WO2013052944A1

WO2013052944A1 - Periodic artifact reduction from biomedical signals

Info

Publication number: WO2013052944A1
Application number: PCT/US2012/059217
Authority: WO
Inventors: Gari CLIFFORD; Julien Oster; Olivier Pietquin; Matthieu GEIST
Original assignee: Isis Innovation Limited
Priority date: 2011-10-06
Filing date: 2012-10-08
Publication date: 2013-04-11

Abstract

The present disclosure provides systems and methods for separating periodic or pseudo periodic artifacts (e.g., magnetohydrodynamic (MHD) effects) from biomedical signals of interest (e.g., electrocardiograms (ECG)) by model-based nonlinear Bayesian filtering or other adaptive processes.

Description

PERIODIC ARTIFACT REDUCTION FROM BIOMEDICAL SIGNALS CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority to co-pending U.S. provisional application entitled "MAGNETOHYDRODYNAMIC EFFECT SEPARATION FROM ELECTROCARDIOGRAM" having serial no. 61/543,925, filed October 6, 2011, the entirety of which is hereby incorporated by reference as if fully set forth herein.

CROSS-REFERENCE TO RELATED DOCUMENTS

[0002] This application incorporates by reference the following papers as if they were fully set forth herein expressly in their entireties:

[0003] (a) "MagnetoHydroDynamic Effect separation from ECG acquired during

MRI by using nonlinear Bayesian filtering" (attached hereto as Appendix A);

[0004] (b) "Comparison of three artificial models of the MHD Effect on the

Electrocardiogram" (attached hereto as Appendix B);

[0005] (c) a poster presentation describing the separation of MHD effects from ECG data (attached hereto as Appendix C);

[0006] (d) "Tracking arrhythmias in the ECG using a switching Kalman filter"

(attached hereto as Appendix D);

[0007] (e) J. Oster, R. Linares, Z. Tse, E. J. Schmidt and G. D. Clifford, "Realistic

MHD modeling based on MRI blood flow measurements." In Proceedings of the annual meeting of the Int. Soc. for Magn. Res. in Med., 2012;

[0008] (f) G. D. Clifford et al. , An artificial vector model for generating abnormal electrocardiographic rhythms. Phys. Meas., 31:595-609, 2010; [0009] (g) R. Sameni, M. B. Shamsollahi, C. Jutten, and G. D. Clifford, "A Nonlinear

Bayesian Filtering Framework for ECG Denoising." IEEE Trans. Biomed. Eng., vol. 54, pp. 2172-2185, 2007;

[0010] (h) O. Sayadi, M Shamsollahi and G. Clifford, "Robust detection of premature ventricular contractions using a wave-based Bayesian framework." IEEE Trans. Biomed. Eng., vol. 57, no. 2, pp. 353-362, 2010; and

[0011] (i) U.S. Patent Application No. 11/470,506, published as US2007/0260151 on

November 8, 2007, by Clifford, filed on September 6, 2006, and having the title "Method and Device for Filtering, Segmenting, Compressing, and Classifying Oscillatory Signals" (hereafter "the Clifford Application") which is also incorporated by reference as if expressly set forth herein in its entirety.

FIELD OF THE DISCLOSURE

[0012] The present disclosure relates generally to signal processing and, more particularly, to systems and methods for separating artifacts from acquired signals.

BACKGROUND

[0013] Electrocardiograms (ECG) are often used in conjunction with magnetic resonance imaging (MRI) techniques. For example, ECG signals can be used to monitor a patient's physiological parameters during an MRI scan. Additionally, ECG signals can be used to synchronize MRI signal acquisition with heart motion.

The ECG signal, however, can be distorted because of various effects associated with the MRI. Thus, there exists a need in the industry for improved signal processing techniques.

SUMMARY

[0015] The present disclosure provides systems and methods for reducing or

separating periodic (or semi-periodic) artifacts from biomedical signals acquired in noisy environments, such as during magnetic resonance imaging (MRI). Examples of artifacts that can be reduced or separated from such an acquired biomedical signal include magnetohydrodynamic (MHD) effects, ballistocardiogram effects and during fetal monitoring. Examples of the biomedical signals include electrocardiograms (ECG), electroencephalogram (EEG), electrooculogram (EOG), electromyogram (EMG), fetal ECG, blood pressure waveform and photoplethysmogram (PPG). Thus, for example, in various embodiments the present disclosure provides systems and methods for reducing or separating artifacts such as magnetohydrodynamic (MHD) effects from electrocardiograms acquired during a magnetic resonance imaging (MRI) examination. In other embodiments, for example, ballistocardiogram effects can be reduced or separated from electroencephalogram (EEG) acquisitions during functional MRI (fMRI).

