Non-invasive methods and systems for measuring myocardial ischemia, stenosis identification, localization, and fractional flow reserve estimation
Through unfiltered broadband cardiac phase gradient data and subspace data set generated by wavelet operators, combined with machine learning, non-invasive coronary stenosis recognition and blood flow reserve fraction estimation are achieved, solving the invasiveness in the prior art The risks and inaccuracy of detection improve detection efficiency and accuracy.
Patent Information
- Application Number
- CN202110321582.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2016-10-17
- Filing Date
- 2017-06-26
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2037-06-26
AI Technical Summary
The existing methods for coronary stenosis and myocardial ischemia detection require invasive surgery, which poses high risks such as stroke and heart attack, and the evaluation process is time-consuming, expensive and inaccurate enough, making it difficult to non-invasively identify high-risk patients.
By acquiring unfiltered broadband heart phase gradient data, a numerical wavelet operator is used to generate residual subspace and noise subspace data sets, combined with machine learning algorithms, the heart structure characteristics are extracted and abnormalities are located, and non-invasive coronary stenosis is achieved. Identification and blood flow reserve score estimates.
It realizes a non-invasive accurate assessment of myocardial ischemia, stenosis and blood flow reserve scores, reduces detection risks, improves detection efficiency and accuracy, and provides a visual diagnostic tool.
Smart Images

Figure CN113040740B_ABST
Abstract
Description
[0001] This case is a divisional application based on the Chinese patent application with patent application number 201780047004.7 filed on June 26, 2017 and invention name “Non-invasive method and system for measuring myocardial ischemia, stenosis identification, positioning and blood flow reserve fraction estimation”.
[0002] Related Applications
[0003] This application claims priority to and the benefit of U.S. Provisional Application No. 62 / 354,673, filed on June 24, 2016, and U.S. Provisional Application No. 62 / 409,176, filed on October 17, 2016, the entire contents of each of which are incorporated herein by reference. Technical Field
[0004] The present disclosure generally relates to non-invasive methods and systems for characterizing cardiovascular circulation. More specifically, the present disclosure relates to non-invasive methods for using unfiltered broadband cardiac phase gradient data to generate residual subspace and noise subspace data for, for example, predicting and locating coronary artery stenosis, locating and / or estimating blood flow reserve fraction, and characterizing myocardial ischemia. Background Art
[0005] Vascular disease is usually manifested as a reduction in blood flow due to atherosclerotic occlusion of a blood vessel. For example, occlusion of the coronary arteries that supply blood to the heart muscle is the main cause of heart disease. Invasive procedures for relieving arterial blockage (such as bypass surgery and stent placement with a catheter) rely on estimates of occlusion characteristics and blood flow through the occluded artery. These estimates are based on measurements of occlusion size and / or blood flow. Unfortunately, current methods for measuring occlusion size and blood flow require invasive procedures, such as coronary angiography, which requires cardiac catheterization. The process involves placing a long, thin, flexible catheter in a blood vessel in the arm, groin (upper thigh), or neck; the catheter is then screwed onto the heart. Through the catheter, the doctor can visually assess the inner diameter of the blood vessel through angiography or fluoroscopy and / or use a small sensor (usually a transducer) on the tip of the guidewire to measure parameters such as pressure, temperature, and flow to determine the severity of the lesion; and blood flow reserve fraction (FFR). These minimally invasive diagnostic tests on the heart carry the risk of stroke, heart attack, damage to the artery / heart where the catheter is inserted, irregular heart rhythm, kidney damage, infection and X-ray radiation exposure. These procedures are time consuming, require expertise to interpret the results and are expensive.
[0006] In the treatment phase, when balloon angioplasty, stenting or drug delivery are performed subsequently, the stenosis geometry is also important. For example, accurate stent placement is essential for reducing the risk of restenosis. Therefore, the decision on whether to use any occlusion relief method and which method should be used is usually based on partial information and does not consider coronary collateralization (coronary collateralization). Ischemic stress (ischemic stress) usually causes an increase in collateral circulation in the small coronary vessels, which sometimes compensates for distal vessel occlusion. In addition, the assessment of treatment success is also problematic, wherein the occlusion opening and stent position must be assessed. A class of methods currently used mainly requires a lengthy process to find and determine the severity of the lesion or multiple lesions, the obstruction to blood flow. Contemporary technology assesses cardiac gradient phase space changes and associates the changes with cardiac computed tomography (CT), myocardial perfusion imaging and cardiovascular angiography. The surface cardiac gradient contains detailed information about the electrophysiology of the recorded chamber. Because the surface cardiac gradient represents the sum of the individual action potentials of each cardiac cell in the syncytium, in theory, any information that could be determined by measuring coordinated cellular action potentials should be available at the surface "global" level. Furthermore, although information related to the effects of myocardial tissue architecture on conduction properties is inherent in the surface cardiac gradient, the challenge is to discern the relevant information from these long, quasi-periodic cardiac gradient signals while excluding noise contamination. Further, there is a distinct lack of non-invasive tools that could be used to enhance the identification of high-risk patients and thus pilot preventive strategies in a non-invasive manner. Summary of the invention
[0007] The present disclosure is helpful for evaluating broadband phase gradient information of cardiac tissue to assess the presence of ischemic heart disease. It is noteworthy that the present disclosure provides an improved effective method for identifying and risk stratifying coronary artery stenosis of the heart using high resolution and broadband cardiac gradients obtained from a patient or subject. In some embodiments, patient data is derived from cardiac gradient waveforms across multiple leads, thereby obtaining high-dimensional data and long cardiac gradient records showing complex nonlinear changes. Through numerical wavelet operators, spatiotemporal analysis is used to study the morphology of cardiac gradient data as a phase space data set by extracting dynamic and geometric properties from the phase space data set. Digital wavelet operators facilitate real-time or near real-time processing of acquired broadband cardiac gradient data sets to generate residual subspace data sets and noise subspace data sets. In some embodiments, the residual subspace data set and the noise subspace data set are generated as multidimensional data sets (e.g., three-dimensional) and are very suitable for image and graphics processing. The morphology of the extracted residual subspace data set and the noise subspace data set is fed to the learning algorithm as parameters and variables to associate them with abnormalities of the heart under study.
[0008] Thus, the present disclosure provides a non-invasive system and method whereby cardiac gradient measurements can be taken and converted to locate and visualize architectural features of the myocardium and characterize abnormalities and cardiovascular function in the heart through rendered images. In addition, the present disclosure provides a system and method for visualizing (e.g., inverse cardiac gradient problems) architectural features of the heart and locating abnormally conducting / functioning cardiac tissue. In addition, the present disclosure provides a system and method for outputting the structural features of the heart and the location of abnormally conducting / functioning cardiac tissue as parameters to a therapy, treatment device, or diagnostic device.
[0009] In one aspect, a method for non-invasively identifying and / or measuring myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve in a mammalian subject or patient is disclosed, the method comprising: simultaneously acquiring a plurality of broadband gradient signals from the subject via at least one electrode (e.g., one or more surface electrodes, non-contact electrodes, or other types of biopotential sensing devices); and determining, via one or more processors, one or more coronary artery physiological parameters of the subject selected from the group consisting of: fractional flow reserve estimate, stenosis value, and myocardial ischemia estimate (e.g., and / or other arterial blood flow characteristics) based on a residual subspace dataset and a noise subspace dataset derived from data associated with the plurality of broadband gradient signals.
[0010] In some embodiments, a residual subspace dataset is determined by: generating a first wavelet signal dataset by performing a first wavelet operation (by, for example, a first phase linear wavelet operator) on data derived from a plurality of broadband gradient signals; generating a second wavelet signal dataset by performing a second wavelet operation (by, for example, a second phase linear wavelet operator) on the first wavelet signal data; and subtracting the values of the first wavelet signal dataset from the values of the second wavelet signal dataset to generate a residual subspace dataset, wherein the residual subspace dataset comprises a three-dimensional phase space dataset in a space-time domain.
[0011] In some embodiments, the method also includes: extracting a first set of morphological features of the three-dimensional phase space data set, wherein the extracted first set of morphological features includes parameters selected from a group consisting of the following parameters: 3D volume value, void volume value, surface area value, principal curvature direction value and Betti value.
[0012] In some embodiments, the extracted first set of morphological features is extracted using an alpha-hull operator.
[0013] In some embodiments, the method also includes: dividing the three-dimensional phase space data set into a plurality of segments, each of the segments comprising a non-overlapping portion of the three-dimensional phase space data set; and extracting a second set of morphological features for each of the plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from a group consisting of the following parameters: 3D volume value, void volume value, surface area value, principal curvature direction value, and Betti value.
[0014] In some embodiments, the extracted second set of morphological features is extracted using an alpha-hull operator.
[0015] In some embodiments, the plurality of segments comprises a number of segments selected from the group consisting of 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20.
[0016] In some embodiments, the noise subspace dataset is determined by: generating a first wavelet signal dataset by performing a first wavelet operation (e.g., a first phase linear wavelet operator) on data derived from the multiple broadband gradient signals; and generating a second wavelet signal dataset by performing a second wavelet operation (e.g., a second phase linear wavelet operator) on the first wavelet signal dataset, the second wavelet signal data including the noise subspace dataset, wherein the noise subspace dataset includes a three-dimensional phase space dataset in the space-time domain.
[0017] In some embodiments, the method further comprises: extracting a set of morphological features of the three-dimensional phase space data set, wherein the extracted set of morphological features comprises parameters selected from a group consisting of: 3D volume value, void volume value, surface area value, principal curvature direction value and Betti value.
[0018] In some embodiments, the extracted set of morphological features is extracted using an alpha-hull operator.
[0019] In some embodiments, the method also includes: dividing the three-dimensional phase space data set into a plurality of segments, each segment comprising a non-overlapping portion of the three-dimensional phase space data set; and extracting a second set of morphological features for each of the second plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from a group consisting of the following parameters: 3D volume value, void volume value, surface area value, principal curvature direction value, and Betti value.
[0020] In some embodiments, the extracted second set of morphological features is extracted using an alpha-hull operator.
