Method and system for quantifying and removing asynchronous noise in biophysical signals

Abstract

The exemplary methods and systems described herein facilitate quantification and removal of asynchronous noise, such as muscle artifact noise contamination, to more accurately assess complex nonlinear variability in quasi-periodic biophysical signal systems, such as acquired cardiac and brain signals.

Description

[0001] Related Applications

[0002] This International PCT Application claims priority to U.S. Provisional Application No. 62 / 686,245, filed on June 18, 2018, entitled “METHODS AND SYSTEMS TO QUANTIFY AND REMOVE ASYNCHRONOUS NOISE IN BIOPHYSICAL SIGNALS,” which is incorporated herein by reference in its entirety. Technical Field

[0003] The present disclosure generally relates to non-invasive methods and systems for characterizing cardiovascular circulation and other physiological systems. More specifically, on the one hand, the present disclosure relates to filtering asynchronous noise from collected biophysical signals (e.g., cardiac signals, brain signals, etc.). On the other hand, the present disclosure relates to quality assessment of collected signals and gating of collected signals for analysis. On the other hand, the present disclosure relates to normalizing a first set of data sets collected by a first set of biophysical signal measuring devices, and normalizing a second set of data sets collected by a second set of biophysical signal measuring devices, so that the first set of data sets and the second set of data sets can be analyzed together in a machine learning operation. Background Art

[0004] Ischemic heart disease (also called cardiac ischemia or myocardial ischemia) is a disease or group of diseases characterized by a reduced blood supply to the heart muscle, usually due to coronary artery disease (CAD). CAD typically develops when the inner walls of the coronary arteries that supply blood to the myocardium, or heart muscle, develop atherosclerosis (hardening or stiffening of the inner wall and a buildup of plaque therein, often accompanied by abnormal inflammation). Over time, CAD can also weaken the heart muscle and lead to, for example, angina, myocardial infarction (heart attack), heart failure, and arrhythmias. An arrhythmia is an abnormal heart rhythm that can include any change in the normal sequence of electrical conduction in the heart, which in some cases can lead to cardiac arrest.

[0005] The assessment of CAD can be complex, and many techniques and tools are used to evaluate the presence and severity of the condition. In terms of electrocardiography, a field of cardiology in which the electrical activity of the heart is analyzed to obtain information about its structure and function, severe ischemic heart disease can alter the ventricular conductivity of the myocardium in the perfusion bed downstream of coronary artery stenosis or occlusion. This pathology can express itself at different locations in the heart and at different stages of severity, making accurate diagnosis challenging. In addition, the conductive properties of the myocardium can vary from person to person, and other factors (such as measurement variability associated with the placement of measurement probes and parasitic losses associated with such probes and their associated components) can also affect the captured biophysical signals during electrophysiological testing of the heart. In addition, when the conductive properties of the myocardium are captured as relatively long cardiac phase gradient signals, they can exhibit complex nonlinear variability that cannot be effectively captured by traditional modeling techniques.

[0006] It is desirable to quantify and filter asynchronous noise and artifacts in acquired biophysical signals (eg, cardiac signals, brain signals, etc.), which may facilitate more accurate assessment of pathologies and conditions. Summary of the invention

[0007] The exemplary methods and systems described herein help to quantify and / or remove asynchronous noise such as skeletal muscle artifact noise contamination to more accurately evaluate complex nonlinear changes in quasi-periodic biophysical signal systems, such as changes in acquired cardiac signals, brain signals, etc. The exemplary methods and systems described herein also help to evaluate the signal quality of acquired signals to select the acquired signals for subsequent analysis.

[0008] As used herein, the term "cardiac signal" (also referred to as a signal of the heart) refers to a signal associated with the function and / or activity of the electrical conduction system of the heart (e.g., causing contraction of the myocardium), and in some embodiments includes an electrocardiogram signal, such as a signal acquired via an electrocardiogram (ECG). Quantification of the level of asynchronous noise (e.g., skeletal muscle-related signal contamination and muscle artifact noise contamination and other asynchronous noise contamination in the acquired signal) can then be used to automatically exclude such asynchronous noise from the measurement of a biophysical signal such as a cardiac signal, for which the presence of such asynchronous noise may negatively impact subsequent analysis of the cardiac signal and / or biophysical signal and / or clinical prediction / estimation of a disease state (which evaluates various periodic features of quasi-periodic biophysical signals).

[0009] As used herein, the term "brain signal" (also referred to herein as a neural signal) refers to a signal associated with brain function / activity, and in some embodiments includes an electroencephalogram signal, such as an electroencephalogram signal acquired via an electroencephalogram (EEG). Quantification of asynchronous noise, such as extraocular muscle noise contamination and facial muscle noise contamination, and other asynchronous noise contamination in the acquired signal can then be used to automatically or manually exclude such asynchronous noise from the measurement of biophysical signals such as brain signals, for which the presence of such asynchronous noise may negatively affect subsequent analysis of brain signals and / or biophysical signals and / or clinical prediction / estimation of disease states (assessment of their various quasi-periodic characteristics).

[0010] For purposes of this disclosure, the term "biophysical signal" is not intended to be limited to cardiac and brain signals, but rather encompasses any mammalian electrical or electrochemical signal that can be sensed, including but not limited to signals associated with the following and electrical signals generated at the cellular level anywhere in the mammalian body: central and peripheral nervous systems (e.g., electrical signals from the brain, spinal cord and / or nerves and their associated neurons), pulmonary, circulatory (e.g., blood), lymphatic, endocrine, digestive, musculoskeletal, urinary, immune, reproductive, epidermal and reproductive systems. Although this disclosure is directed to the beneficial quantification of asynchronous noise in the diagnosis and treatment of cardiac-related pathologies and conditions and / or brain-related pathologies and conditions (including, for example, coronary artery disease and pulmonary hypertension (e.g., pulmonary hypertension, pulmonary hypertension caused by left heart disease, pulmonary hypertension caused by pulmonary disease, pulmonary hypertension caused by chronic blood clots, and pulmonary hypertension caused by other diseases), as well as other cardiac-related conditions and / or diseases and / or brain-related conditions and / or diseases mentioned herein), such quantification may also be applied to the diagnosis and treatment (including drug therapy) of any pathology or condition in which the biophysical signal involves any relevant system of the mammalian body.

[0011] Skeletal muscle-related signals (e.g., characterized by electromyography (EMG)) are typically characterized as "in-band noise" relative to cardiac signals, brain signals, etc. - that is, they typically occur in the same or similar frequency range as the acquired biophysical signal. For example, for cardiac signals, the main frequency component of the generated signal is typically between about 0.5 Hz and about 80 Hz. For brain signals, the frequency component is typically between about 0.1 Hz and about 50 Hz. In addition, depending on the degree of contamination, signals related to skeletal muscle can also have the same or similar amplitude as typical cardiac-based waveforms and brain-based waveforms, etc. In fact, the similarity of skeletal muscle-related signals to cardiac signals, brain signals, etc. may cause significant problems for the automatic diagnostic analysis of biophysical signals. Therefore, quantifying the level of skeletal muscle-related contamination and other asynchronous noise in the measured biophysical signals is crucial for the following: quality assessment of acquired biophysical signals and automatic exclusion of acquired contaminated signals so that they are not used for subsequent analysis, and / or providing information for subsequent analysis to compensate for contamination.

[0012] The key observation when quantifying the level of skeletal muscle related signals in collected biophysical signals such as cardiac signals, brain signals, etc. is that skeletal muscle related signals are not synchronized with cardiac signals, brain signals, etc., because the sources of skeletal muscle related signals and biophysical signals are completely different. For example, cardiac signals are derived from the sum of action potentials of cardiac myocytes, brain signals are derived from the sum of ionic currents within neurons in the brain, and skeletal muscle related signals are derived from the sum of action potentials of primitive muscles (such as pectoralis, biceps, triceps, etc.). It is unlikely that these two sources share a deeper common source that can produce synchronization.

[0013] Therefore, skeletal muscle-related signals (and other asynchronous artifacts) can be quantified by comparing the acquired biophysical signals and cycles therein to an ideal representative biophysical signal for the patient as described herein, and the overall difference between the acquired biophysical signals and the ideal biophysical signals can be interpreted as the contribution of skeletal muscle-related signal contamination (and other asynchronous signals) in the acquired biophysical signals.

[0014] In one aspect, a method for filtering asynchronous noise (skeletal muscle artifact noise and other asynchronous noise) from an acquired biophysical signal dataset is disclosed, the method comprising: receiving a biophysical signal dataset of a subject by a processor; determining, by the processor, at least one template signal vector dataset characteristic of a representative quasi-periodic signal pattern (e.g., a representative heartbeat pattern) of the subject based on a plurality of detected quasi-periodic signal cycles detected in the received biophysical signal dataset; applying, by the processor, the at least one determined template signal vector dataset to one or more denoising vector datasets, wherein the one or more denoising vector datasets collectively have a vector length corresponding to the vector length of the portion to be filtered of the received biophysical signal dataset. , wherein i) the at least one determined template signal vector data set is applied to each detected cycle determined to be present in the portion to be filtered of the received cardiac signal data set; and ii) the length of the at least one determined template signal vector data set is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received biophysical signal data set; and a filtered biophysical signal data set of or a portion of the biophysical signal data set is generated by merging the portion to be filtered of the received biophysical signal data set and one or more generated denoising vector data sets (for example, using a window-based operation that applies a weighted average of the received biophysical signal and one or more generated denoising vectors in the frequency domain).

[0015] In some embodiments, the method also includes: receiving one or more additional biophysical signal data sets by the processor, each of the additional biophysical signal data sets being collected from the subject simultaneously with the biophysical signal data set; determining by the processor at least one template signal vector data set characteristic of a representative quasi-periodic signal pattern of the subject based on a plurality of detected heartbeat cycles detected in each of the one or more additional biophysical signal data sets received; and applying by the processor, for each of the one or more additional biophysical signal data sets received, a plurality of determined template signal vector data sets to one or more denoising vector data sets in a repetitive manner, wherein the one or more denoising vector data sets together have a vector length that is the same as the portion to be filtered of the received additional biophysical signal data set. corresponding vector length, and wherein, i) each of the plurality of determined template signal vector data sets is applied to each detected cycle determined to be present in the portion to be filtered of the received additional biophysical signal data set, and ii) the length of each of the plurality of determined template signal vector data sets is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received additional biophysical signal data set; and generating a filtered biophysical signal data set of or a portion of the biophysical signal data set by merging the portion to be filtered of the received biophysical signal data set and one or more generated denoising vector data sets (e.g., using a window-based operation that applies a weighted average of the received biophysical signal and one or more generated denoising vector data sets in the frequency domain).

[0016] In some embodiments, the step of determining the characteristics of at least one template signal vector data set of the representative quasi-periodic signal pattern includes: determining, by the processor, a plurality of signal features of different characteristics in the received biophysical signal data set or a portion thereof (e.g., the R peak of a cardiac signal); determining, by the processor, a plurality of cyclic regions (e.g., median RR intervals) between each of the plurality of determined signal features (e.g., stored in an MxN matrix, where M is the number of detected cycles and N is approximately 40% of the median RR interval); aligning, by the processor, each of the plurality of cyclic regions with each other in the same aspects of the plurality of signal features or another set of signal features located in each of the plurality of cyclic regions (e.g., for a cardiac signal, a feature may include the initialization of a Q wave, or a peak of an R wave, or a delay estimate by cross-correlation); and determining, by the processor, each point in the at least one template signal vector data set using an average operation or a median operation performed on each set of points in the plurality of cyclic regions.

[0017] In some embodiments, the received biophysical signal dataset includes a cardiac signal dataset, and wherein the multiple signal features are selected from the group consisting of: an R peak in the received cardiac signal dataset or a portion thereof, an S peak in the received cardiac signal dataset or a portion thereof, a T peak in the received cardiac signal dataset or a portion thereof, a Q peak in the received cardiac signal dataset or a portion thereof, or a P peak in the received cardiac signal dataset or a portion thereof.

[0018] In some embodiments, the received biophysical signal dataset comprises a cardiac signal dataset, wherein the plurality of signal features correspond to R-peaks in the received cardiac signal dataset or a portion thereof.

[0019] In some embodiments, the step of determining the at least one template signal vector data set characteristic of the representative quasi-periodic pattern further comprises determining, by the processor, a normalized parameter (eg, a z-score) derived from each of the plurality of cyclic regions.