[0016] In various embodiments, the reduction or separation of an artifact is

accomplished by an adaptive filtering process, such as by nonlinear Bayesian filtering or other adaptive processes. [0017] In an embodiment of the present disclosure, a method is provided for reducing or separating an artifact from a biomedical signal acquired in a noisy environment. The artifact can be periodic or semi-periodic. The method can includes: acquiring a first biomedical signal apart from the noisy environment; modeling the first biomedical signal; acquiring a second biomedical signal of the same type but acquired in the noisy environment, the second biomedical signal including the artifact; modeling the second biomedical signal; and filtering the biomedical signal by determining an estimated signal from a combination of the modeled first and second biomedical signals and then separating the estimated combination of the first and second biomedical signals into an estimated first biomedical signal and an estimated second biomedical signal. The method can be a computer implemented method.

[0018] In an embodiment, a system is provided for reducing an artifact from a biomedical signal acquired in a noisy environment. The artifact can be periodic or semi-periodic. The system can includes: a data acquisition system configured to acquire a first biomedical signal apart from the noisy environment, and to acquire a second biomedical signal of the same type but acquired in the noisy environment, the second biomedical signal including the artifact; and a processing system coupled to the data acquisition system, the processing system being configured to model the first biomedical signal, to model the second biomedical signal, and to filter the biomedical signal by determining an estimated signal from a combination of the modeled first and second biomedical signals and then separating the estimated combination of the first and second biomedical signals into an estimated first biomedical signal and an estimated second biomedical signal. [0019] In any one or more aspects, the system can be a computer system. The processing system can include: a local interface; and a processor, memory, a user interface, and an I/O device, each coupled to the local interface. The processing system can include a mobile application for a mobile device. The data acquisition system and the processing system can be integrated into a single device, or residing on separate devices.

[0020] In an embodiment, a computer readable medium executable on a computer is provided for at least one of filtering, segmenting or classifying a biomedical signal including an artifact acquired in a noisy environment. The computer readable medium can execute the steps of: acquiring a first biomedical signal apart from the noisy environment; modeling the first biomedical signal; acquiring a second biomedical signal of the same type but acquired in the noisy environment, the second biomedical signal including the artifact; modeling the second biomedical signal; and filtering the biomedical signal by determining an estimated signal from a combination of the modeled first and second biomedical signals and then separating the estimated combination of the first and second biomedical signals into an estimated first biomedical signal and an estimated second biomedical signal.

[0021] In any one or more aspects of the embodiments, the biomedical signal can be selected from one or more of an electrocardiogram (ECG), electroencephalogram (EEG), electromyogram (EMG), fetal ECG, blood pressure waveform or photoplethysmograph (PPG). The noisy environment can be a magnetic resonance imaging (MRI) scanner and the first biomedical signal can be acquired outside of the bore of the MRI scanner and the second biomedical signal can be acquired inside the bore of the MRI scanner. The first and second biomedical signals can be each modeled as a pseudo periodic signal. The first and second biomedical signals can be modeled by fitting Gaussian parameters to the biomedical signals. The second biomedical signal can be modeled as a pseudo periodical signal with a periodicity being the same as the first biomedical signal or having a different periodicity. The cycle of the second biomedical signal can be approximated by a sum of Gaussians. The second biomedical signal can be modeled by solving Navier Stokes equations for different sections of an aortic arch of a subject and by propagating the resulting current source to the torso of the subject. The second biomedical signal can be modeled in accordance with a polynomial function, a spline function or a function like f(t) = A exp(-B(t-Tk)sin(2 pi (t-Tk)/(Tk+l - Tk))), Tk being the time of the kth R peaks and Tk < t < Tk+1 and A B being the parameters of this model. The method can further include the step of tracking the estimated first and second biomedical signals for abnormalities in the biomedical signal appearing as deviations from the estimated signals.

In any one or more aspects of the embodiments, the filtering can involve nonlinear Bayesian filtering. The filtering can include Kalman filtering, particle filtering or Gaussian processes. The filtering step can include use of the following observation equations to provide an estimation of a periodic artifact to be reduced or separated from the biomedical signal:

[0023] In any one or more aspects of the embodiments, the filtering can involve use of signal quality indices (SQFs) to automatically adjust the observation covariance noise matrix for Bayesian filtering. The filtering can involve use of a multiple model approach and a switching Kalman Filter to allow estimation of the different parameters for the models and selection of the best model.