[0021] In some embodiments, the number of segments included in the second segment is selected from the group consisting of 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20.
[0022] In some embodiments, the residual subspace data set is associated with a first shape of a first noise geometric profile, and wherein the noise subspace data set is associated with a second shape of a second noise geometric profile corresponding to random noise.
[0023] In some embodiments, the plurality of broadband gradient signals are acquired simultaneously, and the plurality of broadband gradient signals have a lag or deviation of less than about 10 femtoseconds between each signal.
[0024] In some embodiments, each of the plurality of wideband gradient signals is unfiltered prior to and during processing to generate the residual subspace dataset and the noise subspace dataset.
[0025] In some embodiments, each of the plurality of broadband gradient signals includes cardiac data in the frequency domain having frequency components greater than about 1 kHz.
[0026] In some embodiments, each of the plurality of broadband gradient signals comprises heart rate information at a frequency selected from the group consisting of approximately 1 kHz, approximately 2 kHz, approximately 3 kHz, approximately 4 kHz, approximately 5 kHz, approximately 6 kHz, approximately 7 kHz, approximately 8 kHz, approximately 9 kHz, and approximately 10 kHz.
[0027] In some embodiments, each of the plurality of broadband gradient signals includes heart rate information at a frequency between approximately 0 Hz and approximately 50 kHz.
[0028] In some embodiments, each of the plurality of broadband gradient signals includes heart rate information at a frequency between about 0 Hz and about 500 kHz.
[0029] In some embodiments, the method further includes: associating the extracted first set of morphological features with a plurality of candidate models through a first machine learning (e.g., neural network, formula learning, etc.) operation, the plurality of candidate models being associated with estimates of the blood flow reserve fraction, the stenosis value, and the myocardial ischemia estimate; and selecting a candidate model from the plurality of candidate models to determine an output for each of the estimates of the reserve blood flow reserve fraction, the stenosis value, and the myocardial ischemia estimate.
[0030] In some embodiments, the method further comprises: visualizing the determined one or more coronary artery physiological parameters.
[0031] In some embodiments, the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate are presented at one or more corresponding coronary artery regions on a cardiac image (eg, via a standard 17-segment model of the heart).
[0032] In some embodiments, the method further includes: outputting the determined one or more coronary artery physiological parameters to a surgical device or a diagnostic device.
[0033] In some embodiments, one or more of the at least one electrode is selected from the group consisting of a surface electrode, an intracardiac electrode, and a non-contact electrode.
[0034] In some embodiments the method further comprises identifying one or more significant stenoses having an AUC-ROC score ("Area Under the Curve" and "Receiver Operating Characteristic Curve" score) greater than about 0.7 during the validation phase.
[0035] In another aspect, a system for non-invasively identifying and / or measuring myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve, for example, by a broadband biopotential measurement device is disclosed. The system includes a processor and a memory having instructions stored thereon, wherein execution of the instructions causes the processor to simultaneously acquire multiple broadband gradient signals from one or more electrodes (e.g., surface electrodes, non-contact electrodes, or other types of biopotential sensing devices); and determine one or more coronary artery physiological parameters selected from the group consisting of the following parameters based on a residual subspace dataset and a noise subspace dataset: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate, wherein the residual subspace dataset and the noise subspace data are derived from data associated with the multiple broadband gradient signals.
[0036] In some embodiments, execution of the instructions further causes the processor to: visualize the determined coronary artery physiological parameter for presentation on a display.
[0037] In some embodiments, execution of the instructions further causes the processor to: output the determined one or more coronary artery physiological parameters to a treatment device.
[0038] In another aspect, a computer-readable medium is disclosed having instructions stored thereon, wherein execution of the instructions further causes a processor to: determine one or more coronary artery physiological parameters selected from the group consisting of: a blood flow reserve fraction estimate, a stenosis value, and a myocardial ischemia estimate based on a residual subspace dataset and a noise subspace dataset derived from data associated with a plurality of broadband gradient signals simultaneously acquired from at least one surface electrode.
[0039] In some embodiments, further execution of the instructions causes the processor to: visualize the determined coronary artery physiological parameter for presentation on a display.
[0040] In some embodiments, execution of the instructions further causes the processor to: output the determined one or more coronary artery physiological parameters to a treatment device. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The components in the drawings are not necessarily drawn to scale relative to each other, like reference numerals indicate corresponding parts throughout the several views:
[0042] Figure 1 is a diagram of a system for non-invasively determining arterial blood flow characteristics in the heart (eg, presence of myocardial ischemia, stenosis identification, location, and fractional flow reserve estimation) using broadband cardiac gradient data in accordance with an exemplary embodiment.
[0043] Figure 2 is a graph of an example broadband cardiac gradient signal shown as time series data in accordance with an exemplary embodiment.
[0044] Figure 3 is a graph of an example wideband cardiac gradient signal shown in the frequency domain, according to an exemplary embodiment.
[0045] Figure 4 is a time series graph of example broadband cardiac gradient signals associated with a single surface electrode in accordance with an exemplary embodiment.
[0046] Figure 5 is a frequency plot of example broadband cardiac gradient signals associated with three surface electrodes in accordance with an exemplary embodiment.
[0047] Figure 6 is a frequency plot of example ultra-wideband cardiac gradient signals associated with three surface electrodes in accordance with an exemplary embodiment.
[0048] Figure 7 is a diagram of a method of processing broadband cardiac gradient signals to non-invasively identify and / or estimate the extent of myocardial ischemia, stenosis identification and / or location, and fractional flow reserve estimation, according to an exemplary embodiment.
[0049] Figure 8 is a diagram of a method of performing feature topology analysis of, for example, a multidimensional residual subspace dataset, a multidimensional noise subspace dataset, and a multidimensional wavelet-cleaned dataset, according to an exemplary embodiment.
[0050] Fig. 9is a diagram of a method for performing machine learning analysis to create and select a nonlinear model to identify and / or estimate the extent of myocardial ischemia, stenosis identification and / or location, and fractional flow reserve estimation, according to an exemplary embodiment, as described with respect to Figure 8 Described.
[0051] Fig. 10A and 10B is a diagram of an example wavelet transform for generating a multidimensional wavelet-cleaned data set in accordance with an exemplary embodiment.
[0052] Fig.11 is a graph of an example time series data set of a wavelet cleaned data set in accordance with an exemplary embodiment.
[0053] Fig.12 is a diagram of an example wavelet-based operation for generating a multidimensional residual subspace dataset in accordance with an exemplary embodiment.
[0054] Fig.13A and 13B is a diagram of an example wavelet transform for generating a multidimensional wavelet-cleaned data set in accordance with an exemplary embodiment.
[0055] Fig.14 is a depiction of an example multidimensional residual subspace data set, in accordance with an exemplary embodiment.
[0056] Fig.15 is a depiction of an exemplary multidimensional noise subspace data set in accordance with an exemplary embodiment.
[0057] Fig.16 and 17 Each is a depiction of an exemplary multidimensional residual subspace data set, according to an exemplary embodiment. Fig.16 An example multidimensional residual subspace dataset for a subject without ischemia is shown. Fig.17 An example multidimensional residual subspace dataset for an ischemic subject is shown.
[0058] Fig.18 and 19 Each of which is a depiction of an exemplary multidimensional noise subspace data set, according to an exemplary embodiment. Fig.18 An example multidimensional noise subspace dataset for a subject without ischemia is shown. Fig.19 An example multidimensional noise subspace dataset of an ischemic subject is shown.
[0059] Fig. 20 is a diagram of a method of visualizing arterial blood flow characteristics determined in a heart, according to an exemplary embodiment.
[0060] Fig.21 and 22is a diagram showing a method of using Figure 1 Figure 1 shows the results of the analysis of the studies conducted. DETAILED DESCRIPTION
[0061] The components in the drawings are not necessarily drawn to scale relative to each other, like reference numerals indicate corresponding parts throughout the several views.
[0062] Figure 1 is a diagram of a system for non-invasively determining arterial flow characteristics in a heart using wide-band cardiac gradient data, according to an exemplary embodiment. Figure 1 As shown, the system 100 includes a broadband biopotential measurement device 102 and an analysis subsystem 104. The broadband biopotential measurement device 102 collects broadband biopotential signals 112 (shown as 112a, ..., n) (also referred to herein as broadband cardiac gradient signal data 112) from a subject or patient 110 via at least one electrode 106 (shown as surface electrodes 106a, 106b, ..., 106n) and a corresponding common-mode reference lead 108, wherein the at least one electrode 106 and the corresponding common-mode reference lead 108 are both in Figure 1 in a system and both are attached to the surface of a mammalian subject or patient 110 (e.g., the skin of an animal or a human). The broadband biopotential measurement device 102 can be any device configured to capture an unfiltered electrophysiological signal so that the spectral components of the signal are not changed. That is, all captured signals (if not a significant portion of the captured signals) include components that are typically perceived or considered noise, such as components within a frequency range greater than about 1 kHz. To this end, the broadband biopotential measurement device 102 captures, converts, and analyzes the acquired broadband biopotential signal 112 without performing any filtering (by hardware circuitry or digital signal processing) that affects the phase linearity of the broadband biopotential signal 112. That is, only phase deterministic operations, numerical or analytical, are performed in phase space transformation and analysis. Phase distortion is a non-deterministic distortion that causes a shift in the frequency components of the signal.
[0063] An example broadband biopotential measurement device 102 is described in U.S. patent application Ser. No. 15 / 248,838, Publication No. US2017 / 0119272, and entitled “Method and Apparatus for Wide-Band Gradient Signal Acquisition,” the entire contents of which are incorporated herein by reference. In some embodiments, the broadband biopotential measurement device 102 is configured to record unfiltered physiological signals at multiple observation points on a patient or subject (in a stationary position) at a rate of approximately 8 kHz for 210 seconds. The resulting signal record is then securely transmitted to a cloud-based repository, where it is automatically queued for processing. In some embodiments, the resulting signal record is securely transmitted to a cloud-based repository, where it is automatically queued for processing. The processing pipeline derives the phase energy of the chest system by performing a multidimensional (space-time) transformation on the signal, and then reconstructs it into a phase space model of the patient's heart.