[0020] In some embodiments, the step of determining the characteristics of the at least one template signal vector data set of the representative quasi-periodic signal pattern also includes: normalizing, by the processor, the value of each of the multiple cyclic areas or a parameter derived therefrom (e.g., a z-score) to a predetermined ratio (e.g., between "0" and "1" or "-1" and "1", or between a standard deviation value greater than 0 and less than 10).

[0021] In some embodiments, the step of determining the characteristics of the at least one template signal vector data set of the representative quasi-periodic signal pattern also includes: performing a clustering-based analysis (e.g., PCA+DBSCAN) on the multiple cyclic regions by the processor to determine whether there are more than one dominant cyclic form, wherein a template signal vector is determined for each determined dominant cyclic form.

[0022] In some embodiments, the plurality of loop regions include loops that are adjacent to each other.

[0023] In some embodiments, the cycles adjacent to each other partially overlap each other.

[0024] In some embodiments, the cycles adjacent to each other do not overlap with each other.

[0025] In some embodiments, the filtered biophysical signal dataset is generated by using two or more template signal vector datasets from two or more groups of cycles from the plurality of cycle regions, wherein the two or more groups of cycles from the plurality of cycle regions are adjacent to each other.

[0026] In some embodiments, the filtered biophysical signal dataset is generated in near real time while the biophysical signal (eg, heart signal, lung signal, brain signal) is being acquired.

[0027] In some embodiments, the filtered biophysical signal dataset is generated after the biophysical signals are fully acquired.

[0028] In some embodiments, the one or more denoised vector data sets are arranged as a one-dimensional vector.

[0029] In some embodiments, the one or more denoised vector data sets are arranged as an N-dimensional vector, where N corresponds to the number of detected cycles determined to be present in the portion of the received biophysical signal data set to be filtered.

[0030] In some embodiments, the step of applying the multiple determined template signal vector data sets to one or more denoising vector data sets includes: initializing, by the processor, the one or more denoising vector data sets as a one-dimensional vector, the length of which corresponds to the length of the portion of the received biophysical signal to be filtered; and copying, by the processor, the determined template signal vector in a one-dimensional vector so as to align at least one data point associated with a peak (e.g., an R peak or a cardiac signal) of the determined template signal vector with each peak (e.g., an R peak) determined in the received biophysical signal to be filtered.

[0031] In some embodiments, during the copying step, a conflicting portion of the currently copied template signal vector data set is assigned an average value relative to a corresponding portion of a previously copied template signal vector data set that conflicts with the currently copied template signal vector data set.

[0032] In some embodiments, during the copying step, a blank area in the one-dimensional vector between a currently copied template signal vector data set and a previously copied template signal vector data set stores a value interpolated between a last fill value and a next fill value around the blank area.

[0033] In some embodiments, the window-based operation includes: scaling the portion to be filtered of the received biophysical signal dataset by the processor through multiple window functions having a predetermined window length to generate a modified biophysical signal dataset; scaling one or more of the generated denoised vector datasets by the processor through the multiple window functions to generate a modified denoised vector dataset; determining the envelope of the modified denoised vector dataset by the processor (for example, using a low-pass filter); converting the envelope of the modified denoised vector dataset and the portion to be filtered of the received biophysical signal dataset to the frequency domain by the processor through an FFT operation; performing a weighted averaging operation on the FFT envelopes of the modified denoised vector dataset and the modified biophysical signal dataset using static interpolation coefficients or a set of dynamic interpolation coefficients by the processor to generate a result dataset; and converting the result dataset into a time series dataset by the processor through an inverse FFT operation as a filtered biophysical signal dataset of the biophysical signal.

[0034] In another aspect, a method is disclosed for normalizing a first set of data sets acquired by a set of first measurement devices (e.g., by removing asynchronous noise) and a second set of data sets acquired by a second set of measurement devices (e.g., configured to remove certain asynchronous noise) so that the first set of data sets are analyzable together with the second set of data sets in a machine learning operation. The method includes: receiving, by a processor, a set of biophysical signal data sets of a subject acquired by a set of first measurement devices (e.g., each device in the set of first measurement devices has similar or identical noise performance characteristics); determining, by the processor, at least one template signal vector data set characteristic of a representative quasi-periodic signal pattern of the subject based on a plurality of detected quasi-periodic signal cycles detected in the received biophysical signal data sets; applying, by the processor, a plurality of determined template signal vector data sets or vectors selected from a group thereof to one or more denoised vector data sets, wherein the one or more denoised vector data sets collectively have a vector length corresponding to a vector length of a portion of the received biophysical signal data set to be filtered, wherein i) each applied template signal vector data set is applied to each detected cycle determined to be present in the portion of the received biophysical signal data set to be filtered; and ii) the length of each applied template signal vector dataset is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received biophysical signal dataset; and generating a filtered biophysical signal dataset associated with the biophysical signal dataset or a portion thereof as a normalized dataset of the biophysical signal, wherein the filtered biophysical signal is generated by merging the portion to be filtered of the received biophysical signal and one or more generated denoising vectors (for example, using a window-based operation that applies a weighted average of the received biophysical signal and one or more generated denoising vectors in the frequency domain), wherein the normalized dataset associated with the biophysical signal acquired by the first measuring device and the second dataset acquired by the second measuring device are analyzable together as a machine learning training dataset (for example, in the case of skeletal muscle related noise / muscle artifact noise removal).

[0035] In some embodiments, the data set of received biophysical signals includes data captured from a sensor (e.g., in a smart device or a handheld medical diagnostic device) selected from the following groups: a 12-lead surface potential sensing electrode system (e.g., an electrocardiogram system), an intracardiac electrocardiogram, a dynamic electrocardiogram, a 6-lead differential surface potential sensing electrode system, a 3-lead orthogonal surface potential sensing electrode system, and a single-lead potential sensing electrode system.

[0036] In some embodiments, the received data set of biophysical signals includes broadband bio-cardiac phase gradient cardiac signal data (e.g., sampling frequency is above about 1 KHz, above about 10 KHz, above about 40 KHz, above about 80 KHz, above about 500 KHz), which is derived from bio-potential signals simultaneously captured (e.g., with a deflection of less than about 100 microseconds) from multiple surface electrodes placed on the body surface near the heart.

[0037] On the other hand, a method for excluding acquired signals is disclosed, comprising: a processor receiving a biophysical signal dataset of a subject; the processor comparing the received biophysical signal dataset with at least one template signal vector dataset characteristic of a representative quasi-periodic signal pattern in the biophysical signal dataset; and the processor excluding the received biophysical signal dataset based on the comparison result.

[0038] In some embodiments, the step of determining the characteristics of the at least one template signal vector data set of the representative quasi-periodic signal pattern includes: the processor determines a plurality of signal features of different characteristics in the received biophysical signal data set or a portion thereof (e.g., the R peak of a cardiac signal); the processor determines a plurality of cyclic regions (e.g., median RR intervals) between each of the plurality of signal features determined (e.g., stored in an MxN matrix, where M is the number of cycles detected and N is approximately 40% of the median RR interval); the processor aligns each of the plurality of cyclic regions with each other in the same aspects of the plurality of signal features or another set of signal features located in each of the plurality of cyclic regions (e.g., for a cardiac signal, the features may include the initialization of a Q wave, or the peak of an R wave, or a delay estimated by cross-correlation); and the processor determines each point in the at least one template signal vector data set using an average operation or a median operation performed on each set of points in the plurality of cyclic regions.

[0039] In some embodiments, the step of determining the at least one template signal vector data set characteristics of the representative quasi-periodic signal pattern also includes: the processor performing a clustering-based analysis (e.g., PCA+DBSCAN) on the multiple cyclic regions to determine whether there are more than one dominant cyclic form, wherein a template signal vector is determined for each determined dominant cyclic form.

[0040] In some embodiments, the method further comprises: the processor generating a notification of failure in acquiring the biophysical signal dataset, wherein the notification prompts to subsequently acquire the biophysical signal dataset.

[0041] In some embodiments, the method further comprises: the processor transmitting the received biophysical signal data set to an external analysis system via a network, wherein the analysis system is configured to analyze the received biophysical signal data to determine the presence or extent of a pathology or clinical condition.

[0042] In some embodiments, the comparing includes determining the presence of asynchronous noise having a value or energy exceeding a predetermined threshold in the acquired biophysical signal data set.

[0043] In one aspect, a method for quantifying asynchronous noise in acquired biophysical signals is disclosed, the method comprising: a processor receiving a biophysical signal dataset of a subject;

[0044] The processor determines a plurality of signal features of different characteristics in the received biophysical signal data set or a portion thereof (e.g., R peaks of cardiac signals); the processor determines a plurality of cyclic regions (e.g., median RR intervals) between each of the determined plurality of signal features (e.g., stored in an MxN matrix, where M is the number of detected cycles and N is approximately 40% of the median RR interval); the processor aligns each of the plurality of cyclic regions with each other in the same aspects of the plurality of signal features or another set of signal features located in each of the cyclic regions (e.g., for cardiac signals, features may include the initialization of a Q wave, or a peak of an R wave, or a delay estimated by cross-correlation); the processor determines each point in the at least one template signal vector data set using an average operation or a median operation performed on each set of points in the plurality of cyclic regions; and the processor performs a clustering-based analysis (e.g., PCA+DBSCAN) on the plurality of cyclic regions to determine whether there is more than one dominant cyclic morphology, wherein a template signal vector is determined for each determined dominant cyclic morphology.

[0045] A system is disclosed, comprising: one or more processors; and a memory having instructions stored thereon, wherein the instructions are executed by the one or more processors to enable the one or more processors to perform the method as described above.

[0046] A non-transitory computer-readable medium having instructions stored thereon is disclosed, wherein execution of the instructions by one or more processors causes the one or more processors to perform the method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments and, together with the description, serve to explain the principles of the methods and systems embodied herein. The patent or application file contains at least one drawing in color. Copies of this patent publication or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0048] The embodiments of the present invention may be better understood from the following detailed description when read in conjunction with the accompanying drawings. The accompanying drawings include the following figures:

[0049] Figure 1A is a diagram of an example system configured to quantify and remove asynchronous noise and artifact contamination to more accurately assess complex nonlinear changes in quasi-periodic systems, according to an illustrative embodiment.

[0050] Figure 1B is a diagram of an example system configured to exclude acquired biophysical signals based on quantification of asynchronous noise and artifact contamination according to another illustrative embodiment.

[0051] Figure 1C is a diagram of an example system configured to remove asynchronous noise and / or exclude acquired biophysical signals according to another illustrative embodiment.

[0052] Figure 2 An exemplary method of removing asynchronous noise from acquired biophysical signals (eg, acquired cardiac signals, acquired brain signals, etc.) according to an exemplary embodiment is shown.

[0053] Figure 3 is a diagram showing a method according to an exemplary embodiment. Figure 2 FIG. 1 is a diagram of an example implementation method of a process.

[0054] Figure 4 is a flow chart of an example method of looping through a representative data set, according to an illustrative embodiment.

[0055] Figure 5 is a diagram of an example method for quantifying asynchronous noise contamination in biophysical signals, according to an illustrative embodiment.

[0056] Figure 6 is a graph of representative cyclic data set characteristics of a representative quasi-periodic signal pattern (eg, a representative heartbeat pattern in an acquired cardiac signal) in accordance with an illustrative embodiment.

[0057] Figure 7 A diagram showing a method for generating a template-vector dataset (eg, in an acquired cardiac signal) according to an illustrative embodiment.

[0058] Figure 8is an example diagram of an original biophysical signal dataset, a generated biophysical template-vector dataset, and a resulting denoised biophysical signal dataset according to an illustrative embodiment.

[0059] Fig. 9 A diagram illustrating a process for segmenting a biophysical cycle (eg, a cardiac cycle) from a biophysical signal dataset (eg, a cardiac signal dataset) to quantify asynchronous noise contamination in the biophysical signal dataset, according to an illustrative embodiment.

[0060] Fig. 10A According to an exemplary embodiment, Figure 4 A graph showing the results of the normalization process.

[0061] Fig. 10B Shown is a template signal vector data set superimposed on top of a set of stacked cycles for a high noise signal.

[0062] Fig. 10C Shown is a template signal vector data set superimposed on top of a set of stacked cycles for a low noise signal.

[0063] Fig.11 An example output of principal component analysis of a generated circulant matrix is ​​shown in accordance with an illustrative embodiment.

