, black dotted arrow 1106), during early diastole (AVSRE, black dotted arrow 1108) and during late diastole (AVSRA, black dotted arrow 1110) were calculated as half the sum of the instantaneous maximal absolute corresponding values of the LA and LV and further in which the corresponding values left atrial single beat atrio-ventricular peak strain rate during left ventricular systole (LASR-S), left atrial single beat atrio-ventricular peak strain rate during early diastole (LASR-E), left atrial single beat atrio-ventricular peak strain rate during left atrial contraction (LASR-A), left ventricular single beat atrio-ventricular peak strain rate during left ventricular systole (LVSR-S), left ventricular single beat atrio-ventricular peak strain rate during early diastole (LVSR-E), and left ventricular single beat atrio-ventricular peak strain rate during left atrial contraction (LVSR-A) are depicted in accordance with an embodiment of the present disclosure.

[0042] FIG. 12 illustrates diastolic dysfunction classifications in accordance with an embodiment of the present disclosure.

[0043] FIG. 13 illustrates a study protocol in accordance with an embodiment of the present disclosure.

[0044] Like reference numerals refer to corresponding parts throughout the several views of the drawings. DETAILED DESCRIPTION

[0045] Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be apparent to one of ordinary skill in the art that the present disclosure may be practiced without these specific details. In other instances, well-known methods, procedures, components, circuits, and networks have not been described in detail so as not to unnecessarily obscure aspects of the embodiments.

[0046] It will also be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first subject could be termed a second subject, and, similarly, a second subject could be termed a first subject, without departing from the scope of the present disclosure. The first subject and the second subject are both subjects, but they are not the same subject.

[0047] The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the description of the invention and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

[0048] As used herein, the term "if may be construed to mean "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [a stated condition or event] is detected" may be construed to mean "upon determining" or "in response to determining" or "upon detecting [the stated condition or event]" or "in response to detecting [the stated condition or event]," depending on the context. [0049] An aspect of the present disclosure provides methods and apparatus for categorization of the diastolic dysfunction of a heart into a classification in a set of

classifications. In such methods and apparatus, a plurality of ultrasound gray-scale measurement images of the heart across a plurality of heartbeats is obtained. Then, a plurality of cardiac parameters is determined from the measurement images. In this determination, two or more of the images contributes to each cardiac parameter. The parameters are then subjected to three different classifiers, either sequentially or concurrently. The parameters are not subjected to the three different classifiers in any particular order. In some embodiments, the parameters are only subjected to one of the classifiers and the output of this single classifier is used to determine the classification in the set of classifications of the heart. In some embodiments, the parameters are only subjected to two of the classifiers and the combined output of these two classifiers is used to determine the classification in the set of classifications of the heart.

[0050] In preferred embodiments, the parameters are only subjected to all three of the classifiers. Accordingly, in such embodiments, the plurality of cardiac parameters is subjected to a linear discriminant function. In other words, the plurality of cardiac parameters serves as input to the linear discriminant function, where the linear discriminant function has been previously trained to classify into the set of classifications. In this way, a first prediction of the

classification is obtained. The parameters are also subjected to a weighted neighborhood scheme. In other words, the plurality of cardiac parameters serves as input to the weighted neighborhood scheme, where the weighted neighborhood scheme has been previously trained to classify into the set of classifications. In this way, a second prediction of the classification is obtained. The parameters are further subjected to an artificial neural network. In other words, the plurality of cardiac parameters serves as input to the artificial neural network, where the artificial neural network has been previously trained to classify into the set of classifications. In this way, a third prediction of the classification is obtained. The first, second, and third prediction are applied to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart. That is, at least two of the predictions have to predict the same classification in the set of classifications in order for a classification to be called.

[0051] Figure 1 illustrates a computer system 100 that applies the above-described categorization of the diastolic dysfunction of a heart. For instance, it can be used as a system to categorize, or diagnose, a diastolic dysfunction of a heart. Moreover, it can be used as a system to determine that a heart does not have a diastolic dysfunction.

[0052] Referring to Figure 1, in typical embodiments, analysis computer system 100 comprises one or more computers. For purposes of illustration in Figure 1, the analysis computer system 100 is represented as a single computer that includes all of the functionality of the disclosed analysis computer system 100. However, the disclosure is not so limited. The functionality of the analysis computer system 100 may be spread across any number of networked computers and/or reside on each of several networked computers, or be implemented in a cloud computing environment. For instance, all or some of the modules and data illustrated in Figure 1 can be implemented in a virtual machine. One of skill in the art will appreciate that a wide array of different computer topologies are possible for the analysis computer system 100 and all such topologies are within the scope of the present disclosure.

[0053] Turning to Figure 1 with the foregoing in mind, an analysis computer system 100 comprises one or more processing units (CPU' s) 74, a network or other communications interface 84, a user interface 78 (e.g., including a display 82 and keyboard 80 or other form of input device) a memory 92 (e.g., random access memory, volatile memory), one or more magnetic disk storage and/or persistent devices 90 optionally accessed by one or more controllers 88, one or more communication busses 12 for interconnecting the aforementioned components, and a power supply 76 for powering the aforementioned components.

[0054] Data in memory 92 can be seamlessly shared with non-volatile memory 90 using known computing techniques such as caching. In some embodiments, memory 92 and/or memory 90 includes mass storage that is remotely located with respect to the central processing unit(s) 74. In other words, some data stored in memory 92 and/or memory 90 may in fact be hosted on computers that are external to analysis computer system 100 but that can be electronically accessed by the analysis computer system over an Internet, intranet, or other form of network or electronic cable using network interface 84.

[0055] The memory 92 of analysis computer system 100 stores:

• an operating system 40 that includes procedures for handling various basic system

services;

• a categorization module 42 for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications;

• a measurement image dataset 44 that stores a plurality of images (e.g., ultrasound grayscale measurement images), each such image 46 being an image of the heart, the plurality of images taken across a plurality of heartbeats of the heart (e.g., across two or more cardiac cycles);

• a cardiac parameter dataset 48, in which two or more images 46 in the plurality of images contributes to each respective cardiac parameter 50 in the plurality of cardiac parameters, and where each respective cardiac parameter 50 informs the choice of the classification in a set of classifications for a heart;

• a linear discriminant function 52 that uses the cardiac parameter dataset 48 for a heart to determine a classification in the set of classifications as a first prediction result 54;

• a weighted neighborhood scheme 56 that uses the cardiac parameter dataset 48 for a heart to determine a classification in the set of classifications as a second prediction result 58;

• an artificial neural network 60 that uses the cardiac parameter dataset 48 for a heart to determine a classification in the set of classifications as a third prediction result 62;

• a majority voting method 64 that collectively applies the first prediction 54, the second prediction 58, and the third prediction 62 to a majority voting method thereby obtaining the classification 66, in the set of classifications, for the diastolic dysfunction of the heart.

[0056] In some implementations, one or more of the above identified data elements or modules of the analysis computer system 100 are stored in one or more of the previously disclosed memory devices, and correspond to a set of instructions for performing a function described above. The above identified data, modules or programs (e.g., sets of instructions) need not be implemented as separate software programs, procedures or modules, and thus various subsets of these modules may be combined or otherwise re-arranged in various implementations. In some implementations, the memory 92 and/or 90 optionally stores a subset of the modules and data structures identified above. Furthermore, in some embodiments the memory 92 and/or 90 stores additional modules and data structures not described above.

[0057] Now that a system for categorization of the diastolic dysfunction of a heart into a classification has been disclosed, methods for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications (202) is detailed with reference to Figure 2 and discussed below. [0058] The set of classifications discriminated against using the categorization module

42. In some embodiments, the set of classifications constitute degrees of diastolic dysfunction. Diastolic dysfunction is graded from mild (grade I) to severe (grade III) with increasing likelihood of symptomatic heart failure and worse prognosis with higher grade dysfunction. As such, in some embodiments, the set of classifications that are discriminated against by the categorization module 42 consists of a first classification and a second classification (204). In some such embodiments, the first classification is grade I diastolic dysfunction or normal and the second classification is grade II diastolic dysfunction or grade III diastolic dysfunction (206). In other words, if a subject is scored "grade I diastolic dysfunction" or "normal" they are placed in the first classification, and if a subject is scored "grade II diastolic dysfunction" or "grade III diastolic dysfunction" they are placed in the second classification.

[0059] Referring to Table 1 for explanation of parameters as well as Figure 12, in some embodiments, grade I (mild) diastolic dysfunction is defined as a mitral E/A ratio is <0.8, predominant systolic flow in the pulmonary venous flow (S>D), annular e' <8 cm/s (septal and lateral), and E/e' ratio <8 (septal and lateral). A reduced mitral E/A ratio in the presence of normal annular TD velocities can occur in normal old individuals and is typically not used to diagnose diastolic dysfunction.

[0060] In some embodiments, Grade II diastolic dysfunction is defined as mitral E/A ratio is >1, and average E/e' ratio (septal and lateral) is >10. In some patients with moderate diastolic dysfunction, left ventricular (LV) end diastolic pressure is the only pressure that is increased and recognized by Ar-A duration >30 ms.

[0061] In some embodiments, Grade III severe diastolic dysfunction is defined as restrictive LV filling occurs with an E/A ratio >2, DT <160 ms, IVRT <70 ms, systolic filling fraction <40 percent, and average E/e' ratio >13. LV filling may revert to one of impaired relaxation with successful therapy in some patients, whereas in others LV filling remains restrictive. The latter response predicts increased morbidity and mortality.

Table 1 : Conventional echocardiographic parameters.

Table 1 : Conventional echocardiographic parameters.

Table 1 : Conventional echocardiographic parameters.