[0064] The inventors have discovered that broadband biopotential signals having energy and frequency components that are beyond those of a conventional electrocardiogram (ECG) and that are traditionally perceived or viewed as random noise include measurable data of cardiac physiology that can be resolved by genetic algorithms (and other machine learning algorithms) to assess local flow characteristics of the heart, including, for example, estimates of stenosis and identification of ischemia and fractional flow reserve (FFR) of specific arteries and their branches. Removing noise (e.g., by applying cleaning techniques to the data, resulting in the same amount of data as before the noise removal) is a fundamental step in signal processing. However, the exemplary methods and systems process the entire acquired biopotential signal without any noise removal operation. What has heretofore been perceived and / or classified as unwanted noise in broadband data is in many cases a signal of interest. Examples of noise removal that is not performed include, but are not limited to, analog-based low-pass filters, band-pass filters, high-pass filters, and digital-based filters (such as FIR filters, Butterworth filters, Chebyshev filters, and median filters (among others)) configured to change the phase linearity of the processed signal. Note that analog based low pass filters, band pass filters, high pass filters, as well as digital based filters configured for phase linearization may be used. In some embodiments, a signal may be processed through a phase linearization operation to allow analysis of specific aspects of high frequency broadband data.
[0065] As described in U.S. Patent Application No. 15 / 248,838, in some embodiments, the broadband biopotential measurement device 102 is configured to capture one or more biosignals, such as biopotential signals, at a microvolt or sub-microvolt resolution (which is a resolution at or significantly below the background noise of conventional electrocardiogram and biosignal acquisition instruments). In some embodiments, the broadband biopotential measurement device 102 is configured to acquire and record simultaneously sampled broadband phase gradient signals (e.g., broadband cardiac phase gradient signals, broadband brain phase gradient signals), which in some embodiments have a time skew or "lag" of less than about 1 μs, and in other embodiments, have a time skew or lag of no more than about 10 femtoseconds. It is worth noting that the illustrated system minimizes nonlinear distortion (e.g., distortion that can be introduced by certain filters) in the acquired broadband phase gradient signals so as not to affect the information therein.
[0066] Still see Figure 1 , the analysis system 104 is configured to generate a phase space graph to be used in a subsequent phase space analysis 118 described later. The output of the phase space analysis is then evaluated using a machine learning analysis 120 to evaluate parameters 122 associated with the presence of disease or physiological characteristics (e.g., local arterial blood flow characteristics). In some embodiments, the machine learning analysis 120 can use a library 124 of quantified FFR, stenosis, and ischemia data when evaluating the obtained broadband cardiac gradient signal data 112. The output 122 of the processor performing the analysis 104 is then transmitted to a graphical user interface (e.g., a touch screen or other monitor) for visualization. In some embodiments, the graphical user interface is included in a display unit configured to display the parameters 122. In some embodiments, the graphical user interface displays intermediate parameters, such as a 3D phase space graph representation of biopotential signal data and virtual biopotential signal data. In other embodiments, the output of the processor is then transmitted to one or more non-graphical user interfaces (e.g., a printout, a command line, or a text-only user interface), directly to a database or memory device, for example, later retrieval and / or additional analysis, or a combination thereof.
[0067] As used herein, the term "processor" refers to a physical hardware device that executes coded instructions for performing functions on inputs and creating outputs. A processor may include one or more processors, each configured to execute instructions and process data to perform one or more functions associated with a computer for indexing images. The processor may be communicatively coupled to RAM, ROM, storage, databases, I / O devices, and interfaces. The processor may be configured to execute sequences of computer program instructions to perform various processes.
[0068] Example broadband cardiac gradient signal
[0069] Figure 2 (exist Figure 1 ) is a graph of an example broadband cardiac gradient signal 112 shown as time series data, according to an embodiment. Figure 3 (also Figure 1 ) is shown in the frequency domain according to an embodiment Figure 2 FIG. 112 is an example broadband cardiac gradient signal 112. Figure 3 As shown, the broadband cardiac gradient signal 112 has a frequency component greater than 1kHz, which is significantly higher than conventional electrocardiogram measurements. In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 4kHz (e.g., about 0Hz to about 4kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 5kHz (e.g., about 0Hz to about 5kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 6kHz (e.g., about 0Hz to about 6kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 7kHz (e.g., about 0Hz to about 7kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 8kHz (e.g., about 0Hz to about 8kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 9kHz (e.g., about 0Hz to about 9kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component of up to about 10kHz (e.g., about 0Hz to about 10kHz). In some embodiments, the broadband cardiac gradient signal 112 has frequency components up to 50 kHz (eg, about 0 Hz to about 50 kHz).
[0070] Figure 4 is a time series graph of an example broadband cardiac gradient signal 112 associated with a single surface electrode, under an embodiment. The graph shows the signal in mV over time (in seconds).
[0071] Figure 5 is a frequency diagram of an example broadband cardiac gradient signal 112 associated with three surface electrodes, according to an embodiment. As shown, Figure 5 The broadband cardiac gradient signal 112 includes frequency components up to 4 kHz. As further shown, the broadband cardiac gradient signal 112 presented has a power between -20 dB and 20 dB in the frequency domain at frequencies greater than 1 kHz. This portion 502 of the broadband cardiac gradient signal 112 includes topological and functional information about cardiac tissue and its underlying structures, which can be used to determine local blood flow characteristics, such as estimation of local FFR, estimation and identification of local stenosis, and / or estimation of the extent of ischemia.
[0072] Broadband cardiac gradient signals (e.g., having a frequency between about 1 kHz and about 10 kHz) facilitate phase space analysis of unevenly sampled data. In some embodiments, the broadband cardiac gradient signal has a higher sampling rate during intervals of interest and a lower sampling rate during other intervals in order to minimize the size of the resulting data set. Such varying sampling rates can be used in applications where data storage is limited. Many nonlinear functions (e.g., such as those used in phase space analysis) operate more efficiently when identifying amplitudes with points that are unevenly spaced in the time domain. A sampling rate that is much higher than the sampling rate of the highest frequency of interest (e.g., 10 times higher than the sampling rate of the highest frequency of interest) facilitates correct characterization of the shape of the signal system. This is similar to a Lorenz system, where very high frequencies facilitate correct modeling of the shape of the system in phase space. An example of a Lorenz system is described in Lorenz, Edward Norton, "Deterministic non-periodic flow," Journal of the Atmospheric Sciences 20(2), pp. 130-141 (1963), the entire contents of which are incorporated herein by reference.
[0073] Example ultra-wideband cardiac gradient signal
[0074] In some embodiments, the illustrative methods and systems are used to classify ultra-wideband cardiac gradient signals having a negative spectral energy signature up to about 500 kHz (eg, having a frequency between about 1 kHz and about 500 kHz).
[0075] Figure 6 is a frequency diagram of an example ultra-wideband cardiac gradient signal according to an exemplary embodiment. As shown in the figure, Figure 6 The frequency components of the ultra-wideband cardiac gradient signal sampled at a frequency of about 200kHz and including up to about 100kHz (according to the Nyquist sampling theorem). It is noteworthy that, as shown, the presented ultra-wideband cardiac gradient signal includes a negative spectral energy feature (shown as frequency 602) with a frequency greater than about 70-80kHz; the negative spectral energy feature has an energy between about 40dB and about 50dB in the frequency domain. The data indicate that low energy features in the ultra-wideband electrocardiogram may have information that can be used to image the morphology or function of the body and / or for diagnosis.
[0076] Example Processing of Broadband Biopotential Signal Data
[0077] Figure 7is a diagram of a method 700 for processing broadband biopotential signal data 112 (and ultra-wideband biopotential signal data) according to an exemplary embodiment. Figure 7 As shown, method 700 includes acquiring broadband gradient cardiac signal data 112 (shown as “broadband gradient signal unfiltered raw ADC data” 112) and preprocessing 702 the broadband gradient cardiac signal data to generate phase space data sets in phase space analysis (shown as “residual subspace” data set 704 and “noise subspace” data set 706), thereby extracting features of the phase space data sets (704, 704) (operation 708) and evaluating the features of the phase space data sets (704, 704) with a nested nonlinear function 710 to generate stenosis and FFR estimates 122.
[0078] Broadband gradient cardiac signal data 112 may be collected from one or more electrodes (e.g., surface electrodes, non-contact electrodes). In some embodiments, broadband gradient cardiac signal data 112 is collected simultaneously from 1 to about 20 or more electrodes (e.g., 1 electrode, 2 electrodes, 3 electrodes, 4 electrodes, 5 electrodes, 6 electrodes, 7 electrodes, 8 electrodes, 9 electrodes, 10 electrodes, 11 electrodes, 12 electrodes, 13 electrodes, 14 electrodes, 15 electrodes, 16 electrodes, 17 electrodes, 18 electrodes, 19 electrodes, and 20 or more electrodes). In some embodiments, the sampling of these electrodes may have a skew or "lag" of less than 10 femtoseconds. In other embodiments, the sampling of these electrodes may have a skew or lag of less than 100 femtoseconds. In other embodiments, the sampling of these electrodes may have a skew or lag of less than a few picoseconds. In some embodiments, the broadband cardiac gradient signal 112 has a frequency component up to about 10 kHz (e.g., about 0 Hz to about 4 kHz; about 0 Hz to about 5 kHz; about 0 Hz to about 6 kHz; about 0 Hz to about 7 kHz; about 0 Hz to about 8 kHz; about 0 Hz to about 9 kHz; or about 0 Hz to about 10 kHz). In some embodiments, the broadband cardiac gradient signal 112 has a frequency component up to about 50 kHz (e.g., about 0 Hz to about 50 kHz). In some embodiments, the broadband gradient cardiac signal data 112 has a voltage resolution of about 1 / 2 μV sensitivity. In other embodiments, the broadband gradient cardiac signal data 112 has a voltage resolution greater than about 1 / 2 μV sensitivity (e.g., about 1 μV, about 10 μV, about 100 μV, or about 1 mV). In some embodiments, the resolution of the signal data is about 24 bits. In some embodiments, the effective resolution is 20 bits, 21 bits, 22 bits, or 23 bits or more bits. In some embodiments, the effective resolution is less than 20 bits (e.g., 18 bits or 14 bits or less).