[0024] In any one or more aspects of the embodiments, the artifact can be a

magnetohydrodynamic (MHD) effect from an electrocardiogram (ECG) signal acquiring during a magnetic resonance imaging (MRI) technique and the first biomedical signal acquired can be an ECG, the second biomedical signal acquired can also be ECG, the modeling of the second biomedical signal can include modeling for MHD effect, and the filtering can include determining an estimated signal from the combination of the modeled first ECG signal and the modeled second ECG signal modeled for MHD effect and then separating the estimated combination of the first and second ECG signals into an estimated ECG signal and an estimated MHD signal.

Other systems, devices, methods, features, and advantages will be or become apparent to one with skill in the art upon examination of the following drawings and detailed description. It is intended that all such additional systems, methods, features, and advantages be included within this description, be within the scope of the present disclosure, and be protected by the accompanying claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The patent or application file may contain at least one drawing executed in color. Copies of this patent or patent application publication with color drawings(s) will be provided by the Office upon request and payment of any necessary fee.

[0027] Many aspects of the disclosure can be better understood with reference to the following drawings. The components in the drawings are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles of the present disclosure. Moreover, in the drawings, like reference numerals designate corresponding parts throughout the several views.

[0028] FIG. 1 is a flowchart showing one embodiment of a process for separating magnetohydrodynamic (MHD) effects from an electrocardiogram (ECG).

[0029] FIGS. 2A and 2B are flowcharts showing one embodiment of a process for

Bayesian filtering of an acquired ECG signal.

[0030] FIG. 3 is an example of ECG data obtained from a patient in the absence of a magnetic resonance imaging (MRI) magnetic field.

[0031 ] FIG. 4 is an example of ECG data contaminated by MHD effects, obtained from a patient in the presence of a magnetic resonance imaging (MRI) magnetic field.

[0032] FIG. 5 is an example of superimposed data, showing ECG data contaminated by MHD effects in a 1.5 Tesla (T) magnetic field, the MHD effects, and ECG data from which MHD effects are removed. [0033] FIG. 6 is an example of superimposed data, showing ECG data contaminated by MHD effects in a 3.0 T magnetic field, the MHD effects, and ECG data from which MHD effects are removed.

[0034] FIG. 7 is a flow chart depicting an embodiment involving the use of signal quality indices (SQI's) to automatically adjust the observation noise co variance matrix for Bayesian filtering a biomedical signal

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] Reference is now made in detail to the description of the embodiments as illustrated in the drawings. While several embodiments are described in connection with these drawings, there is no intent to limit the disclosure to the embodiment or embodiments disclosed herein. On the contrary, the intent is to cover all alternatives, modifications, and equivalents.

[0036] The various embodiments of the present systems and methods are directed to removing periodic (or semi-periodic) artifacts from biomedical signals of interest. As a non-limiting example, attention is first turned to an embodiment for removing (magnetohydrodynamic) MHD effects from electrocardiogram (ECG) signals acquired from a magnetic resonance imaging (MRI) examination, as shown in FIGS. 1 and 2. As described elsewhere herein other artifact effects can be separated from a biomedical signal of interest acquired in another noisy environment.

[0037] Electrocardiograms are important physiological signals used in diagnosing cardiovascular pathologies. ECGs are electrical recordings resulting from successive depolarization of cardiac cells. Magnetic resonance imaging is a non-invasive radiological technique that allows for depiction of soft tissue and organs of the body. Using MRI it is possible to distinguish pathological tissue from healthy tissue, often without using a contrast agent.

[0038] Sometimes, physiological parameters, such as the ECG, are monitored in patients undergoing an MRI. ECGs are often used to detect cardiac abnormalities. Additionally, acquisition of an ECG during an MRI scan allows for synchronization of MRI acquisition with heart motion, which accounts for organ motion during the scan.

[0039] ECG relies on electromagnetic fields naturally induced by the body. The MRI environment has three major physical characteristics, which affect the ECG signal: high static magnetic field; fast varying magnetic fields (gradients); and radio frequency (RF) pulses. RF pulses induce an electrical field inside the body that can also interfere with the electrical components involved in measuring the ECG signal. Gradients induce electrical fields inside the body whose frequency range overlaps an ECG signal causing ECG signal distortion.