[0062] Table 1 constitutes conventional variables. They include parameters that are either used in the conventional recommended algorithms for the assessment of diastolic function, in addition to other parameters that are derived with the conventional modalities such as Doppler echocardiography, tissue Doppler echocardiography, and 2D-echocardiography. In some embodiments, the parameters of Table 1 are determined as follows. A commercially available echocardiography system equipped with a 2.5-MHz multi -frequency phased array transducer (Vivid 7 or E9, GE-Vingmed, Horton, Norway) is used to obtain images. Digital routine grayscale 2-dimensional loops from apical 2- and 4-chamber views with 3 consecutive beats were obtained with both LV and LA clearly and completely visualized. Left ventricular end diastolic volume (EDV), end systolic volume (ESV), and ejection fraction (EF) were calculated using the biplane Simpson's method of discs and left atrial maximum volume (LAVmax) and minimum volume (LAVmin) were calculated using the biplane area-length method. All measurements were made in >3 consecutive cardiac cycles and average values were used for the final analyses. The pulsed-wave Doppler-derived trans-mitral velocity and tissue Doppler-derived mitral annular velocities were obtained from the apical 4-chamber view. The early diastolic wave velocity (E), late diastolic atrial contraction wave velocity (A), and the E-wave deceleration time (E-DcT) were measured using pulsed-wave Doppler recording. Spectral pulsed-wave tissue Doppler-derived early and late diastolic velocities (e' and a') were averaged from the septal and lateral mitral annular positions. The averaged E/e' ratio was calculated as a Doppler

echocardiographic estimate of the LVFP.

[0063] Referring to element 208 of Figure 4, in some embodiments, the set of classifications comprises four or more classifications (e.g., consist of four classifications). In some such embodiments, the set of classifications consists of four classifications in which the first classification is grade I diastolic dysfunction, the second classification is grade II diastolic dysfunction, the third classification is grade III diastolic dysfunction, and the forth classification is normal (210).

[0064] In some alternative embodiments, the set of classifications consists of three classifications (212). For instance, the first classification is normal or grade I diastolic dysfunction, the second classification is grade II diastolic dysfunction, and the third classification is grade III diastolic dysfunction.

[0065] It will be appreciated that the grades of diastolic dysfunction discussed above and illustrated in Figure 12 can be further segmented into any number of classifications. For instance, grade I diastolic dysfunction can be segregated into two grades, grade II diastolic dysfunction can be segregated into two grades, and so forth. As such, in some embodiments, the set of classifications comprises five or more classifications (214), ten or more classifications, or fifteen or more classifications.

[0066] Obtain images. Referring to element 216 of Figure 2 A, in the method a plurality of ultrasound gray-scale measurement images 46 of the heart across a plurality of heartbeats of the heart is obtained. In typical embodiments, the heart is a human heart of a patient in need of diagnosis.

[0067] In some embodiments, each image 46 in the plurality of images is two- dimensional (218). In some embodiments, the plurality of images provides a three-dimensional image of the heart (220). For instance, in some embodiments, the plurality of images provides a three-dimensional full-volume dataset that is acquired by a fully sampled matrix array transducer such as the X5-1/X3-1 (Philips Medical Systems or 4V, GE Healthcare).

[0068] In some embodiments, the images are not ultrasound images but rather are magnetic resonance images. In some embodiments, the images are not gray-scale images but rather are color-scale images.

[0069] In some embodiments, the plurality of ultrasound gray-scale measurement images are obtained from an echocardiographic quantification system (222). For instance, in some embodiments, the plurality of ultrasound gray-scale measurement images are obtained from an echocardiographic quantification system {e.g., a speckle tracking echocardiographic

quantification system, a tissue Doppler system, etc.) (224). For instance, in some embodiments, the echocardiographic quantification system is a commercially available echocardiography system equipped with a 2.5-MHz multi -frequency phased array transducer (Vivid 7 or E9, GE- Vingmed, Horton, Norway). In some embodiments, the echocardiographic quantification system is an ultrasound machine and transducer (iE33, Philips Medical System, An-dover,

Massachusetts or Vivid 7 or E9, GE Healthcare, Horten, Norway).

[0070] Referring to element 226 of Figure 2B, in some embodiments, the plurality of ultrasound gray-scale measurement images are taken of the heart of a subject in need of classification over at least a two minute period, at least a three minute period, or over a period of time that is between two and eight minutes. The intention is to acquire images that collectively encompass two or more complete cardiac cycles of the heart. As such, referring to elements 228 of Figure 2B, in some embodiments, the plurality of ultrasound gray-scale measurement images are taken of the heart over at least two cardiac cycles, at least a three cardiac cycles, or between three and eight cardiac cycles.

[0071] Advantageously, referring to element 230 of Figure 2B, in some embodiments the plurality of ultrasound gray-scale measurement images comprise a first subset of images and a second subset of images. The first subset of images comprises a first collection of gray-scale 2- dimensional loops from apical 2-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. The second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. In some embodiments, the first subset of images encompasses at least three consecutive heartbeats and the second subset of images also encompasses at least three consecutive heartbeats.

[0072] In some embodiments, the plurality of ultrasound gray-scale measurement images are all gray-scale 2-dimensional loops from apical 2-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. That is, in such embodiments, apical 4-chamber views are not acquired in such embodiments.

[0073] In some embodiments, the plurality of ultrasound gray-scale measurement images are all gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. That is, in such embodiments, apical 2-chamber views are not acquired in such embodiments. [0074] Determine a plurality of cardiac parameters. In some embodiments, cardiac parameters 48 were obtained from the images 50 using speckle tracking echocardiography (STE). STE is a method of identifying a peculiar pattern along a curve in moving images for the assessment of displacement and deformation of myocardial segments throughout the cardiac cycle. In echocardiography, images 46 are characterized by the presence of speckles with a certain persistence which can be tracked for calculation of cardiac parameters 50 (e.g., myocardial deformation parameters) in both the left atrium and the left ventricle. Cardiac parameters 50 include strain and strain rate. In addition, other cardiac parameters such as volumes and volume expansion rates can be also assessed using STE. Cardiac parameters 50 used in some embodiments of the present disclosure are summarized in table 2 and their acquisition and derivation is explained below.

[0075] Image acquisition and STE parameters derivation. In some embodiments frame- by-frame movement assessment of the stable patterns of the left ventricular (LV) and left atrial (LA) speckles was done in the apical 4-chamber images 46 and 2-chamber images 46. The endocardial borders of both the LA and LV were traced at the end-diastolic frame, identified as one frame before mitral valve closure at end-diastole. LA and LV speckle tracking was then performed during the cardiac cycle and the instantaneous changes in volume, volume rates, longitudinal strain and strain rate were obtained from both apical views and averaged.

[0076] Single beat speckle tracking Echocardiography (STE) derived volume and strain measurements. In some embodiments, STE-derived volume curves were used for defining LV- EDV and ESV, LAVmax, and LAVmin. From the strain curves, simultaneous peak left ventricular systolic strain (LV-S) and peak left atrial strain during LV systole (LA-S) were also measured. Atrio-ventricular strain (AV-S) was calculated as the average of the magnitude of global LA strain (LAS) and LV strain (LVS) [AV-S= (LAS + LVS)/2], where both LV-S and LA-S are taken as positive values. From the volume rate and strain rate curves, simultaneous diastolic volume rate at early and late diastole, and strain rate at early and late diastole and in peak ventricular systole of the LV, and of the LA were measured. Finally, atrio-ventricular volume rate at early and late diastole as well as strain rate during early and late diastole and peak ventricular systole (VR-EAV, VR-AAV, SR-EAV, SR-AAV, SR-SAV, respectively) were calculated by averaging the respective LV and LA absolute values. [0077] Referring to element 232 of Figure 2B, a plurality of cardiac parameters is obtained, where two or more images in the plurality of images contributes to each respective cardiac parameter in the plurality of cardiac parameters. Table 2 provides a summary of cardiac parameters that are acquired in accordance with one aspect of the present disclosure.

[0078] Advantageously, these cardiac parameters rely on images from both apical 2- chamber views and apical 4-chamber views. Thus, in such embodiments, the plurality of ultrasound gray-scale measurement images comprise a first subset of images and a second subset of images. The first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. The second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable.

[0079] To compute one cardiac parameter 50 in accordance with the present disclosure, a first left atrial strain curve is computed from the first subset of images. A first left ventricular strain curve is also computed from the first subset of images. A second left atrial strain curve is also computed from the second subset of images. Further, a second left ventricular strain curve is computed from the second subset of images. The first left atrial strain curve and the second left atrial strain curve are averaged thereby forming an averaged left atrial strain curve. The first left ventricular strain curve and the second left ventricular strain curve are averaged thereby forming an averaged left ventricular strain curve. Figure 9 illustrates. In particular, Figure 9 illustrates strain plotted against time during a cardiac cycle for left atrial (LA) (902) and left ventricular LV (red) (904). Thus, curve 902 is average left atrial strain curve and is formed as the average of the first left atrial strain curve and the second left atrial strain curve. As such, curve 902 is a composite of the apical 2-chamber views and apical 4-chamber views. As illustrated in Figure 9, these strain curves include two complete cardiac cycles. These curves are used to compute one cardiac parameter 50 that is used in the present disclosure, atrio-ventricular strain of the heart (AV-S). To compute AV-S from the strain curves depicted in Figure 9, the magnitude of global left atrial strain (LAS) and the magnitude of global left ventricular strain (LVS) is obtained from these curves for each cardiac cycle. For instance, if there are three cardiac cycles, LVS (LV-S) and LAS (LA-S) is obtained from each cardiac cycle. LVS/LAS values are matched to their respective cardiac cycles. In other words, LVS in cardiac cycle 1 is matched to LAS in cardiac cycle 1, and so forth. Atrio- ventricular strain (AV-S) is calculated as half the sum of the instantaneous maximal absolute values of LAS and LVS across two or more cardiac cycles. For instance, consider the case in which there are two cardiac cycles. There will be a first instantaneous maximal absolute values of LAS and LVS for cardiac cycle 1, and a second instantaneous maximal absolute values of LAS and LVS for cardiac cycle 2. In such instances, AVS is calculated as the average of (i) half the sum of the first instantaneous maximal absolute values of LAS and LVS and (ii) half the sum of the second instantaneous maximal absolute values of LAS and LVS. Thus, the cardiac parameter AVS is a parameter whose value leverages multiple cardiac cycles in two different views of the heart (apical 2-chamber views and apical 4-chamber views). As such, a cardiac parameter in the plurality of cardiac parameters is atrio-ventricular strain of the heart (AV-S), is computed as the average of simultaneous peak left ventricular systolic strain (LV-S) and peak left atrial strain during left ventricular systole (LA-S), where the LV-S and LA-S are derived from the averaged left atrial strain curve and the averaged left ventricular strain curve, in accordance with the formula:

AV-S = [(LA-S + LV-S)/2]. In typical embodiments, AV-S is computed across three or more cardiac cycles.