[0079] In some embodiments, phase space plot analysis uses geometrical contrast, which is produced by the interference of the depolarization wave with any other orthogonal leads in the phase plane. The existence of a noise-free subspace allows the phase of these waves to be recorded. Typically, the amplitude produced by this interference can be measured; however, the phase of these orthogonal leads still carries information about the structure and generates geometric contrast in the image. Phase space plot analysis takes advantage of the fact that different bioelectric structures within, for example, the heart and its various types of tissue have different impedances, so spectral and non-spectral conduction delays and bend the trajectory of the phase space track through the heart by different amounts. These slight changes in the trajectory can be normalized and quantized beat-to-beat and corrected for abnormal or poor lead placement, and the normalized phase space integral can be visualized or mapped onto a geometric mesh using a genetic algorithm to map the 17 myocardial segments in the ventricle to various tomographic imaging modes of the heart based on retrospective data.
[0080] Still see Figure 7 , three separate phase space analyses are performed to generate metrics and sets of variables (shown as 712a, 712b, and 712c) that are used in the nonlinear function 710 to generate regional FFR estimates, regional stenosis values, and regional ischemia values 122. Table 1 is an example output matrix 122.
[0081] Table 1
[0082] Segment Blood vessel FFR narrow Ischemia 1 Left main artery (LMA) 0.90 0.50 0.20 2 Proximal left circumflex artery (Prox LCX) 0.85 0.60 0.30 3 Mid LCX 0.93 0.35 0.15 4 Distal left circumflex artery (Dist LCX) 1.00 0.00 0.00 5 Left posterior atrioventricular (LPAV) 1.00 0.00 0.00 6 First obtuse marginal branch (OM1) 0.60 0.95 0.72 7 Second obtuse marginal branch (OM2) 1.00 0.00 0.00 8 The third obtuse marginal branch (OM3) 1.00 0.00 0.00 9 Proximal left anterior descending artery (Prox LAD) 1.00 0.00 0.00 10 Mid LAD 1.00 0.00 0.00 11 Distal left anterior descending artery (Dist LAD) 0.70 0.80 0.63 12 LAD D1 0.00 0.00 0.75 13 LAD D2 0.00 0.00 0.00 14 Proximal right coronary artery (Prox RCA) 0.00 0.00 0.00 15 Mid RCA 0.00 0.00 0.00 16 Distal right coronary artery (Dist RCA) 0.00 0.00 0.18 17 Right acute marginal branch of the posterior descending artery (AcMRPDA) 0.00 0.00 0.00
[0083] As indicated above, Table 1 includes fractional flow reserve (FFR) parameters, estimated stenosis parameters, and estimated ischemia parameters corresponding to multiple segments of major blood vessels in the heart. In some embodiments, the matrix 122 includes blood flow reserve fractional (FFR) parameters, estimated stenosis parameters and estimated ischemia parameters for a standardized myocardial segmental map, which has 17 heart segments, including the left main artery (LMA), the proximal left circumflex artery (Prox LCX), the middle left circumflex artery (Mid LCX), the distal left coronary artery (Dist LCX), the left posterior atrioventricular (LPAV), the first obtuse marginal branch (OM1), the second obtuse marginal branch (OM2), the third obtuse marginal branch (OM3), the proximal left anterior descending artery (Prox LAD), the middle left anterior descending artery (Mid LAD), the distal left anterior descending artery (Dist LAD), the first diagonal branch of the left anterior descending artery (LAD D1), the second diagonal branch of the left anterior descending artery (LAD D2), the proximal right coronary artery (Prox RCA), the middle right coronary artery (Mid RCA), the distal right coronary artery (Dist RCA) and the acute marginal branch of the right posterior descending artery (AcMR PDA). In Table 1, the parameters used for myocardial ischemia estimation, stenosis identification, and / or fractional flow reserve estimation are shown in a range of 0 to 1. Other scales or ranges may be used.
[0084] Example nested function for generating fractional flow reserve (FFR) estimates
[0085] Tables 2-5 show example non-linear functions for generating FFR estimates corresponding to several segments of major blood vessels in the heart. In Table 2, an example function for determining an FFR estimate for the left aorta is provided ("FFR_LEFTMAIN").
[0086] Table 2
[0087]
[0088] As shown in Table 2, the FFR estimate of the left aorta is determined based on the extracted metrics and variables such as the Z component parameter ("noisevectorRz") associated with the noise subspace 706, the Alphahull ratio parameter ("Alpharatio"), and the signal density cloud volume 4 ("DensityV4").
[0089] In Table 3, an example function for determining an FFR estimate for the mid-right coronary artery ("FFR_MIDRCA") is provided.
[0090] Table 3
[0091]
[0092] As shown in Table 3, the FFR estimate of the mid-segment of the right coronary artery is determined based on the extracted metrics and variables such as the Y component parameter associated with the noise subspace 706 ("noisevectorRy"), the Alphahull ratio parameter ("Alpharatio"), and the signal density cloud volume 3 ("DensityV3").
[0093] In Table 4, an example function for determining an estimate of the FFR of the mid-left anterior descending artery ("FFR_MIDLAD") is provided.
[0094] Table 4
[0095]
[0096] As shown in Table 4, the FFR estimate of the middle segment of the left anterior descending artery is determined based on the extracted metrics and variables such as the volume to surface area ratio of cloud cluster 3 ("AspectRatio3") and the wavelet residual mean XYZ ("residueLevelMean").
[0097] In Table 5, an example function for determining an estimate of the FFR of the proximal left circumflex artery ("FFR_PROXLCX") is provided.
[0098] Table 5
[0099]
[0100] As shown in Table 5, the FFR estimate of the proximal left circumflex artery was determined based on the extracted metrics and variables such as the wavelet residual volume XYZ ("residualLevelVolume"), the vector cloud 6 volume ("vectorcloud 6"), and the signal density cloud volume 4 ("DensityV4").
[0101] Example Wavelet Cleaning Operator
[0102] See again Figure 7 , a wavelet operator 714 (shown as "wavelet cleaning" 714) can operate on the broadband gradient signal data 112 (or data derived therefrom). Those skilled in the art will appreciate that other intermediate phase linear processing can be performed on the signal data 112 before being operated on by the wavelet operator 714. In some embodiments, the wavelet operator 714 includes a biorthogonal wavelet 3.3 transform. Fig. 10A and 10Bis a diagram of an example wavelet transform (ie, biorthogonal wavelet 3.3) for generating a multidimensional wavelet-cleaned data set in accordance with an exemplary embodiment. Fig. 10A The decomposition scaling function φ is shown. Fig. 10B The decomposition wavelet function ψ is shown. Fig.11 is a diagram of an example output 1102 of a wavelet cleanup operation. The output 1102 is shown together with an input 1104 of the wavelet cleanup operation. In some embodiments, the output is a time series data set.
[0103] Still see Figure 7 , the output of the wavelet operator 714 is combined and transformed by a phase space transform 718 to produce a multidimensional wavelet cleaned data set 716. Feature topological analysis (also shown in box 718) is performed on the multidimensional wavelet cleaned data set 716 to extract metrics and variables 712a. In some embodiments, the extracted metrics and variables 712a include morphological, topological or functional features of the multidimensional wavelet cleaned data set, including, for example, 3D volume values, void volume values, surface area values, principal curvature direction values and Betti values. In some embodiments, the multidimensional wavelet cleaned data set can be segmented or divided into sub-regions, and the metrics and variables of these sub-regions are extracted for these sub-regions. In some embodiments, the void volume value, surface area value, principal curvature direction value and Betti value are also determined for each sub-region. In some embodiments, the number of sub-regions (also referred to as the number of segments) generated is between about 2 and about 20 (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20). In some embodiments, the number of sub-regions is greater than 20.
[0104] Figure 8 800 is a diagram of a method 800 for performing feature topology analysis of a multidimensional wavelet cleaned data set. In some embodiments, the method 800 can be similarly performed on other data sets, such as those obtained by Figure 7 Each of the phase space analyses (eg, residual subspace analysis and noise subspace analysis) generates a multidimensional noise subspace dataset 712b and a multidimensional wavelet cleaned dataset 712c. Figure 8As shown, in some embodiments, morphological, topological or functional feature extraction analysis (shown as "Topological Analysis" 802) includes using an alpha-hull operator 804 to calculate 3D volumes, voids and surface areas of at least one period of a multidimensional wavelet-cleaned data set as a spatiotemporal data set. In some embodiments, the alpha-hull operator uses a static alpha radius. Further details of the alpha-hull operator are described in Edelsbrunner et al., "Three-dimensional alpha shapes," ACM Transactions on Graphics, Vol. 13(1): 43-72 (1994), the entire contents of which are incorporated herein by reference. Other topological or geometric encapsulation operations may be used, including but not limited to, for example, Delaunay triangulation. Delaunay triangulation is a triangulation of a set of points such that no point in the triangulation is within the circumscribed circle of any triangle and the minimum angle of all angles in each triangle in the triangulation is maximized.
[0105] Still see Figure 8 , the generated multidimensional output of alpha-hull operator (operator) 806 can be further extracted to calculate (operator 808) main direction, curvature direction, Betti number and Betti value. In addition, the data set as the spatiotemporal data set is further segmented (via operator 810) into sub-regions, and the volume, surface area and aspect ratio of these sub-regions are also calculated using alpha-hull operator 812. As shown in the figure, the spatiotemporal data set is divided into 5 regions, 8 regions and 12 regions, wherein the volume, surface area and aspect ratio parameters are calculated for some regions in all these regions. In certain embodiments, three groups of regions including 25 regions can generate 3 parameters for each region to provide 75 metrics or variables 712a. In combination with the volume, surface area and aspect ratio of the calculated alpha-hull output and the main direction, curvature direction, Betti number and its Betti value, there can be 82 metrics or variables 712a. Metrics and variables 712a can be provided as a matrix (shown as "topological matrix" 814).