[0040] In addition to these distortions, the presence of the high static magnetic field in MRI scanners induces another, somewhat-indirect artifact on the ECG signal. This indirect effect, called the magnetohydrodynamic effect, or the Hall effect, arises from blood carrying electrically charged particles, such as iron, whose motion inside a magnetic field creates a current source. The strength of the MRI magnetic field is such that the current source created by the blood flow is about the same magnitude as that of the heart natural electrical activity. It has been shown that the main contribution of the MHD effect is induced by the blood ejection through the aortic arch, because of the geometry of the arch, the diameter of the artery and the blood velocity. Given the timing and the amplitude of this phenomenon, the ST to T wave segment of the ECG is hidden behind a high level of noise and the detection of pathological ventricular repolarization is currently impossible during MRI. This is becoming a growing concern with the development of applications such as very high field cardiac imaging, MR guided surgery, cardiac imaging during exercise stress, or intra-cardiac electrophysiology guided by real-time MRI. The electromagnetic field associated with this current source affects the acquired ECG signal, potentially masking cardiac abnormalities. As a result, it becomes useful to separate the MHD effect from the ECG signal to allow for a more accurate representation of the ECG, and correspondingly a more accurate diagnosis of cardiac abnormalities, should such abnormalities exist.

Various embodiments of the present disclosure, therefore, provide systems and methods for removing contaminants from a time series that is similar in frequency, morphology, amplitude, and timing to the signal of interest. Due to the contaminant's similarity to the signal of interest, these types of contaminants are difficult to remove. Currently existing techniques employ simple template matching (and subtraction) or simple adaptive filtering. However, in the case of template subtraction, the changing morphology and timing of the beat is not modeled well. So, residual errors cause clinically significant distortions in the electrocardiogram. Adaptive filtering carries little understanding of either the signal of interest or the contaminant signal. Thus, adaptive filtering techniques per se may not be ideal. [0042] Various embodiments of the present systems and methods combine a model of both the artifact and the signal of interest, which adapts to sample-by- sample changing dynamics to make estimates of underlying sources of information. Thus, for example, embodiments described herein use models of both a biomedical signal of interest, such as ECG, and artifact sources from the individual whose physiological parameters are being measured. Thus, the resulting signal is custom tailored to each individual.

[0043] In various embodiments, the separation of the two contributions (for example,

ECG and artifact) relies on prior knowledge of the uncontaminated ECG and a model of the artifact. Thus, in any one or more aspects, an ECG signal acquired inside an MR bore can be approximated or modeled by the superposition of an ECG signal acquired without artifact (i.e., without MHD effect) and of an MHD signal. The disclosed systems and methods can also take into account the non-stationary nature of the signals, thereby allowing the parameters of the models to evolve over time. This allows for accurate patient monitoring.

[0044] Specifically, FIG. 1 is a flowchart showing an embodiment of a process

(1000) for separating MHD effects from an ECG acquired during an MRI examination. As shown in FIG. 1 , the process comprises the steps of acquiring

(1001) an ECG from a patient outside of the bore of the magnetic resonance imaging (MRI) scanner. An example of ECG data that has been acquired from outside of the bore is shown in FIG. 3.

[0045] Thereafter, the ECG is modeled (1002). As an example the ECG can be modeled by fitting Gaussian parameters to the ECG. In an aspect, the Gaussian parameters can be fit to the ECG using the methods described in the aforementioned Clifford Application (US published application no. US 2007/026015), which is incorporated by reference as if expressly set forth herein in its entirety. The ECG signal can be modeled as a pseudo periodic signal, whose period cycle is a sum of Gaussians, such that:

where ¾ is the angular position in cylindrical coordinates, the

angular speed with δ being the sampling period and R being the time between adjacent heart beats. ¾ represents the ECG value in mV at time k. a, is the amplitude of the i^'th Gaussian, b_t is the width of the i^th Gaussian, and

with

being the angular position of the i^th Gaussian.

In another aspect, the ECG can be modeled with a dynamical vector cardiograph (VCG) model, that represents each of the three axes of the VCG by a sum of Gaussians and then applies a Dower transform. See, e.g., G. D. Clifford et al, An artificial vector model for generating abnormal electrocardiographic rhythms. Phys. Meas., 31 :595-609, 2010. In order to model a pathological ventricular repolarization, the parameters of the Gaussian representing the T wave can be evolving. The T wave inversion can be modeled by inverting the amplitude of the T wave Gaussians with a logistic function over ten cycles. The prolonged QT interval can be modeled by moving forward angular position of the T wave Gaussians with a logistic function over 10 cycles with an amplitude of OArad.