[0064] Fig.12 is a graph of a distribution of difference scores determined based on a comparison of a representative cycle data set and each estimated cycle, according to an illustrative embodiment.

[0065] Fig.13A , 13B 13C shows a schematic diagram of a circuit diagram of a circuit according to an exemplary embodiment. Figure 1A Example broadband brain phase gradient signal dataset acquired by the measurement system.

[0066] Fig.14 shows a phase space representation according to an exemplary embodiment. Figure 13A-13C Broadband brain phase gradient signals.

[0067] Fig.15 A method for normalizing a first set of data sets collected by a first set of biophysical signal measuring devices and a second set of data sets collected by a second set of biophysical signal measuring devices according to an exemplary embodiment, so that the first set of data sets and the second set of data sets can be analyzed together in a machine learning operation.

[0068] Fig.16 is an example method of analysis by a non-invasive cardiac assessment system according to implementations of the present disclosure. DETAILED DESCRIPTION

[0069] Every feature described herein and every combination of two or more such features is included within the scope of the present invention, provided that the features included in such a combination are not mutually contradictory.

[0070] Example System

[0071] Figure 1A is a diagram of an example system 100 configured to quantify and remove asynchronous noise (e.g., skeletal muscle-related artifact noise contamination) and use such quantification to more accurately assess complex nonlinear changes in quasi-periodic systems, according to an illustrative embodiment. As used herein, the term "removal" refers to a reduction (in whole or in part) of any meaningful noise contamination that improves or facilitates subsequent analysis.

[0072] exist Figure 1A In the present invention, a non-invasive measurement system 102 acquires a plurality of biophysical signals 104 (e.g., phase gradient biopotential signals) from a subject 106 via any number of measurement probes 114 (shown in the system 100 of FIG. 1 as including six probes 114a, 114b, 114c, 114d, 114e, and 114f) to generate a phase gradient biophysical signal dataset 108, which can be used by a non-invasive biophysical signal evaluation system 110 (labeled “non-invasive biophysical signal evaluation system” 110 in FIG. 1 ) to determine a clinical output 112. In some embodiments, the clinical output includes an evaluation of the presence or absence of a disease and / or an estimated physiological characteristic of the physiological system under study.

[0073] In some embodiments, and as Figure 1A As shown, the system 102 is configured to remove asynchronous noise contamination (e.g., via processing 118) from the acquired biophysical signal data set 117 generated by the front-end amplification and digitization operation 116. The removal operation 118 is based on quantification of the asynchronous noise potentially present in the acquired signal 114. Once a representative cycle data set is established, for example, based on some samples of the acquired data set, the processing 118 of removing asynchronous noise can be performed in almost real time. In some embodiments, a few hundred samples can be used to establish a representative cycle data set. In other embodiments, a few thousand samples can be used to establish a representative cycle data set.

[0074] In some embodiments, the measurement system 102 includes a biopotential-based measurement system configured to acquire broadband biopotential biophysical signals. In the case of an electrocardiogram, the measurement system 102 is configured to capture a cardiac-related biopotential or electrophysiological signal of a mammalian subject (e.g., a human) as a broadband cardiac phase gradient signal. Examples of the measurement system 102 are described in U.S. Application Publication No. 2017 / 0119272 and U.S. Patent Application No. 15 / 248,838, which are incorporated herein by reference in their entirety.

[0075] In some embodiments, broadband biopotential biophysical signals are captured as unfiltered mammalian electrophysiological signals, so that the spectral components of the signal remain unchanged. In fact, broadband biopotential biophysical signals are acquired, converted, and even analyzed without filtering (e.g., by hardware circuits and / or digital signal processing techniques, etc.) (e.g., before digitization), otherwise the filtering may affect the phase linearity of the biophysical signal of interest. In some embodiments, broadband biopotential biophysical signals are captured with microvolt or sub-microvolt resolution, which is equal to or significantly lower than the background noise of conventional electrocardiograms and other biophysical signal acquisition instruments. In some embodiments, broadband biopotential biophysical signals are sampled simultaneously with a time skew or "lag" of less than about 1 microsecond, and in other embodiments, broadband biopotential biophysical signals are sampled simultaneously with a time skew (temporal skew) or "lag" of no more than about 10 femtoseconds. It is worth noting that the exemplary system minimizes nonlinear distortion (e.g., nonlinear distortion that can be introduced via certain filters) in the acquired broadband phase gradient signal, thereby not affecting the information therein.

[0076] As described above, the measurement system 102 may be used to capture other mammalian biopotential or electrophysiological signals, such as brain / nerve biopotential signals or other mammalian biopotential signals associated with various biological systems as described elsewhere herein.

[0077] Still reference Figure 1A , the evaluation system 110 is configured to receive (e.g., via a network) the acquired biophysical signal dataset 108 (e.g., a denoised biophysical signal dataset in some embodiments), and generate one or more three-dimensional vector electrocardiogram datasets 122 through a conversion operation 120 (labeled "phase space conversion" 120) for analyzing the phase gradient biophysical signal dataset 108 via, for example, machine learning operations and / or predictor operations (shown as step 124). Examples of conversion operations and machine learning / predictor operations are discussed below and in U.S. Application Publication No. 2013 / 0096394, the entire contents of which are incorporated herein by reference in their entirety.

[0078] In some embodiments, the measurement system 102 is configured to evaluate the signal quality of the acquired biophysical signals and to exclude acquired signal data sets based on such evaluation. Figure 1B 1 is a diagram of an example system configured to reject acquired biophysical signals based on quantification of asynchronous noise and artifact contamination according to another illustrative embodiment. In some embodiments, the measurement system 102 is configured to perform an asynchronous noise removal operation 118 and a signal quality assessment operation 130 based on quantification of asynchronous noise.

[0079] Because in some embodiments, the acquired biophysical signals 108 are clinically analyzed on a system separate from the measurement system 102 (e.g., by the evaluation system 110), the signal quality check ensures that the acquired biophysical signal dataset 108 is suitable for subsequent clinical analysis. Near real-time operation can facilitate the non-invasive measurement system 102 to prompt the re-acquisition of the biophysical signal dataset, thereby ensuring that the acquired biophysical signal dataset is not contaminated by asynchronous noise (e.g., noise associated with skeletal muscle) before it is further processed and analyzed or can be used for further processing and analysis for clinical evaluation.

[0080] In some embodiments, the signal quality assessment operation is performed in near real time, such as less than about 1 minute or less than about 5 minutes, so that the system can prompt the re-acquisition of the biophysical signal data set. The near real-time assessment allows the biophysical signal data set to be re-acquired before the patient leaves the test room where the test is performed.

[0081] In some embodiments, the non-invasive measurement system 102 is configured to generate a notification 126 (at Figure 1B The notification may be a visual output, an audio output, or a tactile output provided to a technician near the patient.

[0082] In some embodiments, the excluded biophysical signal data set may be stored (128) for further troubleshooting analysis (132) of the defect that caused the acquired signal to be excluded. To this end, the excluded biophysical signal data set is not used in subsequent analysis (e.g., 120, 124) to generate clinical output results 112.

[0083] Figure 1C is a diagram of an example system 100 configured to quantify asynchronous noise, such as skeletal muscle-related artifact noise contamination, and use such quantification to remove such contamination and / or exclude acquired biophysical signals, according to another illustrative embodiment. Figure 1C, the evaluation system 110 is shown as being configured to further pre-process 134 the received biophysical signal data set 108 by excluding the received biophysical signal data set and / or removing asynchronous noise from the received biophysical signal data set. The pre-processing operation 132 can be performed instead of the asynchronous noise removal operation 118 (as performed on the measurement system 102) and / or the signal quality evaluation operation 130 (as performed on the measurement system 102), or as an additional quality operation to the asynchronous noise removal operation 118 and / or the signal quality evaluation operation 130.

[0084] Asynchronous noise removal

[0085] Figure 2 An example method 118 for removing asynchronous noise from acquired biophysical signals is shown in accordance with an example embodiment. Figure 2 As shown, the method 118 includes the following steps: receiving (step 202) a biophysical signal data set (e.g., data set 108) of the subject 106 by a processor. As described above, the removal operation 118 can be performed by the measurement system 102 and / or the evaluation system 110. For cardiac signals, a handheld device or other device can be used to collect the patient's resting thoracic physiological signals, for example, from a set of six probes or electrodes (e.g., probes 114a-114f) arranged along three orthogonal axes corresponding to the X, Y and Z channels. The electrodes as part of the non-invasive measurement system 102 can collect the phase gradient biophysical signal data set 108 without the use of ionizing radiation, contrast agents, motion or pharmacologic stressors. In some embodiments, the non-invasive measurement system 102 samples for a duration between about 30 seconds and about 1400 seconds at about 8kHz, preferably for about 210 seconds. The collected data points are transmitted as part of the data set 108 to the evaluation system 110 and evaluated by the analysis engine therein using a machine learning algorithm / predictor.

[0086] Other conventional electrode sets and electrographic acquisition methods may be used with the methods and systems disclosed herein.

[0087] Still refer to Figure 2 , method 118 further includes determining (step 204) at least one template signal vector data set (also referred to as a "representative vector data set") characteristics of a representative heartbeat pattern of the subject based on multiple detected heartbeat cycles detected in the received cardiac signal data set (e.g., set 108), via a processor.

[0088] Still reference Figure 2, method 118 also includes applying, by a processor, the determined template signal vector dataset to one or more denoised vector datasets. In some embodiments, the template signal vector dataset is applied to each detected cycle determined to be present in the portion to be filtered of the received biophysical signal dataset (e.g., dataset 108). In some embodiments, the length of the template signal vector dataset is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received biophysical signal dataset (e.g., dataset 108). The denoised vector datasets collectively have a vector length corresponding to the vector length of the portion to be filtered of the received biophysical signal dataset (e.g., dataset 108).

[0089] Still reference Figure 2 , the method 118 also includes generating (step 208) by the processor a filtered biophysical signal dataset (also referred to as a denoised signal dataset) of the acquired biophysical signal or a portion thereof by merging a portion of the received biophysical signal dataset (e.g., dataset 108) corresponding to the portion to be filtered with the one or more generated denoised vector datasets. In some embodiments, the merging operation is performed using a window-based operation that applies a weighted average of the received biophysical signal dataset (e.g., dataset 108) and the one or more generated denoised vector datasets in the frequency domain.

[0090] In other embodiments, the merging operation is performed in the time domain.

[0091] Figure 3 is a diagram showing a method according to an exemplary embodiment. Figure 2 118 is a diagram of an example implementation method 118 for processing. Method 118 includes creating (step 302) one or more representative cyclic data sets, each of which characterizes a representative quasi-periodic signal pattern of a subject, by a processor based on multiple detected quasi-periodic signal cycles detected in the received (one or more) biophysical signals (or data sets related thereto). For cardiac signals, the representative quasi-periodic signal pattern can be characterized as a representative heartbeat pattern. As used herein, the term quasi-periodic generally refers to a characteristic of a signal system that cycles with at least two frequency components, the ratio of which is not a rational number. The representative cyclic data set is also referred to herein as a template signal vector data set. Figure 4 is a flow chart of an example method of looping through a representative data set in accordance with an illustrative embodiment. Figure 6 is a graph of a representative cyclic data set 602 (shown as 602a, 602b) characteristic of a representative quasi-periodic pattern (e.g., a representative heartbeat pattern in an acquired cardiac signal). Figure 4 and Figure 6 .

[0092] Still reference Figure 3 , steps 304 to 312 describe an example for applying, by a processor, a template signal vector data set (e.g., a representative cycle data set) to the one or more noise reduction vector data sets (e.g., a template signal vector data set) to generate a denoised signal data set. Figure 3 As shown, step 304 includes initializing a "template vector" data set, whose length corresponds to the length of the input raw signal data set. That is, the length of the initialized template vector data set is the same as the length of the input raw signal data set. For example, the raw signal data set collected at 8KHz over a period of 210 seconds generates a template vector data set with a length of 1,680,000 samples for each collected data channel via step 304.

[0093] Step 306 includes populating, by a processor, the template vector dataset with (one or more) representative vector datasets. That is, in some embodiments, for each detected cycle in the original signal dataset, the method 118 includes placing or copying the representative vector dataset 602 in the template vector dataset. Each representative vector dataset 602 is placed so that the determined peak (e.g., R-peak) of the representative vector dataset 602 is aligned with the corresponding peak (e.g., R-peak) of each detected cycle at the same or similar time index.