[0106] It should be understood that other topological features may be extracted in addition to or in lieu of those discussed herein. These features may include properties such as energy, area variation, or geometric features such as size.
[0107] It should be understood that other metrics and variables may be extracted and used depending on the number of operations performed, and that the examples provided herein are for illustrative purposes only.
[0108] Example residual subspace analysis and topology extraction
[0109] See again Figure 7 , note the second phase space analysis, which is performed to determine the metrics and variables 712b of the multidimensional residual subspace data set 704.
[0110] Fig.12 According to an exemplary embodiment, for generating Figure 7 FIG. 1200 is a diagram of an example wavelet-based operation 1202 of a multidimensional residual subspace dataset 704. Figure 7 and Fig.12 As shown in FIG. , a multidimensional residual subspace dataset 704 is generated as the residuals (eg, Fig.12 1202). The first wavelet operator may be, for example, a wavelet cleanup 714 using a biorthogonal wavelet 3.3 operator. The second wavelet operator may be an inverse biorthogonal wavelet 3.7 operator 720. Fig.13A and 13B is a diagram of an example wavelet transform (ie, inverse biorthogonal wavelet 3.7) for generating a multi-dimensional residual subspace data set in accordance with an exemplary embodiment. Fig.13A The decomposition scaling function φ is shown. Fig. 13B The decomposition scaling function ψ is shown. It should be possible to use other phase linear wavelet operators.
[0111] Still see Figure 7 , each residual output of the wavelet operator 714 and the wavelet operator 720 for each gradient signal is combined and transformed by phase space transformation to generate a multidimensional residual subspace data set 704. Feature topological analysis (also shown in box 722) is performed on the multidimensional wavelet residual data set to extract metrics and variables 712b. The extracted metrics and variables 712b may include morphological, topological or functional features of the multidimensional wavelet residual data set, including, for example, 3D volume values, void volume values, surface area values, principal curvature direction values and Betti numerical values. In some embodiments, the multidimensional wavelet cleaned data set may be segmented or divided into sub-regions, and the metrics and variables of these sub-regions are extracted for these sub-regions. In some embodiments, the void volume value, surface area value, principal curvature direction value and Betti numerical value are also determined for each sub-region. In some embodiments, the number of sub-regions (also referred to as the number of segments) generated is between about 2 and about 20 (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20). In some embodiments, the number of sub-regions is greater than 20. In some embodiments, the above-described method can be performed. Figure 8 Similar or identical topology extraction analysis as described.
[0112] Fig.14 704 is a depiction of an example residual subspace 704 obtained by subtracting a wavelet model (e.g., 714 and 720) using biorthogonal 3.3 and inverse biorthogonal 3.7 operations. The residual subspace represents those parts of the biosignal that are actually too complex and nonlinear to fit (i.e., represent) a single wavelet function. The residual subspace is processed, converted into representative features, and used to study the dynamic and geometric properties of cardiac gradient data.
[0113] Fig.16 is a depiction of an example dynamic phase space volume object that has been colored by the residual subspace. The phase space volume object is generated by superimposing the values of the residual subspace as a color intensity map onto the input broadband gradient signal data. The lack of intense coloring indicates the absence of ischemic myocardial tissue. That is, Fig.16 is an example dynamic phase space volume object for a healthy person.
[0114] Fig.17 1704 is a depiction of an example dynamic phase space volume object associated with an ischemic patient. That is, the dynamic phase space volume object is generated using broadband gradient signal data of a patient diagnosed with ischemic myocardium. The dynamic phase space volume object has been colored by the residual subspace by superimposing the values of the residual subspace as a color intensity map onto the input broadband gradient signal data. Intense coloring (corresponding to arrow 1702) indicates the presence of ischemic myocardium.
[0115] Example Noise Subspace Analysis and Topology Extraction
[0116] See again Figure 7 , note the third phase space analysis, which is performed to determine the metrics and variables 712c of the multidimensional noise subspace data set 706. Figure 7As shown, the multidimensional noise subspace data set 706 can be calculated by subtracting the input broadband gradient signal data 112 (or a data set derived therefrom) and the output of the wavelet cleaned signal data 716 by a subtraction operator 724. The output of the subtraction operation is combined and transformed by a phase space transform to produce a multidimensional residual subspace data set 706. Feature topological analysis (also shown in box 726) is performed on the multidimensional noise subspace data set to extract metrics and variables 712c. The extracted metrics and variables 712c may include morphological, topological or functional features of the multidimensional wavelet residual data set, including, for example, 3D volume values, void volume values, surface area values, principal curvature direction values and Betti values. In some embodiments, the multidimensional wavelet cleaned data set may be segmented or divided into sub-regions, and the metrics and variables of these sub-regions are extracted for these sub-regions. In some embodiments, the void volume value, surface area value, principal curvature direction value and Betti value are also determined for each sub-region. In some embodiments, the number of sub-regions (also referred to as the number of segments) generated is between 2 and about 20 (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20). In some embodiments, the number of sub-regions is greater than 20. In some embodiments, the above-described method can be performed. Figure 8 Similar or identical topology extraction analysis as described
[0117] Fig.15 is a depiction of the noise subspace 706, which is the result of subtracting the biorthogonal 3.3 wavelet model (e.g., 714) from the input broadband gradient data 112. Similar to the residual subspace 704, it contains complex dynamic information. Specifically, the noise subspace contains chaotic information that cannot be effectively captured in the model. The noise subspace is processed, converted into representative features, and used to study the dynamic and geometric properties of the cardiac gradient data.
[0118] Fig.18 is a depiction of an example noise subspace phase space object that has been colored by the noise subspace. The phase space object is generated by superimposing the values of the noise subspace as a color intensity map onto a derivative transform (e.g., numerical fractional derivatives) of the input broadband gradient signal data (or its derivative data). Fig.18 As shown, the lack of intense staining indicates the absence of ischemic myocardium. Fig.18 is an example noise subspace phase space object for a healthy person.
[0119] Fig.191904 is a depiction of an example noise subspace phase space object associated with an ischemic patient. The phase space object has been colored by the noise subspace by superimposing the values of the noise subspace as a color intensity map onto a derivative transform (e.g., numerical fractional derivative) of the input broadband gradient signal data (or derived data). Intense coloring (corresponding to arrow 1902) indicates the presence of ischemic myocardium.
[0120] Example Machine Learning Analysis
[0121] Fig. 9 is a diagram of a method of performing machine learning analysis to create and select a nonlinear model to identify and / or estimate the extent of myocardial ischemia, identify one or more stenoses, and / or locate and / or estimate fractional flow reserve, according to an exemplary embodiment, as described with respect to Figure 8 As described. Fig. 9 As shown, angiographic dataset 902 and blood flow fraction dataset 904 are used (via operation 906) to create a candidate nonlinear model to identify stenosis (and / or estimate its extent) and estimate blood flow reserve fraction 914. Examples of the generation of noninvasive models for, for example, estimating cardiac chamber size and mechanical function are described, for example, in U.S. Application No. 14 / 295,615, entitled “Noninvasive electrocardiographic method for estimating mammalian cardiacchamber size and mechanical function,” the entire contents of which are incorporated herein by reference.
[0122] In some embodiments, a machine learning algorithm 908 is then used to select a family of nonlinear models 910 from candidate nonlinear models using broadband gradient cardiac signal data 912 of patients or subjects with a certain degree of stenosis and ischemia. In some embodiments, the machine learning algorithm is based on a regression random forest algorithm or a modified variation thereof. In some embodiments, the machine learning algorithm is based on a deep learning algorithm.
[0123] In some embodiments, the machine learning stage calls a meta-genetic algorithm to automatically select a subset of features extracted from a large pool. The feature subset is then used by the AdaBoost algorithm to generate a predictor to diagnose significant coronary artery disease in a patient population representing positive and negative cases. The performance of the candidate predictor is determined by validating a previously unseen patient pool. A further description of the AdaBoost algorithm is provided in Freund, Yoav and Robert E. Schapire's "A decision-theoretic generalization of on-line learning and an application to boosting," (published in the European Conference on Computational Learning Theory, Springer, Berlin, Heidelberg (1995), the entire contents of which are incorporated herein by reference.
[0124] In some embodiments, the spatiotemporal volume can be mapped to a complex phase space difference in 12-dimensional space. Non-Fourier integration that creates a 12-dimensional spatiotemporal density metric can be used to extract spatial variations in the phase space matrix. These metrics for the ventricles are modeled using a genetic algorithm to link 17 nonlinear nested sine-Gaussian equations for 17 segments of the ventricles in the coronary artery region as perfusion blockages. Perfusion images are scored visually using a 17-segment model of the left ventricle and a 5-point grade (0 = normal tracer uptake, 1 = mild reduction, 2 = moderate reduction, 3 = severe reduction, 4 = no uptake). The amount of ischemic myocardial tissue (IM) is calculated as the total difference score (the difference between the summed stress score and the total rest score) divided by 80. Patients are classified as: no ischemia or suspected (IM<5%), mild ischemia (5%≤IM<10%), and moderate / severe ischemia (IM≥10%). The output of these equations provides the amount and location of ischemic myocardial tissue.
[0125] In some embodiments, a modified matching pursuit (MMP) algorithm is used to operate on the broadband biopotential data 112 to create a sparse mathematical model. Details of the MMP algorithm are provided in Mallat et al., "Matching Pursuits with Time-Frequency Dictionaries", IEEE Transactions on Signal Processing, Vol. 41(12), pp. 3397-2415 (1993), the entire contents of which are incorporated herein by reference.
[0126] The features of the model including residual quantization may be included in the feature set. In feature extraction operation 706 ( Figure 7 ) can be extracted from the model to determine the geometric and dynamic characteristics of the model. These subspaces can include, but are not limited to, complex subharmonic frequency (CSF) trajectories, quasi-periodic and chaotic subspaces, low / high energy subspaces, and fractional derivatives of low / high energy subspaces. These subspaces are examples of subspace families that characterize the dynamics of a system, whether pathological or normal.