Table 2: Speckle tracking echocardiography derived parameters

PARAMETER DESCRIPTION

AVVR-S (or AVVRS) Average simultaneous single beat atrio- ventricular peak volume rate during left ventricular systole

AVVR-E (or AVVR.E) Average simultaneous single beat atrio- ventricular peak volume rate during early diastole

AVVR-A (or AVVR.A) Average simultaneous single beat atrio- ventricular peak volume rate during left atrial contraction

AV-VRE/VRA (or AVVR- Ratio between average simultaneous single beat atrioE/VRA, or AV-E/A, or AV- ventricular peak volume rate during early diastole to average E.A.) simultaneous single beat atrio- ventricular peak volume rate during atrial systole

AV-VRE/SRE (or AVVR- Ratio between average simultaneous single beat atrioE/SRE, or AV-E/Ep, or ventricular peak strain rate during early diastole to average AV.E.Ep.) simultaneous single beat atrio-ventricular peak strain rate during atrial contraction

TLSV TLSV- Total left ventricular Stroke volume:

Difference between atrio-ventricular volume at left ventricular end-diastole and atrio-ventricular volume at left ventricular end-systole

TLEF Total left sided ejection fraction: calculated as TLSV/TLV- d

[0080] Referring to Figure 8, the cardiac parameter left atrio-ventricular end diastolic volume (TLH-diast) of the heart is computed as follows. First, as was the case for computation of AVS above, the plurality of images 44 include images from both apical 2-chamber views and apical 4-chamber views. Thus, in such embodiments, the plurality of ultrasound gray-scale measurement images comprise the first subset of images and the second subset of images. The first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. The second subset of images comprises a second collection of gray-scale 2- dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable. Figure 8 illustrates volumes plotted against time (volume curves) during a cardiac cycle for the left atrial (LA) (802), left ventricular (LV) (804), as well as total left heart volume (TLH) (806), and further in which the atrio-ventricular volume at left ventricular end-systole TLH-syst (TLV-s) and the atrio-ventricular volume at left ventricular end-diastole (TLH-diast (TLV-d) are illustrated in accordance with some

embodiments.

[0081] A respective left atrio-ventricular end diastolic volume is computed from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the first subset of images thereby obtaining a first plurality of left atrio-ventricular end diastolic volumes. A respective left atrio-ventricular end diastolic volume is also computed from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the second subset of images thereby obtaining a second plurality of left atrio-ventricular end diastolic volumes. The first plurality of left atrio-ventricular end diastolic volumes and the second plurality of left atrio-ventricular end diastolic volumes are averaged thereby forming the averaged left atrio-ventricular end diastolic volume (TLH-diast). In practice, curve 806 is the average of (i) volumes plotted against time (volume curves) during a plurality of cardiac cycles from the first subset of images and volumes plotted against time (volume curves) during a plurality of cardiac cycles from the first second of images. As such, curve 806 encompasses average total left heart volume from both the apical 4- chamber views and apical 2-chamber views. Thus, in some embodiments, the averaging the first plurality of left atrio-ventricular end diastolic volumes and the second plurality of left atrioventricular end diastolic volumes is accomplished by taking timepoints TLH-diast off of line 806 of Figure 8.

[0082] Average left atrio-ventricular end systolic volume (TLH-syst) is acquired in a similar manner. A respective left atrio-ventricular end systolic volume from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the first subset of images is measured thereby obtaining a first plurality of left atrio-ventricular end systolic volumes. A respective left atrio-ventricular end systolic volume from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the second subset of images is obtained thereby obtaining a second plurality of left atrio-ventricular end systolic volumes. The first plurality of left atrio-ventricular end systolic volumes and the second plurality of left atrio-ventricular end systolic volumes are averaged thereby forming an averaged left atrio-ventricular end systolic volume (TLH-syst). In practice, curve 806 of Figure 8 is the average of (i) volumes plotted against time (volume curves) during a plurality of cardiac cycles from the first subset of images and volumes plotted against time (volume curves) during a plurality of cardiac cycles from the first second of images. As such, curve 806 encompasses average total left heart volume from both the apical 4-chamber views and apical 2-chamber views. Thus, in some embodiments, the averaging the first plurality of left atrio-ventricular end systolic volumes and the second plurality of left atrio-ventricular end systolic volumes is accomplished by taking timepoints TLH-syst off of line 806 of Figure 8.

[0083] Referring to Figure 10, the computation of three additional cardiac parameters 50 of Table 2, AVVR-S, AVVR-E, and AVVR-A, are described. A first curve comprising a volume rate plotted against time during a plurality of cardiac cycles for left atrial from the first subset of images (apical 2-chamber view) is computed. A second curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left ventricular from the first subset of images is also computed. A third curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left atrial from the second subset of images is computed. A fourth curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left ventricular from the second subset of images is computed.

[0084] The first curve and the third curve are averaged thereby forming a first averaged curve comprising a volume rate plotted against time during the plurality of cardiac cycles for the left atrial of the heart (curve 1002 in Figure 10). The second curve and the fourth curve are averaged together thereby forming a second averaged curve comprising a volume rate plotted against time during the plurality of cardiac cycles for the left ventricular of the heart (curve 1004 in Figure 10). An average simultaneous single beat atrio-ventricular peak volume rate during left ventricular systole (AVVR-S / AVVRS / AVVR.S) is computed, as a first cardiac parameter 50 of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial (LAVR-S) and left ventricular (LVVR-S) drawn from the denoted positions in the cardiac cycle of the first averaged curve 1002 and the second averaged curve 1004. In practice, several pairs of LAVR-S / LVVR-S are drawn from curves 1002 / 1004, each pair representing a consecutive cardiac cycle and the half the sum of these pairs are averaged together to compute AVVR-S.

[0085] An average simultaneous single beat atrio-ventricular peak volume rate during early diastole (AVVR-E / AVVR.E / AVVRE) is computed as a second cardiac parameter 50 of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial (LAVR-E) and left ventricular (LVVR-E) drawn from the denoted positions in the cardiac cycle of the first averaged curve 1002 and the second averaged curve 1004. In practice, several pairs of LVVR-E / LAVR-E) are drawn from curves 1002 / 1004, each pair representing a consecutive cardiac cycle and the half the sum of these pairs are averaged together to compute AVVR-E.

[0086] An average simultaneous single beat atrio-ventricular peak volume rate is computed during left atrial contraction (AVVR-A / AVVR. A / AVVRA), as a third cardiac parameter 50 of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial (LAVR-A) and left ventricular (LVVR-A) drawn from the first averaged curve 1002 and the second averaged curve 1004. In practice, several pairs of LVVR-A / LAVR- A) are drawn from curves 1002 / 1004, each pair representing a consecutive cardiac cycle and the half the sum of these pairs are averaged together to compute AVVR-A.

[0087] Referring to Figure 11, the computation of three additional cardiac parameters 50 of Table 2, AVSR-S, AVSR-E, and AVSR-A, are described. A first curve comprising a strain rate plotted against time during a plurality of cardiac cycles for left atrial is computed from the first subset of images (apical 2-chamber view). A second curve comprising a strain rate plotted against time during the plurality of cardiac cycles for left ventricular from the first subset of images (apical 4-chamber view). A third curve comprising a strain rate plotted against time during the plurality of cardiac cycles for left atrial is computed from the second subset of images. A fourth curve comprising a strain rate plotted against time during the plurality of cardiac cycles for left ventricular from the second subset of images. The first curve and the third curve are averaged together thereby forming a first averaged curve comprising a strain rate plotted against time during the plurality of cardiac cycles for the left atrial of the heart (curve 1102 of Figure 11). The second curve and the fourth curve are averaged together thereby forming a second averaged curve comprising a strain rate plotted against time during the plurality of cardiac cycles for the left ventricular of the heart (curve 1104 of Figure 11).

[0088] An average simultaneous single beat atrio-ventricular peak strain rate during left ventricular systole (AVSRS / AVSR-S / AVSR.S) is computed, as a first cardiac parameter 50 of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial (LASR-S) and left ventricular (LVSR-S) drawn from the first averaged curve and the second averaged curve. In practice, several pairs of LASR-S / LVSR-S are drawn from curves 1102 / 1104, each pair representing a consecutive cardiac cycle and the half the sum of these pairs are averaged together to compute AVSRS.