[0094] Figure 7 A diagram 700 is shown of a method for generating a template vector dataset by a processor according to an exemplary embodiment. Figure 7 , graph 700 includes: i) a graph of a template vector dataset 702 populated with representative vector datasets 602 (shown as 704a, 704b, 704c, 704d, 704e, and 704f), and ii) a graph of a received biophysical signal dataset (e.g., dataset 108) for a given measurement channel, with detected cycles identified. Figure 7 As shown, a representative vector dataset (e.g., 704a-704f) is placed / copied in the template vector dataset 702 so that the peaks of the representative vector dataset (e.g., 704a-704f) (e.g., the maximum peak corresponding to the R peak (each peak shown as 706a-706f) in the acquired cardiac signal) are aligned with the peaks of the biophysical signal dataset 108.

[0095] There may be more than one representative vector data set, each representative vector data set corresponding to an evaluated quasi-periodic signal pattern (e.g., a heartbeat pattern of a cardiac signal). When there is more than one representative cycle data set, in some embodiments, method 118 also includes placing the representative cycle data set selected to correspond to (i.e., more closely match) a given current cycle in the original signal data set.

[0096] Still reference Figure 7 When placing or reproducing (reproducing, copying) the representative cycle data set 602 in the template vector data set 702, if the placed representative cycle data set conflicts with the existing data in the template vector data set, the overlapping part of the existing data sample (in Figure 7 If there are gaps in the template vector data set 702 (e.g., shown as 710a, 710b), the gaps (e.g., 710a, 710b) may be filled in the template vector data set by the processor, where the data values ​​are determined, for example, by an optional interpolation operation (e.g., between the last fill value and the next fill value around the blank area), and the overlapped portions of the new data samples are averaged. Figure 3 This is shown as step 308.

[0097] Once the template vector dataset 702 has been created, the method 118 includes initiating a denoising process via the processor ( Figure 3 In some embodiments, step 310 includes selecting a window size of a window on which the denoising operation is performed and values ​​of interpolation coefficients to control the effect of the template vector dataset on the original signal dataset.

[0098] In some embodiments, each window has a window size of about 0.25 seconds. In other embodiments, a window size of less than about 0.25 seconds is used. In other embodiments, a window size of greater than about 0.25 seconds is used.

[0099] In some embodiments, the static value of the interpolation coefficient is about 0.75 (i.e., the effect attributable to the template vector data set is about three times the effect of the original signal data set). In other embodiments, the value of the interpolation coefficient is less than about 0.75. In other embodiments, the value of the interpolation coefficient is greater than about 0.75. The values ​​of the interpolation coefficient and the window size can be evaluated based on the need to eliminate noise relative to maintaining signal variability. In some embodiments, the window size or interpolation coefficient is allowed to change dynamically, for example, based on an evaluation of the signal, for example, relative to the level of automatically quantified contamination (e.g., contamination related to bone-muscle) in the signal.

[0100] Still reference Figure 3, step 310 includes initializing a blank / empty vector dataset that will eventually become the denoised signal dataset 114. In some embodiments, the number of samples of the initialized denoised signal vector dataset is the same as the number of samples of the original signal dataset 114 of interest, i.e., the same number of samples of the original signal 114 that will be used in the subsequent machine learning analysis.

[0101] Still reference Figure 3 , step 312 includes creating, by the processor, a window function for each window on which the noise reduction operation is performed. In some embodiments, for each window, step 312 includes creating a Hamming window data set centered on the middle portion of the window. When the Hamming window data set is not placed at the exact middle sample of the signal data set, the Hamming window data set is placed asymmetrically relative to the entire signal data set, so that samples with equal distances from the middle position of the window have equal values ​​in the Hamming window data set. In some embodiments, the window function enhances the ability of the FFT operation to extract spectral data from the signal by reducing the effects of leakage that may occur during the FFT operation of the data. In any case, the window function can attenuate or remove high-frequency components that are caused by discontinuities when using the window for analysis and data discretization. Other types of window functions can be applied, including, for example, Hann windows, Blackman windows, Harris windows, sine windows, Nuttall windows, triangular windows, and combinations thereof. Step 312 also includes multiplying each of the original signal data set 108 and the filled template vector data set 702 with a window function to generate a modified original signal data set and a modified template vector data set.

[0102] Still refer to Figure 3 , method 118 includes calculating (step 314) the envelope of the modified template vector data set by the processor by performing a low-pass filtering operation on the data set. In some embodiments, a Butterworth filter is used. In some embodiments, the Butterworth filter is a 5th order filter with a normalized cutoff frequency of 0.025. In some embodiments, a Chebyshev filter is used.

[0103] Step 316 includes the processor performing a fast Fourier transform (eg, discrete FFT) on each of the modified template vector envelope data set and the modified original signal data set to transform each of them into the frequency domain.

[0104] Step 318 includes merging, by the processor, the modified template vector envelope data set and the modified original signal data set in the frequency domain. In some embodiments, a weighted average operation of the modified template vector envelope data set and the modified original signal data set in the frequency domain is performed. In some embodiments, the weights used in the interpolation are interpolation coefficients that are initially set to control the effect of the template vector on the original signal.

[0105] Step 320 includes performing an inverse Fourier transform operation by the processor to transform the resulting data back to the time domain. The resulting data is assigned as the current window of the noise reduction signal data set. The process is repeated for all windows or a portion thereof to fill the rest of the noise reduction signal data set.

[0106] Figure 8 is an example diagram of an original biophysical signal dataset 108 (e.g., shown as “Original Signal” 108 ), a generated template vector dataset 702 (e.g., shown as “Template Vector” 702 ), and a resulting denoised signal dataset 802 (e.g., shown as “Noised Signal”).

[0107] like Figure 8 As shown, the original signal dataset 108 is heavily contaminated by noise (e.g., noise associated with skeletal muscles). In some embodiments, application of the template vector dataset 702 completely removes the noise while retaining some high-frequency information in the QRS waveform (e.g., shown at notch 808 occurring around time index 57.8 seconds), but the template vector dataset 702 does not include most of the variability (i.e., cardiac signal variability) present in the original signal dataset. The denoised signal dataset 802 includes variability (i.e., cardiac signal variability) information as included in the original signal dataset 108.

[0108] In practice, the method described herein comprises generating a filtered cardiac signal (i.e., a denoised signal) or a portion thereof of a cardiac signal by combining the portion of the received biophysical signal to be filtered (e.g., as a modified original signal) and the one or more generated denoising vectors (e.g., as a modified template vector envelope).

[0109] Determining representative cycles of quasi-periodic signal patterns

[0110] As mentioned above, Figure 4 4 is a diagram of a method 400 for determining a template signal vector data set of a representative cycle of a quasi-periodic signal pattern (e.g., a representative cycle of a heartbeat pattern). The method 400 can be part of the evaluation or quantification of skeletal muscle related artifacts and noise contamination in a biophysical signal (e.g., a cardiac signal, a brain signal, etc.).

[0111] like Figure 4As shown, method 400 includes: first, detecting (step 402) a peak (e.g., an R peak corresponding to the maximum depolarization of a cardiac signal) of the entire biophysical signal data set (e.g., data set 108) or a portion thereof by a processor. In some embodiments, a Pan-Tompkins algorithm is used to detect the peak, for example, "A Real Time QRS Detection Algorithm" disclosed by Pan & Tompkins in IEEE Transactions on Biomedical Engineering, Vol. 32-3, pp. 230-236, 1985, which is incorporated herein by reference in its entirety. In other embodiments, other algorithms may be used to detect peaks in a cardiac signal data set. Examples include “Hilbert transform based adaptive ECG R-peak detection technique” described by Makwana et al. in International Journal of Electrical and Computer Engineering, Vol. 2, No. 5, p. 639 (2012), “Smart ECG Monitoring Patch with Built-in R-Peak Detection for Long-Term HRV Analysis” described by Lee et al. in Annals of Biomedical Engineering, 44(7), 2292-3201 (2016); and “Detection of R-Peaks in ECG Signal by Adaptive Linear Neuron (ADALINE)” proposed by Kim et al. in MATEC Web of Conferences, 54, 10001 (2016), each of which is incorporated herein by reference in its entirety.

[0112] In some embodiments, the system is configured to evaluate the number of cycles and cycle boundaries in a biophysical signal dataset (e.g., dataset 108), and then a subset of cycles in the determined grouping of adjacent cycles is used to determine a template signal vector dataset of representative cycles. In some embodiments, the system is configured to evaluate the boundaries of cycles in the entire biophysical signal dataset or in a portion desired to be analyzed. In other embodiments, the system is configured to evaluate the boundaries of cycles for a predefined number of adjacent cycles in a portion of interest of a biophysical signal dataset (e.g., dataset 108).

[0113] In some embodiments, the adjacent / neighboring cycle may be defined as ±1, 2, .. 10 cycles around the middle cycle of a set of determined cycles. In other embodiments, the adjacent / neighboring cycle may be defined as +1, 2, .. 20 cycles relative to the starting cycle of a set of determined cycles. In other embodiments, the adjacent / neighboring cycle may be defined as -1, 2, .. 20 cycles relative to the last cycle in a set of determined cycles.

[0114] To this end, multiple template signal vector data sets can be generated, and each template signal vector data set can be applied separately to the loop used to generate it. For example, in the case where neighborhood grouping 1 consists of loops 1...10 and results in template vector #1, only the analysis of loops 1-10 is evaluated for template vector #1 (e.g., as described herein); in the case where neighborhood grouping 2 consists of loops 5...15 (e.g., has some loops and results in template vector #2), the analysis of loops 5-10 is evaluated for template vector #1 and template vector #2 (e.g., by the average value of vectors #1 and #2); in the case where neighborhood grouping 3 consists of loops 10...20 (e.g., has some loops and results in template vector #2), the analysis of loops 10-15 is evaluated for template vector #2 and template vector #3 (e.g., as described herein) (e.g., by the average value of vectors #2 and #3).

[0115] Indeed, in some embodiments, analysis is performed until all cardiac cycles (eg, 3.5 minutes of PSR recording x 60 BPM = 210) or the portion of interest of the biophysical signal dataset has been evaluated.

[0116] This neighborhood approach can reduce sensitivity to long-term variations by incorporating only local cycles into the template, but may also reduce robustness to noise because fewer component cycles are used or analyzed. In fact, using the full recording can capture natural cardiac variations in the entire data set, but can also produce non-noise-based deviations between the template and the test cardiac cycle. By using all signals in the neighborhood (but in adjacent groupings), all signals (and the inherent variations of the acquired signals) can still be evaluated and sensitivity can be improved locally.

[0117] The number of neighborhood sizes may be 10, for example, as described above, or it may be a user defined parameter. In some embodiments, the neighborhood size is determined based on some evaluated change in the signal. In practice, the number of neighborhood sizes may be 2, 3, 4, 5, 6, 7, 8, 9, 10. In some embodiments, the number of neighborhood sizes may be greater than 10, for example, between 10 and 15. In some embodiments, the number of neighborhood sizes may be greater than 15, for example, between 15-25. In some embodiments, the number of neighborhood sizes may be greater than 25, for example, between 25-50.

[0118] In some embodiments, neighborhoods or groupings of loops are defined by offset sizes. In some embodiments, the offset size is the distance in index count from a reference point in one loop to the next loop. The reference point can be a middle point, a starting point, or an end point in the loop. In the example above, where loops are defined from 1..10, 5..15, 10..20, etc., the offset size is 5 (each reference point is at the starting point, middle, or end point).

[0119] In some embodiments, depending on the offset size and neighborhood size, each given loop may have one or more template signal vector data sets compared to it to determine metrics (e.g., mean, median, mode, and other metrics discussed) for the template signal vector data set. The metrics may then be combined to provide a modified score for the template signal vector data set.

[0120] For example, a template signal vector data set may be defined as the preceding and following neighboring points of a reference point defined at the middle of a given defined cycle. A template signal vector data set may be generated (e.g., based on a mean, mode, median, etc.) based solely on the preceding and following neighboring points (rather than based on a reference point defined at the middle of the cycle). Once the template signal vector data set is generated, the template signal vector data set is compared to the middle of the given defined cycle to determine a score for the score.

[0121] In some embodiments, the analysis can be iterative, with scores from subsequent loops being combined. For example, in loop #1, score #1 is determined for loop #2. Then, for loop #2, score #2 is determined based on a local score determined only from loop #2, which is then combined with the score from loop #1. Then, for loop #3, score #3 is determined based on a local score determined only from loop #3, which is then combined with the scores from loops #1 and #2. This iterative analysis can be applied to all or part of the input data set of interest.