[0127] Fig. 20 is a diagram of a method of visualizing estimated arterial blood flow characteristics in a heart according to an exemplary embodiment. Fig. 20 As shown, the visualization engine 2002 receives the determined arterial blood flow characteristics and presents the characteristics on a 3D visualization output. In some embodiments, the visualization engine 2002 provides a system-level view of all arterial blood flow characteristics and their interactions in a graphical user interface (GUI). In some embodiments, the GUI presents the cascading effects of upstream modifications of arterial blood flow on downstream circulation.
[0128] Experimental data
[0129] A coronary artery disease learning and formulation development study conducted under a clinical protocol acquired resting phase signals from human subjects prior to coronary angiography. The acquired signals were evaluated using the non-invasive acquisition and analysis methods described herein to detect the presence or absence of significant coronary artery disease in symptomatic adult patients or subjects. In addition, the acquired signals were evaluated to assess left ventricular ejection fraction and identify the location of significant coronary artery disease. A comparative paired trial design was used to evaluate the performance of the non-invasive acquisition and analysis methods described herein, and the results are shown in FIG. Fig.21 and Fig. 22 shown.
[0130] Further description of this clinical protocol is provided in U.S. Provisional Application No. 62 / 340,410, entitled “Method and System for Collecting Phase Signals for PhaseSpace Tomography Analysis,” the entire contents of which are incorporated herein by reference.
[0131] Fig.21 and 22is a graph showing the results of the study conducted on 523 human subjects according to an exemplary embodiment. The data presented relate to a prospective, non-randomized trial to improve the non-invasive acquisition and analysis methods described herein to detect and assess significant coronary artery disease (CAD) (i.e., ≥70% stenosis or <0.80 reduced blood flow velocity fraction) during catheterization using paired phase signals and assessed clinical outcome data.
[0132] In the data presented, a dataset of 429 subjects (out of 523) was used as a training dataset, and a dataset of 94 subjects was used as a validation population to assess the sensitivity and specificity of the non-invasive acquisition and analysis methods described herein. Fig.21 ), which provided a ROC curve of 0.80, with a positive predictor value (PPV) of 47% and a negative predictor value (NPV) of 96% compared with angiographic results. Fig. 22 ), which provided a ROC curve of 0.78 with a positive predictor value (PPV) of 49% and a negative predictor value (NPV) of 92% compared with angiographic findings. Candidate predictors A and B are internal parameters used in the machine training process (e.g., training a classifier).
[0133] Compared to the diagnostic performance of non-invasive myocardial perfusion imaging for detecting obstructive coronary artery disease using single photon emission computed tomography, cardiac magnetic resonance, and positron emission tomography as published in J. Am. Coll. Cardiol. 8: 59 (19), 1719-28 (May 2012) (shown as "SPECT"), the non-invasive acquisition and analysis methods described herein (shown as "cPSTA") performed quite well. These solutions routinely achieved AUC-ROC scores greater than 0.7 in the validation phase, performing as well or better than previous manually guided methods. Table 6 below shows the diagnostic performance between the above studies and the studies herein.
[0134] Table 6
[0135]
[0136]
[0137] Having thus described several embodiments of the present disclosure, it will be apparent to those skilled in the art that the foregoing detailed disclosure is intended to be presented by way of example only and not by way of limitation. Many advantages of non-invasive methods and systems for locating abnormalities in the heart have been discussed herein. Although not explicitly stated herein, various changes, improvements, and modifications will occur and are contemplated for those skilled in the art. These changes, improvements, and modifications are intended to be proposed hereby and fall within the spirit and scope of the present disclosure. In addition, the order of the described processing elements or processing sequences, or the use of numbers, letters, or other names, is not intended to limit the claimed processing to any order unless expressly specified in the claims. Therefore, the present disclosure is limited only by the appended claims and their equivalents.
[0138] For example, additional examples of phase space processing that can be used with the example methods and systems are described in the following documents: U.S. Provisional Patent Application No. 62 / 184,796, entitled “Latent teratogen-induced heart deficits are unmasked postnatally with mathematical analysis and machine learning on ECG signals”; U.S. Patent Application No. 15 / 192,639, entitled “Methods and Systems Using Mathematical Analysis and Machine Learning to Diagnose Disease”; U.S. Patent Application No. 14 / 620,388, published as US2015 / 0216426, entitled “Method and system for characterizing cardiovascular systems from single channel data”; U.S. Patent Application No. 14 / 596,541, issued as US9,597,021, entitled “Noninvasive method for estimating glucose, glycosylated hemoglobin and other blood constituents”; U.S. patent application number 14 / 077,993, publication number US2015 / 0133803, and invention name “Noninvasive electrocardiographic method for estimating mammalian cardiac chamber size and mechanical function”; U.S. patent application number 14 / 295,615, and invention name “Noninvasive electrocardiographic method for estimating mammalian cardiac chamber size and mechanical function”;U.S. Patent Application No. 13 / 970,582, Publication No. US9,408,54, “Non-invasive method and system for characterizing cardiovascular systems and all-cause mortality and sudden cardiac death risk”; U.S. Patent Application No. 15 / 061,090, Publication No. US2016 / 0183822, “Non-invasive method and system for characterizing cardiovascular systems”; U.S. Patent Application No. 13 / 970,580, Publication No. US9,289,150, “Non-invasive method and system for characterizing cardiovascular systems”; U.S. Patent Application No. 62 / 354,668, “Method and System for Phase Space Analysis to Determine Arterial Flow Characteristics”; and U.S. Patent Application No. 61 / 684,217, “Non-invasive method and system for characterizing cardiovascular systems” systems", each of which is incorporated herein by reference in its entirety.;
[0139] Although the method and system have been described in conjunction with preferred embodiments and specific examples, it is not intended to limit the scope to the particular embodiments set forth, as the embodiments herein are intended in all respects to be illustrative rather than restrictive.
[0140] For example, the exemplary methods and systems may be used to generate stenosis and FFR outputs for use with an interventional system configured to use the FFR / stenosis outputs to determine and / or modify a plurality of stents and their in vivo placement procedures.
[0141] Unless otherwise expressly stated, it is not intended that any method described herein be construed as requiring that its steps be performed in a specific order. Therefore, in the event that a method claim does not actually recite the order in which its steps are to be followed, or in the event that the steps are not specifically stated in the claim or specification to be limited to a specific order, no order is intended to be inferred in any way. This applies to any possible non-explicit basis for interpretation, including: logical issues regarding the arrangement of steps or operational flows; plain meaning derived from grammatical organization or punctuation; the number or type of embodiments described in the specification.
[0142] Throughout this application, various publications are referenced. The disclosures of these publications are hereby incorporated by reference in their entireties into this application in order to more fully describe the state of the art to which the methods and systems pertain.
[0143] Various modifications and variations will be apparent to those skilled in the art.
Claims
1. A system for detecting one or more conditions in a human subject, comprising: an analysis system configured with computer-readable instructions to acquire an electrophysiological signal dataset, the electrophysiological signal dataset (i) associated with the subject and (ii) having been acquired by a measurement device configured to non-invasively acquire the signal dataset as a long cardiac gradient recording exhibiting complex nonlinear variations via one or more sensors, the analysis system using the signal dataset to detect the one or more pathologies of the subject, The detection operations include: generating, by a processor performing a transformation operation, a residual subspace dataset and a noise subspace dataset based on the broadband gradient cardiac signal data in the signal dataset, wherein characteristic topological analysis is performed on the residual subspace dataset and the noise subspace dataset using an alpha-hull operator or Delauney triangulation; and Extracting, by the processor, one or more geometric features and / or dynamic characteristics of an alpha shape or a Delaunay triangle mesh from the residual subspace dataset and the noise subspace dataset through the feature topology analysis, wherein the geometric features and / or dynamic characteristics include at least one of the following: The 3D volume value of the alpha shape or Delaunay triangle mesh, The void volume value of the alpha shape or Delaunay triangle mesh, The surface area value of the alpha shape or Delaunay triangle mesh, the principal curvature direction values of the alpha shape or Delaunay triangle mesh, and The Betti values of the alpha shape or Delaunay triangle mesh, One or more geometric features and / or dynamic characteristics of the extracted alpha shape or Delaunay triangle mesh are used as a variable representing the subject in a machine learning operation to detect the one or more conditions of the subject.
2. The system according to claim 1, wherein: The residual subspace dataset and the noise subspace dataset are generated in the following manner: A phase space model of the subject's heart is generated by the processor by performing a multi-dimensional transformation on the signal dataset to derive phase energy of the thoracic system and reconstructing it. 3 . The system of claim 1 , wherein the signal dataset comprises cardiac data acquired through two or more channels, each biopotential signal being acquired over multiple heart beats as the long cardiac gradient record.
4. The system of claim 2, wherein the signal dataset is acquired for at least 210 seconds, and wherein a portion of the long cardiac gradient recording is modeled to generate the residual subspace dataset and the noise subspace dataset.
5. The system of claim 1, wherein the signal data set comprises a plurality of broadband gradient signals acquired simultaneously through a plurality of channels, the plurality of channels comprising two or more channels, wherein a time lag or deviation between each of the channels is less than 10 femtoseconds. 6 . The system of claim 5 , wherein each of the plurality of broadband gradient signals is unfiltered prior to and during the transform operation to generate the signal data set. 7 . The system of claim 5 , wherein each of the plurality of broadband gradient signals comprises cardiac data having frequency components greater than 1 kHz in the frequency domain.
8. The system of claim 5, wherein each of the plurality of broadband gradient signals comprises cardiac data in the frequency domain having frequency components up to 8 kHz.
9. The system of claim 1, wherein the one or more sensors comprise at least one of a surface electrode and a non-contact electrode.
10. The system of claim 1, wherein the machine learning operation comprises an artificial neural network algorithm or a regression random forest algorithm.
11. The system of claim 1, wherein the measurement device is configured to transmit the signal data set to the analysis system via a network.
12. The system of claim 11, further comprising: A cloud-based repository, wherein the measurement device is configured to transmit the signal dataset to the cloud-based repository over a network, wherein the analysis system is configured to retrieve the acquired signal dataset from the cloud-based repository for analysis.
