CN104665768B

CN104665768B - The monitoring of physiological parameter

Info

Publication number: CN104665768B
Application number: CN201410525844.8A
Authority: CN
Inventors: 罗翰·班纳吉; 阿尼鲁达·辛哈; 艾什瓦尔雅·维斯瓦纳坦; 阿尼班·杜塔·乔杜里; 安鹏·派尔; 安莱格·库马尔
Original assignee: Tata Consultancy Services Ltd
Current assignee: Tata Consultancy Services Ltd
Priority date: 2013-10-03
Filing date: 2014-10-08
Publication date: 2019-07-23
Anticipated expiration: 2034-10-08
Also published as: CN104665768A

Abstract

The present invention relates to the monitorings of physiological parameter.There is illustrated the methods for monitoring physiological parameter associated with subject using handheld apparatus (130).In the implementation, this method includes obtaining multiple sample photoplethaysmography features associated with the sample subject i.e. sample PPG feature from the video of the physical feeling (132) of sample subject.At least one correlated samples PPG feature associated with physiological parameter is selected from multiple sample PPG features based on the standard value of the physiological parameter for subject.In addition, determining the mathematical model for indicating the correlativity between correlated samples PPG feature and physiological parameter based on the standard value of at least one correlated samples PPG feature and physiological parameter.The mathematical model can be disposed for monitoring physiological parameter in real time.

Description

The monitoring of physiological parameter

Technical field

This theme is usually directed to monitoring physiological parameter, and particularly supervises not exclusively to handheld apparatus is used Survey physiological parameter.

Background technique

The certain of the people for being directed to respiratory rate, pulse frequency, blood pressure and heart etc. are usually carried out in clinical setting The monitoring of physiological parameter and vital sign.Generally it has been found that if people recognizes that his or her physiological parameter is just monitored, Then this may cause the people go short of or deliberately and physiological parameter may due to people the state of mind and change, thus may It can not reflect the actual medical situation of the people.It therefore, it has been developed to several non-invasive technologies for monitoring physiological parameter.

One this non-invasive technology is photoplethysmography (photoplethysmography, PPG).PPG is related to Optical means and based in the vascular system under skin blood volume variation dynamic work.Traditionally, with various sides Formula realizes that PPG is measured with the video of the subject for example by recording physiological parameter to be measured in a non contact fashion and monitored this A little physiological parameters.

Summary of the invention

A method of for monitoring physiological parameter associated with subject, the side using handheld apparatus 130 Method from the video of the physical feeling 128 of sample subject the following steps are included: obtain associated more with the sample subject A sample photoplethaysmography feature, that is, multiple samples PPG feature；Step is selected, for based on for the sample subject The standard value of the physiological parameter to select associated with the physiological parameter at least one from the multiple sample PPG feature A correlated samples PPG feature；And the standard value based at least one correlated samples PPG feature and the physiological parameter come The mathematical model for indicating the correlativity between the correlated samples PPG feature and the physiological parameter is determined, wherein the number Model is learned to be disposed for monitoring the physiological parameter in real time.

A method of for monitoring physiological parameter associated with subject, the side using handheld apparatus 130 Method from the video of the physical feeling 128 of sample subject the following steps are included: obtain associated more with the sample subject A sample photoplethaysmography feature, that is, multiple samples PPG feature；Step is selected, for based on for the sample subject The standard value of the physiological parameter to select associated with the physiological parameter at least one from the multiple sample PPG feature A correlated samples PPG feature；And it is based only upon the standard value of at least one correlated samples PPG feature and the physiological parameter Determine the mathematical model for indicating the correlativity between at least one correlated samples PPG feature and the physiological parameter, Wherein the mathematical model is disposed for monitoring the physiological parameter in real time, and the physiological parameter is blood pressure i.e. BP and electrocardio Figure is at least one of ECG feature.

A kind of modeling 100, for monitoring physiological parameter associated with subject, the modeling 100 includes: Processor 102；Processing module 110 is connected to the processor 102, and the processing module is for obtaining and sample subject Associated multiple sample photoplethaysmography features, that is, multiple samples PPG feature, wherein the multiple sample PPG feature be from It is extracted in the video of the physical feeling 128 of the sample subject；Feature selection module 112 is connected to the processing Device 102, the feature selection module based on the standard value of the physiological parameter from the multiple sample PPG feature for being selected Select at least one correlated samples PPG feature associated with the physiological parameter；And modeling module 116, it is connected to described Processor 102, the modeling module are used for the standard based at least one correlated samples PPG feature and the physiological parameter Value indicates the mathematical model of the correlativity between the correlated samples PPG feature and the physiological parameter to determine, wherein institute Mathematical model is stated for monitoring the physiological parameter in real time.

A kind of physiologic parameter monitoring device 130, for monitoring physiological parameter associated with subject, the physiological parameter Monitoring device 130 includes: processor；Monitoring modular 144 is connected to the processor, for performing the following operation: obtaining table Show the correlativity between correlated samples photoplethaysmography feature i.e. correlated samples PPG feature and the physiological parameter to be monitored Mathematical model, wherein the correlated samples PPG feature be the influence based on the physiological parameter to multiple sample PPG features and from Selected in the multiple sample PPG feature；It is determined and the survey according to the video of the physical feeling 134 of test subject The associated test PPG feature of subject is tried, wherein the video is the camera using the physiologic parameter monitoring device 130 Taken by 136；And it is monitored based on the test PPG feature and the mathematical model for the test subject The physiological parameter.

Detailed description of the invention

It is described in detail with reference to attached drawing.In the accompanying drawings, the leftmost number of appended drawing reference identifies the appended drawing reference The attached drawing occurred first.Throughout the drawings using identical number to refer to identical feature and component.

Fig. 1 show it is according to the realization of this theme, be connected with modeling it is associated with subject for monitoring The physiologic parameter monitoring device of physiological parameter.

Fig. 2 shows according to the realization of this theme, for determining to monitor physiological parameter associated with subject The method of mathematical model.

Fig. 3 show it is according to the realization of this theme, for being selected from sample PPG feature used when determining mathematical model The method of correlated samples PPG feature.

Fig. 4 show it is according to the realization of this theme, for monitoring life associated with subject using handheld apparatus The method for managing parameter.

Specific embodiment

This theme is related to monitoring physiological parameter associated with subject using handheld apparatus.

Traditionally, photoplethysmography (PPG) is realized in various ways, to measure and monitor physiological parameter.In minority In several traditional technologies, the video of the subject of physiological parameter is measured from certain distance shooting, analyzes the video then with true Determine physiological parameter.Thus, certain other traditional technologies can be related to the position for determining that the body of subject is in contact with video camera Position, then using the video measure physiological property.

In order to measure physiological parameter, usually by the way that Video segmentation framing is extracted PPG waveform from video, then in frame Predetermined set in analyze the peak frequencies of one or more quantized color values.Traditional technology generally relates to record video to obtain Obtain the video recording apparatus of PPG waveform.It usually for example will be used to measure such as pulse blood oxygen instrument of heart rate or based on the biography of sound The sensing device of sensor etc. is used in conjunction with the video recording apparatus for monitoring physiological parameter.Usually by sensing device institute The feature and PPG waveform measured is used to improve the measurement accuracy of physiological parameter with being combined.However, although by removing The measurement for using sensing device to make physiological parameter other than PPG waveform is quite accurate, but with for monitoring physiology in this way The equipment of parameter is associated at high cost.In addition, the equipment is not portable and may be not used to mobile realization.At it In its traditional technology, more than two PPG waveforms are extracted from video, and these PPG waveforms are used to carry out physiological parameter Measurement is to improve accuracy when measurement physiological parameter.However, PPG waveform processing and analysis using a large amount of computing resources and when Between, lead to technology trouble and time-consuming in this way.

This subject specification is to use physiologic parameter monitoring device (referred to herein as device) associated with subject to monitor Physiological parameter method and apparatus.According to one aspect, this theme is related to realizing and the following photoplethaysmography being combined (PPG) realization of technology: modeling the correlativity between PPG waveform and the given value of physiological parameter, then by the mould Type is deployed on device to monitor physiological parameter.Physiological parameter for example may include heart or pulse frequency, respiratory rate, blood pressure or Heart based on electrocardiogram (ECG) feature.In this example, device can be such as smart phone or tablet personal computer (PC) the handheld apparatus such as.