[0089] An average simultaneous single beat atrio-ventricular peak strain rate during early diastole (AVSRE / AVSR-E / AVSR.E) is computed, as a second cardiac parameter 50 of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial (LASR-E) and left ventricular (LVSR-E) drawn from the first averaged curve and the second averaged curve. In practice, several pairs of LVSR-E / LASR-E are drawn from curves 1102 / 1104, each pair representing a consecutive cardiac cycle and the half the sum of these pairs are averaged together to compute AVSRE.

[0090] An average simultaneous single beat atrio-ventricular peak strain rate during left atrial contraction (AVSRA / AVSR-A / AVSR.A) is computed, as a third cardiac parameter 50 of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial (LASR-A) and left ventricular (LVSR-A) drawn from the first averaged curve 1102 and the second averaged curve 1104. In practice, several instantaneous pairs of LASR-A / LVSR-A are drawn from curves 1102 / 1104, each pair representing a consecutive cardiac cycle and the half the sum of these pairs are averaged together to compute AVSRA.

[0091] In some embodiments, the plurality of cardiac parameters 50 comprises any five, any six, any seven, any eight, or any nine cardiac parameters 50 of Table 2. In some

embodiments, the plurality of cardiac parameters comprises all the cardiac parameters of Table 2. In some embodiments, the plurality of cardiac parameters 50 comprises any combination of cardiac parameters 50 of Table 2 as well as cardiac parameters 50 that are not listed in Table 2, such as radial and circumferential strain parameters and LV twist mechanics. In some embodiments, the plurality of cardiac parameters 50 comprises two or more, three or more, four or more, five or more, six or more, seven or more, eight or more, or nine or more of the cardiac parameters 50 of Table 2 as well as radial strain parameters. In some embodiments, the plurality of cardiac parameters 50 comprises two or more, three or more, four or more, five or more, six or more, seven or more, eight or more, or nine or more of the cardiac parameters 50 of Table 2 as well as circumferential strain parameters. In some embodiments, the plurality of cardiac parameters 50 comprises two or more, three or more, four or more, five or more, six or more, seven or more, eight or more, or nine or more of the cardiac parameters 50 of Table 2 as well as LV twist mechanics. In some embodiments, the plurality of cardiac parameters comprises all the cardiac parameters of Table 2.

[0092] In typical embodiments, the linear discriminant function 52, weighted

neighborhood scheme 56, and artificial neural network 60 described below are trained using a cohort population from each classification in the set of classifications. In some embodiments, this training of the linear discriminant function 52, weighted neighborhood scheme 56, and artificial neural network 60 is performed using any five, any six, any seven, any eight, or any nine cardiac parameters 50 of Table 2. In some embodiments, this training makes use of all the cardiac parameters of Table 2. In typical embodiments, the plurality of cardiac parameters 50 applied in the method describe in Figure 2 are the same cardiac parameters 50 that the linear discriminant function 52, weighted neighborhood scheme 56, and artificial neural network 60 are trained on. In typical embodiments, the linear discriminant function 52, weighted neighborhood scheme 56, and artificial neural network 60 are trained on the same exact cardiac parameters 50 using cohort populations. In some embodiments, the linear discriminant function 52, weighted neighborhood scheme 56, and artificial neural network 60 are trained on different cardiac parameters 50 using cohort populations.

[0093] Subject the plurality of cardiac parameters to a linear discriminant function.

Once the images of a heart of a subject have been taken and the cardiac parameters 50 have been derived from these images, it is possible to categorize the heart into a classification in the set of classifications. Referring to element 240 of Figure 2C, one classifier that is used in some embodiments for such classification is a kernel function such as a support vector machine.

Application of the derived cardiac parameters to the linear discriminant function results in the obtaining of a first prediction of the classification in the set of classifications as output from the linear discriminant function. [0094] In embodiments where the set of classifications consists of more than two classifications, the linear discriminant function is enacted as a plurality of support vector machines, implemented on a one-versus-all or a one-versus-one basis, where each respective support vector machine in the plurality of support vector machines distinguishes between a different pair of classification in the set of classification and the plurality of support vector machines thereby represents each possible pair of classifications in the set of classifications

(242).

[0095] Support vector machines (SVMs), illustrated in Figure 5, are described in

Cristianini and Shawe-Taylor, 2000, "An Introduction to Support Vector Machines," Cambridge University Press, Cambridge; Boser et al., 1992, "A training algorithm for optimal margin classifiers," in Proceedings of the 5^th Annual ACM Workshop on Computational Learning Theory, ACM Press, Pittsburgh, Pa., pp. 142-152; Vapnik, 1998, Statistical Learning Theory, Wiley, New York; Mount, 2001, Bioinformatics: sequence and genome analysis, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y.; Duda, Pattern Classification, Second Edition, 2001, John Wiley & Sons, Inc., pp. 259, 262-265; and Hastie, 2001, The Elements of Statistical Learning, Springer, New York; and Furey et al., 2000, Bioinformatics 16, 906-914, each of which is hereby incorporated by reference in its entirety. When used for classification, SVMs separate a given set of binary labeled data training set (e.g., "Grade I diastolic

dysfunction" versus "Grade II diastolic dysfunction") with a hyper-plane that is maximally distant from the labeled data. For cases in which no linear separation is possible, SVMs can work in combination with the technique of ^'kernels^' , which automatically realizes a non-linear mapping to a feature space. The hyper-plane found by the SVM in feature space corresponds to a non-linear decision boundary in the input space.

[0096] Subject the plurality of cardiac parameters to a weighted neighborhood scheme.

Referring to element 244 of Figure 2C, another classifier that is used in some embodiments for categorization of a heart into a classification in a set of classifications is a neighborhood scheme (e.g., a random forest or a k-nearest neighbor algorithm) thereby obtaining a second prediction of the classification in the set of classifications. Random forests, illustrated in Figure 4, are described in Breiman, 1999, "Random Forests—Random Features," Technical Report 567, Statistics Department, U.C. Berkeley, September 1999, which is hereby incorporated by reference in its entirety. Nearest neighbor classifiers are memory-based and require no classifier to be fit. Given a query point xo (which is the set of cardiac parameters 50) the k training points x(r), r,... , k closest in distance to xo are identified and then the point xo is classified using the k nearest neighbors. Ties can be broken at random. In some embodiments, Euclidean distance in feature space is used to determine distance as:

[0097] Subject the plurality of cardiac parameters to an artificial neural network.

Referring to element 246 of Figure 2C, another classifier that is used in some embodiments for categorization of a heart into a classification in the set of classifications is artificial neural network (Figure 3) thereby obtaining a third prediction of the classification in the set of classifications. Artificial neural networks (ANNs) or connectionist systems are a computational model which is based on a large collection of connected simple units called artificial neurons, loosely analogous to axons in a biological brain. Connections between neurons carry an activation signal of varying strength. If the combined incoming signals are strong enough, the neuron becomes activated and the signal travels to other neurons connected to it. See Aleksander and Morton, 1995, An Introduction to Neural Computing, Intl Thomson Computer Pr, which is hereby incorporated by reference.

[0098] Apply the prediction. Referring to element 228 of Figure 2C, the first prediction, the second prediction, and the third prediction are collectively application to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart. In some embodiments, the majority voting method deems the classification to be the classification in the set of classifications that matches two or more of the group consisting of the first prediction, the second prediction and the third prediction (250). If the majority voting method does not achieve two or more of the group consisting of the first prediction, the second prediction and the third prediction, meaning that each classifier voted for a different classification in the set of classifications, then the classification is not called. In such situations, conventional diagnostic methods are invoked or the subject is reimaged and the classification is repeated using the ensemble classifier.

[0099] Example 1 - Precision Phenotyping in Heart Failure: Pattern Clustering of

Large-scale Cardiac Ultrasound Data for Assessing Diastolic Function. An ensemble-learning model was designed by combining three different methods (random forest, artificial neural network and support vector machines) and final predictions were estimated using a voting method. Variable importance was estimated using information gain algorithm. Information gain measure was used to select features to train a machine-learning model with high accuracy for elevated LVFP prediction (10 fold cross validation area under the curve, AUC: 0.84). The model performed with comparable accuracy in the validation cohort (AUC: 0.88). The resulting algorithm correctly classified 80% of patients to PCWP > and < 18 mmHg (kappa= 0.51, p < 0.001). Incorporation of STE data to develop accurate, machine-learning based phenotyping algorithms is useful for the development of rapid automated techniques for noninvasive assessment of LVFP.

[00100] STE provides large sets of spatial and temporal measurements; novel big data analytic approaches may, therefore, be well suited for STE databases for pattern recognition and precise staging of cardiac muscle dysfunction. In this example we hypothesized that the cumulative information obtained using STE-based measurements is similar to that obtained using conventional 2D echocardiograms and Doppler measurements for characterizing LV diastolic function and LV filling pressures. To this effect, STE-derived parameters were measured in an exploratory subset of HF patients for understanding the relationships between STE and conventional variables, and then subsequently tested in a separate validation group of patients with invasive pressure measurements.

[00101] Study population. Patients for exploratory and validation cohorts (Fig. 13) were recruited from two centers, Ain Shams University Hospital, Cairo, Egypt (CAI), and Icahn School of Medicine at Mount Sinai, New York, NY, USA (NY). Local ethics committee of both institutions approved the study. A single specialist analyzed echocardiographic studies from both institutions (Mount Sinai core lab, NY).

[00102] Exploratory group: In the period between June 2013 and March 2014, registries of the echocardiography laboratories of CAI and NY were reviewed for cases referred for assessment of LV systolic and diastolic function. Patients were excluded if they had poor echocardiographic images, inadequate visualization of left ventricle and left atrial biplane views, inadequate data for assessing LV diastolic function and filling pressures, systemic co-morbidities (e.g. malignancies, terminal hepatic failure, and end-stage chronic renal disease on dialysis), more than mild degree of valve disease, and pericardial diseases. As such, 130 patients from both centers were identified and were included in the exploratory cohort. [00103] Validation group: Forty-four patients were prospectively identified with heart failure symptoms who were undergoing left and right heart catheterization. The exclusion criteria used in the exploratory cohort was also observed for the validation group.