13. A method for detecting one or more conditions in a human subject, comprising: acquiring, by a processor, from a storage area network, an electrophysiological signal dataset, the electrophysiological signal dataset (i) associated with a human subject and (ii) having been acquired by a measurement device configured to non-invasively acquire the signal dataset as a long cardiac gradient recording exhibiting complex nonlinear variations via one or more sensors, detecting one or more conditions in the subject using the signal dataset, wherein detecting comprises: generating, by a processor performing a transformation operation, a residual subspace dataset and a noise subspace dataset based on the broadband gradient cardiac signal data in the signal dataset, wherein characteristic topological analysis is performed on the residual subspace dataset and the noise subspace dataset using an alpha-hull operator or Delauney triangulation; and Extracting, by the processor, one or more geometric features and / or dynamic characteristics of an alpha shape or a Delaunay triangle mesh from the residual subspace dataset and the noise subspace dataset through the feature topology analysis, wherein the geometric features and / or dynamic characteristics include at least one of the following: The 3D volume value of the alpha shape or Delaunay triangle mesh, The void volume value of the alpha shape or Delaunay triangle mesh, The surface area value of the alpha shape or Delaunay triangle mesh, the principal curvature direction values of the alpha shape or Delaunay triangle mesh, and The Betti values of the alpha shape or Delaunay triangle mesh, One or more geometric features and / or dynamic characteristics of the extracted alpha shape or Delaunay triangle mesh are used as a variable representing the subject in a machine learning operation to detect the one or more conditions of the subject.
14. The method according to claim 13, wherein The residual subspace dataset and the noise subspace dataset are generated in the following manner: A phase space model of the subject's heart is generated by the processor by performing a multi-dimensional transformation on the signal dataset to derive phase energy of the thoracic system and reconstructing it.
15. The method of claim 13, wherein the signal dataset comprises cardiac data acquired through at least one channel, each biopotential signal being acquired over multiple heart beats as the long cardiac gradient record.
16. The method of claim 14, wherein the electrophysiological signal dataset is acquired for at least 210 seconds, and wherein a portion of the long cardiac gradient recording is modeled to generate the residual subspace dataset and the noise subspace dataset.
17. The method of claim 13, wherein the signal dataset comprises a plurality of broadband gradient signals acquired simultaneously through a plurality of channels, the plurality of channels including the at least one channel, wherein a time lag or deviation between each of the channels is less than 10 femtoseconds.
18. The method of claim 17, wherein each of the plurality of broadband gradient signals is unfiltered prior to and during the transform operation to generate the signal data set.
19. The method of claim 17, wherein each of the plurality of broadband gradient signals comprises cardiac data having frequency components greater than 1 kHz in the frequency domain.
20. The method of claim 17, wherein each of the plurality of broadband gradient signals comprises cardiac data in the frequency domain having frequency components up to 8 kHz.
21. A method for non-invasively identifying and / or measuring or estimating the extent of myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve, the method comprising: acquiring, by one or more processors, a data set associated with a plurality of broadband gradient signals from a subject simultaneously via at least one electrode; generating, by the one or more processors, a three-dimensional phase space dataset from a dataset associated with the plurality of broadband gradient signals, the three-dimensional phase space dataset comprising a residual subspace dataset and a noise subspace dataset; extracting, by the one or more processors, a first set of morphological features from the three-dimensional phase space dataset using an alpha-hull operator, wherein the extracted first set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti number value; determining, by the one or more processors, one or more coronary artery physiological parameters of the subject selected from the group consisting of: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate, wherein the one or more coronary artery physiological parameters are determined by inputting the first set of morphological features into one or more machine learning models; and The one or more coronary physiological parameters are output via one or more of the output data sets in a report and / or display, wherein the output is used to diagnose the heart disease.
22. The method according to claim 21, wherein The residual subspace dataset is determined by: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; generating a second wavelet signal data set by performing a second wavelet operation on the first wavelet signal data; and subtracting the values of the first wavelet signal dataset from the values of the second wavelet signal dataset to generate the residual subspace dataset, The residual subspace data set is in the spatiotemporal domain.
23. The method of claim 21, further comprising: dividing the three-dimensional phase space data set into a plurality of segments, each segment comprising a non-overlapping portion of the three-dimensional phase space data set; and A second set of morphological features is extracted for each of the plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value. The method of claim 23 , wherein the extracted second set of morphological features is extracted using the alpha-hull operator.
25. The method of claim 23, wherein the plurality of segments comprises a number of segments selected from the group consisting of 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20.
26. The method of claim 21, wherein: The noise subspace dataset is determined by: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; and generating a second wavelet signal dataset by performing a second wavelet operation on the first wavelet signal dataset, wherein the second wavelet signal dataset includes the noise subspace dataset; The noise subspace dataset is in the spatiotemporal domain.
27. The method of claim 21, wherein: The residual subspace dataset is associated with a first noise geometric profile of a first shape, and wherein the noise subspace dataset is associated with a second noise geometric profile corresponding to a second shape of the noise.
28. The method of claim 21, wherein the plurality of broadband gradient signals are acquired simultaneously, the plurality of broadband gradient signals having a lag or deviation of less than 10 femtoseconds between each signal.
29. The method of claim 21, wherein each of the plurality of broadband gradient signals is not filtered by any nonlinear phase operation before and during processing to generate the residual subspace dataset and the noise subspace dataset.
30. The method of claim 21, wherein each of the plurality of broadband gradient signals comprises cardiac data in the frequency domain having frequency components greater than 1 kHz.
31. The method of claim 21, wherein each of the plurality of broadband gradient signals comprises heart rate information at a frequency selected from the group consisting of: 1 kHz, 2 kHz, 3 kHz, 4 kHz, 5 kHz, 6 kHz, 7 kHz, 8 kHz, 9 kHz, and 10 kHz.
32. The method of claim 21, wherein each of the plurality of broadband gradient signals comprises heart rate information at a frequency between 0 Hz and 50 kHz.
33. The method of claim 21, wherein each of the plurality of broadband gradient signals comprises heart rate information having a frequency between 0 Hz and 500 kHz.
34. The method of claim 21, further comprising: associating the extracted first set of morphological features with a plurality of candidate models associated with the determined one or more coronary artery physiological parameters through a first machine learning operation; as well as A candidate model from the plurality of candidate models is selected to determine an output for each of the estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate.
35. The method of claim 34, wherein the determined one or more coronary artery physiological parameters are presented at one or more corresponding coronary artery regions on the cardiac image.
36. The method of claim 21, further comprising: The determined one or more coronary artery physiological parameters are output to a surgical device or a diagnostic device.
37. The method of claim 21, wherein one or more of the at least one electrode is selected from the group consisting of a surface electrode and a non-contact electrode.
38. The method of claim 21, further comprising the step of identifying one or more significant stenoses, said identification having an AUC-ROC score greater than 0.7 during a validation phase.
39. A system for non-invasively identifying and / or measuring or estimating the extent of myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve, the system comprising: at least one processor; and A memory having instructions stored thereon, wherein when the instructions are executed by the processor, the processor performs the following operations: A dataset of multiple broadband gradient signals is acquired simultaneously from multiple electrodes; generating a three-dimensional phase space dataset based on the dataset of the plurality of broadband gradient signals, the three-dimensional phase space dataset comprising a residual subspace dataset and a noise subspace dataset; extracting a first set of morphological features from the three-dimensional phase space dataset using an alpha-hull operator, wherein the extracted first set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti number value; determining one or more coronary artery physiological parameters selected from the group consisting of: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate, wherein the one or more coronary artery physiological parameters are determined by inputting the first set of morphological features into one or more machine learning models; and The one or more coronary physiological parameters are output via one or more of the output data sets in a report and / or display, wherein the one or more output coronary physiological parameters are used to diagnose a heart disease.
40. The system of claim 39, wherein: When the instruction is executed by the at least one processor, the at least one processor will further perform the following operations: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; generating a second wavelet signal data set by performing a second wavelet operation on the first wavelet signal data; and subtracting the values of the first wavelet signal dataset from the values of the second wavelet signal dataset to generate the residual subspace dataset, The residual subspace data set is in the spatiotemporal domain.
41. The system of claim 39, wherein the instructions, when executed by the at least one processor, further cause the at least one processor to: dividing the three-dimensional phase space data set into a plurality of segments, each segment comprising a non-overlapping portion of the three-dimensional phase space data set; and A second set of morphological features is extracted for each of the plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value.
42. The system of claim 39, wherein the instructions, when executed by the at least one processor, further cause the at least one processor to: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; and generating a second wavelet signal dataset by performing a second wavelet operation on the first wavelet signal dataset, wherein the second wavelet signal dataset includes the noise subspace dataset; in, The noise subspace dataset is in the spatiotemporal domain.
43. The system of claim 39, wherein: The residual subspace dataset is associated with a first noise geometric profile of a first shape, and wherein the noise subspace dataset is associated with a second noise geometric profile corresponding to a second shape of the noise.
44. The system of claim 39, wherein the system simultaneously acquires the plurality of broadband gradient signals, the plurality of broadband gradient signals having a lag or deviation of less than 10 femtoseconds between each signal.
45. The system of claim 39, wherein each of the plurality of broadband gradient signals is not filtered by any nonlinear phase operation before and during processing to generate the residual subspace dataset and the noise subspace dataset.
46. The system of claim 39, wherein each of the plurality of broadband gradient signals comprises cardiac data in the frequency domain having frequency components greater than 1 kHz.
47. The system of claim 39, wherein the instructions, when executed by the at least one processor, further cause the at least one processor to: associating, by a first machine learning operation, the extracted first set of morphological features with a plurality of candidate models associated with estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate; and A candidate model from the plurality of candidate models is selected to determine an output for each of the estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate.
48. The system of claim 47, wherein the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate are presented at one or more corresponding coronary artery regions on a cardiac image.
49. The system of claim 39, wherein the instructions, when executed by the at least one processor, further cause the at least one processor to: The determined one or more coronary artery physiological parameters are output to a surgical device or a diagnostic device.