According to realization, using sampling apparatus, the body of sample subject is for example shot using the camera of sampling apparatus The video at position.For example, sampling apparatus can be the handheld apparatus of smart phone or tablet personal computer (PC) etc.. In the example, the video of physical feeling is shot in the case where making camera lens of the physical feeling against camera.For example, can be with The finger tip of subject or the video of ear-lobe are shot to measure physiological parameter.In addition, handling the video to obtain sample PPG waveform. It in one example, can be by being handled for the quantized color value of each frame in video video, then determining frame The frequency of the quantized color value of each frame in predetermined set, to obtain sample PPG waveform.Then, based on the frame in the set Frequency determines sample PPG waveform.

Once be judged as PPG waveform be it is coherent, then multiple sample PPG features are extracted from sample PPG waveform.In example In, sample PPG feature can be extracted in the time domain；However, in another example, sample PPG feature can be extracted in a frequency domain. In another example, sample PPG feature can be extracted in time domain and frequency domain.In one case, extracted in the time domain Sample PPG feature (also known as temporal signatures) may include for the peaking frequency in sample PPG waveform peak to peak time interval, Pulse spacing, time to peak, Delta Time, the trough to trap for indicating the time needed for sample PPG waveform reaches peaking frequency Time, fall time, trap to decrease amount, the rate of rise, descending slope and the area below the period.In addition, in this example, In a frequency domain extracted sample PPG feature (also known as frequency domain character) may include the position of peak frequency, leading peak frequency and Closely the distance between peak frequency, spectral centroid and leading peak frequency field width.According to example, as some sample PPG Feature, it is also contemplated that the age of the weight of such as subject associated with examinee, the height of examinee and examinee Deng physical trait.

According to the aspect of this theme, based on the practical given value of the physiological parameter of modeling to be realized come from multiple sample PPG One or more correlated samples PPG features are selected in feature.These practical given values of physiological parameter are known as standard value (ground truth value).In the implementation, can influence based on physiological parameter to PPG feature select correlated samples PPG feature.Therefore, in order to determine relevant sample PPG feature, determine that correlation is graded for each sample PPG feature, and can To select correlated samples PPG feature compared with threshold correlation grading based on correlation grading.In this example, for sample The correlation grading of PPG feature can use the variation of physiological parameter to indicate the variation of sample PPG feature.According to the side Face, correlated samples PPG feature can be with the physiological parameter to be monitored share can distinguish relationship and with can area otherwise Indicate the feature of physiological parameter.

According to the aspect of this theme, the selection of PPG feature can be related to the method for two steps.In the first step, it determines Correlativity between PPG feature and standard value.It in the second step, can the standard value based on PPG feature and physiological parameter Between the intensity of correlativity select related PPG feature.A part of selection as related PPG feature, can be by institute The entire set-partition of the feature of extraction is at test set and one or more training sets.In this example, it can be mentioned from training set Related PPG feature is taken, and test set can be used to determine the correlation of selected PPG feature and the accuracy of the selection. In the training stage, the standard value of PPG feature and physiological parameter is known, and is determined respectively based on PPG feature and standard value The value of the related coefficient of PPG feature.Then the gain factor of the PPG feature is determined using the value of the related coefficient of PPG feature. In this example, gain function curve can be used to determine the value of gain factor.It, can be to gain function song in the example The slope of line is finely adjusted with the optimal value of the gain factor of each PPG feature of determination.It uses in test phase and so obtains most Excellent gain factor.In the training stage, by PPG feature multiplied by their optimum gain factor, training is subsequently used for for estimating to give birth to Manage the sorter model of parameter.On the other hand, dduring test, can by optimum gain factor multiplied by each PPG feature to estimate Physiological parameter.

Therefore, in the implementation, multiple respective phases of PPG feature in training set are determined based on PPG feature and standard value Relationship number.The related coefficient can capture the relationship between PPG feature and the standard value of physiological parameter.In this example, phase relation Number can be maximum information coefficient (MIC) value and can be determined based on MIC technology.Once it is determined that the MIC of PPG feature Value, then can determine the intensity of the correlativity between each PPG feature and standard value.Therefore, related coefficient and increasing can be based on Beneficial function determines multiple respective gain factors of PPG feature.In this example, gain function can be S type (sigmoid) gain Function.

As will be appreciated, gain function, thus gain factor can be based on the threshold of such as MIC value of PPG feature Value enhances or highlights the high PPG feature of intensity of correlativity.Therefore, in the implementation, by each PPG feature multiplied by corresponding Gain factor to select related PPG feature.In this example, PPG feature can be selected based on the threshold value of gain factor.Another In the case of one, PPG feature can be selected based on the threshold value of PPG feature.In the case that it is above the two, multiply by PPG feature In the case where gain factor to be worth low (for example, below threshold value of gain factor), the value of PPG feature is suppressed, is down to The threshold value of PPG feature is hereinafter, and can abandon these PPG features.It is thereby possible to select value is greater than threshold value or gain factor Value be greater than threshold value PPG feature be used as correlation PPG feature.

Then, selected correlation is executed using (for example, previously having selected from extracted PPG feature) test set The test of feature.In the implementation, it will be used for the PPG feature in test set for each selected gain factor of PPG feature, to survey Whether examination has accurately selected to be selected as relevant PPG feature based on gain factor.It in this example, can will be in test set PPG feature is multiplied with for gain factor determined by training set respectively.Based on the multiplication, it can be determined that whether such as from training set Conducted in select like that, select identical PPG feature as correlation PPG feature from test set.

In addition, disposing related PPG feature selected above according to realization to estimate in real time and monitor physiological parameter. In embodiment, the standard value based on correlated samples PPG feature and physiological parameter can determine mathematical model.The mathematical model The relationship between correlated samples PPG feature and the standard value of physiological parameter can be captured.According to one aspect, supervision can be used to learn Habit technology determines mathematical model based on the standard value of correlated samples PPG feature and physiological parameter.So determination can be used Mathematical model carrys out the standard value based on PPG waveform and PPG feature assessment physiological parameter, and vice versa.

In the implementation, before further deployment mathematical model, the accuracy of mathematical model can be checked.In this example, The physiology ginseng for the value range for indicating that the measured value of physiological parameter is present in can be estimated using mathematical model in experimental enviroment Number segmentation (bin).Estimated physiological parameter can be segmented and be compared with the practical given value of physiological parameter, with judgement Whether mathematical model is accurate.In the case where mathematical model inaccuracy, realize the training of mathematical model to improve accuracy.Example Such as, more PPG waveforms for various sample subjects can be obtained and handled in a manner of identical with the above, To improve mathematical model.

In embodiment, it can be filled in the physiological compensation effects for monitoring physiological parameter associated with subject is tested Set setting mathematical model.In the implementation, physiological parameter is monitored in order to use the device for being deployed with mathematical model, can be used The camera of the device tests the video of subject to shoot.In the implementation, it can use device processing video then to obtain It must extract and test PPG waveform based on test PPG feature.In one example, with for obtain sample PPG waveform described in Identical mode obtains test PPG waveform from video.In addition, test PPG feature can be identical as sample PPG feature.Another In the case of, which can extract PPG feature corresponding with correlated samples PPG feature.In addition, being based on test feature and mathematics Model, the device with estimating physiological parameters and can monitor the physiological parameter.In this example, the device and its interior disposed mathematics Model can be segmented physiological parameter estimated by physiological parameter with indicator.Therefore, in the example, instead of quantitative measurment, The property of the estimation carried out based on mathematical model can be indicative.In this case, it is mentioned according to the estimation of this theme For following method, wherein utilizing this method, such as the physiological parameter and situation of subject can be monitored, to keep tracking examinee Medical conditions, allow in time provide medical rescue appropriate to subject.

By selecting several correlated samples PPG features, physiology ginseng in the entire set from the extracted PPG feature of video The accuracy of several estimation and its monitoring is quite high.In addition, due to the feature that analyze and handle during estimating physiological parameters Quantity is less, therefore related computing resource and time are greatly decreased when monitoring physiological parameter.Therefore, it might even be possible to will be this Model is deployed on the low device of process performance.As a result, being easy extension according to the monitoring of the physiological parameter of this theme and can be with With high degree of availability.In addition, include the accuracy when physical trait of sample subject also improves estimating physiological parameters, because For these factors in estimating physiological parameters in view of influencing physiological parameter.Further, since being filled without using from other sensings The input set, thus by these physical traits with from the extracted feature of PPG waveform be combined use provide such as blood pressure and The accurate estimation of the physiological parameter of ECG feature etc..Therefore, in terms of this theme, PPG feature can be based only upon to estimate simultaneously Monitor physiological parameter.For example, the standard value of sample PPG feature and the physiological parameter to be monitored can be based only upon to determine mathematical modulo Type.As a result, this theme provides the accurate measurements of physiological parameter, while can set equipment used in this monitoring to just Take formula and be easy manipulation, for example, by using mobile phone etc. handheld apparatus form.