Echocardiographic examinations were performed by an investigator blinded to the exploratory group analyses (AMSO) and were acquired using the same standardized protocol.

Echocardiographic examinations were performed simultaneously to the right heart

catheterization studies. PCWP and left ventricular end diastolic pressure (LVEDP), were measured by an investigator blinded to echocardiographic data (OR). See Nagueh, 2009, "Echocardiographic assessment of left ventricular relaxation and cardiac filling pressures," Current heart failure reports 6(3): 154-159, which is hereby incorporated by reference.

[00104] Echocardiographic Examination

[00105] Two-dimensional echocardiography: All echocardiographic studies were performed with a commercially available echocardiography system equipped with a 2.5-MHz multi -frequency phased array transducer (Vivid 7 or E9, GE-Vingmed, Horton, Norway).

Digital routine gray-scale 2-dimensional loops from apical 2- and 4-chamber views with three consecutive beats were obtained with both LV and LA clearly and completely visualized. Left ventricular end diastolic volume (EDV), end systolic volume (ESV), and ejection fraction (EF) were calculated using the biplane Simpson's method of discs and left atrial maximum volume (LAVmax) and minimum volume (LAVmin) were calculated using the biplane area-length method. All measurements were made in >3 consecutive cardiac cycles and average values were used for the final analyses.

[00106] Pulsed-Wave Doppler Examination: The pulsed-wave Doppler-derived trans- mitral velocity and tissue Doppler-derived mitral annular velocities were obtained from the apical 4-chamber view. The early diastolic wave velocity (E), late diastolic atrial contraction wave velocity (A), and the E-wave deceleration time (E-DcT) were measured using pulsed-wave Doppler recording. Spectral pulsed-wave tissue Doppler-derived early and late diastolic velocities (e' and a') were averaged from the septal and lateral mitral annular positions. The averaged E/e' ratio was calculated as a Doppler echocardiographic estimate of the LVFP.

[00107] Speckle tracking echocardiography. Two-dimensional cardiac performance analysis software (2D CPA, TOMTEC®) was used for a simultaneous frame-by-frame movement assessment of the stable patterns of left ventricular (LV) and left atrial (LA) speckles in apical 4-chamber and 2-chamber views. The endocardial borders of both the LA and LV were traced at the end-diastolic frame, identified as one frame before mitral valve closure at end- diastole. The software allowed delineating end-points at the annulus for both LV and the LA tracking lines, with a specific marker point that indicated the position of the annulus. LA and LV speckle tracking was then performed during the cardiac cycle and the instantaneous changes in volume, volume rates, longitudinal strain and strain rate were obtained from both apical views and averaged.

[00108] Single beat speckle tracking derived volume and strain measurements: STE- derived volume curves were used for defining LV EDV and ESV, LAVmax, and LAVmin, and total left heart volume during ventricular systole (TLVs) and diastole (TLVd) (Figure 8). From the strain curves, simultaneous peak left ventricular systolic strain (LV-S) and peak left atrial strain during LV systole (LA-S) were also measured. (Figure 9). Atrio-ventricular strain (AV-S) was calculated as the average of the magnitude of global LA strain (LAS) and LV strain (LVS) [AV-S= (LAS + LVS)/2], where both LV-S and LA-S are taken as positive values. From the volume rate and strain rate curves, simultaneous diastolic volume rate at early and late diastole, and strain rate at early and late diastole and in peak ventricular systole of the LV (VR-ELV, VR- ALV, SR-ELV, SR-ALV, SR-SLV, respectively), and of the LA (VR-ELA, VR-ALA, SR-ELA, SR- ALA, SR-SLA, respectively) were measured (Figures 10 and 11). Finally, atrial -ventricular volume rate at early and late diastole as well as strain rate during early and late diastole and peak ventricular systole (VR-EAV, VR-AAV, SR-EAV, SR-AAV, SR-SAV, respectively) were calculated by averaging the respective LV and LA absolute values.

[00109] Cardiac Catheterization Studies. Right and left cardiac catheterization were done simultaneously to the echocardiographic examinations in the validation group for invasive pressures measurement using a fluid-filled balloon-tipped catheter. See Swan et al. 1970, "Catheterization of the heart in man with use of a flow-directed balloon-tipped catheter," The New England Journal of Medicine 283(9) :447-451, which is hereby incorporated by reference. Fluoroscopically verified mean PCWP and LVEDP were obtained at end expiration with the zero-level set at the midaxillary line and represent the average of 5 cardiac cycles. See Nagueh et al, 1997, "Doppler tissue imaging: a noninvasive technique for evaluation of left ventricular relaxation and estimation of filling pressures," Journal of the American College of Cardiology. 30(6): 1527-1533, which is hereby incorporated by reference. Significant elevation of LVFP was defined as PCWP>18 mmHg ^{12 13}. See, Forrester et al, 1976, "Medical therapy of acute myocardial infarction by application of hemodynamic subsets (first of two parts)," The New England journal of medicine, 295(24): 1356-1362; and Nohria et a/., 2003, "Clinical assessment identifies hemodynamic profiles that predict outcomes in patients admitted with heart failure," Journal of the American College of Cardiology, 41(10): 1797-1804, each of which is hereby incorporated by reference.

[00110] Statistical analysis. All analyses were performed with commercially available software (SPSS version 21.0; SPSS, Inc, Chicago, IL, USA, and R: A language and environment for statistical computing, version 3.0.1; R Foundation for Statistical Computing, Vienna, Austria). A p-value of <0.05 was considered statistically significant. Categorical variables were expressed as number (%) and were compared using chi² test. Continuous variables were expressed as mean±SD and were compared using the independent sample t-test, or the Mann- Whitney U-test if not normally distributed.

[00111] Extracting STE-derived data for modeling. Correlations between conventional versus STE parameters were checked using linear regression, expressed as Pearson correlation coefficient (r) and the maximal information coefficient (MIC). An absolute difference between Pearson's r² and MIC (MIC - r²) of > 0.1 was used as a marker of non-linear correlations, which were subsequently confirmed by visual inspection. Reshef et al., 2011, "Detecting novel associations in large data sets," 2011;334(6062): 1518-1524, which is hereby incorporated by reference. Correlations were used to extract STE-derived data that correspond to the

conventional parameters used for echocardiographic assessment of diastolic function and LVFP. STE correspondents were then used to construct t a clustering model for the assessment of diastolic function. Variables extracted were then used to train and test machine learning model for automated classification of cardiac imaging data into different grades of diastolic

dysfunction.

[00112] Clustering models: agglomerative hierarchical clustering, was used for classifying patients and their echocardiographic variables (nine conventional and Doppler variables and nine STE variables). These continuous variables were standardized to a mean=0 and a standard deviation=l . Hierarchical clustering was then performed with the hclust function in R, with the dissimilarity matrix given by Euclidean distance. Subsequent leaf ordering and the distance between the conventional variables and their corresponding STE-derived variables were then visualized by means of a heat map produced by the heatmap function in R. Statistical significance of distance between parameters within clusters was derived using pvclust package in R, with 1000 bootstrapping. See, Suzuki et al, 2006, "Pvclust: an R package for assessing the uncertainty in hierarchical clustering," Bioinformatics 22(12): 1540-1542. The approximately unbiased probability (AU) and the bootstrap probability (BP) were calculated and dendrogram leaflets with AU of >95% were considered statistically significant.

[00113] Next, all STE data (11 variables) were used to construct a two-step Doppler independent clustering model derived by SPSS two-step cluster analysis function (HC. R., 2004, "Clustering analysis for researchers," Morrisville, NC: Lulu Press, hereby incorporated by reference), a method suitable for large data sets which classifies data in two steps, by creating small pre-clusters based on calculation of BIC and a log-likelihood distance criterion (Step 1), which are then merged into distinct groups by finding the greatest change in distance between the two closest clusters in each hierarchical clustering stage (Step 2). Clusters were

characterized for their diastolic dysfunction and LVFP according to values of the conventional variables.

[00114] The robustness and stability of the final clustering solution and their

characteristics were re-evaluated by saving the model and testing it in the validation group.