50. A non-transitory computer-readable medium having instructions stored thereon, wherein execution of the instructions causes a processor to: acquiring a data set of multiple broadband gradient signals acquired simultaneously from multiple electrodes; generating a three-dimensional phase space dataset based on the dataset of the plurality of broadband gradient signals, the three-dimensional phase space dataset comprising a residual subspace dataset and a noise subspace dataset; extracting a first set of morphological features from the three-dimensional phase space dataset using an alpha-hull operator, wherein the extracted first set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti number value; determining one or more coronary artery physiological parameters selected from the group consisting of: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate, wherein the one or more coronary artery physiological parameters are determined by inputting the first set of morphological features into one or more machine learning models; and The one or more coronary physiological parameters are output via one or more of the output data sets in a report and / or display, wherein the one or more output coronary physiological parameters are used to diagnose a heart disease.
51. The computer-readable medium of claim 50, wherein: When the instructions are executed by the processor, the processor will also perform the following operations: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; generating a second wavelet signal data set by performing a second wavelet operation on the first wavelet signal data; and subtracting the values of the first wavelet signal dataset from the values of the second wavelet signal dataset to generate the residual subspace dataset, The residual subspace data set is in the spatiotemporal domain.
52. The computer-readable medium of claim 50, wherein the instructions, when executed by the processor, further cause the processor to: dividing the three-dimensional phase space data set into a plurality of segments, each segment comprising a non-overlapping portion of the three-dimensional phase space data set; and A second set of morphological features is extracted for each of the plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value.
53. The computer-readable medium of claim 50, wherein the instructions, when executed by the processor, further cause the processor to: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; and generating a second wavelet signal dataset by performing a second wavelet operation on the first wavelet signal dataset, wherein the second wavelet signal dataset includes the noise subspace dataset; in, The noise subspace dataset is in the spatiotemporal domain.
54. The computer-readable medium of claim 50, wherein: The residual subspace dataset is associated with a first noise geometric profile of a first shape, and wherein the noise subspace dataset is associated with a second noise geometric profile corresponding to a second shape of the noise.
55. The computer-readable medium of claim 50, wherein the plurality of broadband gradient signals are acquired simultaneously with a lag or offset of less than 10 femtoseconds between each signal.
56. The computer-readable medium of claim 50, wherein each of the plurality of wideband gradient signals is not filtered by any nonlinear phase operation before and during processing to generate the residual subspace dataset and the noise subspace dataset.
57. The computer-readable medium of claim 50, wherein each of the plurality of broadband gradient signals comprises cardiac data in the frequency domain having frequency components greater than 1 kHz.
58. The computer-readable medium of claim 50, wherein the instructions, when executed by the processor, further cause the processor to: associating, by a first machine learning operation, the extracted first set of morphological features with a plurality of candidate models associated with estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate; and A candidate model from the plurality of candidate models is selected to determine an output for each of the estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate.
59. The computer-readable medium of claim 58, wherein the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate are presented at one or more corresponding coronary artery regions on a cardiac image.
60. The computer-readable medium of claim 50, wherein the instructions, when executed by the processor, further cause the processor to: The determined one or more coronary artery physiological parameters are output to a surgical device or a diagnostic device.
61. A method for non-invasively identifying and / or measuring or estimating the extent of myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve, the method comprising: simultaneously acquiring a plurality of broadband gradient signals from a subject via at least one electrode; as well as determining, by one or more processors, one or more coronary artery physiological parameters of the subject selected from the group consisting of: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate based on a three-dimensional phase space dataset derived from data associated with the plurality of broadband gradient signals, wherein the three-dimensional phase space dataset comprises a residual subspace dataset and a noise subspace dataset, and a first set of morphological features is extracted from the three-dimensional phase space dataset using an alpha-hull operator, wherein the extracted first set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value; The one or more coronary artery physiological parameters are determined by inputting the first set of morphological features into one or more machine learning models.
62. The method of claim 61, wherein The residual subspace dataset is determined by: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; generating a second wavelet signal data set by performing a second wavelet operation on the first wavelet signal data; and subtracting the values of the first wavelet signal dataset from the values of the second wavelet signal dataset to generate the residual subspace dataset, The residual subspace data set is in the spatiotemporal domain.
63. The method of claim 62, further comprising: dividing the three-dimensional phase space data set into a plurality of segments, each segment comprising a non-overlapping portion of the three-dimensional phase space data set; and A second set of morphological features is extracted for each of the plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value.
64. The method of claim 63, wherein the extracted second set of morphological features is extracted using an alpha-hull operator.
65. The method of claim 63, wherein the plurality of segments comprises a number of segments selected from the group consisting of 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20.
66. The method of claim 61, wherein The noise subspace dataset is determined by: generating a first wavelet signal data set by performing a first wavelet operation on data derived from the plurality of broadband gradient signals; and generating a second wavelet signal dataset by performing a second wavelet operation on the first wavelet signal dataset, wherein the second wavelet signal dataset includes the noise subspace dataset; The noise subspace dataset is in the spatiotemporal domain.
67. The method of claim 66, further comprising: dividing the three-dimensional phase space data set into a plurality of segments, each segment comprising a non-overlapping portion of the three-dimensional phase space data set; as well as A second set of morphological features is extracted for each of the plurality of segments, wherein the extracted second set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value.
68. The method of claim 67, wherein the extracted second set of morphological features is extracted using an alpha-hull operator.
69. The method of claim 67, wherein the plurality of segments comprises a number of segments selected from the group consisting of 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, and 20.
70. The method of claim 61, wherein The residual subspace dataset is associated with a first noise geometric profile of a first shape, and wherein the noise subspace dataset is associated with a second noise geometric profile corresponding to a second shape of the noise.
71. The method of claim 61, wherein the plurality of broadband gradient signals are acquired simultaneously, the plurality of broadband gradient signals having a lag or deviation of less than 10 femtoseconds between each signal.
72. The method of claim 61, wherein each of the plurality of wideband gradient signals is unfiltered prior to and during processing to generate the residual subspace dataset and the noise subspace dataset.
73. The method of claim 61, wherein each of the plurality of broadband gradient signals comprises cardiac data in the frequency domain having frequency components greater than 1 kHz.
74. The method of claim 61, wherein each of the plurality of broadband gradient signals comprises heart rate information at a frequency selected from the group consisting of: 1 kHz, 2 kHz, 3 kHz, 4 kHz, 5 kHz, 6 kHz, 7 kHz, 8 kHz, 9 kHz, and 10 kHz.
75. The method of claim 61, wherein each of the plurality of broadband gradient signals comprises heart rate information at a frequency between 0 Hz and 50 kHz.
76. The method of claim 61, wherein each of the plurality of broadband gradient signals comprises heart rate information at a frequency between 0 Hz and 500 kHz.
77. The method of claim 61, further comprising: associating, by a first machine learning operation, the extracted first set of morphological features with a plurality of candidate models, the plurality of candidate models being associated with estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate; as well as A candidate model from the plurality of candidate models is selected to determine an output for each of the estimates of the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate.
78. The method of claim 61, further comprising: The determined one or more coronary artery physiological parameters are visualized.
79. The method of claim 61, wherein the fractional flow reserve estimate, the stenosis value, and the myocardial ischemia estimate are presented at one or more corresponding coronary artery regions on a cardiac image.
80. The method of claim 61 , further comprising: The determined one or more coronary artery physiological parameters are output to a surgical device or a diagnostic device.
81. The method of claim 61, wherein one or more of the at least one electrode is selected from the group consisting of a surface electrode and a non-contact electrode.
82. The method of claim 61, further comprising the step of identifying one or more significant stenoses, said identification having an AUC-ROC score greater than 0.7 during a validation phase.
83. A system for non-invasively identifying and / or measuring or estimating the extent of myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve, the system comprising: processor; as well as A memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause a computing device to perform the method of any one of claims 61-82.
84. A non-transitory computer-readable medium having instructions stored thereon, wherein the instructions, when executed by a processor, cause the processor to perform the method of any one of claims 61-82.
85. A system for non-invasively identifying and / or measuring or estimating the extent of myocardial ischemia, identifying one or more stenoses, and / or locating and / or estimating fractional flow reserve, the system comprising: processor; and A memory having instructions stored thereon, wherein execution of the instructions causes the processor to: Simultaneously acquiring multiple broadband gradient signals from at least one electrode; and determining one or more coronary artery physiological parameters selected from the group consisting of: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate based on a three-dimensional phase space data set derived from data associated with the plurality of broadband gradient signals, wherein the three-dimensional phase space dataset comprises a residual subspace dataset and a noise subspace dataset, and a first set of morphological features is extracted from the three-dimensional phase space dataset using an alpha-hull operator, wherein the extracted first set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value; The one or more coronary artery physiological parameters are determined by inputting the first set of morphological features into one or more machine learning models.
86. The system of claim 85, wherein execution of the instructions further causes the processor to: The determined coronary artery physiological parameters are visualized for presentation on a display.
87. The system of claim 85 or 86, wherein execution of the instructions further causes the processor to: The determined one or more coronary physiological parameters are output to a treatment device.
88. A computer-readable medium having instructions stored thereon, wherein execution of the instructions further causes a processor to: determining one or more coronary artery physiological parameters selected from the group consisting of: a fractional flow reserve estimate, a stenosis value, and a myocardial ischemia estimate based on a three-dimensional phase space dataset derived from data associated with a plurality of broadband gradient signals simultaneously acquired from at least one surface electrode, wherein the three-dimensional phase space dataset comprises a residual subspace dataset and a noise subspace dataset, and a first set of morphological features is extracted from the three-dimensional phase space dataset using an alpha-hull operator, wherein the extracted first set of morphological features comprises parameters selected from the group consisting of: a 3D volume value, a void volume value, a surface area value, a principal curvature direction value, and a Betti value; The one or more coronary artery physiological parameters are determined by inputting the first set of morphological features into one or more machine learning models.
89. The computer-readable medium of claim 88, wherein further execution of the instructions causes the processor to: The determined coronary physiological parameters are visualized for presentation on a display.
90. The computer-readable medium of claim 88 or 89, wherein execution of the instructions further causes the processor to: The determined one or more coronary physiological parameters are output to a treatment device.