[0122] In fact, the system can choose to apply the template only to a single loop at, for example, the exact center of the neighborhood. As discussed, each single loop is then evaluated against a single template, and that single template is unique among all possible templates. This type of analysis provides a different advantage of looking at the local effects of a loop.

[0123] Method 400 includes using (step 404) the detected peak positions by a processor to determine a median peak-to-peak interval (e.g., a median RR peak of a cardiac signal) and setting a cyclic region around each peak (e.g., an R peak of a cardiac signal). For a cardiac signal, the cyclic region is set around the R peak and includes both the P wave and the complete T wave. Fig. 9 A diagram showing the process of a processor segmenting a biophysical signal cycle from a biophysical signal data set to quantify skeletal muscle related noise contamination in the biophysical signal data set. Fig. 9 As shown, the detected peak positions (e.g., shown as 902a-902g) are used to determine the median peak-to-peak interval (e.g., the median RR peak of the cardiac signal shown as 904a-904g) and set a cyclic region (e.g., shown as 906a-906f) around each peak (e.g., the R-peak value of the cardiac signal shown as 908a-908g). Fig. 9 It is further shown that the cycle region is set around the R peak and includes the P wave (e.g., shown as 910a-910g) and the completion of the T wave (e.g., shown as 912a-912g) of the cardiac signal. In some embodiments, the range is about -20% to about +20% of the median interval (e.g., shown as 912a, 912b). The processor can store each cycle region (e.g., 906a-906f) in a matrix (also referred to as a "circular matrix"). The circular matrix can be MxN, where M is the number of cycles detected and N is 40% of the median peak-to-peak interval (e.g., the median RR interval of the cardiac signal), where 40% of the peak-to-peak interval represents the entire time "width" of the cycle. Specifically, once the median peak-to-peak interval (e.g., the median RR interval of the cardiac signal) is known in the entire data set, the signal can be divided into two halves, for example, to obtain the "20%" before and after the peak (e.g., R peak) in time to capture other waves (e.g., T and P waves of the cardiac signal). Of course, other loop region lengths can also be used for various different waves in heart signals and brain signals, etc.

[0124] Reference Figure 4 , method 400 includes normalizing (step 406 ), by the processor, each cycle to remove any offset. Fig. 10A The present invention is shown according to an exemplary embodiment. Figure 4 The result of normalization processing is shown in Figure 1. Fig. 10AIn the embodiment, the processor normalizes each cycle region (e.g., 906a-906f) of the biophysical signal data set (e.g., data set 108) to remove any offset so that the average value of each cycle region is zero. As shown, the normalized cardiac signal data set can have a range of "1" to "-1", but the range can vary depending on the distribution of the data.

[0125] In some embodiments, the centering operation includes an operation to time-align the same features (e.g., peaks) in multiple waveforms. Examples of these features include the onset of the Q wave for cardiac signals, the peak of the R wave, or a delay estimate determined by a cross-correlation operation, etc. In some embodiments, the amplitude normalization operation uses features of the QRS waveform as a basis for determining the gain term (e.g., a short average can be taken before the QRS).

[0126] In other embodiments, each cycle is normalized according to a z-score. The z-score value for a given data point in the template signal vector data set can be calculated as the difference between the value of the given data point and the mean value of a set of cycles, and then the difference is normalized by the standard deviation between the given data point and the same index data value of the set of cycles. In some embodiments, the z-score can be output as a cycle variability score. Cycle variability can refer to the degree of variability between cycles in the collected biophysical data set, which can be attributed to asynchronous noise, etc.

[0127] Still reference Figure 4 , the method 400 further includes the processor performing a principal component analysis (PCA) on the generated circulant matrix to extract the first two principal components. Fig.11 An example output of a principal component analysis operation performed on the generated circulant matrix is ​​shown.

[0128] Still reference Figure 4, method 400 also includes performing, by the processor, a clustering operation on the output of the principal component analysis (step 410). Examples of clustering operations that can be used include the DBSCAN algorithm described in "A density-based algorithm for discovering clustering in large spatial databases with noise," by Ester, Kriegel, Sander, and Xu, published in the Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, pages 226-231, which is incorporated herein by reference in its entirety. In some embodiments, the clustering operation is configured to be performed on the first two PCA components, which in some embodiments represent cycles in a two-dimensional space. If the algorithm detects a second dominant cluster that represents more than 10% of the total number of cycles, it indicates the presence of a second dominant cyclic morphology, such as premature ventricular contraction. Note that based on DBSCAN's identification, Fig.11 There are no multiple distinct cyclic morphologies. The dataset visually appears to have two levels due to the EMG levels in the signal.

[0129] Still refer to Figure 4 , method 400 includes extracting (step 412) representative cycles by a processor based on all or some of the cycles corresponding to the dominant PCA clusters (e.g., detected by DBSCAN). The representative cycles can be extracted in one or several ways, each with different characteristics. In some embodiments, each data point in the representative cycle will reflect the underlying distribution, where the distribution consists of that time point in all M cycles. For example, taking the average (for each N, over all M points) has a low-pass filtering effect (removing high-frequency information and noise), while taking the median retains high-frequency information in a non-linear manner. The different effects of the compression techniques (mean vs. median) are due to changing the underlying distribution. If the M points are normally distributed, the mean and median have the same results, but begin to differ with more complex distributions (such as those with non-zero skewness, especially in combination with negative kurtosis, or the presence of multimodality).

[0130] As mentioned above, Figure 6An example mean representative cycle (e.g., 602a) and a median representative cycle (e.g., 602b) for the same base cycle are shown. It is observed that fragmentation (e.g., high-frequency components retained in the median cycle but removed in the mean cycle, such as shown by arrow 604) is retained in the median cycle, between 3000-3500 samples, while the mean cycle has removed this feature (604). In addition, throughout the representative cycle, some high-frequency noise can be seen in the median beat (medianbeat), but not in the mean beat (mean beat). The use of the median operation can retain high-frequency features that may not exist in the mean representation due to changes in QRS morphology over time or due to time tailing (timesmearing) related to beat detection alignment. In addition, a function can be used to describe the distribution. Such a function will create a spectral mask (mask) that can remove the desired removal characteristics or enhance the characteristics that need to be preserved (e.g., the pattern of the kernel density estimate of the basic distribution).

[0131] In other words, the mean beat can be used to generate a "cleaner" representation of the cycle (i.e., less high-frequency components, where the high-frequency components include both signal and noise characteristics), while the median beat contains high-frequency components. Depending on the situation, one of these methods may be more appropriate. For example, the median beat may be used when i) it is desirable to ensure that high-frequency components of the biophysical signal characteristics are captured and preserved for analysis (even in the presence of some high-frequency noise that may obscure the analysis), or ii) the high-frequency noise in the signal is minimal or low.

[0132] Determine the representative cycle of the quasi-periodic signal cycle Figure 4 The processing can be part of a large number of studies to quantify skeletal muscle related noise and other asynchronous noise contamination in acquired biophysical signals.

[0133] In some embodiments, portions of the resulting windows that are adjacent windows within a group of windows are combined and evaluated (eg, to generate a template signal or for use in rejecting a signal). Fig. 10B Shown is a template signal vector data set superimposed on top of a set of stacked cycles for a high noise signal. Fig. 10C Shown is a template signal vector data set superimposed on top of a set of stacked cycles for a low noise signal. Fig. 10A Based on a single PSR record, and Fig. 10B Based on the second single PSR record.

[0134] Once the loops are identified (e.g., in each of these cases), the identified loops can be stacked (i.e., plotted or arranged on top of each other). For example, loop 1 data point 1 is placed at x=1, and loop 1 data point 6000 is placed at x=6000; then loop 150 (e.g.) data point 1 is also placed at x=1, and loop 150 data point 6000 is also placed at x=6000.

[0135] exist Fig. 10B and 10C As shown, once the data is stacked, a template signal vector data set 702 corresponding to the template for a given cycle may be generated.

[0136] It is worth noting that Fig. 10B and 10C An intermediate output of the embodiment is shown. Fig. 10B shows that the technique is able to extract meaningful template vectors in very noisy situations (in which case the typical cycles are not visually apparent), while Fig. 10C It is shown that the technique is able to extract typical cycles under ideal conditions.

[0137] Quantification of skeletal muscle artifact noise contamination in biophysical signals

[0138] Figure 5 is a diagram of an example method 500 for quantifying, by a processor, skeletal muscle related artifact noise contamination in acquired biophysical signals in accordance with an illustrative embodiment.

[0139] Method 500 includes: Figure 4 Steps 402-412 are discussed, and also include the step of quantifying, by the processor, the distribution of the differences between the determined representative cycle data set and the original signal data set(s).

[0140] The method 500 also includes comparing each detected cycle in the original signal data set cycle with the representative cycle data set. The comparison is performed by first phase aligning the representative cycle with each cycle examined (step 502). In some embodiments, a method such as finding the maximum value of the cross-correlation is used.

[0141] The comparison also includes determining (step 504) the difference between the representative cycle data set and the cycle for which the phase alignment is checked. In some embodiments, a method such as a correlation between two signals is used. In other embodiments, the median absolute error is used. In other embodiments, the mean absolute error is used. If there is more than one representative cycle data set (e.g., which is detected by clustering the two-dimensional PCA output), the corresponding representative cycle data set that best matches the given cycle is used.

[0142] The comparison also includes distinguishing outlying cycles from inlying cycles based on a difference score (the difference score is determined, for example, using a distribution-based filter) (step 506). In some embodiments, the distribution-based filter is configured to identify cycles whose standard deviation is greater than 1 from the mean. Fig.12 is a plot of the distribution of the difference scores determined based on the comparison of the representative cycle data set and each estimated cycle as a function of the cycle metric. Fig.12 As shown, inner loops are identified as being within one standard deviation of the mean of the distribution (e.g., within the region indicated by 1202) (as indicated by line 1204), and outer loops are identified as being outside one standard deviation from the mean (e.g., in the region indicated by 1206). A final assessment of the contamination of the biophysical signal by skeletal muscle-related noise can be made by taking a representative value (e.g., mean or median) of the inner difference score.

[0143] Without wishing to be bound by a particular theory, the presence of external cycles can be attributed to several factors, including noise caused by the biophysical signal being studied and physiological variability of the underlying physiological system. For cardiac signals, external cycles may be attributed to variations in the length and / or energy of depolarization or repolarization cycles, among other factors.

[0144] discuss

[0145] As mentioned above, quantification of asynchronous noise contamination (e.g., skeletal muscle-related artifacts and noise contamination) in biophysical signals (e.g., cardiac signals) can be complicated. For example, skeletal muscle-related artifacts and noise may appear as in-band noise relative to the biophysical signal, that is, it may occur in the same frequency range as the dominant component of the biophysical signal, typically about 0.5 Hz-80 Hz for cardiac signals and about 0.1-50 Hz for brain signals. In addition, EMG may also have amplitudes similar to typical cardiac or brain waveforms.

[0146] The similarity of skeletal muscle-related artifacts and noise contamination to biophysical signals may render automated diagnostic analysis of such signals problematic, and thus, quantifying the levels of skeletal muscle-related artifacts and noise contamination in signals may facilitate automated exclusion of signals that may be unsuccessful in subsequent analysis and / or compensation for such contamination in subsequent analysis.

[0147] When quantifying the level of skeletal muscle related artifacts and noise in biophysical signals, especially cardiac signals, it was observed that the skeletal muscle related artifacts and noise were not synchronized with the biophysical signals. Because the two have different sources (i.e., the cardiac signal originates from the sum of action potentials of cardiomyocytes, while the EMG originates from the sum of action points of primitive muscles (such as pectoralis, biceps, triceps, etc.), it is unlikely that these sources share a deeper common source that could produce synchronization. In fact, the skeletal muscle related artifacts and noise can be quantified by comparing each cardiac cycle of this patient to an idealized cardiac cycle, where the overall difference can be explained by the presence of skeletal muscle related artifacts and noise contamination in the biophysical signals.

[0148] Quantification of skeletal muscle related artifacts and noise becomes problematic (eg, skeletal muscle related artifacts and noise are within the band of the physical signal), and thus skeletal muscle artifact and noise denoising is also a challenge.