[0047] Upon modeling (1002) the ECG, another ECG is acquired (1003) from the patient from inside of the bore of the MRI scanner. An example of ECG data that has been acquired from inside of the bore is shown in FIG. 4. As shown in FIG. 4, the ECG data exhibits contamination from MHD effects.

[0048] Thereafter, an MHD effect is modeled ( 1004). For example the MHD effect can be modeled as a pseudo periodical signal with periodicity being the same as the ECG signal, though its periodicity need not be the same as the ECG signal. In an aspect the model of the MHD effect can be computed by solving the Navier Stokes equations for different sections of the aortic arch (which has been shown to be the main contributor of the artifact) and by propagating the resulting current source to the torso. For example, the MHD effect can be modeled by using 4D blood flow measurements from a phase contrast MRI technique. The blood flow can be measured in sections of the aortic arch (such as 4 sections). 3D models of the aortic arch and the human torso can be used to project the current source onto the torso for computing the biopotential on the torso at the different electrode positions. See, e.g., J. Oster, R. Linares, Z. Tse, E. J. Schmidt and G. D. Clifford, "Realistic MHD modeling based on MRI blood flow measurements." In Proceedings of the annual meeting of the Int. Soc. for Magn. Res. in Med., 2012. See also, for example, the poster presentation, Appendix C hereto. In various embodiments, the MHD periodicity can be approximated by fitting model parameters, for example a sum of Gaussians. The number of Gaussians can vary between the subjects as the effect is strongly influenced by blood flow characteristics. For example, the blood flow can be more or less laminar, and the presence of vortices induces the presence of high frequencies on the MHD effect and increases the number of Gaussians required for its approximation.

[0049] It should be appreciated that, for other embodiments, the MHD effect can also be modeled in accordance with the methods set forth in Appendix B hereto, "Comparison of three artificial models of the MHD Effect on the Electrocardiogram." These methods also include polynomial functions, spline functions or functions like

being the time of the k^th R peaks and T_k < t < T_k+i and A B being the parameters of this model.

[0050] The out-of-bore ECG acquisition (1001), the ECG modeling (1002), the in- bore ECG acquisition (1003), and the MHD modeling (1004) can be seen as the initialization process, which is performed at the beginning of an MRI examination.

[0051] After the initialization process, the patient's ECG signal can be filtered. For example, the patient's ECG signal can be filtered using Bayesian filtering (1005) or other adaptive processes. Exemplary Bayesian filtering processes include the Kalman Filter (such as an Extended Kalman Filter), the particle filter and Gaussian processes. One embodiment of a Bayesian filtering (1005) process is shown in FIG. 2.

Bayesian filtering aims at recursively estimating a set of hidden variables, x, given a sequence of noisy observation,^. The observations are related to the hidden state (the artifact to be removed) by a supposedly known observation equation and the evolution state can be determined by the evolution equation, described in more detail below. [0052] As shown in FIGs. 2A and 2B, one embodiment of a Bayesian filtering (1005) process begins with the inputting (2000) of data. This data includes patient ECG acquired (2005) during an MRI scan, the modeled (1002) ECG signal, and the modeled (1004) MHD signal. Given these inputs (2000), an estimated ECG+MHD signal is calculated (2001) from the acquired (2005) ECG, using the modeled (1002) ECG signal and the modeled (1004) MHD signals, respectively. The resulting output (2002) from the calculated (2001) ECG+MHD signals is then separated (2003) into an estimated ECG signal and an estimated MHD signal, which are subsequently outputted (2004). Results from such a Bayesian filtering (1005) process are shown in FIGS. 5 and 6, for field strengths of 1.5T and 3.0T, respectively.