These and other advantage of the invention will be described in more detail in conjunction with attached drawing.Although can with it is any number of not Same computing system, environment and/or structure realize the various aspects of the system and method for monitoring physiological parameter, but with These embodiments are illustrated in the context of lower device.

Fig. 1 show it is according to the embodiment of this theme, for the ease of monitoring company, physiological parameter institute associated with subject The modeling 100 connect.In the implementation, modeling 100 can be based on photoplethaysmography (PPG) technology and physiological parameter Given value determines the correlativity between PPG waveform and physiological parameter.Then the correlativity can be used to supervise in real time Survey physiological parameter.In this example, modeling 100 can be embodied as work station, such as desktop computer or laptop Deng personal computer, multicomputer system, network computer, minicomputer or server.

In one implementation, modeling 100 includes processor 102 and memory 104.Processor 102 can be realized For one or more microprocessors, microcomputer, microcontroller, digital signal processor, central processing unit, state machine, patrol It collects circuit and/or handles any device of signal based on operational order.In addition to other performance, processor 102 can also be matched It is set to extraction and executes the computer-readable instruction stored in memory 104.Memory 104 can connect to processor 102, It and may include any computer-readable medium known in the art, wherein the computer-readable medium is for example including such as The volatile memory of static random access memory (SRAM) and dynamic random access memory (DRAM) etc. and/or The non-volatile of read-only memory (ROM), erasable programmable ROM, flash memory, hard disk, CD and tape etc. is deposited Reservoir.

In addition, modeling 100 may include module 106 and data 108.Module 106 and data 108 can connect to place Manage device 102.Among other things, module 106 further includes routine, the journey for carrying out special duty or realizing special abstract data type Sequence, object, component, data structure etc..Furthermore it is possible to by module 106 be embodied as signal processor, state machine, logic circuit and/ Or any other device or component of signal are handled based on operational order.

In the implementation, module 106 includes processing module 110, feature selection module 112, test module 114, modeling module 116 and other modules 118.Other modules 118 may include supplementing the journey of application program or function that modeling 100 is carried out Sequence or coded command.In addition, data 108 include processing data 120, characteristic 122, modeling data 124 in the realization With other data 126.Among other things, other data 124 are also used as storing as one in execution module 106 The result of a or multiple modules and the repository of data that can handle, receive or generate.It is although showing data 108 in modeling The inside of system 100, it is to be understood that, data 108 may reside within the external repository that can be connected to modeling 100 (in figure It is not shown) in.Therefore, modeling 100 can be set communicated with external repository to obtain letter from data 108 The interface (not shown) of breath.

In addition, modeling 100 can connect to sampling apparatus 128 related to sample subject to obtain in order to work The PPG waveform of connection.At work, the shooting of sampling apparatus 128 will standard value to the physiological parameter that correlativity is modeled The video for the sample subject known.Such as it should be appreciated that standard value is the practical given value of physiological parameter.It is blood pressure in physiological parameter Example in, standard value can be the value of systolic pressure and diastolic pressure.In the ECG feature that physiological parameter is for monitoring heart Another example in, standard value can be the value of such as QRS complex, the interval PR, the interval RR and the ECG feature at the interval QT.Cause This, modeling 100 can model correlativity based on video and standard value.It can be by the standard of sample subject Value is stored in modeling data 124.

In addition, modeling 100 is connect with physiologic parameter monitoring device 130, wherein the physiologic parameter monitoring device 130 makes With the physiological parameter of correlativity and the test subject of monitoring patient etc..In this example, physiologic parameter monitoring device 130 Can be has for providing the handheld apparatus of the processor of process performance.For example, physiologic parameter monitoring device 130 can be Mobile phone, personal digital assistant (PDA), smart phone or tablet personal computer.

During operation, the physical feeling 132 of sample subject is placed on to camera of sampling apparatus 128 etc. In video capture device 134.In one example, sampling apparatus 128 can be handheld apparatus, and may include for example moving Any device of mobile phone, personal digital assistant (PDA), smart phone or tablet personal computer etc..For example, sampling apparatus 128 portable videos that can be smart phone or portable video photokymograph etc. with video recording capabilities are clapped Take the photograph device, i.e. field camera.In this example, in the case where the flash lamp of video capture device 134 is connected, subject can Physical feeling to be positioned to be in contact with the camera lens of video capture device 134 or on the contrary.For example, subject can be by him The finger tip of hand be placed in video capture device 134 to shoot video.In another example, it can be clapped from the ear-lobe of examinee Take the photograph video.Then, sampling apparatus 128 can be obtained by the quantized color value by Video segmentation framing and for example based on each frame Sample PPG waveform, to handle the video.

In the implementation, a part as the processing of video, sampling apparatus 128 can determine one of each frame in multiple windows A or multiple quantized color values.In this example, sampling apparatus 128 can determine quantized color value for particular color model.Example Such as, in the case where color model is RGB (RGB) color model, sampling apparatus 128 can be determined for each frame it is red at Divide, the average value of green components and any ingredient in blue component, and the value can be quantized color value.In color model It is in another situation of hue saturation brightness (HSV) model, sampling apparatus 128 can determine the color of color model for each frame The average value of any ingredient in phase constituent, saturation degree ingredient and luminance components, and the average value can be quantized color value. In addition, sampling apparatus 128 can check the validity of captured video, for example check whether video has for being modeled Enough clarity and illumination.

Then, in the implementation, sampling apparatus 128 can choose the frame set of the frame with predetermined quantity to obtain sample PPG waveform.In this example, sampling apparatus 128 can quantized color value application short time discrete Fourier transform (STFT) technology to frame With the frequency of the quantized color value in each frame of determination.Based on the frequency in each frame, sampling apparatus 128 can be obtained for frame set Sample PPG waveform.

In a further implementation, sampling apparatus 128 can obtain sample PPG waveform and be provided to sample PPG waveform Before modeling 100, the continuity of video is checked.Therefore, sampling apparatus 128 can obtain several frame set from video (each set is referred to as window), and continuity analysis is carried out based on selected window.In the realization, sampling apparatus 128 can The peak frequency of quantized color value with the position of the peak frequency based on the quantized color value in selected window and for other windows Position determine continuity.For example, sampling apparatus 128 can determine that drift motion of the peak frequency between each window of video determines view The continuity of frequency.In the example, if the drift of peak frequency is within a predetermined range, it is judged as that video is continuity. Then, as described above, sampling apparatus 128 selects the frame set of the frame with predetermined quantity from video, and based in each frame The frequency of quantized color value determine PPG waveform from the set.

In the implementation, PPG waveform is provided to modeling 100 by sampling apparatus 128.In a further implementation, sampling apparatus 128 provide sample subjects captured video, and the processing module 110 of modeling 100 with above with reference to sampling cartridge It sets the 128 identical modes and handles video, to obtain PPG waveform.In the realization, in this example, sampling apparatus 128 can To be simply the video recording apparatus 134 of camera etc..

In addition, processing module 110 can analyze sample PPG waveform and obtain multiple sample PPG from sample PPG waveform Feature.It in this example, may include set or the frequency domain spy of temporal signatures from the extracted sample PPG feature of sample PPG waveform The set or the two of sign.For example, the set of temporal signatures may include for the peak-to-peak of the peaking frequency in sample PPG waveform Time interval, pulse spacing, indicate the time to peak of time, time diastole needed for sample PPG waveform reaches peaking frequency, Area below pulse height and sample PPG waveform.

It can understand that processing module 110 carried out determines sample PPG according to PPG waveform with the help of illustrated below Feature.Consider to obtain the case where sample PPG feature is to determine the model of blood pressure for estimating subject.In this case, In order to obtain PPG feature from sample PPG waveform, such as systole phase peak (T is determined in the time domain_sn,A_sn), valley point (T_vn,A_vn) and again Fight incisura (T_dn,A_dn).In the example, T indicates the amplitude at moment and the features described above of A expression sample PPG waveform.Example Such as, processing module 110 minimum point, can represent based on the local maximum from PPG waveform and for example the function of PPG waveform To determine systole phase peak and valley point.In addition, processing module 110 can indicate PPG waveform by determining first in the example Function derivative, then one PPG waveform of identification systole phase peak and the valley point of adjacent PPG waveform peak between first Local maximum, to determine dicrotic notch.