[00115] Predictive model for diastolic dysfunction. Meta-learning or ensemble learning (Zhou , 2012, "Ensemble Methods: Foundations and Algorithms," Chapman Hall, hereby incorporated by reference) is an artificial intelligence algorithm development strategy that combines multiple classes of algorithms in an efficient way of performing a classification task. See Vilalta and Drissi, "A Perspective View and Survey of Meta-Learning," Artificial

Intelligence Review 18(2):77-95; Chan and Stolfo, 1995, "A comparative evaluation of voting and meta-learning on partitioned data," Paper presented at: ICML1995; and Seewald and Fiirnkranz, "An Evaluation of Grading Classifiers," In: Hoffmann F, Hand DJ, Adams N, Fisher D, Guimaraes G, eds, Advances in Intelligent Data Analysis: 4th International Conference, IDA 2001 Cascais, Portugal, September 13-15, 2001 Proceedings. Berlin, Heidelberg: Springer Berlin Heidelberg; 2001 : 115-124, each of which is incorporated by reference. In this study, a binary class prediction model was designed to predict a given cardiac imaging data as diastolic dysfunction or controls, conforms to the diastolic dysfunction classification guidelines set by American Society of Echocardiography /European Society of Echocardiography. See Nagueh et al., 2016, "Recommendations for the Evaluation of Left Ventricular Diastolic Function by Echocardiography: An Update from the American Society of Echocardiography and the

European Association of Cardiovascular Imaging, "J Am Soc Echocardiogr. 29(4):277-314; Nagueh et al., 2009, "Recommendations for the evaluation of left ventricular diastolic function by echocardiography," Eur J Echocardiogr. 10(2): 165-193; and Nagueh et al., 2009,

"Recommendations for the evaluation of left ventricular diastolic function by

echocardiography," J Am Soc Echocardiogr. 22(2): 107-133, each of which is hereby

incorporated by reference. Binary prediction learning algorithms using singleton algorithms using relatively small number (<100 cases per group) is prone to overfitting. See Rokach, 2010, Pattern Classification Using Ensemble Methods, World Scientific Publishing Co., Inc.; and Frey et al, 2014, "Big Data Deep Phenotyping: Contribution of the EVIIA Genomic Medicine

Working Group," Yearbook of Medical Informatics 9(1):206-211, each of which is hereby incorporated by reference. There are multiple ways to improve these situations including improvement of single algorithms. These include bagging (Breiman L., "Bagging predictors," Machine Learning 24(2): 123-140, hereby incorporated by reference), boosting (Freund, 1995, "Boosting a weak learning algorithm by majority," Inf. Comput. 121(2):256-285.) or both, or using multiple independent learners and evaluate the results based on concordance estimations, run the prediction task and gather final results based on approximation from individual learners etc. See Alceu et al., 2014 "Dynamic selection of classifiers-A comprehensive review," Pattern Recogn. 47(11):3665-3680; and Micha et al, 2014 "A survey of multiple classifier systems as hybrid systems, Inf. Fusion 16:3-17, which is hereby incorporated by reference. To address these, an ensemble learning strategy was used with a voting method to develop our algorithm. In a recent pilot study, it was determined that combining random forests, artificial neural networks, and support vector machines using majority voting is an effective approach to phenotype athlete's heart and hypertrophic cardiomyopathy using high-dimensional cardiac imaging data. Here this ensemble model is used to predict diastolic dysfunction. Several prediction tasks in biology and medicine including prediction of arterial waveform analyses (Almeida et al. 2013, "Machine learning techniques for arterial pressure waveform analysis," J Pers Med. 3(2):82-101, which is hereby incorporated by reference) have used ensemble learning. We have also employed a k-fo\d cross validation approach to assessing sample-induced bias and error rates; final results are reported based on 10-fold cross-validation.

[00116] Construction of predictive models and feature selection. For machine learning, data was tabulated using R and machine learning was performed using Weka. Feature importance was derived using information gain criterion and ranked using Ranker method and was implemented using the Weka workbench. True positive (TP) rate, false positive (FP) rate, precision, recall, F-measure, and receiver operator characteristic curve (ROC-curve) were compiled to evaluate the models. Three different algorithms that belong to three different genres of machine learning were used. The model outputs were compared using the area under the curve (AUC) estimated from the ROC-curve. Data pre-processing was carried out in

MICROSOFT EXCEL Version 12.0 and custom Perl scripts were used to prepare input files for machine learning. Statistical analyses were performed using R: a language and environment for statistical computing, version 3.1.2. Machine learning algorithms and feature selection methods were implemented using R packages and Weka. See Frank et al., 2004, "Data mining in bioinformatics using Weka," Bioinformatics 20(15):2479-2481 ; and Hall et al., 2009, "The WEKA Data Mining Software: An Update," SIGKDD Explor Newsl 1 1 : 10-18; each of which is hereby incorporated by reference.

[00117] Results. The demographic, clinical, and echocardiographic data for patients from exploratory and validation groups are summarized in table 3.

Table 3 : Demographic, clinical and echocardiographic variables for both study groups

Exploratory group Validation group

(n=130) (n=44)

Age (years) 53.6±16.4 57.7±7.9

Sex (male/female) 94(72)736(28) 29(66)715(34)

Heart rate (beat/m) 77.3±13.5 78.5±14.9

NYHA class n(%)

I/II/III/IV 35(28)/61(46)/31(22)73(3) 3(7)/28(64)/12(27)/l(2)* Table 3 : Demographic, clinical and echocardiographic variables for both study groups

Exploratory group Validation group (n=130) (n=44)

NYHA>2 34(26) 13(30) Mean blood pressure (mmHg) 94.5±19 103.8±22* Risk Factors: n(%)

Diabetes 38(29) 17(39) Hypertension 56(43) 27(61)* Smoking 38(29) 10(23) Hyperlipidemia 23(18) 4(9) >2 risk factors: 44(34) 20(45) Type of presentation

Dilated cardiomyopathy 58(45) 20(45) Ischemic heart disease 23(18) 24(55) Restrictive cardiomyopathy 11(8)

Hypertension 14(11)

others 24(18)

Treatment

Beta Blockers 62(48) 25(57)

Renin Angiotensin-Aldosterone 44(34) 17(39) blockers

Spironolactone 23(18) 3(7)*

Furosemide 46(35) 15(34)

Digoxin 15(12) 1(2)

Statins 47(36) 16(36) Table 3 : Demographic, clinical and echocardiographic variables for both study groups

Exploratory group Validation group

(n=130) (n=44)

Nitrates 22(17) 24(55)*

Aspirin 50(38) 29(66)*

Clopidogrel 11(8) 13(30)*

Calcium channel blockers 10(7) 5(11)

Conventional variables

LAVmax (ml) 66±25.8 62.4±22.3

LAVmin (ml) 30.7±21.6 24.9±15.1

ESV (ml) 70.2±58 54.1±29

EF (%) 53.1±16 55.3±12.6

EF>50% / EF<50% 73/57 25/19

E (cm/s) 82.4±22.4 76.8±18.9

A (cm/s) 68.6±29.8 77.1±26.2

E-DcT (msec) 180.6±74.7 192±66.4

e' (cm/s) 7.7±4 7.1±2 a' (cm/s) 7.8±3.1 8.6±2.7

E/A 1.41±0.65 1.15±0.67*

E/e' 13.1±6.6 11.5±4.1

[00118] In Table 3, Nominal data were expressed as n (%) and continuous data were expressed as mean±SD. NYHA is the New York heart association functional class. LAVmax, is maximal left atrial volume in milliliters, LAVmin, is minimal left atrial volume, EDV, left ventricular end diastolic volume, ESV, left ventricular end systolic volume, EF, left ventricular ejection fraction, TLVs, total left heart volume during ventricular systole, total left heart volume during ventricular diastole, E, mitral flow early diastolic velocity, A, mitral flow late diastolic velocity, E-DcT, mitral E-wave deceleration time, e', tissue Doppler derived mitral annular early diastolic velocity, a', tissue Doppler derived mitral annular late diastolic velocity, s', tissue Doppler derived mitral annular ejection systolic velocity, *p<0.05.

[00119] Both groups were similar in age, gender distribution, NYHA class and all risk factors except hypertension which was more prevalent in the validation cohort (p=0.028). Both groups had similar degree of LV remodeling and severity of diastolic function.

[00120] Dependencies between conventional Doppler based and STE-based indices were examined. The distribution of SR-EAV, SR-AAV, SR-SAV, VR-EAV, VR-AAV, VR-E/VR-AAV, SR- E/SR-AAV, VR-E/SR-EAV corresponded with that of e', a', s', E, A velocities, and E/A, e'/a', and E/e' ratios, respectively. In addition, 2D and STE-derived LV and LA volumes as well as TLVs and TLVd were well correlated.

[00121] Clustering models.

[00122] Exploratory group. A clustering model was constructed using the conventional variables E, A, e', a', and LAVmax and the ratios E/A, e'/a', and E/e', in addition to their correlations' derived corresponding STE variables. In this model it was found that all STE variables showed significant proximity to their conventional counterparts within the clustering dendrograms (AU>95%), suggesting a high statistical level of overlap between the STE-derived variables and the conventional volume and Doppler variables (log likelihood ratio: -1413, BIC: - 4287.5, distance: minimum 1.2, median 5.522, mean 5.671, maximum 13.380).

[00123] Moreover, a two-step clustering model using all STE variables independent of their correlations with the conventional variables divided the patients into 3 different groups (log. likelihood ratio: -451, BIC: -1939.3). These clusters were found to show progressive worsening of diastolic functions and LVFP as suggested by the conventional variables E/A, e', LAVmax, and E/e' and accompanied with progressively increasing age, more prevalence of hypertension, increasing NYHA class, and worsening EF. [00124] Validation: The Doppler independent clustering model when applied to the validation group reproduced the same features of clustering as seen in the exploratory cohort. The clinical and functional characteristics of the exploratory cohorts were also retained in the validation cohort. Although patient's age, risk factors, and NYHA class were not statistically different between clusters, systolic and diastolic function parameters was significantly different between the groups. More importantly, the values of PCWP and LVEDP corresponded to the severity of diastolic function.

[00125] Machine learning and feature selection. The overall accuracy of the model was determined using different voting approaches and used Majority voting based method based on conservative estimates of AUC.

[00126] By combining three different algorithms (RF, ANN and SVM) using majority- voting method we created two sets of models (See Table 4).