[0053] In a non-limiting aspect, the ECG signal can be first recorded during

installation of the patient, that is before entering the MRI bore. A number of cycles, for example 10 ECG cycles, can be recorded outside of the MRI bore and used in order to compute an ECG template and initialize Gaussian parameters by, for example, computing the Gaussian parameters of the mean ECG cycle. See, e.g., R. Sameni, M. B. Shamsollahi, C. Jutten, and G. D. Clifford, "A Nonlinear Bayesian Filtering Framework for ECG Denoising." IEEE Trans. Biomed. Eng., vol. 54, pp. 2172-2185, 2007. Once the subject is inserted inside of the MRI bore, a number of cycles, for example 10 cycles, can be extracted and used to compute a mean template. This template corresponds to the sum of the ECG and MHD. An MHD template can be estimated by subtracting the ECG template. The MHD Gaussian parameters can then be initialized in the same way as for the ECG. A Bayesian filter, for example an Extended Kalman Filter (EKF), can then be applied for separating an estimated ECG and an estimated MHD and denoising. See, e.g., O. Sayadi, M Shamsollahi and G. Clifford, "Robust detection of premature ventricular contractions using a wave-based Bayesian framework." IEEE Trans. Biomed. Eng., vol. 57, no. 2, pp. 353-362, 2010.

[0054] By tracking the estimated ECG signal and the estimated MHD signal over time, abnormalities in the ECG, such as ischemia, will appear as deviations from the estimated signals. It should also be noted that cardiac output, ejection time, and ejection fraction, which are all useful clinical parameters, can be estimated with the resolution of the MHD. An example of simulating MHD effect from blood flow measurements is shown in Appendix B hereto, "Comparison of three artificial models of the MHD Effect on the Electrocardiogram."

[0055] During an MRI acquisition, the MHD effect overlaps the ECG signal in the frequency domain. In an exemplary embodiment, the Kalman Filter based technique can be used to separate MHD and ECG dynamics, since they are mathematically modeled. This approach takes into consideration that both ECG and MHD are non- stationary. Since the MHD contribution occurs simultaneously with the clinically important ECG waves, the separation of the dynamics is difficult, if its dynamics are not temporally separated. In an embodiment, to overcome this difficulty, a new observation equation is introduced in which a synthetic signal is created by subtracting from the raw ECG observation the prior information on the ECG signal (e.g., the sum of Gaussians of the ECG model given the synthetic phase signal). This new observation gives an approximation of the MHD effect.

[0056] The evolution equations can be written as follows:

The observation equations are given by:

These evolution equations account for the evolution of the ECG signal over time by estimating the parameters of the Gaussians that represent the ECG signal. The evolution equations, Eq. (2), and the observation equations, Eq. (3), are explained in greater detail in Appendix A hereto, "MagnetoHydroDynamic Effect separation from ECG acquired during MRI by using nonlinear Bayesian filtering," which shows experimental data using the methods described herein.

[0059] In this embodiment, for the evolution equation for the MHD contribution, the

Gaussian parameters are not included even though the signal is non-stationary, however the uncertainty in the noise allows its consideration in this model.

[0060] Bayesian filtering techniques rely on some parameter adjustments. Their initialization should be done such that they reflect properly the problem encountered. The observation noise covariance matrix of the Bayesian filter reflects the level of trust in the measurements. In the case where the measurements are biomedical signal acquisitions, which are non-stationary, the level of trust in the measurements can vary dramatically during an examination given patient motion and other external factors. In an aspect, signal quality indices (SQI's) can be used to adjust automatically the observation noise covariance matrix for optimal Bayesian filtering. As a non- limiting example, when using an Extended Kalman Filter (EKF) the observation and evolution equations can be linearized as:

with

We can then apply standard Kalman equations:

that are a function of time and that can be adjusted to an estimated noise level by use of a Signal Quality Index (SQI):

Ideally the energy of is a linear function of with

cte being a constant.

Thus when running an Extended Kalman Filter with an adaptive observation noise covariance matrix, automatic adjustment or adaption can be done with amplification signal

[0061] FIG. 7 is a flow chart that depicts an aspect of use of a Signal Quality Index to automatically adjust the observation noise covariance matrix to optimize Gaussian Filtering. An acquired biomedical signal is input (701) into the filtering scheme. A Signal Quality Index (SQI) can be computed and used to estimate a noise level (702) in the inputted biomedical signal. Meanwhile the biomedical signal can be analyzed (703) with a Gaussian Filtering technique, such as an Extended Kalman Filter (EKF). The observation noise covariance matrix of the filter can be adjusted (704) to the estimated noise level. For example, the adaption of the noise level can be done with the above amplification signal. This technique, thus, allows for optimal analysis of biomedical signals even when the conditions of recording are changing (and thus also the level of noise).