[0149] By leveraging the same observations for quantifying skeletal muscle-related artifacts and noise, a time series dataset of representative cycles can be generated and subjected to either frequency-based analysis or time-based analysis to remove or reduce skeletal muscle-related artifacts and noise and other asynchronous contamination.

[0150] Indeed, in some embodiments, a sample-by-sample comparison of the original signals can be performed in the frequency domain, followed by a frequency domain denoising operation between the signals to derive a denoised signal based on a spectral mask determined from the representative cycle vector and used to mask the noise characteristics of the original signal in the frequency domain. The exemplary denoising method exploits the robust information contained in the representative cycles as well as the information about the biophysical signal variations contained in the original data.

[0151] The exemplary methods and systems are described above with respect to cardiac signals. Note that the exemplary methods and systems can be applied to brain signals and other biophysical signals. Fig.13A , 13B and 13C shows the Figure 1A An example broadband brain phase gradient signal data set acquired by the measurement system 102. Fig.14 shows the phase space representation of Figures 13A-13C In fact, the broadband brain phase gradient signal is a quasi-periodic system and is similar to the cardiac broadband cardiac phase gradient signal in this respect, and the exemplary methods and systems can be applied to this signal.

[0152] Device normalization

[0153] In another aspect, an asynchronous contamination removal operation as described herein can be used to normalize cardiac signals acquired from multiple different acquisition platforms; for example, prior to performing machine learning-based disease association on data acquired from such platforms. Normalization is driven at least in part by knowledge of theoretical topological differences and insights from deep learning. Device normalization can be applied to data acquired from multiple acquisition devices (e.g., acquisition devices from two or more different generations) to improve similarity between signal groups that can improve machine learning training processing (as guided by machine learning and electrical engineering theory).

[0154] Fig.15 Schematic diagram of a method for normalizing a first set of data sets collected using a first set of biophysical signal measuring devices and a second set of data sets collected using a second set of biophysical signal measuring devices so that the first set of data sets can be analyzed together with the second set of data sets in a machine learning operation. Fig.15 As shown, a first set of cardiac signal data sets (e.g., shown as 1502) of a first group of subjects (e.g., shown as 1504) are collected using a first set of biophysical signal measurement devices (e.g., shown as 1506). A second set of cardiac signal data sets (e.g., shown as 1508) of a second group of subjects (e.g., shown as 1510) are collected using a second set of biophysical signal measurement devices (e.g., shown as 1512).

[0155] In some embodiments, as described with respect to FIG. 1 , a first set of cardiac signal data sets (e.g., 1502) is processed by a processor to remove asynchronous noise contamination, thereby improving the similarity between the first set of cardiac signal data sets and the second set of cardiac signal data sets and facilitating the use of the first set of cardiac signal data sets and the second set of cardiac signal data sets as the same training data sets for machine learning operations.

[0156] In some embodiments, the second set of cardiac signal data sets is processed by a processor (e.g., 1508) to remove asynchronous noise contamination, such as Figure 1A In order to improve the similarity between the first set of cardiac signal data sets and the second set of cardiac signal data sets.

[0157] Experimental Results

[0158] A Coronary Artery Disease-Learning Algorithm Development (CADLAD) study involving two different phases was conducted to support the development and testing of machine learning algorithms. In Phase 1, paired clinical data was used to guide the design and development of preprocessing, feature extraction, and machine learning steps. That is, the collected clinical research data was divided into three batches: training (50%), validation (25%), and confirmation (25%). Similar to the steps described above for processing signals from patients for analysis, each signal was first preprocessed to clean and normalize the data. After these processes, a set of features was extracted from the signal, each of which was paired with a representation of the true situation - for example, a binary classification of whether there was significant CAD. The final output of this phase is a fixed algorithm implemented in a measurement system. In Phase 2 of the CADLAD study, a machine learning algorithm will be used to provide a determination of significant CAD for a data pool of previously untested clinical data. Disease criteria were established according to the definitions in the American College of Cardiology (ACC) clinical guidelines, specifically as a stenosis greater than 70% as determined by angiography or a fraction-flow less than 0.80 as determined using flow-line techniques.

[0159] For some studies, a first set of cardiac signal data sets associated with early acquisition hardware (eg, “Gen 1”) (eg, measurement system 104) was processed to remove asynchronous noise contamination (eg, Figure 1A ), so as to use the first set of cardiac signal data sets and the second set of cardiac signal data sets (acquired by later acquisition hardware, such as "Gen 2") (e.g., measurement system 104) as training data sets for machine learning operations in the CADLAD study. Further description of example early acquisition hardware (e.g., including a unipolar sensing front end) can be found in U.S. patent application publication number 2017 / 0119272, which is incorporated herein by reference in its entirety, and further description of subsequent acquisition hardware (e.g., including a bipolar sensing front end) can be found in U.S. patent application publication number 15 / 911,047, which is also incorporated herein by reference in its entirety.

[0160] In some embodiments, the evaluation system 110 automatically and iteratively explores combinations of features in various functional arrangements, with the goal of finding those combinations that can successfully match predictions based on the features. In order to avoid overfitting of the solution to the training data, a validation set is used as a comparator. Once a candidate predictor is developed, it is manually applied to a validation data set to evaluate the predictor performance for data that has not yet been used to generate the predictor. If the data set is large enough, the performance of the selected predictor for the validation set will be close to the performance of the predictor for new data.

[0161] Fig.1616 is an example method 1600 for generating and analyzing a phase space volume object 122 by a non-invasive cardiac assessment system 110 according to an embodiment of the present disclosure. Other embodiments will become apparent to one of ordinary skill in the art based on this disclosure. In some embodiments, the method 1600 includes removing a baseline drift from a raw differential channel signal of a phase gradient biophysical signal data set 108 (operation 1602). In some embodiments, the raw differential channel signal is derived from six signals sampled simultaneously by the measurement system 102.

[0162] In some embodiments, six simultaneously sampled signals are captured from a stationary subject as a raw differential channel signal data set, wherein the signal embeds inter-lead timing and phase information of the acquired signal, which is specific to the subject. The geometric contrast produced by the interference of other orthogonal leads on the phase plane of the depolarization wave can be used, which can facilitate the superposition of phase space information on the three-dimensional representation of the heart. The noise-free subspace further facilitates the observation of the phases of these waves. That is, the phases of the orthogonal leads carry information about the structure and produce geometric contrast in the image. Phase-contrast takes advantage of the fact that different bioelectric structures have different impedances, so spectral and non-spectral conduction can cause the trajectory of the phase space track through the heart to be delayed and bent by different amounts. These subtle changes in the trajectory can be normalized and quantified beat by beat and corrected for abnormal or poor lead placement, and the normalized phase space integral can be mapped to a geometric grid for visualization.

[0163] In some embodiments, the original differential channel signal data set is normalized and the baseline drift is removed using a modified moving average filter. For example, in some embodiments, the baseline drift is extracted from each original differential channel signal using a median filter of the order of 1500 milliseconds, smoothed with a 1 Hz low-pass filter, and the baseline drift is subtracted from the signal. Then, the bias is removed from the resulting signal by subtracting the estimated value of the signal from the maximum probability density calculated using the kernel smoothing function. All or a portion of the signal can be divided by their respective interquartile ranges to complete the normalization process.

[0164] In some embodiments, method 1600 then includes reconstructing (operation 1604) the noise-free model signal by decomposing and selecting a set of candidate basis functions to create a sparse mathematical model. In some embodiments, a modified matching pursuit (MMP) algorithm is used to find the noise-free model of the original differential channel signal. Other sparse approximation algorithms can be used, including but not limited to evolvable mathematical models, symbolic regression, orthogonal matching pursuit, LASSO, linear models optimized using cyclic coordinate descent, orthogonal search, fast orthogonal search, and cyclic coordinate descent. In some embodiments, the reconstruction operation 504 generates a model as a function of a weighted sum of basis functions, wherein the basis function terms are sequentially attached to an initially empty basis to approximate the target function while reducing the approximation error.

[0165] Then, in some embodiments, method 1600 includes selecting (operation 506) a subspace component (e.g., a low energy frequency subspace component) from the selected basis functions and coefficients. The low energy subspace component includes a model reconstructed by using only the low amplitude subset coefficients (frequency components) that contribute the least to the modeling error X%. In some embodiments, the low energy subspace component includes higher order candidates that are subsequently selected in phase space coordinates as part of a sparse representation of the signal. That is, the last 5%, 10%, 15%, 20%, 25%, 30% of the candidates last selected by sparse approximation (as higher order candidates) are used. Other percentage values ​​may be used.

[0166] In some embodiments, method 1600 then includes: reconstructing (operation 1608) a predefined set of n-order fractional calculus result sets (e.g., via numerical fractional calculus operations) to generate a three-dimensional point cloud that partially defines the phase space volume object 122. In some embodiments, the fractional calculus operations are based on the Grünwald-Letnikov fractional differentiation method. In certain embodiments, the fractional derivative operations are based on the fractional linear multi-step method of Lubich. In some embodiments, the fractional calculus operations are based on the fractional Adams-Moulton method. In some embodiments, the fractional calculus operations are based on the Riemann-Liouville fractional differentiation method. In some embodiments, the fractional derivative operations are based on the Riesz fractional derivative method. Other methods of performing fractional calculus may be used.

[0167] In some embodiments, method 1600 then includes performing (1610) a triangulation operation to generate surface features (i.e., faces) of the point cloud. In some embodiments, an Alpha Hull triangulation with a predetermined radius (α) is performed on the reconstructed noise-free model signal. In other embodiments, Delaunay triangulation, alpha shape, ball pivoting, mesh generation, Convex Hull triangulation, etc. are used.

[0168] In some embodiments, the method 1600 then includes calculating (1612) one or more values ​​for each vertex in the point cloud. In some embodiments, the vertex values ​​are scaled within a range of renderable colors. In some embodiments, the vertex value is the variance between the modeled channel data set (e.g., an X-axis data set, a Y-axis data set, or a Z-axis data set) and the baseline raw channel data set (e.g., a corresponding X-axis data set, a Y-axis data set, or a Z-axis data set). In some embodiments, the variance is determined by subtracting the data points of the baseline raw channel data set from the corresponding data points of the modeled channel data set. In some embodiments, the modeled channel data set is based on a sparse approximation of the baseline raw channel data set to generate a reconstructed noise-free signal of the baseline raw channel data set. In some embodiments, each face of the phase space volume object 122 is assigned a face color value that is triangularly interpolated between adjacent boundary vertex color values ​​(e.g., 3 boundary vertex colors).

[0169] In some embodiments, various views of the phase space volume object 122 are captured to be presented as calculated phase space tomographic images, for example, to a physician via a web portal to assist the physician in assessing the presence or absence of pulmonary hypertension. In some embodiments, the phase space volume object or the calculated phase space tomographic image is evaluated by a trained neural network classifier configured to assess the presence or absence of pulmonary hypertension. In some embodiments, the calculated tomographic image (e.g., a collection of two-dimensional views) is presented together with the results of the machine-generated prediction to assist the physician in making a diagnosis.

[0170] In other embodiments, the phase space volume object 122 is analyzed in a subsequent machine learning operation (e.g., an image-based machine learning operation or a feature-based machine learning operation) to determine one or more coronary artery physiological parameters. In some embodiments, the evaluation system 110 is configured to determine a volume metric (e.g., an alpha hull volume) of the phase space volume object 122. In some embodiments, the evaluation system 110 is configured to determine multiple different bodies (e.g., different volumes) of the generated phase space volume object 122. In some embodiments, the evaluation system 110 is configured to evaluate the maximum color change (e.g., color gradient) of the generated phase space volume object 122. In some embodiments, all of these features are evaluated as mathematical features based on the phase space volume object 122.

[0171] In some embodiments, the mathematical features of the phase space volume object 122 are extracted along with hundreds of other different mathematical features that represent specific aspects of the collected biophysical signals. The feature extraction engine of the evaluation system 110 can extract each feature into a specific formula / algorithm. In some embodiments, when the feature extraction process is applied to the input biophysical signal, the output is a matrix of all calculated features, the matrix including, for example, a list of more than hundreds of real numbers, one number for each feature, where each feature represents one or more aspects of the dynamic, geometric, fractional calculus, chaotic, and / or topological properties of the signal.