Based on above-mentioned parameter associated with PPG waveform, various sample PPG features are determined.These samples PPG feature is for example It may include the paddy amplitude (A measured at valley point_vn), the systole phase peak amplitude that is measured at each systole phase peak (A_sn), the dicrotic notch amplitude (A that is measured at dicrotic notch_dn) and indicate between systole phase peak and dicrotic notch The paddy of the systole phase area of area below PPG waveform and dicrotic notch and next PPG waveform peak as a PPG waveform peak The dicrotic notch area of the area below PPG waveform between point.In this example, following equation can be used to indicate to shrink Phase area and dicrotic pulse area.

Wherein P indicates the formula for being used for PPG waveform.

In addition, may include for example as contraction based on above-mentioned parameter sample PPG feature obtained in the example The gross area below PPG waveform measured by the summation of phase area and dicrotic pulse area and, for example, opposite as dicrotic pulse area The area ratio measured by the ratio of systole phase area.In addition, sample PPG feature for example may include adjacent as two Determined by time interval between the systole phase peak of PPG waveform peak it is peak-to-peak every, as according to measured by the paddy of PPG waveform Pulse height determined by the amplitude at systole phase peak and as pulse determined by the time between the valley point at the adjacent peak PPG Interval.In one example, the gross area, area ratio, peak-to-peak between, pulse height and pulse is determined based on following equation Every.

The gross area=systole phase area+dicrotic pulse area

Area ratio=dicrotic pulse area/systole phase area

It is peak-to-peak every=T_sn+1–T_sn

Pulse height=A_sn–A_vn

Pulse spacing=T_vn+1–T_vn

In addition, in this example, sample PPG feature may include as same PPG waveform peak systole phase peak and valley point it Between time difference determined by time to peak, indicate time difference between the dicrotic notch and systole phase peak of same PPG waveform Delta Time.In addition, sample PPG feature may include Augmentation index and reflection index.Following equation can be used as example To determine time to peak, Delta Time, Augmentation index and reflection index.

Time to peak=T_sn–T_vn

Delta Time=T_dn-T_sn

Augmentation index=(A_dn–A_vn)/(A_sn–A_vn)

Reflection index=1-Augmentation index

Consider to obtain sample PPG feature to determine another situation for estimating the model of the ECG feature of subject.Showing In example, also in this case, in order to obtain PPG feature from sample PPG waveform, contraction is determined according to sample PPG waveform Phase peak (T_sn,A_sn), valley point (T_vn,A_vn) and dicrotic notch (T_dn,A_dn), wherein T indicates the moment and A indicates to be directed to sample PPG The amplitude of the features described above of waveform.Based on the coordinate of systole phase peak, valley point and dicrotic notch, obtain associated with PPG waveform Various sample PPG features.

In this example, ECG feature assessment in this case, sample PPG feature may include following: as two phases Peak-to-peak interval determined by time interval between the systole phase peak of adjacent PPG waveform peak；As adjacent PPG waveform peak valley point it Between time measured by the pulse spacing；Determined by amplitude as the systole phase peak measured according to the paddy of PPG waveform Pulse height；Indicate the time to peak of the time difference between the systole phase peak and valley point of same PPG waveform peak；As same PPG Delta Time measured by time difference between the dicrotic notch and systole phase peak of waveform peak.In one example, using with Described above identical each equation determines these samples PPG feature.

In addition, sample PPG feature may include following in the case where ECG feature assessment: as same PPG waveform peak Valley point and dicrotic notch between time interval determined by the dicrotic pulse time；Indicate a PPG waveform peak systole phase peak and The fall time of time interval between the valley point of adjacent PPG waveform peak；Indicate the dicrotic notch of a PPG waveform peak and adjacent The dicrotic pulse of time interval between the valley point of PPG waveform peak is to minimum time；For the PPG waveform from valley point to systole phase peak The rate of rise of PPG waveform measured by rising part；And the systole phase peak to valley point for adjacent PPG waveform peak The descending slope of PPG waveform measured by the sloping portion of PPG waveform.In this example, it is determined respectively based on following equation Dicrotic pulse time, fall time, dicrotic pulse to minimum time, the rate of rise and descending slope.

The dicrotic pulse time=T_dn–T_vn

Fall time=T_vn+1–T_sn

Dicrotic pulse is to minimum time=T_vn+1–T_dn

The rate of rise=(A_sn–A_vn)/(T_sn–T_vn)

Descending slope=(A_vn+₁–A_sn)/(T_vn+₁–T_sn)

In addition, processing module 110 can be for example from the PPG feature extracted in frequency domain in amplitude-frequency curve.In example In, processing module 110 can extract the position of leading peak frequency, dominate peak frequency and closely the distance between peak frequency, frequency spectrum The width of mass center and leading peak frequency field is as frequency domain character.In this example, in order to obtain frequency domain character, processing module 110 can be divided into the frame in Sample video the non-overlap rectangular window of 1024 or 256 samples, using side as described above Formula obtains sample PPG waveform.

In addition, according to one aspect, it is also contemplated that special as sample PPG to physical trait associated with sample subject Sign.For example, physical trait may include the age of the weight of subject, the height of subject and sample subject.Show described In example, such as it will be understood that by described above, processing module 110 can obtain sample in the time domain or in a frequency domain or in the two This PPG feature.Processing module 110 can will in the time domain and/or in a frequency domain PPG characteristic storage obtained in processing data In 120, and these features can form the entire set for the PPG feature for obtaining or extracting from Sample video.

In addition, according to the aspect of this theme, feature selection module 112 can be based on will be to the physiology that correlation is modeled The standard value of parameter selects one or more correlated samples PPG features from multiple sample PPG features.In the implementation, feature Selecting module 112 can influence based on physiological parameter to PPG feature select correlated samples PPG feature, vice versa.

In addition, in the implementation, before selecting correlation PPG feature in the set from PPG feature, processing module 110 can be with It goes to intermediate pseudo- peak or trough point to remove noise divided by from PPG feature from PPG feature.Otherwise, practical peak or trough point may be by In Noise environment and lose completely, and may cause during extracting PPG feature occur PPG feature incorrect meter It calculates.In this example, processing module 110 can create the PPG feature of two clusters.In addition, being based on histogram analysis, processing module 110 can initialize mass center to carry out clustering.Then, processing module 110 can be equal using 2 after aggregation density estimation Value is clustered to remove incorrect PPG feature.In another situation, processing module 110 can be using k mean algorithm to be collected Group's mass center.In addition, processing module 110 can use Xie-Beni index to remove incorrect PPG feature and can be used for Carry out the set of the PPG feature of the selection of correlation PPG feature.

In addition, in the implementation, feature selection module 112 can select one or more phases from multiple sample PPG features Close sample PPG feature.According to the aspect of this theme, feature selection module 112 can be followed for from extracted PPG feature Two step methods of correlation PPG feature are selected in entire set.In the first step, feature selection module 112 can determine PPG Correlativity between feature and the standard value of physiological parameter.In addition, in the second step, feature selection module 112 can be with base The intensity of correlativity between PPG feature and the standard value of physiological parameter selects related PPG feature.

According to realization, a part of the selection as related PPG feature, feature selection module 112 can will be extracted These training sets and test set at test set and one or more training sets and are stored in place by the entire set-partition of PPG feature It manages in data 120.In this example, feature selection module 112 can extract correlation PPG feature from training set, and use test The accuracy for collecting to determine the selection of related PPG feature.

Therefore, feature selection module 112 can be directed to the spy of multiple PPG in training set based on PPG feature and standard value Sign respectively determines related coefficient.The related coefficient can capture the relationship between PPG feature and the standard value of physiological parameter.Showing In example, feature selection module 112 can be determined as the MIC value of related coefficient based on maximum information coefficient (MIC) technology.? In example, feature selection module 112 can construct the grid with all size to find between data pair, i.e. PPG feature and Maximum mutual information between standard value.For each pair of data (x, y), if I is the mutual information of grid G, for sample The MIC of the set D of the paired data of size n and sizing grid (xy), feature selection module 112 can be made based on following relationship Related coefficient, i.e. MIC value are determined for example.