[0062] In addition, model-based Bayesian filtering techniques rely on the prior knowledge of a system. For example, in the aspect herein involving acquisition of ECG signals, this prior knowledge is the dynamics of the ECG signal represented in one aspect by a sum of Gaussians. The occurrence of pathological rhythms can be followed by a drastic change of the ECG morphology in which case the prior knowledge will not correspond to the signal acquired in the noisy environment. In an aspect a multiple model approach, during which the best-fitted model will be automatically selected, can be employed for the analysis of such biomedical signals. For example, different models can reflect a normal case, a recurrent pathologic rhythm such as Premature Ventricular Rhythm, and an unexpected pathological rhythm or any abnormality which can be represented by a dummy model and can allow detection of abnormalities in the signal. In an aspect, a switching Kalman Filter can be used to allow estimation of the different model parameters and the selection of the best model in parallel. The best model can then be applied in the filtering step.

In an exemplary embodiment, in an aspect of acquiring ECG signals, the ECG signal acquired (1001) outside of the noisy environment can be modeled (1002) not just for a normal case, but also for example for Premature Ventricular Contraction (PVC) and for an X factor mode, i.e. and unknown beat (whether artifacts or unknown pathological beat). Thus, in this embodiment, the acquired ECG signal (1001) is modeled for not just one mode but for the three different modes. The types of modes and the number of modes modeled are not limited to these particular three modes. Instead any number and types of modes can be modeled. For each new ECG signal acquired inside the noisy environment the ECG is filtered. In the aspect of a Kalman filter, for each new ECG sample acquired from the noisy environment the Kalman filter is computed for each of the three modes and the likelihood is then used to choose which of the three modes is best suited to the observation. The switching Kalman filter is then allowed to switch to the most likely mode and filter (1005) the ECG signal using the most appropriate prior knowledge (i.e., the modeled signal acquired outside of the noisy environment of the most likely of the three modes). An example of an application of a switching Kalman Filter is described in greater detail in Appendix D hereto, "Tracking arrhythmias in the ECG using a switching Kalman filter."

[0064] The processes of FIGS. 1 and 2, and their component steps, may be

implemented in hardware, software, firmware, or a combination thereof. In the preferred embodiment(s), these processes are implemented in software or firmware that is stored in a memory and that is executed by a suitable instruction execution system. If implemented in hardware, as in an alternative embodiment, these processes can be implemented with any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit(s) having logic gates for implementing logic functions upon data signals, an application specific integrated circuit (ASIC) having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), etc.

[0065] Any process descriptions or blocks in flow charts should be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps in the process, and alternate implementations are included within the scope of the preferred embodiment of the present disclosure in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present disclosure. The processes of FIGS. 1 and 2 may be implemented as a computer program, which comprises an ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this document, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium. More specific examples (a nonexhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM) (electronic), a readonly memory (ROM) (electronic), an erasable programmable read-only memory (EPROM or Flash memory) (electronic), an optical fiber (optical), and a portable compact disc read-only memory (CDROM) (optical). Note that the computer- readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory. [0067] Although exemplary embodiments have been shown and described, it will be clear to those of ordinary skill in the art that a number of changes, modifications, or alterations to the disclosure as described may be made. For example, while one specific embodiment of removing MRI-related artifacts from ECG signals is described in detail, this disclosure can more broadly be seen as teaching systems and methods for reducing periodic artifacts from biomedical signals, and even more broadly as removing noise on any oscillatory signal for which an approximate model can be constructed, whether as sums of Gaussians or otherwise. Thus, for example, the disclosed systems and methods can be used to remove ballistocardiogram effects from electroencephalogram (EEG) acquisitions during functional MRI (fMRI), or any other bioelectric signals acquired during MRI (e.g., electrooculogram (EOG), electromyogram (EMG), fetal ECG, etc).

[0068] All such changes, modifications, and alterations should therefore be seen as within the scope of the disclosure.

Claims

What is claimed:

1. A method for reducing an artifact from a biomedical signal acquired in a noisy

environment comprising:

acquiring a first biomedical signal apart from the noisy environment;

modeling the first biomedical signal;

acquiring a second biomedical signal of the same type but acquired in the noisy environment, the second biomedical signal including the artifact;

modeling the second biomedical signal; and

filtering the biomedical signal by determining an estimated signal from a combination of the modeled first and second biomedical signals and then separating the estimated combination of the first and second biomedical signals into an estimated first biomedical signal and an estimated second biomedical signal.