[0172] In some embodiments, a machine learning algorithm (e.g., a meta-genetic algorithm) is used to generate a predictor that links together various aspects of a phase space model (e.g., a pool of features) representing an entire patient population of positive (i.e., disease) cases and negative (i.e., disease-free) cases to detect the presence of myocardial tissue associated with pulmonary hypertension. In some embodiments, the performance of a candidate predictor is evaluated by a validation process for a previously unseen patient population. In some embodiments, a machine learning algorithm calls a meta-genetic algorithm to automatically select a subset of features from a large pool. This subset of features is then used by an adaptive boosting (AdaBoost) algorithm to generate a predictor to diagnose pulmonary hypertension for an entire patient population representing positive and negative cases. The performance of a candidate predictor is determined by validation of a previously unseen patient pool. A further description of the AdaBoost algorithm is provided in "A decision-theoretic generalization of on-line learning and an application to boosting" by Freund, Yoav, and Robert E. Schapire et al., European Conference on Computational Learning Theory, Springer, Berlin, Heidelberg (1995), which is incorporated herein by reference in its entirety.

[0173] In some embodiments, the assessment system 110 generates one or more images of a representation of the phase space volume object 122, wherein vertices, face triangulations, and vertex colors are presented. In some embodiments, multiple views of the representation are generated and included in a report. In some embodiments, the one or more images are presented as a three-dimensional object that can be rotated, scaled, and / or translated based on user input. In practice, this presentation can be used by a skilled operator for visual assessment to determine whether a subject has pulmonary hypertension.

[0174] Neural Network Classification

[0175] The trained neural network classifier can directly evaluate the three-dimensional phase space volume object or the calculated phase space tomographic image to determine the presence or absence of pulmonary hypertension. In some embodiments, the neural network classifier can be a neural network trained on a set of grayscale tomographic images that are paired with coronary angiography results that are evaluated for the presence or absence of pulmonary hypertension. In some embodiments, a neural network-based nonlinear classifier is used. In some embodiments, the neural network-based nonlinear classifier is configured to map each pixel from the generated tomographic image to a binary disease state prediction (i.e., whether the condition exists or not) or an estimated physiological characteristic. In some embodiments, the neural network weights that control this mapping are optimized using a gradient descent technique.

[0176] Examples of disease state predictions may include, but are not limited to, the presence or absence of significant coronary artery disease, the presence or absence of pulmonary hypertension, the presence or absence of pulmonary arterial hypertension, the presence or absence of pulmonary arterial hypertension caused by left heart disease, the presence or absence of pulmonary arterial hypertension caused by lung disease, the presence or absence of pulmonary arterial hypertension caused by chronic blood clots, etc.

[0177] Examples of physiological characteristics assessed may include, but are not limited to, fractional flow reserve, degree of stenosis, degree of ischemia, blood glucose levels, cardiac chamber size, and mechanical function, among others.

[0178] Other examples of treatments that can be used with the exemplary methods and systems are described in the following patent documents: U.S. Patent No. 9,289,150, entitled "Non-invasive Method and System for Characterizing Cardiovascular Systems"; U.S. Patent No. 9,655,536, entitled "Non-invasive Method and System for Characterizing Cardiovascular Systems"; U.S. Patent No. 9,968,275, entitled "Non-invasive Method and System for Characterizing Cardiovascular Systems"; U.S. Patent No. 8,923,958, entitled "System and Method for Evaluating an Electrophysiological Signal"; U.S. Patent No. 9,408,543, entitled "Non-invasive Method and System for Characterizing Cardiovascular Systems and All-Cause Mortality and Sudden Cardiac Death Risk"; U.S. Patent No. 9,955,883, entitled "Non-invasive Method and System for Characterizing Cardiovascular Systems and All-Cause Mortality and Sudden Cardiac Death Risk”; U.S. Patent No. 9,737,229, entitled “Noninvasive Electrocardiographic Method for Estimating Mammalian Cardiac Chamber Size and Mechanical Function”; U.S. Patent No. 10,039,468, entitled “Noninvasive Electrocardiographic Method for Estimating Mammalian Cardiac Chamber Size and Mechanical Function”;U.S. Patent No. 9,597,021, entitled “Noninvasive Method for Estimating Glucose, Glycosylated Hemoglobin and Other Blood Constituents”; U.S. Patent No. 9,968,265, entitled “Method and System for Characterizing Cardiovascular Systems From Single Channel Data”; U.S. Patent No. 9,910,964, entitled “Methods and Systems Using Mathematical Analysis and Machine Learning to Diagnose Disease”; U.S. Patent Application Publication No. 2017 / 0119272, entitled “Method and Apparatus for Wide-Band Phase Gradient Signal Acquisition”; U.S. Patent Application Publication No. WO2017 / 033164, entitled “Method and Apparatus for Wide-Band Phase Gradient Signal Acquisition” Acquisition”; U.S. patent application with publication number 2018 / 0000371 and invention name “Non-invasive Method and System for Measuring Myocardial Ischemia, Stenosis Identification, Localization and Fractional Flow Reserve Estimation”; PCT application with publication number WO2017 / 221221 and invention name “Non-invasive Method and System for Measuring Myocardial Ischemia, Stenosis Identification, Localization and Fractional Flow Reserve Estimation”; U.S. patent number 10,292,596 and invention name “Method and System for Visualization of Heart Tissue at Risk”;U.S. Patent Application No. 16 / 402,616, entitled “Method and System for Visualization of Heart Tissue at Risk”; U.S. Patent Application Publication No. 2018 / 0249960, entitled “Method and System for Wide-band Phase Gradient Signal Acquisition”; U.S. Patent Application No. 16 / 232,801, entitled “Method and System to Assess Disease Using Phase Space Volumetric Objects”; PCT Application No. IB / 2018 / 060708, entitled “Method and System to Assess Disease Using Phase Space Volumetric Objects”; U.S. Patent Application Publication No. US2019 / 0117164, entitled “Methods and Systems of De-Noising Magnetic-Field Based Sensor Data of Electrophysiological Signals”; U.S. Patent Application No. 16 / 232,586, entitled “Method and System to Assess Disease Using Phase Space Volumetric Objects” Phase Space Tomography and Machine Learning”; PCT application number PCT / IB2018 / 060709, entitled “Method and System to Assess Disease Using Phase Space Tomography and Machine Learning”; U.S. patent application Ser. No. __ / __ / __, entitled “Methods and Systems to Quantify and Remove Asynchronous Noise in Biophysical Signals”, filed concurrently with the present application (attorney docket number 10321-033us1, and claiming priority to U.S. Provisional Application No. 62 / 686,245);U.S. Patent Application No. __ / __ / __, entitled “Method and System to Assess Disease Using Phase Space Tomography and Machine Learning” (Agent Docket No. 10321-034pv1, claiming priority to U.S. Provisional Application No. 62 / 784,984); U.S. Patent Application No. 16 / 429,593, entitled “Method and System to Assess Pulmonary Hypertension Using Phase Space Tomography and Machine Learning”; U.S. Patent Application No. __ / __ / __, entitled “Method and System for Automated Quantification of Signal Quality” (Agent Docket No. 10321-036pv1, claiming priority to U.S. Provisional Application No. 62 / 784,962); U.S. Patent Application No. __ / __ / __, entitled “Method and System to Configure and Use Neural Network To Assess Medical Disease” (attorney docket number 10321-037pv1, claiming priority to U.S. Provisional Application No. 62 / 784,925); U.S. Patent Application No. __ / __ / __, entitled “Method and System to Assess Disease Using Phase Space Volumetric Object and Machine Learning” (attorney docket number 10321-038pv1, claiming priority to U.S. Provisional Application No. 62 / 785,158); U.S. Patent Application No. 15 / 653,433, entitled “Discovering Novel Features to Use in Machine Learning Techniques, such as Machine Learning Techniques for Diagnosing Medical Conditions”; U.S. Patent Application No. 15 / 653,431, entitled “Discovering Genomes to Use in Machine Learning Techniques”;U.S. Provisional Application No. __ / __ / __ (Agent Docket No. 10321-041pv1), filed concurrently therewith, entitled “Method and System to Assess Disease Using Dynamical Analysis of Cardiac and Photoplethysmographic Signals,” all of which are incorporated herein by reference in their entirety. ;

[0179] Unless otherwise expressly stated, it is not intended that any method described herein be interpreted as requiring that its steps be performed in a particular order. Therefore, in the absence of a method claim actually reciting the order in which its steps are to be followed, or in the absence of other specific statements in the claims or specification that the steps are to be limited to a particular order, it is not intended that the order is intended in any way. This applies to any possible non-express basis for interpretation, including: logical issues regarding the arrangement of steps or operational flows; simple meaning derived from grammatical organization or punctuation; the number or type of embodiments described in the specification.

[0180] Although the above-described methods and systems have been described in conjunction with certain embodiments and specific examples, it is not intended to limit the scope to the specific embodiments set forth, as the embodiments herein are intended in all respects to be illustrative rather than restrictive.

[0181] The methods, systems, and processes described herein can be used to generate stenosis and FFR outputs for use in conjunction with processes such as placing a vascular stent in a blood vessel such as an artery of a mammalian (e.g., human) subject and other interventional and surgical systems or processes. In one embodiment, the methods, systems, and processes described herein can be configured to use the FFR / stenosis output to determine and / or modify the number of stents to be placed in a mammal (e.g., a human) during surgery, including the optimal location of the stent in a given blood vessel, etc.

[0182] Examples of other biophysical signals that may be analyzed in whole or in part using the exemplary methods and systems include, but are not limited to, electrocardiogram (ECG) datasets, electroencephalogram (EEG) datasets, gamma synchronization signal datasets, respiratory function signal datasets, pulse oximetry signal datasets, perfusion data signal datasets, quasi-periodic biosignal datasets, fetal ECG datasets, blood pressure signals, cardiac magnetic field datasets, and heart rate signal datasets.

[0183] Exemplary analyses can be used to identify various pathologies and conditions including, but not limited to, heart disease, arrhythmias, diabetic autonomic neuropathy, Parkinson's disease, forms of epilepsy, brain damage, altered cognitive status, stability of the heart at different heart rates, medications, ischemia, silent ischemia, atrial fibrillation, ventricular fibrillation, ventricular tachycardia, vascular occlusion, pulmonary hypertension, attention deficit, and the like.

Claims

1. A method for filtering asynchronous noise from a collected biophysical signal data set, the method comprising: receiving, by a processor, a biophysical signal dataset of a subject; determining, by the processor, at least one template signal vector data set characteristic of a representative quasi-periodic signal pattern of the subject based on a plurality of detected quasi-periodic cycles detected in the received biophysical signal data set; Applying, by the processor, the at least one determined template signal vector data set to one or more denoising vector data sets, wherein the one or more denoising vector data sets collectively have a vector length corresponding to a vector length of a portion to be filtered of the received biophysical signal data set, wherein i) the at least one determined template signal vector data set is applied to each detected cycle determined to be present in the portion to be filtered of the received cardiac signal data set; and ii) the length of the at least one determined template signal vector data set is varied to match the vector length of a corresponding detected cycle of the portion to be filtered of the received biophysical signal data set; and A filtered biophysical signal dataset of the biophysical signal dataset or a portion thereof is generated by merging the portion to be filtered of the received biophysical signal dataset and the one or more generated denoised vector datasets.

2. The method according to claim 1, further comprising: receiving, by the processor, one or more additional biophysical signal data sets, each of the additional biophysical signal data sets being acquired from the subject concurrently with the biophysical signal data set; determining, by the processor, at least one template signal vector data set characteristic of a representative quasi-periodic signal pattern of the subject based on a plurality of detected heartbeat cycles detected in each of the received one or more additional biophysical signal data sets; The processor applies multiple determined template signal vector data sets to one or more denoising vector data sets in a repetitive manner for each of the one or more received additional biophysical signal data sets, wherein the one or more denoising vector data sets collectively have a vector length corresponding to the vector length of the portion to be filtered of the received additional biophysical signal data set, and wherein i) each of the multiple determined template signal vector data sets is applied to each detected cycle determined to be present in the portion to be filtered of the received additional biophysical signal data set, and ii) the length of each of the multiple determined template signal vector data sets is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received additional biophysical signal data set.

3. The method according to claim 1 or 2, wherein the step of determining the at least one template signal vector data set characteristic of the representative quasi-periodic signal pattern comprises: determining, by the processor, a plurality of signal features of different characteristics in the received biophysical signal data set or a portion thereof; determining, by the processor, a plurality of cyclic regions between each of the determined plurality of signal features; aligning, by the processor, each of the plurality of cyclic regions with one another on the same aspect of the plurality of signal features or another set of signal features located in each of the cyclic regions; as well as Each point in the at least one template signal vector data set is determined by the processor using a mean operation or a median operation performed on each set of points in the plurality of cycle regions.