MIC (D)=max_xy<B(n){M(D)_x,y}………….(1)

In above-mentioned relation formula (1), expression formula { M (D)_x,yMeasurement data is to the normalized mutual information between (x, y).Separately Outside, in relational expression (1), sizing grid (xy) is less than B (n), and wherein B (n) is the function of sample size and can for example lead to Following relationship is crossed to provide.

B (n)=n^0.6

In addition, for the different distributions of grid G M (D) can be provided as example by following formula.

Once it is determined that the MIC value of PPG feature, then feature selection module 112 can determine each PPG feature and standard value it Between correlativity intensity.Therefore, according to one aspect, feature selection module 112 can be based on related coefficient and gain function Gain factor is respectively determined for multiple PPG features.

In this example, gain function can be S type gain function, and the value of PPG feature can be made from-∞~∞ transformation It is 0~1.In the example, feature selection module 112 can determine gain factor as example based on following S type function (G_n)。

In above expression, w_nRelated coefficient, such as MIC value that can be PPG feature associated with ECG, and 0.5 can be the threshold value of the related coefficient.Although threshold value to be selected as to the centre of maximum MIC value (that is, 1) in above situation, But in other examples, threshold value can be selected as to non-zero .5.In the example, gain factor can be based on MIC obtained Value is distributed weight to each PPG feature relative to standard value.For example, if MIC value obtained is high, is greater than about 0.5, root According to for G_nEquation, gain factor becomes close to 1, and if MIC value obtained is low, is less than about 0.5, for should The gain factor of PPG feature is close to zero.In addition, the curve of constant m control gain function is marking and drawing increasing relative to related coefficient Slope or steepness in the case where beneficial factor.In fact, as pass through above equation it is clear that the value of m can determine gain because Several values.For example, the function forms horizontal line at m=0, to obtain all values for related coefficient, gain factor is equal It is 0.5.This, which is understood to be, is equal to that there is no feature selecting standards.

Therefore, in the implementation, each PPG feature can be multiplied to select by feature selection module 112 with gain factor respectively Related PPG feature.In view of the above example of the relational expression, increasing is specified by selection gain function slope of a curve constant m Beneficial factor.In the realization, feature selection module 112 can be such that value increases with predetermined increment, so that it is determined that the optimal value of m, And thereby determine that the optimal value of the gain function for each PPG feature.Gradient constant m is known as by this be incremented by of predetermined step width The fine tuning of gradient constant m.

According to realization, in order to determine that the optimal value of gain function, feature selection module 112 can roll over verification technique using k. According to the technology, in this example, sorter model is can be used in feature selection module 112, by the value to gradient constant m into Row fine tuning determines PPG feature using training dataset based on the different value of gradient constant m.In this example, classifier mould Type can be the model based on support vector machines (SVM) and one of the model for being based on adaptive neural network (ANN).Institute It states in example, based on the accuracy of identified PPG feature, can determine the value of gain function.In the example, it can incite somebody to action Identified PPG feature is compared the determination accuracy to determine PPG feature with known standard value.Furthermore, it is possible to select needle To the gain factor of accurately determining PPG feature as optimum gain factor.

In another example, based on accurately determining PPG feature, the optimal value of gradient constant m can be determined.In this feelings It, can be according to for gain factor G based on the optimal value of gradient constant m under condition_nEquation determine the value of gain factor.This Outside, in a further implementation, instead of sorter model, feature selection module 112 can be used regression model and be used as predicting to give birth to The fallout predictor model for managing the value of parameter, to be finely adjusted by the value to gradient constant m come the value to determine PPG feature.? In one situation, regression model can be one of linear regression model (LRM), nonlinear regression model (NLRM) and polynomial regression model.

In addition, once it is determined that gain factor, feature selection module 112 can be selected based on the threshold value of gain factor PPG feature.In another situation, feature selection module 112 can select PPG feature based on the threshold value of PPG feature.For example, Feature selection module 112 by PPG feature multiplied by be worth low (such as below threshold value of gain factor) gain factor in the case where, The value of PPG feature is suppressed, is down to the threshold value of PPG feature hereinafter, and can abandon these PPG features.Therefore, feature It is special as correlation PPG greater than the PPG feature that threshold value or the value of gain factor are greater than threshold value that selecting module 112 can choose value Sign.In this example, in the case where providing the intensity of the correlativity between PPG feature and standard value by related coefficient, increase Beneficial factor amplifies intensity value and provides the intensity based on correlativity come the convenience for selecting related PPG feature and accurate side Formula.

In other examples, feature selection module 112 can be based on Pearson came (Pearson) product moment correlation coefficient (PPMCC) concept selects related PPG feature.Feature selection module 112 can determine any between PPG feature and standard value Linearly or nonlinearly relationship, and correspondingly select correlation PPG feature.In addition, in this example, feature selection module 112 can be with Related PPG feature selecting is carried out using statistical and analytical tool.For example, maximum asymmetric score can be used in statistical and analytical tool (MAS) technology, maximal margin value (MEV) technology and minimum unit value (MCV) technology.

Then, test model 114 may be implemented using (for example, previously having selected from extracted PPG feature) test Collection is to test selected correlated characteristic.In the implementation, test module 114 can will be directed to each PPG by feature selection module 112 The selected gain factor of feature is for the PPG feature in test set, to test whether accurately to have selected to be selected as correlation PPG feature.In this example, test model 114 can be by the PPG feature in test set respectively multiplied by in test set Gain factor determined by PPG feature.Based on the multiplication, test model 114 be can decide whether as carried out from training set It selects to select identical PPG feature as correlation PPG feature like that, from test set.

After having selected correlated samples PPG feature, in embodiment, modeling module 116 can be based on correlated samples PPG The standard value of feature and physiological parameter determines mathematical model.Such as it should be appreciated that so determining mathematical model captures related sample Relationship between this PPG feature and physiological parameter.According to one aspect, supervised learning technology can be used and be based on correlated samples PPG The standard value of feature and physiological parameter determines mathematical model.In this case, due to not deposited between standard value and PPG feature The relationship between the two is modeled in direct relation, therefore using supervised learning technology.In one example, it models The learning art based on recurrence, the learning art based on support vector machines (SVM) can be used, based on artificial neuron in module 116 The learning art of network (ANN) or any other this learning art for determining mathematical model.

In addition, as previously described, in the case where physiological parameter is blood pressure, standard value can be systolic pressure and diastolic pressure Value.In another example that physiological parameter includes for monitoring the ECG feature of heart, it is multiple that standard value can be such as QRS The value of the ECG feature at multiplex, the interval PR, the interval RR and the interval QT etc..Mathematical model can be stored in modeling by modeling module 116 In data 122.According to another realization, feature selecting is carried out instead of using accurate standard value, modeling module 116 can will be marked The entire set of quasi- value splits into range or segmentation and determines mathematical model based on these segmentations.

In addition it is possible to use mathematical model carrys out the standard value based on PPG waveform and PPG feature assessment physiological parameter.

Although above specification provides obtain sample PPG waveform for a sample subject to build in other implementations Modular system 100 can obtain sample PPG waveform for multiple sample subjects, and be used in a manner of identical with the above Different sample PPG waveforms are to determine mathematical model.In this case, due to based on associated from different sample subjects Standard value and PPG waveform determine mathematical model, therefore the adaptability of mathematical model is high, and can be used for accurately estimating and supervising Survey physiological parameter.

In the implementation, before further deployment mathematical model is to estimate and monitor physiological parameter, modeling module 116 can be with Determine the accuracy of mathematical model.It in this example, can be in the proving ring for for example developing the modeling 100 disposed in environment The inspection of the accuracy of mathematical model is carried out in border.In one case, modeling module 116 can will be from for physiological parameter The set of the PPG waveform of subject known to standard value PPG feature obtained is provided to mathematical model.Correspondingly, the mathematics Model can be segmented with estimating physiological parameters, i.e. value range present in the measured value of physiological parameter.Modeling module 116 can be into one Estimated physiological parameter is segmented and is compared with the practical given value of physiological parameter to judge whether mathematical model is quasi- by step Really.Mathematical model inaccuracy in the case where, modeling 100 obtain for various sample subjects more PPG waveforms with Training mathematical model, to improve the accuracy of mathematical model.