2. The method according to claim 1, wherein the biomedical signal is selected from one or more of an electrocardiogram (ECG), electroencephalogram (EEG),

electromyogram (EMG), fetal ECG, blood pressure waveform or

photoplethysmograph (PPG).

3. The method according to claim 1 or 2, wherein the noisy environment is a magnetic resonance imaging (MRJ) scanner and the first biomedical signal is acquired outside of the bore of the MRI scanner and the second biomedical signal is acquired inside the bore of the MRI scanner.

4. The method according to any of claims 1-3, wherein the first and second biomedical signals are each modeled as a pseudo periodic signal.

5. The method according to any of the foregoing claims, wherein the first and second biomedical signals are modeled by fitting Gaussian parameters to the biomedical signals.

6. The method according to any of the foregoing claims, wherein the second biomedical signal is modeled as a pseudo periodical signal with a periodicity being the same as the first biomedical signal.

7. The method of claim 5, wherein the periodicity of the second biomedical signal is approximated by a sum of Gaussians.

8. The method of claim 1 , wherein the second biomedical signal is modeled by solving Navier Stokes equations for different sections of an aortic arch of a subject and by propagating the resulting current source to the torso of the subject.

9. The method of claim 1, wherein the second biomedical signal is modeled in

accordance with a polynomial function, a spline function or a function like f(t) = A exp(-B(t-Tk)sin(2 pi (t-Tk)/(Tk+l - Tk))), Tk being the time of the kth R peaks and Tk < t < Tk+1 and A B being the parameters of this model.

10. The method of any of the foregoing claims, further including the step of tracking the estimated first and second biomedical signals for abnormalities in the biomedical signal appearing as deviations from the estimated signals.

1 1. The method of any of the foregoing claims, wherein the filtering involves non-linear Bayesian filtering.

12. The method of any of the foregoing claims, wherein the filtering includes Kalman filtering, particle filtering or Gaussian processes

13. The method of any of the foregoing claims, wherein the filtering step includes use of the following observation equations that provide an estimation of a periodic artifact to be reduced or separated from the biomedical signal:

14. The method of any of the foregoing claims: wherein the artifact is a magnetohydrodynamic (MHD) effect from an electrocardiogram (ECG) signal acquiring during a magnetic resonance imaging (MRI) technique; and wherein the first biomedical signal acquired is an ECG, the second biomedical signal acquired is also an ECG, the modeling of the second biomedical signal includes modeling for MHD effect, and the filtering includes determining an estimated signal from the combination of the modeled first ECG signal and the modeled second ECG signal modeled for MHD effect and then separating the estimated combination of the first and second ECG signals into an estimated ECG signal and an estimated MHD signal.

15. The method of any of the foregoing claims, wherein the filtering involves use of signal quality indices to automatically adjust the observation covariance noise matrix for Bayesian filtering.

16. The method of any of the foregoing claims, wherein the filtering involves use of a multiple model approach and a switching Kalman Filter to allow estimation of the different parameters for the models and selection of the best model.

17. The method of any of the foregoing claims further including outputting the filtered biomedical signal, and wherein the method is a computer implemented method.

18. A computer readable medium executable on a computer for at least one of filtering, segmenting and classifying a biomedical signal including an artifact acquired in a noisy environment, the computer readable medium executing the steps of:

acquiring a first biomedical signal apart from the noisy environment;

modeling the first biomedical signal;

modeling the second biomedical signal; and

19. A system for reducing an artifact from a biomedical signal acquired in a noisy

environment, comprising:

a data acquisition system configured to acquire a first biomedical signal apart from the noisy environment, and to acquire a second biomedical signal of the same type but acquired in the noisy environment, the second biomedical signal including the artifact; and

a processing system coupled to the data acquisition system, the processing system being configured to model the first biomedical signal,to model the second biomedical signal, and to filter the biomedical signal by determining an estimated signal from a combination of the modeled first and second biomedical signals and then separating the estimated combination of the first and second biomedical signals into an estimated first biomedical signal and an estimated second biomedical signal.

20. The system of claim 19, the processing system comprising:

a local interface; and

a processor, memory, a user interface, and an I/O device, each coupled to the local interface.

21. The system of claim 19 or 20, the processing system comprising a mobile application for a mobile device.

22. The system of claim any one of claims 19-21, the data acquisition system and the processing system being integrated into a single device, or residing on separate devices.

23. The system of any of claims 19-22, wherein the system is a computer system.