Table 4: Evaluation of diastolic dysfunction predictive models using full features and reduced features a) Model statistics using full set of features

Groups Class TPR FPR Precision Recall F-Measure ROCArea

Exploratory cohort (n=130)

10-f CV CONTROL 0.904 0.25 0.825 0.904 0.863 0.847

DDF 0.75 0.096 0.857 0.75 0.8 0.847 WA 0.837 0.183 0.839 0.837 0.836 0.847

Validation cohort (n=44)

CONTROL 0.968 0.231 0.909 0.968 0.937 0.868

DDF 0.769 0.032 0.909 0.769 0.833 0.868

WA 0.909 0.172 0.909 0.909 0.907 0.868 Table 4: Evaluation of diastolic dysfunction predictive models using full features and reduced features b) Model statistics after feature reduction

Exploratory cohort (n

10-f CV CONTROL 0.836 0.214 0.836 0.836 0.836 0.853

DDF 0.786 0.164 0.786 0.786 0.786 0.853

WA 0.814 0.193 0.814 0.814 0.814 0.853 Validation cohort (n=44)

CONTROL 0.968 0.308 0.882 0.968 0.923 0.881

DDF 0.692 0.032 0.9 0.692 0.783 0.881

WA 0.886 0.226 0.888 0.886 0.882 0.881

[00127] In Table 4, 10-fCV = 10-fold cross-validation; WA=weighted average; TPR=True Positive Rate (sensitivity) is the number of detected positive examples (DDF) among all positive examples; FPR=False Positive Rate (specificity) is the proportion of detected negative examples among all negative examples (CONTROLS); Precision=The number of DDF examples among all examples classified as DDF; Recall=same as sensitivity and a component of F-Measure; F- measure=weighted harmonic mean of precision and recall (2 * precision * recall / (precision + recall); AUROC=Area Under the Received Operator Curve.

[00128] A model was created using full set of features («=15), the model had a 10-fold cross validation 0.847 in exploratory cohort and 0.868 in validation. After feature reduction (n=l 1) two new sets of models were generated with the AUCs as 0.853 and 0.881 for exploratory and validation set respectively. Consistent with theoretical estimations (Guyon et al, 2003, "An introduction to variable and feature selection," J. Mach. Learn. Res. 3 : 1157- 1182, which is hereby incorporated by reference), these models showed better performance using reduced set of features. [00129] Discussion. Clinical evaluation of cardiac biomechanics generates the significant amount of information. This example demonstrates that big data analytics, and machine learning frameworks can automate the assessment of LV diastolic function using predictive models. The key findings of this example are: 1) a high statistical level of overlap was seen between STE- derived data and conventional echocardiographic methods of diastolic function assessment, 2) STE data clustered the patients into three different groups that corresponded to worsening severity of diastolic functions grades as suggested by the conventional parameters, and, 3) a linear multivariate model built only from STE data had good diagnostic accuracy in predicting LV filling pressures. These findings suggest that the information content of STE variables corresponds to that derived from 2D and Doppler-based analysis and can provide an independent assessment of diastolic function and LV filling pressures. The emergence of novel big data analysis solutions for characterization of new patterns in speckle tracking derived cardiac function assessment will open up new opportunities for efficient and accurate characterization of heart failure phenotypes in clinical practice.

[00130] Conventional assessment of diastolic function: strengths and pitfalls. The assessment of LV diastolic functions is essential in the course of management of patients with heart failure. See Yancy 2013, et al., ACCF/AHA guideline for the management of heart failure: a report of the American College of Cardiology Foundation/ American Heart Association Task Force on Practice Guidelines," Journal of the American College of Cardiology

62(16):el47-239, which is hereby incorporated by reference. Methods used in that regard have seen an evolution over the decades from clinical evaluation, through invasive pressure measurements, and non-invasive advanced imaging, with echocardiography commonly used at the bedside. However, no single echocardiographic parameter has been shown to adequately address diastolic function, and conventional approaches incorporate several 2-dimensional, Doppler, and tissue Doppler variables in decision trees based algorithms. Although currently recommended algorithms continue to be successfully implemented in clinical practice, they require a high level of training and expertise to be effectively used. See Nagueh et al.,

"Recommendations for the evaluation of left ventricular diastolic function by

echocardiography," 2009, J Am Soc Echocardiogr. 22(2): 107-133. Moreover, echocardiography acquisition frequently requires using different 2D and pulsed Doppler approaches and measurements which require considerable laboratory effort at standardization. In addition, these parameters are compiled from different cardiac cycles at different cardiac locations and thus susceptible to time and hemodynamic load-related measurement variances. In contrast, the instantaneous integration of atrial and ventricular function and geometry from a single heart beat may be useful to overcome variability related to heart rate and respiratory load-related changes. Several STE parameters such as strain rate during early diastole (SR-E) have been previously investigated as surrogate variables for measuring LVFP or to assess of LV diastolic function. See Chen, 2014, "Evaluation of left ventricular diastolic function by global strain rate imaging in patients with obstructive hypertrophic cardiomyopathy: a simultaneous speckle tracking echocardiography and cardiac catheterization study," Echocardiography 31(5):615-622; Kuwaki et al., "Redefining diastolic dysfunction grading: combination of E/A </=0.75 and deceleration time >140 ms and E/epsilon' >/=10" JACC. Cardiovascular imaging 7(8):749-758; and Wang et al, 2007, "Global diastolic strain rate for the assessment of left ventricular relaxation and filling pressures," Circulation. 115(11): 1376-1383. Such STE-based assessments are attractive alternative ways of assessing cardiac function because multivariate assessment of cardiac function can be performed using just gray scale based cardiac ultrasound motion data with multiple spatial assessments that can be integrated for more than one cardiac chamber.

[00131] Precision medicine: role of STE-derived large data analytics platforms. The field of big data analytics has operationalized precision medicine as an approach to establishing clinical phenotypic characterization of different diseases while taking into account genetic and environmental variability. Phenomapping approaches using unbiased cluster analysis has been recently proposed for meaningful categorization of patients with heart failure. As shown in our study, STE is capable of generating useful data from a single echocardiographic loop; and can, therefore, improve imaging based cardiovascular phenotypic characterization. Moreover, recent standardization efforts (Yang et al., 2015, "Improvement in Strain Concordance between Two Major Vendors after the Strain Standardization Initiative," J Am Soc Echocardiogr. 28(6):642- 648 e647; and Voigt et al., 2015 "Definitions for a common standard for 2D speckle tracking echocardiography: consensus document of the EACVI/ASE/Industry Task Force to standardize deformation imaging," J Am Soc Echocardiogr 28(2): 183-193 have revealed that STE measurements are more reproducible than conventional 2D and Doppler indices. The emergence of automated STE approaches may further improve efficiency and reduce inter- and intra- observer variability. In the near future, the ability to extract large scale information from echocardiography database will enable the emergence of machine learning models capable of capturing data for automated analyses, so that once images are uploaded, information will be automatically extracted and analyzed for providing decisions in real time. This may be helpful in increasing diagnostic throughput and efficiency in the face of the growing burden of

cardiovascular disease in the community and the existing work-shortage in the field. The data generated using innovative approximation methods can be further extended using machine learning algorithms as predictive models. These models can be now deployed using cloud- computing technologies to aid echocardiographers to do complex cardiac phenotyping easily. The machine-learning model discussed in this example is one of the first attempts to automate the task and predict diastolic dysfunction. Adding more samples and expanding to the multiclass predictions can further improve the model.

[00132] This example utilized novel clustering approaches for analyzing large-scale STE data for characterizing diastolic function grades and LV filling pressures. However, STE measurements in this example were obtained from biplane views for simultaneously enabling chamber quantification using the biplane-Simpson method. This allowed extraction of functional and geometric measurements that correspond to conventional 2D and Doppler-based functional assessments. The 3-chamber view of LV and LA was not included and, therefore, the data may not be truly representative of global LV and LA mechanics. Secondly, only longitudinal velocity, strain and strain rate parameters were used in the STE database because longitudinal LV mechanical parameters are better standardized and currently more reproducible. The incremental value of radial and circumferential strain parameters and LV twist mechanics was not tested in the present investigation. Finally, the sample size was small for subgroup analysis for patients with preserved or reduced ejection fraction and the prognostic information of clustered groups was not evaluated in the current study.

[00133] Medical imaging in the era of precision medicine aims for accurate phenotyping of diseases that can benefit from early targeted therapies. Cardiac biomechanics generate high level of information that can be used for fully automated diastolic function assessment. STE data contains large-scale data with a high level of information overlap with the existing 2D and Doppler-based indices of diastolic function. Cluster patterns of STE-based data may be useful for phenotypic characterization of LV diastolic functions in patients with heart failure.

Incorporation of STE data to develop accurate, machine-learning based phenotyping algorithms will be useful for development of rapid automated techniques for noninvasive assessment of LVFP.

CONCLUSION

[00134] The foregoing description, for purpose of explanation, has been described with reference to specific implementations. However, the illustrative discussions above are not intended to be exhaustive or to limit the implementations to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings. The implementations were chosen and described in order to best explain the principles and their practical applications, to thereby enable others skilled in the art to best utilize the implementations and various implementations with various modifications as are suited to the particular use contemplated.

Claims

What is claimed is:

1. An apparatus for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications, the apparatus comprising one or more processors and memory storing one or more programs for execution by the one or more processors, the one or more programs singularly or collectively executing a method comprising:

A) obtaining a plurality of ultrasound gray-scale measurement images of the heart across a plurality of heartbeats of the heart;

B) determining a plurality of cardiac parameters from the plurality of measurement images, wherein two or more images in the plurality of images contributes to each respective cardiac parameter in the plurality of cardiac parameters;

C) subjecting the plurality of cardiac parameters to a linear discriminant function thereby obtaining a first prediction of the classification in the set of classifications;

D) subjecting the plurality of cardiac parameters to a weighted neighborhood scheme thereby obtaining a second prediction of the classification in the set of classifications;

E) subjecting the plurality of cardiac parameters to an artificial neural network thereby obtaining a third prediction of the classification in the set of classifications; and

F) collectively applying the first prediction, the second prediction, and the third prediction to a majority voting method thereby obtaining the classification, in the set of classifications, for the diastolic dysfunction of the heart.

2. The apparatus of claim 1, wherein the linear discriminant function is a kernel function.

3. The apparatus of claim 2, wherein the kernel function is a support vector machine.

4. The apparatus of any one of claims 1-3, wherein the weighted neighborhood scheme is a random forest or a k-nearest neighbor algorithm.