In addition, mathematical model is provided at physiologic parameter monitoring device 130 (hereinafter referred to as device 130) about deployment, with Monitoring physiological parameter associated with test subject.It in other examples, can be all as may be mounted at using mathematical model Application program on such as handheld apparatus of device 130, such as can download application program provides.In an example In, physiological parameter may include blood pressure associated with test subject and ECG feature.In this example, in monitoring blood pressure In the case of, diastolic pressure associated with subject is tested and systolic pressure can be monitored.In another situation of monitoring ECG feature, The ECG feature at QRS complex associated with subject is tested, the interval PR, the interval RR and the interval QT etc. can be monitored.

In addition, in this example, mathematical model can be stored in the modeling data 140 of device 130 by device 130.In reality In existing, physiological parameter is monitored in order to use the device for being deployed with mathematical model 130, the camera 138 of device 130 can be used To shoot the video of the finger of test subject or the physical feeling 136 of ear-lobe etc..In the implementation, the monitoring of device 130 Module 144 can by with above with reference in a manner of identical described in sampling apparatus 128 (for example, based on quantized color value and its peak frequency Rate) processing video, PPG waveform is tested to obtain.

In addition, monitoring modular 144 can extract test PPG feature from test PPG waveform.In this example, monitoring modular 144 can extract and utilize the correlated samples that processing module 110 is previously identified and is stored in feature selecting data 122 The corresponding PPG feature of PPG feature.Therefore, device 130 can obtain phase from the feature selecting data 122 in modeling 100 Close sample PPG feature and by these characteristic storages in the feature selecting data 142 on device 130.In addition, being based on test feature And mathematical model, monitoring modular 144 with estimating physiological parameters and can monitor the physiological parameter.

In this example, in order to monitor physiological parameter, monitoring modular 144 can be estimated to indicate what physiological parameter was likely to be present in It is worth the physiological parameter segmentation of range.Therefore, in the example, instead of quantitative measurment, the life carried out using monitoring modular 144 The property of estimation and the monitoring of reason parameter can be indicative.In this case, monitoring modular 114 can be provided based on life The value range that parameter is present in is managed to monitor the mode of the medical conditions of the subject in such as stipulated time section.Therefore, one In a example, the medical conditions of subject can be tracked, so that medical rescue appropriate can be provided to subject in due course.

In an example of the BP value of 144 monitoring and test subject of monitoring modular, physiological parameter segmentation can be for " very It is low ", " low ", " normal ", "high" and " very high ".In the example, diastolic pressure be less than about 50 millimetress of mercury (mmHg) or In the case that systolic pressure is less than about 70mmHg, monitoring modular 144 monitors that the BP level of test subject falls in " very low " point In section.In addition, diastolic pressure substantially in about 50~65mmHg in the range of or systolic pressure substantially in the model of about 70~100mmHg In the case where in enclosing, the BP for testing subject is fallen in " low " segmentation, and in diastolic pressure substantially in the model of about 65~90mmHg In enclosing or systolic pressure substantially in the range of about 100~135mmHg in the case where, the BP for testing subject falls in " normal " divide Duan Zhong.In addition, diastolic pressure substantially in about 90~100mmHg in the range of or systolic pressure substantially about 135~160mmHg's In the case where in range, the BP level for testing subject can be considered as falling in "high" segmentation, and be greater than about in diastolic pressure In the case that 100mmHg or systolic pressure are about 160mmHg or more, the BP level for testing subject can be considered as falling in " very In height " segmentation.

Consider that monitoring modular 144 estimates another situation of the ECG feature as a part of monitoring physiological parameter.In this feelings Under condition, physiological parameter segmentation can be referred to as " very low ", " low ", " normal ", "high" and " very high " again.In an example In, in the case where the interval RR is less than about 0.6 millisecond (ms), monitoring modular 144 may determine that associated with test subject ECG feature be " very low ", and the interval PR be less than about the interval 120ms, QRS be less than about the interval 60ms, QT be less than about The interval 350ms or RR substantially in the range of about 0.6~0.8m in the case where, ECG feature can be " low ".In addition, institute State in example, the interval PR substantially in the range of about 120~200ms, the interval QRS substantially in the range of about 60~100ms, The interval QT substantially in the range of about 350~470ms or the interval RR substantially in the range of about 0.8~1 second (s) the case where Under, it can be fallen in " normal " segmentation for the ECG feature of test subject.In addition, being greater than about between 200ms, QRS at the interval PR Every the greater than about interval 100ms, QT be greater than about the interval 470ms or RR substantially in the range of about 1~1.2s in the case where, sentence Break to be fallen in "high" segmentation for the ECG feature of test subject, and in the case where the interval RR is greater than about 1.2s, judgement It is fallen in " very high " segmentation for these ECG features.

In addition, although illustrating the estimation of physiological parameter with reference to physiologic parameter monitoring device 132, it is also possible to be in modeling The monitoring of physiological parameter is realized in system 100 in real time.In this case, the modeling 100 for being stored with mathematical model receives The PPG feature extracted from test video, and can estimating physiological parameters in real time.

Fig. 2 shows for determining the method 200 to monitor the mathematical model of physiological parameter associated with subject, scheme 3 show the method 300 for selecting correlated samples PPG feature from sample PPG feature, and Fig. 4 is shown according to this theme Embodiment, monitor life associated with subject for using the physiologic parameter monitoring device 130 of such as handheld apparatus The method for managing parameter.The sequence for describing these methods be not intended to be construed as it is restrictive, and can group in any order Any amount of the method block is closed to realize these methods or any alternative.Furthermore it is possible to without departing from institute here Individual blocks are deleted in the case where the spirit and scope for the theme stated from these methods.Furthermore, it is possible to any hardware appropriate, Software, firmware or their combination realize these methods.

It can be in general these methods of described in the text up and down of computer executable instructions.Refer in general, computer is executable Order may include routine, programs, objects, component, data structure, the mistake for carrying out specific function or realizing special abstract data type Journey, module, function etc..These methods can also be practiced in a distributed computing environment, wherein in the distributed computing environment, Function is carried out using the remote processing device connected via communication network.In a distributed computing environment, computer is executable Instruction can be located in local and remote computer storage media the two including memory storage device.

In the implementation, one or more of method described here can be implemented at least partially as being embedded into non-transient Computer-readable medium and the instruction that can be performed by one or more computing devices.In general, the processor of such as microprocessor It receives and instructs from the non-transient computer-readable media of such as memory, and execute these instructions, thus carry out including here One or more methods of one or more of the method.It can be used in various known computer-readable mediums Arbitrary medium stores and/or sends these instructions.

With reference to the explanation of Fig. 2, Fig. 3 and Fig. 4, for simplicity, here without discussing modeling 100,128 and of sampling apparatus The details of each component of physiologic parameter monitoring device 132.These details can be as provided in explanation described in reference diagram 1 Understood.

With reference to Fig. 2, in block 202, sample photoplethaysmography (PPG) waveform associated with sample subject is obtained. In the implementation, the view of the physical feeling 132 of at least one sample subject is shot via the camera 134 of sampling apparatus 218 Frequently.For example, the video of the finger tip 132 of subject can be shot.Furthermore, it is possible to pass through the quantized color for each frame in video Value handles the frequency of the quantized color value of each frame in the video, the then predetermined set of determining frame, to obtain sample PPG wave Shape.In this example, the frequency for the frame being then based in the set determines sample PPG waveform.In the implementation, it can use sampling Device 128 carries out the processing for video to obtain PPG waveform.In a further implementation, modeling 100 can be from adopting Sampling device 128 obtains video and handles the video to obtain PPG waveform.

In block 204, sample PPG feature associated with sample subject is extracted from sample PPG waveform.In example In, sample PPG feature may include the set of temporal signatures or the set of frequency domain character or the two.For example, temporal signatures Set may include the peak to peak time interval for the peaking frequency in sample PPG waveform, the pulse spacing, indicate sample PPG wave Face below the time to peak of time needed for shape reaches peaking frequency, time diastole, pulse height and sample PPG waveform Product.In the example, sample PPG feature can be extracted in the time domain.In another example, sample can be extracted in a frequency domain PPG feature.In addition, in this example, extracted sample PPG feature may include the position of leading peak frequency, master in a frequency domain Lead peak frequency and the closely width of the distance between peak frequency, spectral centroid and leading peak frequency field.As substitution or Furthermore it is possible in view of physical trait associated with sample subject is as sample PPG feature.For example, these physical traits The height of weight, subject, the age of subject and other this kind of bodies associated with sample subject including subject Feature.

In block 206, one or more correlated samples PPG features are selected from sample PPG feature.It can be based on and sample The standard value of at least one associated physiological parameter of subject, for example based on physiological parameter to PPG feature influence (otherwise also So) select related PPG feature.Standard value can be interpreted as to the practical given value for the physiological parameter to be monitored.Join in physiology Number is under the example of blood pressure, and standard value can be the value of systolic pressure and diastolic pressure.Physiological parameter in monitoring includes for supervising It surveys in another example of the ECG feature of heart, standard value can be such as between QRS complex, the interval PR, the interval RR and QT Every the value of equal ECG feature.The selection of correlated samples PPG feature is described in detail with reference to Fig. 3.

In block 208, the number of the physiological parameter is determined based on the standard value of correlated samples PPG feature and each physiological parameter Learn model.Mathematical model indicates the correlativity between correlated samples PPG feature and standard value.In addition, in this example, can make Mathematical model is determined with supervised learning technology.For example, supervised learning technology may include learning art based on recurrence, be based on The learning art of support vector machines (SVM) and the learning art for being based on artificial neural network (ANN).

In block 210, accuracy of the mathematical model for example in terms of estimating and monitoring physiological parameter is checked.In this example, It can be segmented in experimental enviroment using mathematical model with estimating physiological parameters.Physiological parameter segmentation indicates the measurement of physiological parameter The value range that value is present in.Estimated physiological parameter can be compared with the practical given value of physiological parameter, to sentence Whether disconnected mathematical model is accurate.In the case where mathematical model inaccuracy, it is accurate to improve that the training of mathematical model may be implemented Property.

In block 212, after having passed through accuracy testing, mathematical model is provided to be for example deployed in physiological compensation effects At device 132.

Fig. 3 shows the method 300 for selecting correlated samples PPG feature from sample PPG feature.

In block 302, Noise and incorrect PPG feature can be removed from extracted PPG feature.

It in block 304, can be by the entire set-partition of extracted feature at test set and one or more training sets. In this example, correlation PPG feature can be extracted from training set, and test set can be used to determine selected PPG feature Correlation and the selection accuracy.

In block 306, multiple PPG features in training set are directed to standard value based on PPG feature, and respectively determination is related Coefficient.The related coefficient can capture the relationship between PPG feature and the standard value of physiological parameter.In this example, the phase relation Number can be maximum information coefficient (MIC) value and can be determined based on MIC technology.

In block 308, can be determined based on related coefficient and gain function for multiple respective gains of PPG feature because Number.In this example, gain function can be S type gain function.Furthermore, it is possible to the selection of the gradient constant based on gain function come Select gain factor.In the realization, the optimal of gain function can be determined based on the optimal value of the slope of gain function Value.

In a block 310, each PPG is directed to by being finely adjusted to parameter associated with gain factor determined above Feature determines optimum gain factor.In this example, Cross-Validation technique can be rolled over using k to determine optimum gain function.According to Training dataset can be used in this example to be finely adjusted by the value to gradient constant m (namely based on slope in the technology The different value of constant m) PPG feature is determined using sorter model.In this example, sorter model can be based on support to The model of amount machine (SVM) and one of the model for being based on adaptive neural network (ANN).In another example, classifier mould Type can be regression model.

In the realization, based on the accuracy of identified PPG feature, the value of gain function can be determined.Described In example, identified PPG feature can be compared to the determination accuracy to determine PPG feature with known standard value.This Outside, it can choose the gain factor for accurately determining PPG feature as optimum gain factor.In another example, it is based on Accurately determining PPG feature, can determine the optimal value of gradient constant m.In this case, based on the optimal of gradient constant m Value, can be according to for gain factor G_nEquation determine the value of gain factor.

In block 312, each PPG feature is executed to the selection of related PPG feature multiplied by optimum gain factor respectively.

In a block 314, it can be selected from extracted feature based on the product of optimum gain factor and each PPG feature Related PPG feature.In this example, PPG feature can be selected based on the threshold value of gain factor.It in another case, can be with base PPG feature is selected in the threshold value of PPG feature.In the case that it is above the two, by PPG feature and be worth it is low (for example, gain because Below several threshold values) gain factor be multiplied in the case where, the value of PPG feature is suppressed, be down to the threshold value of PPG feature with Under, and these PPG features can be abandoned.It is thereby possible to select value is greater than threshold value or the value of gain factor is greater than threshold value PPG feature is used as correlation PPG feature.In one example, the threshold value of PPG feature and the product of gain factor may be about 0.001。

In block 316, executed using (for example, previously having been selected from extracted PPG feature) test set selected Correlated characteristic test.In the implementation, the spy of the PPG in test set will be used for for each selected gain factor of PPG feature Sign is selected as relevant PPG feature to test whether accurately to have selected based on gain factor.In this example, it can incite somebody to action PPG feature in test set is respectively multiplied by for gain factor determined by test set.Based on the multiplication, it can be determined that whether such as Select identical PPG feature as correlation PPG feature like that, from test set from training set is selected.

In addition, Fig. 4 is shown for being monitored physiological parameter associated with subject using physiologic parameter monitoring device 130 Method 400.

With reference to Fig. 4, in block 402, at least one survey is shot via the camera 138 of physiologic parameter monitoring device 130 Try the video of the physical feeling 136 of subject.It in this example, can be by determining the position of finger tip at against camera 138 Camera lens and be switched to the flash lamp of camera 138 connect to shoot the video of finger tip.

In block 404, video is handled to obtain testing photoelectronic capacity trace (PPG) waveform from the video.In this example, with Identical mode described in sample PPG waveform is obtained in the block 202 of Fig. 2 with reference Fig. 1, obtains test PPG wave from video Shape.

In block 406, test PPG feature is extracted from test PPG waveform.In this example, test PPG feature can be with sample This PPG feature is identical.In another example, PPG feature corresponding with correlated samples PPG feature can be extracted as test PPG feature.

In block 408, estimated based on extracted PPG feature and mathematical model corresponding at least one physiological parameter It counts and monitors at least one physiological parameter.In this example, the measured value institute of physiological parameter can be indicated for physiological parameter estimation The physiological parameter segmentation of existing value range.Therefore, it in the example, instead of quantitative measurment, is carried out based on mathematical model Estimation property can be it is indicative.

Although illustrating for the method and system for monitoring the physiological parameter of subject using handheld apparatus It realizes, it is to be understood that, this theme is not necessarily confined to the special characteristic or method.On the contrary, these special characteristics and Method is as monitoring the implementation of physiological parameter of subject and disclosed for using handheld apparatus.

Claims

1. one kind is for monitoring the modeling method of physiological parameter associated with subject, institute using handheld apparatus (130) State modeling method the following steps are included:

Step is obtained, the video for the physical feeling (128) from sample subject obtains associated with the sample subject Multiple sample photoplethaysmography features, that is, multiple samples PPG feature；

Select step, for based on the standard value of the physiological parameter for the sample subject come from the multiple sample At least one correlated samples PPG feature associated with the physiological parameter is selected in PPG feature, wherein the physiological parameter Standard value is the practical given value of the physiological parameter；And

Being determined based at least one correlated samples PPG feature to the standard value of the physiological parameter indicates the related sample The mathematical model of correlativity between this PPG feature and the physiological parameter, wherein the mathematical model is disposed for real When monitor the physiological parameter,

Wherein the acquisition step includes:

Multiple windows are obtained from the video, each window includes the frame of predetermined quantity；

For each frame in the multiple window, at least one quantized color value is determined for one or more color model；With And

Based at least one quantized color value described in each frame, to the conclusive window of the predetermined quantity from the multiple window Continuity analysis is carried out to determine the continuity of the video, wherein in response to obtaining the conclusive window of the predetermined quantity, Carry out the continuity analysis.

2. modeling method according to claim 1, wherein the step of obtaining the multiple sample PPG feature include: when From the multiple sample PPG feature of the video extraction in one of domain and frequency.

3. modeling method according to claim 1, wherein the physiological parameter include indicate heart electrocardiogram i.e. At least one of ECG and respiratory rate.

4. modeling method according to claim 1, wherein the multiple sample PPG feature includes temporal signatures and frequency domain The set of at least one of feature.

5. modeling method according to claim 1, wherein the multiple sample PPG feature includes tested with the sample The associated physical trait of body.

6. modeling method according to claim 5, wherein the physical trait include the sample subject height, The age of the weight of the sample subject and the sample subject.

7. modeling method according to claim 1, wherein the step of determining the mathematical model is based on supervised learning skill What art carried out.

8. modeling method according to claim 1, wherein further comprising the steps of:

Test PPG feature associated with the test subject is obtained from the video of the physical feeling (134) of test subject； And

The physiological parameter for the test subject is monitored based on the test PPG feature and the mathematical model.

9. modeling method according to claim 1, wherein the selection step includes:

Correlation grading is respectively determined for the multiple sample PPG feature, wherein correlation grading indicates the multiple The respective relationship with the physiological parameter of sample PPG feature；And

Based on the respective correlation grading of the multiple sample PPG feature and threshold correlation grading come from the multiple sample At least one correlated samples PPG feature is determined in this PPG feature.

10. modeling method according to claim 1, wherein the selection step includes:

Respectively determining for the multiple sample PPG feature indicates between sample PPG feature and the standard value of the physiological parameter Relationship related coefficient；And

Gain factor is respectively determined for the multiple sample PPG feature based on the related coefficient；

Wherein, the selection step is carried out based on the gain factor.

11. modeling method according to claim 10, wherein the related coefficient is maximum information coefficient i.e. MIC.

12. modeling method according to claim 10, wherein the step of determining the gain factor is based on the gain of S type What function carried out.

13. modeling method according to claim 10, wherein the step of determining the gain factor includes: to be tested based on k folding The accuracy of card technology is finely adjusted gradient constant, that is, m associated with the gain factor, wherein the fine tuning is to make It is carried out with one of regression model and sorter model.

14. modeling method according to claim 13, wherein the regression model is linear regression model (LRM), non-linear time Return one of model and polynomial regression model.

15. modeling method according to claim 13, wherein the sorter model is the model based on support vector machines Model i.e. based on SVM and the model based on adaptive neural network are one of the model based on ANN.

16. modeling method according to claim 10, wherein the selection step includes:

The multiple sample PPG feature is multiplied with the respective gain factor respectively；And

The correlated samples are selected from the multiple sample PPG feature based on the threshold value of each sample PPG feature after multiplication PPG feature.

17. modeling method according to claim 10, wherein further comprising the steps of: being based on using processor (102) Each gain factor determines the respective true correlation of correlated samples PPG feature.

18. one kind is for monitoring the modeling method of physiological parameter associated with subject, institute using handheld apparatus (130) State modeling method the following steps are included:

The standard value of at least one correlated samples PPG feature and the physiological parameter is based only upon to determine that expression is described at least The mathematical model of correlativity between one correlated samples PPG feature and the physiological parameter, wherein the mathematical model quilt Deployment for monitoring the physiological parameter in real time, the physiological parameter be in electrocardiogram i.e. ECG feature and respiratory rate at least One,

Wherein the acquisition step includes:

19. modeling method according to claim 18, wherein the selection step includes: to obtain and the physiology to be monitored The standard value of parameter has at least one correlated samples PPG feature of correlativity, wherein at least one described related sample This PPG feature is according to the correlativity between PPG feature and the standard value of the physiological parameter and based on the related pass Be determined by gain factor and selected in the multiple sample PPG feature.

20. a kind of modeling (100), for monitoring physiological parameter associated with subject, modeling (100) packet It includes:

Processor (102)；

Processing module (110) is connected to the processor (102), and the processing module is for obtaining and sample subject phase Associated multiple sample photoplethaysmography features, that is, multiple samples PPG feature, wherein the multiple sample PPG feature is from institute It states extracted in the video of the physical feeling (128) of sample subject；

Feature selection module (112) is connected to the processor (102), and the feature selection module is used to be based on the life The standard value of reason parameter comes from the multiple sample PPG feature selection and the physiological parameter, and associated at least one is related Sample PPG feature, wherein the standard value of the physiological parameter is the practical given value of the physiological parameter；And

Modeling module (116) is connected to the processor (102), and the modeling module is used for based at least one described phase The standard value of sample PPG feature and the physiological parameter is closed to determine and indicate the correlated samples PPG feature and physiology ginseng The mathematical model of correlativity between number, wherein the mathematical model is used to monitor the physiological parameter in real time,

Wherein obtain the multiple sample photoplethaysmography feature, that is, multiple samples PPG feature associated with sample subject Include:

21. modeling (100) according to claim 20, wherein the processing module (110) performs the following operation:

The video of the physical feeling (128) of the sample subject is obtained from sampling apparatus (126)；And

The video is handled to determine sample PPG waveform.

22. modeling (100) according to claim 20, wherein the processing module (110) time domain and frequency extremely The multiple sample PPG feature is obtained from the video in one few.

23. modeling (100) according to claim 20, wherein the modeling module (116) is based on supervised learning skill Art determines the mathematical model.

24. modeling (100) according to claim 20, wherein the feature selection module (112) carries out following behaviour Make:

Correlation grading is respectively determined for the multiple sample PPG feature, wherein correlation grading indicates each sample PPG The relationship of feature and the physiological parameter；And

The respective correlation grading of the multiple sample PPG feature is compared with threshold correlation grading, with selection At least one correlated samples PPG feature.

25. modeling (100) according to claim 20, wherein the feature selection module (112) carries out following behaviour Make:

The pass indicated between PPG feature and the standard value of the physiological parameter is respectively determined for the multiple sample PPG feature The related coefficient of system；

Gain factor is respectively determined for the multiple sample PPG feature based on the related coefficient；And

The correlated samples PPG feature is selected based on the gain factor.

26. modeling (100) according to claim 25, wherein further include test module (114), the test mould Block (114) is connected to the processor (102), for determining the correlated samples PPG feature based on each gain factor Respective true correlation.

27. modeling (100) according to claim 25, wherein the feature selection module (112) carries out following behaviour Make:

28. modeling (100) according to claim 25, wherein the related coefficient be maximum information coefficient i.e. MIC。

29. modeling (100) according to claim 25, wherein the feature selection module (112) is increased based on S type Beneficial function determines the gain factor.

30. modeling (100) according to claim 25, wherein the feature selection module (112) is based on k folding and tests The accuracy of card technology is finely adjusted gradient constant, that is, m associated with the gain factor, wherein the fine tuning is to make It is carried out with one of regression model and sorter model.

31. a kind of physiologic parameter monitoring device (130), for monitoring physiological parameter associated with subject, the physiology ginseng Counting monitoring device (130) includes:

Processor；

Monitoring modular (144), is connected to the processor, for performing the following operation:

Acquisition expression correlated samples photoplethaysmography feature, that is, between correlated samples PPG feature and the physiological parameter to be monitored The mathematical model of correlativity, wherein the correlated samples PPG feature is based on the physiological parameter to multiple sample PPG features Influence and selected in the multiple sample PPG feature；

Test PPG associated with the test subject is determined according to the video of the physical feeling (134) of test subject Feature, wherein the video is taken by camera (136) using the physiologic parameter monitoring device (130)；And

The physiological parameter for the test subject is monitored based on the test PPG feature and the mathematical model,

The multiple sample PPG feature is wherein obtained in the following way:

32. physiologic parameter monitoring device (130) according to claim 31, wherein the monitoring modular (144) obtain with The standard value for the physiological parameter to be monitored has at least one correlated samples PPG feature of correlativity, wherein the physiology is joined Several standard values is the practical given value of the physiological parameter, and the correlated samples PPG feature is according to PPG feature and the life Manage the correlativity between the standard value of parameter and the gain factor based on determined by the correlativity and from the multiple Selected in sample PPG feature.

33. a kind of non-transient computer-readable media has and is performed so that modeling (100) executes following steps Computer-readable instruction collection:

Select step, for based on the standard value of the physiological parameter for the sample subject come from the multiple sample PPG At least one correlated samples PPG feature associated with the physiological parameter is selected in feature, wherein the mark of the physiological parameter Quasi- value is the practical given value of the physiological parameter；And

Being determined based at least one correlated samples PPG feature to the standard value of the physiological parameter indicates the related sample The mathematical model of correlativity between this PPG feature and the physiological parameter, wherein the mathematical model for supervising in real time The physiological parameter is surveyed,

Wherein the acquisition step includes:

34. non-transient computer-readable media according to claim 33 also makes the modeling (100):

The phase indicated between PPG feature and the standard value of the physiological parameter is respectively determined for the multiple sample PPG feature The related coefficient of pass relationship；

Gain factor is respectively determined for the multiple sample PPG feature based on the related coefficient and S type gain function；And

The correlated samples PPG feature is selected based on the gain factor.