4. The method according to claim 3, wherein: The received biophysical signal dataset includes a cardiac signal dataset, and wherein the plurality of signal features are selected from a group consisting of: an R peak in the received cardiac signal dataset or a portion thereof, an S peak in the received cardiac signal dataset or a portion thereof, a T peak in the received cardiac signal dataset or a portion thereof, a Q peak in the received cardiac signal dataset or a portion thereof, or a P peak in the received cardiac signal dataset or a portion thereof.

5. The method according to claim 3, wherein: The received biophysical signal dataset comprises a cardiac signal dataset, and wherein the plurality of signal features correspond to R-peaks in the received cardiac signal dataset or a portion thereof.

6. The method according to claim 3, wherein: The step of determining the at least one template signal vector data set characteristic of the representative quasi-periodic signal pattern further comprises: A normalization parameter derived from each of the plurality of cycle regions is determined by the processor.

7. The method according to claim 3, wherein: The step of determining the at least one template signal vector data set characteristic of the representative quasi-periodic signal pattern further comprises: The value of each of the plurality of cycle regions or a parameter derived therefrom is normalized by the processor to a predetermined ratio.

8. The method of claim 3, wherein the step of determining the at least one template signal vector data set characteristic of the representative quasi-periodic signal pattern further comprises: A cluster-based analysis is performed by the processor on the plurality of cyclic regions to determine whether there is more than one dominant cyclic morphology, wherein a template signal vector is determined for each determined dominant cyclic morphology.

9. The method according to claim 3, wherein: The plurality of loop regions include loops that are adjacent to each other.

10. The method according to claim 9, wherein: The cycles adjacent to each other partially overlap each other.

11. The method according to claim 9, wherein: The cycles adjacent to each other do not overlap each other.

12. The method according to claim 9, wherein: The filtered biophysical signal dataset is generated by using two or more template signal vector datasets from two or more groups of cycles from the plurality of cycle regions, wherein the two or more groups of cycles from the plurality of cycle regions are adjacent to each other.

13. The method according to claim 1, wherein: The filtered biophysical signal dataset is generated in near real time while the biophysical signals are being acquired.

14. The method according to claim 1, wherein: After the biophysical signals are completely acquired, the filtered biophysical signal dataset is generated.

15. The method according to claim 1, wherein: The one or more denoised vector data sets are arranged as a one-dimensional vector.

16. The method according to claim 1, wherein: The one or more denoised vector data sets are arranged as an N-dimensional vector, where N corresponds to the number of detected cycles determined to be present in the portion of the received biophysical signal data set to be filtered.

17. The method of claim 1, wherein the step of applying the plurality of determined template signal vector data sets to one or more noise reduction vector data sets comprises: Initializing, by the processor, the one or more denoised vector data sets as a one-dimensional vector having a length corresponding to a length of the portion of the received biophysical signal to be filtered; as well as The determined template signal vector is replicated, by the processor, in the one-dimensional vector to align at least data points associated with peaks of the determined template signal vector with each peak determined in the received biophysical signal to be filtered.

18. The method according to claim 17, wherein: During said copying of the determined template signal vector, a conflicting portion of said currently copied template signal vector data set is assigned an average value relative to a corresponding portion of a previously copied template signal vector data set that conflicts with said currently copied template signal vector data set.

19. The method according to claim 17, wherein: During said copying of the determined template signal vector, a blank area in said one-dimensional vector between a currently copied template signal vector data set and a previously copied template signal vector data set stores a value interpolated between a last filling value and a next filling value around said blank area.

20. The method of claim 1, further comprising a window-based operation, wherein the window-based operation comprises: scaling, by the processor, the portion of the received biophysical signal data set to be filtered by a plurality of window functions having predetermined window lengths to generate a modified biophysical signal data set; scaling, by the processor, the one or more generated denoised vector data sets by the plurality of window functions to generate modified denoised vector data sets; determining, by the processor, an envelope of the modified denoised vector data set; converting, by the processor, the envelope of the modified denoised vector data set and the portion to be filtered of the received biophysical signal data set to the frequency domain by an FFT operation; Performing a weighted average operation on the FFT envelope of the modified denoised vector dataset and the modified biophysical signal dataset using a static interpolation coefficient or a set of dynamic interpolation coefficients by the processor to generate a result dataset; as well as The processor converts the result data set into a time series data set through an inverse FFT operation as a filtered biophysical signal data set of the biophysical signal.

21. The method according to claim 1, wherein: The data set of received biophysical signals includes data captured from sensors selected from the group consisting of: a 12-lead surface potential sensing electrode system, an intracardiac electrocardiogram, a dynamic electrocardiogram, a 6-lead differential surface potential sensing electrode system, a 3-lead orthogonal surface potential sensing electrode system, and a single-lead potential sensing electrode system.

22. The method according to claim 1, wherein: The received data set of biophysical signals includes wideband cardiac phase gradient cardiac signal data derived from biopotential signals simultaneously captured at a plurality of surface electrodes placed on a body surface near the heart.

23. The method of claim 1, further comprising: A notification of failure in acquiring a biophysical signal dataset is generated by the processor, wherein the notification prompts that subsequent acquisition of a biophysical signal dataset is to be performed.

24. The method of claim 1, further comprising: The received biophysical signal data set is transmitted by the processor over a network to an external analysis system, wherein the analysis system is configured to analyze the received biophysical signal data to determine the presence or extent of a pathology or clinical condition.

25. A method for normalizing a first set of data sets collected by a first set of measurement devices and a second set of data sets collected by a second set of measurement devices so that the first set of data sets are analyzed together with the second set of data sets in a machine learning operation, the method comprising: receiving, by a processor, a set of biophysical signal data sets of a subject acquired by a set of first measurement devices; determining, by the processor, at least one template signal vector data set characteristic of a representative quasi-periodic signal pattern of the subject based on a plurality of detected quasi-periodic signal cycles detected in the received biophysical signal data set; Applying, by the processor, a plurality of determined template signal vector data sets or vectors selected from a group thereof to one or more denoised vector data sets, wherein the one or more denoised vector data sets collectively have a vector length corresponding to a vector length of a portion to be filtered of a received biophysical signal data set, wherein i) each applied template signal vector data set is applied to each detected cycle determined to be present in the portion to be filtered of the received biophysical signal data set; and ii) the length of each applied template signal vector data set is varied to match the vector length of a corresponding detected cycle of the portion to be filtered of the received biophysical signal data set; and generating a filtered biophysical signal dataset associated with the biophysical signal dataset or a portion thereof as a normalized dataset of the biophysical signal, wherein the filtered biophysical signal is generated by combining the portion to be filtered of the received biophysical signal and the one or more generated noise reduction vectors, Among them, the normalized data set associated with the biophysical signal collected by the first measuring device is analyzed together with the second data set collected by the second measuring device as a machine learning training data set.

26. The method according to claim 25, wherein: The data set of received biophysical signals includes data captured from sensors selected from the group consisting of: a 12-lead surface potential sensing electrode system, an intracardiac electrocardiogram, a dynamic electrocardiogram, a 6-lead differential surface potential sensing electrode system, a 3-lead orthogonal surface potential sensing electrode system, and a single-lead potential sensing electrode system.

27. The method according to claim 25, wherein: The received data set of biophysical signals includes wideband cardiac phase gradient cardiac signal data derived from biopotential signals simultaneously captured at a plurality of surface electrodes placed on a body surface near the heart.

28. The method of claim 25, further comprising: A notification of failure in acquiring a biophysical signal dataset is generated by the processor, wherein the notification prompts that subsequent acquisition of a biophysical signal dataset is to be performed.

29. The method of claim 25, further comprising: The received biophysical signal data set is transmitted by the processor over a network to an external analysis system, wherein the analysis system is configured to analyze the received biophysical signal data to determine the presence or extent of a pathology or clinical condition.

30. A method for excluding collected biophysical signals, comprising: receiving, by a processor, a biophysical signal dataset of a subject; determining, by the processor, at least one template signal vector data set characteristic of a representative quasi-periodic pattern of the subject based on a plurality of detected quasi-periodic cycles detected in the received biophysical signal data set; applying, by the processor, the at least one determined template signal vector data set to one or more denoised vector data sets, wherein the one or more denoised vector data sets collectively have a vector length corresponding to a vector length of the portion to be filtered of the received biophysical signal data set, wherein i) the at least one determined template signal vector data set is applied to each detected cycle determined to be present in the portion to be filtered of the received biophysical signal data set; and ii) the length of the at least one determined template signal vector data set is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received biophysical signal data set; comparing, by the processor, the received biophysical signal dataset with at least one template signal vector dataset characteristic of a representative quasi-periodic pattern in the biophysical signal dataset; and The received biophysical signal data set is excluded by the processor based on the comparison result.

31. The method of claim 30, wherein the step of determining the at least one template signal vector data set characteristic of the representative quasi-periodic pattern comprises: determining, by the processor, a plurality of signal features of different characteristics in the received biophysical signal data set or a portion thereof; determining, by the processor, a plurality of cyclic regions between each of the determined plurality of signal features; aligning, by the processor, each of the plurality of cyclic regions with one another in the same aspect of the plurality of signal features or in the same aspect of another set of signal features located in each of the cyclic regions; as well as Each point in the at least one template signal vector data set is determined by the processor using a mean operation or a median operation performed on each set of points in the plurality of cycle regions.

32. The method of claim 31 , wherein determining the representative quasi-periodic pattern The step of determining at least one template signal vector data set characteristic further comprises: A cluster-based analysis is performed by the processor on the plurality of cyclic regions to determine whether there is more than one dominant cyclic morphology, wherein a template signal vector is determined for each determined dominant cyclic morphology.

33. The method of claim 30, further comprising: A notification of failure in acquiring a biophysical signal dataset is generated by the processor, wherein the notification prompts that subsequent acquisition of a biophysical signal dataset is to be performed.

34. The method of claim 30, further comprising: The received biophysical signal data set is transmitted by the processor over a network to an external analysis system, wherein the analysis system is configured to analyze the received biophysical signal data to determine the presence or extent of a pathology or clinical condition.

35. The method of claim 30, wherein the comparing comprises: It is determined whether there is asynchronous noise in the acquired biophysical signal data set whose value or energy exceeds a predetermined threshold.

36. A method for quantifying asynchronous noise in an acquired biophysical signal, the method comprising: receiving, by a processor, a biophysical signal dataset of a subject; determining, by the processor, at least one template signal vector data set characteristic of a representative quasi-periodic signal pattern of the subject based on a plurality of detected quasi-periodic cycles detected in the received biophysical signal data set; as well as applying, by the processor, the at least one determined template signal vector data set to one or more denoised vector data sets, wherein the one or more denoised vector data sets collectively have a vector length corresponding to a vector length of the portion to be filtered of the received biophysical signal data set, wherein i) the at least one determined template signal vector data set is applied to each detected cycle determined to be present in the portion to be filtered of the received cardiac signal data set; and ii) the length of the at least one determined template signal vector data set is varied to match the vector length of the corresponding detected cycle of the portion to be filtered of the received biophysical signal data set; Wherein determining at least one template signal vector data set characteristic of the representative quasi-periodic signal pattern comprises: determining, by the processor, a plurality of signal features of different characteristics in the received biophysical signal data set or a portion thereof; determining, by the processor, a plurality of cyclic regions between each of the determined plurality of signal features; aligning, by the processor, each of the plurality of cyclic regions with one another in the same aspect of the plurality of signal features or in the same aspect of another set of signal features located in each of the cyclic regions; determining, by the processor, each point in the at least one template signal vector data set using a mean operation or a median operation performed on each set of points in the plurality of cycle regions; and A cluster-based analysis is performed by the processor on the plurality of cyclic regions to determine whether there is more than one dominant cyclic morphology, wherein a template signal vector is determined for each determined dominant cyclic morphology.

37. A system for processing acquired biophysical signals, comprising: one or more processors; as well as A memory having instructions stored thereon, wherein the instructions are executable by the one or more processors to cause the one or more processors to perform the method according to any one of claims 1 to 36.

38. A non-transitory computer readable medium having instructions stored thereon, wherein execution of the instructions by one or more processors causes the one or more processors to perform the method of any one of claims 1 to 36.

Images