5. The apparatus of any one of claims 1-4, wherein each image in the plurality of images is two- dimensional.

6. The apparatus of any one of claims 1-4, wherein the plurality of images provides a three- dimensional image of the heart.

7. The apparatus of any one of claims 1-6, wherein the set of classifications consists of a first classifications and a second classifications.

8. The apparatus of claim 7, wherein the first classification is grade I diastolic dysfunction or normal and the second classifications is grade II diastolic dysfunction or grade III diastolic dysfunction.

9. The apparatus of claim 8, wherein the plurality of cardiac parameters comprises any five cardiac parameters of Table 2.

10. The apparatus of claim 8, wherein the plurality of cardiac parameters comprises any seven cardiac parameters of Table 2.

11. The apparatus of claim 8, wherein the plurality of cardiac parameters comprises all the cardiac parameters of Table 2.

12. The apparatus of claim 8, wherein the majority voting method deems the classification to be the classification in the set of classifications that matches two or more of the group consisting of the first prediction, the second prediction and the third prediction.

13. The apparatus of any one of claims 1-6, wherein the set of classifications comprises four or more classifications.

14. The apparatus of claim 13, wherein the set of classifications consists of a first classification, a second classification, a third classification, and a fourth classification.

15. The apparatus of claim 14, wherein the first classification is grade I diastolic dysfunction, the second classification is grade II diastolic dysfunction, the third classification is grade III diastolic dysfunction, and the forth classification is normal.

16. The apparatus of claim 13, wherein the linear discriminant function is a plurality of support vector machines, implemented on a one-versus-all or a one-versus-one basis, wherein each respective support vector machine in the plurality of support vector machines distinguishes between a different pair of classifications in the set of classifications and wherein the plurality of support vector machines thereby represents each possible pair of classifications in the set of classifications.

17. The apparatus of claim 16, wherein the plurality of support vector machines is implemented on a one-versus-one basis.

18. The apparatus of claim 15, wherein the plurality of cardiac parameters comprises any five cardiac parameters of Table 2.

19. The apparatus of claim 15, wherein the plurality of cardiac parameters comprises any seven parameters of Table 2.

20. The apparatus of claim 15, wherein the plurality of cardiac parameters comprises all the parameters of Table 2.

21. The apparatus of claim 15, wherein the majority voting method deems the classification to be the classification in the set of classifications that matches two or more of the group consisting of the first prediction, the second prediction and the third prediction.

22. The apparatus of any one of claims 1-6, wherein the set of classifications consists of three classifications.

23. The apparatus of any one of claims 1-6, wherein the set of classifications comprises five or more classifications.

24. The apparatus of any one of claims 1-9, wherein the plurality of ultrasound gray-scale measurement images are obtained from an echocardiographic quantification system.

25. The apparatus of claim 24, wherein the echocardiographic quantification system is a speckle tracking echocardiographic quantification system.

26. The apparatus of claims 24, wherein the echocardiographic quantification system is a tissue Doppler system.

27. The apparatus of any one of claims 1-26, wherein the plurality of ultrasound gray-scale measurement images are taken of the heart over at least a two minute period.

28. The apparatus of any one of claims 1-26, wherein the plurality of ultrasound gray-scale measurement images are taken of the heart over at least a three minute period.

29. The apparatus of any one of claims 1-26, wherein the plurality of ultrasound gray-scale measurement images are taken of the heart over a period of time that between two minutes and eight minutes.

30. The apparatus of any one of claims 1-25, wherein

the plurality of ultrasound gray-scale measurement images comprise a first subset of images and a second subset of images,

the first subset of images comprises a first collection of gray-scale 2-dimensional loops from apical 2-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable, and

the second subset of images comprises a second collection of gray-scale 2-dimensional loops from apical 4-chamber views of the heart in which the left ventricular and the left atrial chambers of the heart are distinguishable.

31. The apparatus of claim 30, wherein

the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure comprising:

computing a first left atrial strain curve from the first subset of images;

computing a first left ventricular strain curve from the first subset of images; computing a second left atrial strain curve from the second subset of images; computing a second left ventricular strain curve from the second subset of images; averaging the first left atrial strain curve and the second left atrial strain curve thereby forming an averaged left atrial strain curve; and

averaging the first left ventricular strain curve and the second left ventricular strain curve thereby forming an averaged left ventricular strain curve; wherein

a cardiac parameter in the plurality of cardiac parameters is atrio- ventricular strain of the heart (AV-S), wherein AV-S is computed as the average of simultaneous peak left ventricular systolic strain (LV-S) and peak left atrial strain during left ventricular systole (LA-S), wherein the LV-S and LA-S are derived from the averaged left atrial strain curve and the averaged left ventricular strain curve, in accordance with the formula:

AV-S = [(LA-S + LV-S)/2].

32. The apparatus of claim 30, wherein a cardiac parameter in the plurality of cardiac parameters is an averaged left atrio- ventricular end diastolic volume (TLH-diast) of the heart, and wherein the determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure comprising:

computing a respective left atrio- ventricular end diastolic volume from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the first subset of images thereby obtaining a first plurality of left atrio- ventricular end diastolic volumes;

computing a respective left atrio- ventricular end diastolic volume from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the second subset of images thereby obtaining a second plurality of left atrio- ventricular end diastolic volumes; and averaging the first plurality of left atrio-ventricular end diastolic volumes and the second plurality of left atrio-ventricular end diastolic volumes thereby forming the averaged left atrioventricular end diastolic volume (TLH-diast).

33. The apparatus of claim 30, wherein a cardiac parameter in the plurality of cardiac parameters is an averaged left atrio-ventricular end systolic volume of the heart, and wherein the

determining the plurality of cardiac parameters from the plurality of measurement images comprises a first procedure comprising:

computing a respective left atrio-ventricular end systolic volume from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the first subset of images thereby obtaining a first plurality of left atrio-ventricular end systolic volumes;

computing a respective left atrio-ventricular end systolic volume from each cardiac cycle in at least three consecutive cardiac cycles encompassed by the second subset of images thereby obtaining a second plurality of left atrio-ventricular end systolic volumes; and

averaging the first plurality of left atrio-ventricular end systolic volumes and the second plurality of left atrio-ventricular end systolic volumes thereby forming an averaged left atrioventricular end systolic volume (TLH-syst).

34. The apparatus of claim 30, wherein

computing a first curve comprising a volume rate plotted against time during a plurality of cardiac cycles for left atrial from the first subset of images;

computing a second curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left ventricular from the first subset of images;

computing a third curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left atrial from the second subset of images;

computing a fourth curve comprising a volume rate plotted against time during the plurality of cardiac cycles for left ventricular from the second subset of images; averaging the first curve and the third curve thereby forming a first averaged curve comprising a volume rate plotted against time during the plurality of cardiac cycles for the left atrial of the heart;

averaging the second curve and the fourth curve thereby forming a second averaged curve comprising a volume rate plotted against time during the plurality of cardiac cycles for the left ventricular of the heart;

computing an average simultaneous single beat atrio- ventricular peak volume rate during left ventricular systole (AVVR-S / AVws / AVVR.S), as a first cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve;

computing an average simultaneous single beat atrio- ventricular peak volume rate during early diastole (AVVR-E / AVVR.E / AVVRE), as a second cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve; and

computing an average simultaneous single beat atrio- ventricular peak volume rate during left atrial contraction (AVVR-A / AVVR. A / AVVRA), as a third cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve.

The apparatus of claim 30, wherein

the determining the plurality of cardiac parameters from the plurality of measurement ges comprises a first procedure comprising:

computing a first curve comprising a strain rate plotted against time during a plurality of cardiac cycles for left atrial from the first subset of images;

computing a second curve comprising a strain rate plotted against time during the plurality of cardiac cycles for left ventricular from the first subset of images;

computing a third curve comprising a strain rate plotted against time during the plurality of cardiac cycles for left atrial from the second subset of images; computing a fourth curve comprising a strain rate plotted against time during the plurality of cardiac cycles for left ventricular from the second subset of images;

averaging the first curve and the third curve thereby forming a first averaged curve comprising a strain rate plotted against time during the plurality of cardiac cycles for the left atrial of the heart;

averaging the second curve and the fourth curve thereby forming a second averaged curve comprising a strain rate plotted against time during the plurality of cardiac cycles for the left ventricular of the heart;

computing an average simultaneous single beat atrio- ventricular peak strain rate during left ventricular systole (AVSRS / AVSR-S / AVSR. S), as a first cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve;

computing an average simultaneous single beat atrio- ventricular peak strain rate during early diastole (AVSRE / AVSR-E / AVSR.E), as a second cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve; and

computing an average simultaneous single beat atrio- ventricular peak strain rate during left atrial contraction (AVSRA / AVSR-A / AVSR. A), as a third cardiac parameter of the heart, as half the sum of the instantaneous maximal absolute corresponding values of the left atrial and left ventricular drawn from the first averaged curve and the second averaged curve.

36. A method for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications, the method comprising:

A) obtaining a plurality of ultrasound gray-scale measurement images of the heart of the across a plurality of heartbeats of the heart;

B) determining a plurality of cardiac parameters from the plurality of measurement images, wherein two or more images in the plurality of images contributes to each respective cardiac parameter in the plurality of cardiac parameters; C) subjecting the plurality of cardiac parameters to a linear discriminant function thereby obtaining a first prediction of the classification in the set of classifications;

37. A non-transitory computer readable storage medium storing one or more programs for execution by one or more processors in a computer system for categorization of the diastolic dysfunction of a heart into a classification in a set of classifications, the one or more programs comprising instructions for: