ECG lead reconstruction using machine learning

Abstract

1. A method for reconstructing a 12-lead standard electrocardiogram (ECG) system signal using an M-lead system, the method comprising: recording signals acquired by the 12-lead standard ECG system; recording signals acquired by the M-lead system; and using the recorded signals to train a machine learning model for generating a reconstructed 12-lead standard ECG system signal using the M-lead system.

Description

Technical Field

[0001] Cross - Reference to Related Applications This patent application claims the benefit of the filing date of U.S. Provisional Patent Application No. 63 / 071,803, filed on August 28, 2020, entitled "ELECTROCARDIOGRAM LEAD RECONSTRUCTION USING MACHINE LEARNING", the content of which is hereby incorporated by reference in its entirety.

[0002] This disclosure generally relates to the field of electrocardiograms, and more specifically, to systems and methods for electrocardiogram (ECG) lead reconstruction using machine learning.

Background Art

[0003] An ECG is a graph of the voltage over time of the electrical activity of a subject's heart, using electrodes placed on the subject's skin. The electrodes detect small electrical changes resulting from the depolarization and subsequent repolarization of the myocardium during each cardiac cycle, i.e., the heartbeat. Irregular ECG patterns can suggest various heart abnormalities. In a conventional 12 - lead ECG system, 10 electrodes are placed at standard positions on the subject's torso and limbs. Then, the overall magnitude of the subject's heart potential is measured from 12 different angles, i.e., "leads", and recorded over a certain period (e.g., 10 seconds), obtaining the overall magnitude and direction of the heart's electrical activity over the entire cardiac cycle.

[0004] The three main components of an electrocardiogram are the P wave corresponding to atrial depolarization, the QRS complex corresponding to ventricular depolarization associated with relatively small atrial repolarization, and the T wave representing ventricular repolarization. During each heartbeat, a healthy heart has an orderly progression of depolarization, which produces a characteristic ECG trace. An ECG can convey a lot of information about the heart's structure and electrical function and is thus a useful diagnostic tool.

Brief Description of the Drawings

[0005] To provide a more complete understanding of this disclosure and its features and benefits, the following description is referenced in conjunction with the attached figures, where similar reference numbers in the figures represent the same parts. [Figure 1A] This block diagram shows various features of a system for ECG read reconstruction according to the embodiments described herein. [Figure 1B] This block diagram shows various features of a system for ECG read reconstruction according to the embodiments described herein. [Figure 2A] An artificial neural network (ANN) for use in a system for ECG read reconstruction, characterized by the embodiments described herein, is shown. [Figure 2B] This specification describes alternative embodiments of ANN for use in systems for ECG read reconstruction, characterized by the embodiments described herein. [Figure 3A] The output of the regressor in response to unknown or untrained events can be improved using an expert committee featuring the embodiments described herein for implementing a system for ECG read reconstruction. [Figure 3B] The output of the regressor in response to unknown or untrained events can be improved using an expert committee featuring the embodiments described herein for implementing a system for ECG read reconstruction. [Figure 4] The characteristics of the embodiments described herein indicate that the quality of reconstructions developed using a system for ECG read reconstruction can be evaluated or assessed on the fly based on the inherent characteristics of the relationships between selected signals in a 12-lead standard ECG system. [Figure 5] This specification describes a configuration in which the regressor weights can be adapted to improve the quality of reconstruction provided by the system for ECG read reconstruction, such that it is evaluated or assessed based on the unique characteristics of the relationships between various leads in a 12-lead standard ECG system, as described in the embodiments described herein. [Figure 6] This specification shows how a ranking list of various lead configurations specific to a particular human subject may be generated based on the mathematical precision of the lead configuration, according to the features of the embodiments described herein. [Figure 7] This specification describes exemplary ranking techniques that can be used to rank various lead configurations specific to a particular human subject, as characterized by the embodiments described herein. [Figure 8A] This specification summarizes alternative techniques for ECG read reconstruction based on a priori clustering of input data using fuzzy C-means (FCM), as characterized by the features of the alternative embodiments described herein. [Figure 8B] This specification summarizes alternative techniques for ECG read reconstruction based on a priori clustering of input data using fuzzy C-means (FCM), as characterized by the features of the alternative embodiments described herein. [Figure 8C] This specification summarizes alternative techniques for ECG read reconstruction based on a priori clustering of input data using fuzzy C-means (FCM), as characterized by the features of the alternative embodiments described herein. [Figure 9A] This flowchart illustrates a technique for ECG read reconstruction according to the features of the embodiments described herein. [Figure 9B] This flowchart illustrates a technique for ECG read reconstruction according to the features of the embodiments described herein. [Figure 10] This flowchart illustrates a technique for on-the-fly analysis of an ECG read reconstruction system, as characterized by the embodiments described herein. [Figure 11] This is a block diagram of a computer system that may be used to implement all or some parts of a system for ECG read reconstruction, as characterized by the embodiments described herein. [Modes for carrying out the invention]

[0006] Description of Exemplary Embodiments For the purposes of this disclosure, the phrase "A and / or B" means (A), (B), or (A and B). For the purposes of this disclosure, the phrase "A, B, and / or C" means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C). When used in relation to measurement ranges, the term "between" includes the ends of the measurement range. When used herein, the notation "A / B / C" means (A), (B), and / or (C).

[0007] In this specification, the phrases “in an embodiment” or “in embodiments” may refer to one or more of the same or different embodiments, respectively. Furthermore, when used in reference to embodiments of this disclosure, terms such as “comprising,” “including,” and “having” are synonymous. In this disclosure, perspective-based descriptions such as “above,” “below,” “up,” “down,” and “side” may be used, but such descriptions are used to facilitate consideration and are not intended to limit the applicability of the disclosed embodiments. The accompanying drawings are not necessarily drawn to scale. Unless otherwise noted, the use of “first,” “second,” “third,” and other ordinal adjectives to describe common objects simply indicates that different instances of similar objects are being referred to, and is not intended to suggest that the objects described in this way must be in a given order, temporally, spatially, in order, or in any other manner.

[0008] The following detailed description refers to the accompanying drawings, which form part of it, illustrating possible embodiments as examples. It is understood that other embodiments may be utilized and that structural or logical modifications may be made without departing from the scope of this disclosure. Therefore, the following detailed description should not be construed as restrictive.

[0009] The following disclosure describes various exemplary embodiments and examples for implementing the features and functionalities of this disclosure. Certain components, arrangements, and / or features are described below in relation to various exemplary embodiments, but these are merely examples used to simplify this disclosure and are not intended to limit it. It will be understood, of course, that in developing any actual embodiment, many implementation-specific decisions must be made, including compliance with system, business, and / or legal constraints, which may vary from implementation to implementation, in order to achieve the developer's specific goals. Furthermore, while such development efforts can be complex and time-consuming, they will nevertheless be recognized as routine work for those skilled in the art who are interested in this disclosure.

[0010] In this specification, spatial relationships between various components and spatial orientations of various aspects of components may be referenced, as shown in the accompanying drawings. However, as will be recognized by those skilled in the art after a complete reading of this disclosure, the devices, components, members, apparatus, and others described herein may be arranged in any desired orientation. Therefore, the terms “up,” “down,” “up,” “down,” or other similar terms used to describe spatial relationships between various components or the spatial orientation of aspects of such components should be understood to describe relative relationships between components or the spatial orientation of aspects of such components, respectively, since the components described herein may be oriented in any desired direction. When used to describe a range of dimensions or other properties (e.g., time, pressure, temperature, length, width, etc.) of an element, operation, and / or state, the phrase “between X and Y” refers to a range including X and Y.

[0011] Furthermore, this disclosure may repeat reference numbers and / or reference letters in various embodiments. This repetition is for simplification and clarity and does not in itself determine the relationships between the various embodiments and / or configurations considered. Exemplary embodiments that may be used to implement the features and functionalities of this disclosure are described herein with more specific reference to the accompanying drawings.

[0012] The embodiments described herein include a system and method for reconstructing a standard 12 - lead (using 10 electrodes) ECG trace using machine learning with two or more leads. According to the features of the embodiments described herein, one aspect of the system and method involves optimizing the placement of electrodes on the torso and / or limbs of a particular human subject to ensure the highest quality reconstruction for that particular individual. Additionally and / or alternatively, local qualitative confidence values of the reconstruction may be provided to verify the performance of the system.

[0013] The 10 electrodes in a 12 - lead standard ECG system are shown in Table 1 below.

[0014]

Table 1

[0015] The 12 leads of a 12 - lead standard ECG system include the limb leads I, II, and III, the augmented limb leads aVR, aVL, and aVF, and the precordial leads V1, V2, V3, V4, V5, and V6. In a 12 - lead standard ECG system, each lead corresponds to one or a combination of the electrodes. For example: I = LA - RA II = LL - RA III = LL - LA

[0016] For illustrative purposes only, the ECG reconstruction system and method described herein may be described with reference to an M - lead system (formed using X electrodes) combined with an artificial neural network (ANN), where M is equal to 3 and X is equal to 4, provided that it will be recognized that more or fewer leads and corresponding electrodes and machine - learning techniques other than ANN may be used without departing from the spirit or scope of the embodiments described herein.

[0017] One exemplary embodiment is a technique for reproducing the ECG signal generated by a standard 12-lead ECG system (formed by 10 electrodes) using a 3-lead system (formed by 4 electrodes) combined with an artificial neural network (ANN) containing a trained model. The electrode arrangement for implementing the 3-lead system can be personalized for a specific human subject during model training. As a result, a free-roaming ECG system can be implemented using the 3-lead system without sacrificing the accuracy obtained by a standard 12-lead ECG system.

[0018] Figure 1A shows a functional block diagram of system 100 for performing ECG lead reconstruction using a subset of M leads of the enhanced ECG system 102. According to the embodiments described herein, the enhanced ECG system 102 includes electrodes from a 12-lead standard ECG system (represented by electrode 103A in Figure 1A), as well as additional electrodes (represented by electrode 103B in Figure 1A). As shown in Figure 1A, the enhanced ECG system 102 includes nine additional electrodes (for a total of 19 electrodes), but it should be noted that more or fewer (including zero) additional electrodes may be included without departing from the spirit or scope of the embodiments described herein, and that the number and arrangement of additional electrodes may be selected to optimize the operation of system 100 and increase the number of subsets to what can be evaluated for use, as described below. Additionally and / or alternatively, the arrangement of additional electrodes may be influenced by human subject characteristics (e.g., phenotype) and / or suspected pathological conditions, if that information is available. In certain embodiments, M is equal to 3, however, it will be recognized that subsets containing more or fewer leads may be adopted without departing from the spirit or scope of the embodiments described herein. It will be further recognized that the steps performed by the system 100 as described below are carried out for each of the configurations of N or M leads.

[0019] As shown in Figure 1A, during the initial training session performed by system 100, the electrodes constituting the enhanced ECG system 102 are placed on the torso and limbs of a human subject 104, with the electrodes constituting the 12-lead standard ECG system (e.g., electrode 103A) placed in standard positions, and each of the additional electrodes (e.g., electrode 103B) placed in a different position. Signals acquired from the standard 12-lead system 106 and signals acquired from the M-lead system or configuration 108, which may consist of three leads of the enhanced ECG system 102 (constituting one of N M-lead systems or configurations), are recorded, and a portion of each signal (e.g., about 16 seconds in one embodiment) is input to the training module 110. In certain embodiments, signals from all electrodes of the enhanced ECG system 102 are recorded using a single device. In an alternative embodiment, signal 106 may be acquired using an analog front-end ("AFE") that provides clinical-grade data (for example, using a commercially available ECG cart), while signal 108 may be acquired in parallel using one or more devices having non-clinical-grade AFEs.

[0020] As described, the recorded signals may be used by the training module 110 to obtain coefficients for a trained model 112, which is used by the reconstruction module 114 to reconstruct the signal generated by the standard 12-lead system 106 from the signal generated by the M-lead system 108. In particular, in the shown embodiment, the reconstruction module 114 may apply the remaining portion of the recorded signal from the 3-lead system 108 (e.g., about 1-2 minutes) to the trained model 112 and output the reconstructed 12-lead ECG signal to the evaluation module 116. Once the system (and in particular the model 112) is proven robust (e.g., after a large number of cases have been analyzed), it will be recognized that training may be unnecessary, as the coefficients of the model can be inferred from the prior information.

[0021] The evaluation module 116 refers to the 12-lead ECG signal 106 and checks the accuracy, reliability, and / or trustworthiness of the reconstruction obtained using the M-lead system 108. For example, in a particular embodiment, the evaluation module 116 may compare the reconstructed 12-lead ECG signal output from the reconstruction module 114 with the remainder of the original 12-lead ECG signal 106 obtained by a standard 12-lead system, using several figures of merit ("FoM").

[0022] The FoM calculated by the evaluation module 116 for each of the N configurations are input to a ranking module 118 that ranks the different configurations of the M-lead system 108 (they are recorded, used for training, and evaluated in the same manner as described above with respect to the recorded standard 12-lead system signals 106), and outputs an individual ranking of the configurations / positions 120 for the subject 104. In one exemplary embodiment, the ranking is performed using the FoM for each of the N M-lead configurations, however, it will be recognized that any number of other ranking methods may be employed by the ranking module 118 without departing from the spirit or scope of the embodiments described herein, including methods of giving more weight to certain derivation characteristics or some FoMs than others (for example, based on the subject's known medical condition or history).

[0023] In certain embodiments, one or more M-lead combinations / configurations that produce the most accurate, reliable, and / or trustworthy (or one of them) reconstructed 12-lead ECG signals, along with associated models, are selected as a personalized system for the user, for example, for use at home and / or while on the go.

[0024] As described above, in certain embodiments, a first portion of each signal (e.g., 16 seconds) may be used by the training module 110 to train a model 112 including an ANN in certain embodiments, while the remaining portion of each signal is used for reconstruction (reconstruction module 114), evaluation (evaluation module 116), and ranking (ranking module 118). Additionally and / or alternatively, all of each signal may be used to perform ranking. In addition, in certain embodiments, the reliability, or credibility, of a reconstruction associated with a particular M-lead combination may be verified based on known constraints, such as that the reconstruction satisfies mathematical relationships between certain signals in a 12-lead standard system or specific relationships between chest leads.

[0025] Figure 1B is a block diagram showing one embodiment of on-the-fly evaluation, or assessment, and fitting of the M-lead system 130, which may include, for example, one of the M-lead configurations selected from the individual positional ranking 120 (Figure 1A) and the corresponding model, during the use of the system 130 by the subject. As shown in Figure 1B, the accuracy, reliability, and / or credibility of the reconstructed standard 12-lead ECG signal from the M-lead system 130 is assessed by the on-the-fly verification module 132 based on verifying / evaluating the inherent relationships between some of the signals of the 12-lead standard ECG system (e.g., limb leads and / or chest leads), as described. In addition, as a result of the assessment, as described in detail below (see, for example, Figure 5), the on-the-fly verification module 132 may provide calibration data to the self-calibration module 134 to calibrate the M-lead system 130. Additionally and / or alternatively, following an assessment by the on-the-fly verification module 132 (e.g., as shown in Figure 4), as described in more detail below, a local confidence value may be assigned to the reconstruction based on the precision level of the following equation, which indicates the robustness of the reconstruction within a specific time or time window.

[0026] Figure 2A shows an artificial neural network (ANN) 200 that can be used to implement a model for reconstructing a standard 12-lead ECG system using only three leads formed by four electrodes, as shown in Figure 1A. As shown in Figure 2A, in one exemplary embodiment, the ANN 200 is a single-output ANN that includes a multilayer perceptron network (MLP) comprising an input layer 202 with three inputs corresponding to each of the leads of the 3-lead system, a hidden layer 204, and an output layer with a single output corresponding to one of the 12 reconstructed leads of the standard 12-lead ECG system. The model will be recognized as containing 12 single-output ANNs 200, one for each of the 12 reconstructed leads.

[0027] Figure 2B shows an alternative embodiment of the ANN, indicated by reference numeral 220 in Figure 2B, which may be used to implement a model for reconstructing a 12-lead standard ECG system using only three leads formed by four electrodes, such as system 100 (Figure 1A). As shown in Figure 2B, in one embodiment, ANN 220 is a multi-output ANN including a multilayer perceptron network (MLP) comprising an input layer 222 having three inputs corresponding to each of the leads of a 3-lead system, a hidden layer 224, and an output layer having multiple outputs, each corresponding to one of the 12 reconstructed leads of a 12-lead standard ECG system. It will be recognized that any of the ANN configurations 200, 220, or the alternative configurations may be advantageously used to implement the embodiments as described herein.

[0028] While single-output ANNs like ANN200 offer faster convergence, and therefore systems deploying single-output ANNs are trained more quickly, it will be recognized that the advantage of using multi-output ANNs like ANN220 is that deviations in the mathematical relationships between outputs can be fed back into the ANN to update the weights. In addition, as illustrated and described herein, ANN200 includes a single hidden layer 204, but the number of neurons in the hidden layer should not be interpreted as being limited or restricted to a specific number. Similarly, as illustrated and described herein, ANN220 includes a single hidden layer 224, but the number of hidden layers should not be interpreted as being limited or restricted to a specific number. Furthermore, while the embodiments described herein take acquired reads into consideration, additional inputs related to acquired reads (e.g., angles and magnitudes of cardiac vectors) and / or additional inputs related to the subject (e.g., sex, age, known conditions and / or comorbidities) may be implemented / added as inputs to the ANN (e.g., ANN200, 220) if desired, in order to accelerate network convergence.

[0029] To enhance the robustness of the regression, expert committees containing different regressions may be implemented. Figures 3A and 3B illustrate how expert committees can be used to improve the output of the regressor for previously unknown or untrained events (such as ectopic heartbeats or pathological conditions). In one embodiment, such an expert committee is implemented as a group of ANNs (instead of a single ANN) for reconstructing each lead of a 12-lead standard ECG system from an M-lead system. In the embodiments shown in Figures 3A and 3B, an expert committee is deployed that considers 20 runs of the regressor involving 12 ANNs, one per lead, and the average of each lead is used to reconstruct the lead signal. In one embodiment, as shown in Figure 3A, a reconstruction algorithm including a network 300 containing an expert committee trained using only normal heartbeats can successfully reconstruct leads containing ventricular premature contractions (PVCs) in addition to normal pulses, as represented by graphs 302A-302F. It will be recognized that other pathological conditions can be successfully reconstructed using a system trained using only normal heartbeats.

[0030] Figure 3A shows the use of 20 runs (in this case, 20 ANNs), but it should be noted that more or fewer ANNs may be deployed. Graphs 302A–302F show the original read signals versus the reconstructed read signals for reads I, II, III, aVL, aVR, and aVF, respectively. Additionally and / or alternatively, different and / or additional regressors (e.g., linear regression, CNN, binary tree) may be used to increase the robustness of the committee (or final regressor).

[0031] Figure 4 shows a system 400 that can evaluate or assess the accuracy, reliability, and / or credibility of read reconstruction based on the inherent characteristics of the relationships between some of the signals in a 12-lead standard ECG system. As shown in Figure 4, system 400, which may be used in whole or in part to implement the on-the-fly verification module 132 (Figure 1B), is based on the following four equations: I - II + III = 0

number

[0032] As mentioned above, following the assessment, a local confidence value can be assigned to the reconstruction based on the level of precision of the above equations indicating the robustness of the reconstruction. In one embodiment, the confidence value may be normalized to a value between a first value (e.g., 0) indicating that the reconstruction is very unreliable and a second value (e.g., 1) indicating that the reconstruction is very reliable, with the value between the first and second values ​​indicating the relative reliability / unreliability of the reconstruction. In another embodiment, the confidence value may include one of several values, each indicating the relative reliability and / or acceptability of the reconstruction. Additionally and / or alternatively, the reliability of the acquired signal may be considered when assessing the reconstructed signal and assigning a confidence value. For example, if the acquired signal is quite noisy, the confidence level of the reconstruction may be lower than if the acquired signal were less noisy. The same can be said for situations where large motions are detected (e.g., using an accelerometer), in which case a signal acquired under high-motion conditions may result in the reconstruction being considered less reliable than a reconstruction performed using a signal acquired under more static conditions. In certain embodiments, anterior chest leads may also be evaluated to enhance the reliability of the assessment of the reconstructed signal. It will be recognized that this evaluation (or assessment) may be performed throughout the operation of the system described herein. Information from auxiliary sensors may also be considered, such as in the case of accelerometers that can detect large or violent movements that may degrade / compromise the reconstruction.

[0033] Figure 5 shows a system 500 that can adapt the regressor weights according to the reconstruction quality based on the inherent relationships between different signals in a 12-lead standard system. An on-the-fly verification module 132 (Figure 1B) and a self-calibration module 134 (Figure 1A) can be implemented using all or part of system 500. This adaptation, or self-calibration, may be triggered in response to a confidence value below a certain threshold being assigned to the reconstruction. In addition, confidence values ​​may be used directly for the weight / self-calibration process adaptation. The system can continue to adapt its weights in runs where deviations are used as error signals. In certain embodiments, weight adaptation is restricted to avoid global false minimums, such as when all outputs are zeroed out.

[0034] Figure 6 shows how an individual positional ranking 120 of several (e.g., N) M-lead combinations / configurations (including a 3-lead configuration in the shown embodiment) specific to each subject may be generated by a ranking module 118 (Figure 1A) based on the mathematical precision of the various configurations, in order to ensure the best possible signal reconstruction for a particular subject. Following the mathematical ranking, rules of thumb can be applied to change the order and / or determine which of the configurations should be used based on various limitations and other considerations. Such limitations may be defined by physiological or anatomical constraints and practical effects (e.g., to fit the subject's anatomical structure and / or to limit the distance between electrodes), as well as device shape factors. In particular, Figure 6 shows a ranking list 120 of a subset of possible 3-lead system configurations / combinations ranked in order of the configuration scores (Figure 7) of a human subject 104. Since the degradation of performance is relatively smooth (i.e., configurations separated by only 10 positions in the ranking list 120 typically have approximately equal performance), the list 120 can be used to select one or more 3-lead combinations / configurations to be used by subject 104, for example in a freely moving device, based on the anatomical limitations of subject 104 (e.g., breast tissue) and / or clinical considerations (e.g., certain features of the ECG are of interest, and a given orientation is known to contain more information than others). In one embodiment, the ability to select more than one M lead configurations of similar performance (i.e., similarly ranked) allows the subject to change the electrode placement over time, thereby potentially preventing irritation or damage to the subject's skin.

[0035] Figure 7 shows an exemplary ranking algorithm for implementation by a ranking module 118 (Figure 1A) for ranking M-inductive system configurations according to embodiments described herein. As previously stated, there are many different ways of ranking system configurations, and the ranking method shown in Figure 7 is just one example of how such ranking can be achieved. As shown in Figure 7, in one exemplary embodiment, for the present embodiment, the FoM tensor 700 is defined to include N possible electrode configurations × 12 leads × Y FoMs. In a particular embodiment, Y is equal to 5, however, it will be recognized that more or fewer than 5 FoMs may be defined. In one exemplary embodiment, the FoMs considered may include root mean square (RMS), cross-correlation (CC), maximum absolute distance (MAD), sum of squared distances (SSD), and signal-to-noise ratio (SNR). A unique score 708 for each configuration is derived by extracting the worst FoM for all reads of the configuration 710, determining a Z score (i.e., distance from the mean of the standard deviation) for each FoM 712, and combining the Z scores into a single value which is the configuration score of the configuration 714. The configurations are then ranked by their configuration scores to determine which configuration provides the best reconstruction for a particular subject, thereby generating a ranked list 120.

[0036] It will be recognized that several configurations exist that may offer an acceptable reconfiguration, and that, as discussed above, other factors, in addition to ranking, such as physiological constraints, practical effects, and / or the pathological conditions being investigated, may also influence which configuration is ultimately selected as optimal for a particular application.

[0037] Figures 8A–8C summarize alternative embodiments of a reconstruction algorithm based on a priori clustering of input data using Fuzzy C-Means (FCM). As shown in Figures 8A–8C, the splitting or clustering of the input data can be performed "automatically" by the reconstruction algorithm during the training phase through an optimization process based on machine learning principles. After the data is split, specific regressors can be applied to each cluster and trained specifically for each of its sets. Using fuzzy clusters instead of classical clusters allows for a smoother transition, and membership principles are assigned instead of static and / or exclusive labels.

[0038] In other words, each data sample is assigned a degree of belonging to each cluster based on how well each cluster represents the sample. For example, a data sample may have a degree of belonging to cluster A (30%), cluster B (20%), cluster C (50%), and cluster D (0%). Missing reads are reconstructed by combining individual regressors with the same weights. In an exemplary embodiment, the C mean is applied to the clustering, and a specific linear regressor is applied to each cluster. Once the reads are reconstructed, it is determined whether known relationships between derivations (e.g., III = II - I, aVL = 1 / 2(I - III), AVR = -1 / 2(I + II), aVF = 1 / 2(II + III)) are satisfied. This determination is used to assign confidence levels that define the robustness and / or reliability of the reconstruction, as described above with reference to the ANN embodiment.

[0039] In exemplary embodiments, four models may be constructed to separately represent the repolarization and depolarization of the cardiac atria and ventricles; however, since there are regions where they coexist, the system may provide combinations of models in specific regions of activated ventricular pulse types. In one embodiment, the approach is based on statistical analysis rather than on a priori models.

[0040] Figures 9A and 9B are flowcharts illustrating the operation of a technique for ECG lead reconstruction according to the features of the embodiments described herein, which may be implemented by system 100 (Figure 1A). Referring to Figure 9A, step 900 involves applying the ECG system to a human subject. For example, in the embodiment shown in Figure 1A, an enhanced ECG system, including 19 electrodes including the electrodes of a standard 12-lead ECG system and 9 additional electrodes, is applied to the limbs and torso of the subject according to the embodiments described herein.

[0041] In step 902, signals acquired by a standard 12-lead ECG system and signals acquired by a selected M-lead ECG system, which constitute a subset of the reads from the extended ECG system, are recorded.

[0042] In step 904, portions of the recorded signal are used to train a machine learning model to generate a reconstructed 12-lead signal using an M-lead system. In one embodiment, the first approximately 16 seconds of the recorded signal are used to train the machine learning model in this manner.

[0043] In step 906, the accuracy, reliability, and / or trustworthiness of the reconstruction may be evaluated by referring to the signal acquired by the 12-lead standard ECG system.

[0044] In step 908, a configuration score indicating the evaluated accuracy, reliability, and / or credibility of the reconstruction may be assigned to the M-guided ECG system. In certain embodiments, the configuration score may be assigned by reference to the FoM of the reconstruction.

[0045] The steps shown in Figure 9A may be performed for each of the N M-lead configurations, each of which contains a subset of the leads from the enhanced ECG system, and so on. When applied to a specific human subject, a configuration score is assigned to each M-lead configuration, indicating the accuracy, reliability, and / or credibility of the reconstruction enabled by the configuration.

[0046] Referring now to Figure 9B, in step 920, all N M-inductive configurations are ranked based on the configuration score assigned to each configuration (e.g., based on FoM and / or some other ranking method).

[0047] In step 922, at least one of the N M-lead configurations is selected based on the ranking of the selected system for use in relation to a human subject. For example, the highest-ranking configuration or one of the higher-ranking configurations may be selected. The selected M-lead configuration may be deployed in the shape factor of the freely moving device for use by a human subject. As described above, in certain embodiments, more than one similar-ranking M-lead configurations may be identified to allow the subject to change the electrode placement over time, thereby reducing the possibility of irritation or damage to the subject's skin.

[0048] Figure 10 is a flowchart illustrating techniques for on-the-fly analysis and potential fitting of an ECG lead reconstruction system, as characterized by embodiments described herein (for example, as shown in Figure 1B). In step 1000, the accuracy, reliability, and / or credibility of the reconstruction provided by the M-lead ECG reconstruction system may be assessed or verified on the fly based on the inherent characteristics of the relationships between some of the signals in a 12-lead standard ECG system (as described with reference to Figure 4), the relationships between anterior chest derivations, or information from additional sensors such as an accelerometer or estimates of electrode contact impedance.

[0049] In step 1002, a local confidence value indicating the assessed accuracy / reliability and / or credibility of the reconstruction may be assigned to the M-guided ECG reconstruction system. The confidence value can be used to determine the confidence level that may be imposed on the reconstruction at a particular moment or time window.

[0050] In step 1004, the results of the assessment performed in step 1000 are used by the M-guided ECG reconstruction system to perform self-calibration (as described with reference to Figure 5), thereby potentially improving its operation.

[0051] It will be noted that the steps shown in Figure 10 can be performed "on the fly," the assigned confidence value can be associated with the time or time window of the reconstructed signal, and therefore can change over time as described above. It will also be noted that the steps shown in Figure 10 are optional, independent, and can be omitted in relation to systems whose accuracy, reliability, and / or credibility are already established.

[0052] Figure 11 is a block diagram showing an exemplary system 1100 that may be configured to implement at least a portion of the system for ECG read reconstruction, more specifically, as shown in the above-described diagrams, according to the features of the embodiments described herein. As shown in Figure 11, the system 1100 may include at least one processor 1102, for example, a hardware processor 1102, coupled to memory elements 1104 via a system bus 1106. Thus, the system may store program code and / or data in the memory elements 1104. The processor 1102 may also execute program code accessed from the memory elements 1104 via the system bus 1106. In one embodiment, the system may be implemented as a computer suitable for storing and / or executing program code. However, it should be understood that the system 1100 may be implemented in any form of system including a processor and memory that can perform the functions described herein.

[0053] In some embodiments, the processor 1102 can execute software or algorithms for performing activities such as those considered herein, in particular activities relating to ECG read reconstruction according to the features of the embodiments described herein. The processor 1102 may include any combination of hardware, software, or firmware that provides programmable logic, including, in non-limiting examples, a microprocessor, a DSP, a field-programmable gate array (FPGA), a programmable logic array (PLA), an integrated circuit (IC), an application-specific IC (ASIC), or a virtual machine processor. The processor 1102 may include a cloud processor. The processor 1102 may be communicatively coupled to the memory element 1104, for example in a direct memory access configuration (DMA), so that the processor 1102 can read from and write to the memory element 1104.

[0054] In general, the memory element 1104 may include any suitable volatile or non-volatile memory technology, including double data-rate (DDR) random access memory (RAM), synchronous RAM (SRAM), dynamic RAM (DRAM), flash, read-only memory (ROM), optical media, virtual memory area, magnetic or tape memory, or any other suitable technology. Unless otherwise specified, any memory element considered herein should be interpreted as being encompassed within the broad term “memory.” Information to be measured, processed, tracked, or transmitted to any component of system 1100 can all be provided in any database, register, control list, cache, or storage structure that can be referenced within any suitable time frame. Any such choice of storage device may be included within the broad term “memory” as used herein. Similarly, any potential processing elements, modules, and machines described herein should be interpreted as being encompassed within the broad term “processor.” Each of the elements shown in this figure may also include a suitable interface for receiving, transmitting, and / or otherwise communicating data or information in a network environment, and therefore they can communicate with systems having hardware similar to or identical to another of these elements, for example.

[0055] In certain exemplary implementations, a mechanism for implementing a system for ECG read reconstruction as outlined herein may be implemented by encoded logic in one or more tangible media, which may include non-temporary media, such as embedded logic provided to an ASIC, DSP instructions, and software (potentially including object code and source code) executed by a processor or other similar machine. In some of these examples, a memory element, such as the memory element 1104 shown in Figure 11, can store data or information used for the operations described herein. This includes the memory element storing software, logic, code, or processor instructions to be executed to perform the activities described herein. The processor can execute any type of instruction associated with the data or information to achieve the operations detailed herein. In one embodiment, a processor, such as the processor 1102 shown in Figure 11, can convert an element or article (e.g., data) from one state or thing to another. In another embodiment, the activities outlined herein may be implemented with fixed logic or programmable logic (e.g., software / computer instructions executed by a processor), and the elements identified herein may be any type of programmable processor, programmable digital logic (e.g., FPGA, DSP, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)), or ASIC including digital logic, software, code, electronic instructions, or any preferred combination thereof.

[0056] The memory element 1104 may include one or more physical memory devices, such as local memory 1108 and one or more mass storage devices 1110. Local memory may refer to RAM or other non-persistent memory devices commonly used during the actual execution of program code. Mass storage devices may be implemented as hard drives or other persistent data storage devices. The processing system 1100 may also include one or more cache memories (not shown) that provide temporary storage for at least some program code in order to reduce the number of times the program code must be retrieved from the mass storage device 1110 during execution.

[0057] As shown in Figure 11, the memory element 1104 may store the ECG read-reconstruction module 1120. In various embodiments, the module 1120 may be stored in local memory 1108, one or more mass storage devices 1110, or separately from local memory and mass storage devices. It should be understood that the system 1100 may further run an operating system (not shown in Figure 11) that can facilitate the execution of module 1120. Implemented in the form of executable program code and / or data, module 1120 may be read from and written to by the system 1100, and / or executed by, for example, the processor 1102. In response to reading from, writing to, and / or execution of module 1120, the system 1100 may be configured to perform one or more operation or method steps described herein, such as those shown in and described with reference to Figures 9 and 10.

[0058] The input / output (I / O) devices, designated as input device 1112 and output device 1114, may be optionally coupled to the system. Examples of input devices include, but are not limited to, a keyboard, a pointing device such as a mouse, etc. Examples of output devices include, but are not limited to, a monitor or display, a speaker, etc. In some implementations, the system may include a device driver (not shown) for output device 1114. The input and / or output devices 1112, 1114 may be coupled to system 1100 directly or through an intermediary I / O controller.

[0059] In one embodiment, the input and output devices may be implemented as a combined input / output device (shown in Figure 11 by dashed lines surrounding input device 1112 and output device 1114). An example of such a combined device is a touch-sensitive display, sometimes also referred to as a “touchscreen display” or simply a “touchscreen.” In such an embodiment, input to the device may be provided by the movement of a physical object, such as a stylus or a user’s finger, on or near the touchscreen display.

[0060] Optionally, a network adapter 1116 may also be coupled to system 1100, enabling system 1100 to be coupled to other systems, computer systems, remote network devices, and / or remote storage devices through an intervening private or public network. The network adapter may comprise a data receiver for receiving data transmitted to system 1100 by systems, devices, and / or networks, and a data transmitter for transmitting data from system 1100 to systems, devices, and / or networks. Modems, cable modems, and Ethernet cards are examples of different types of network adapters that may be used in system 1100.

[0061] Example 1 is a method for reconstructing a 12-lead standard electrocardiogram (ECG) system signal for a human subject using an M-lead system, the method comprising: recording a signal acquired by a 12-lead standard ECG system; recording a signal acquired by an M-lead system; and using the recorded signals to train a machine learning model for generating a 12-lead standard ECG system signal reconstructed using an M-lead system.

[0062] In Example 2, the method of Example 1 may further include an M-lead system comprising M leads that constitute a subset of the leads of the enhanced ECG system, wherein the enhanced ECG system includes a 12-lead standard ECG system.

[0063] In Example 3, the method of Example 2 may further include an enhanced ECG system comprising at least one additional electrode.

[0064] In Example 4, any of the methods from Examples 1 to 3 may further include evaluating the performance of the machine learning model by comparing the recorded 12-lead standard EDG system signal with the reconstructed 12-lead standard ECG system signal.

[0065] In Example 5, any of the methods from Examples 1 to 4 may further include a machine learning model that includes an artificial neural network (ANN), where a portion of each of the recorded signals is used to train the coefficients of the ANN.

[0066] In Example 6, the method of Example 5 may further include an ANN that includes multiple output ANNs.

[0067] In Example 7, the method of Example 5 may further include an ANN comprising multiple single-output ANNs.

[0068] In Example 8, the method of Example 5 may further include an ANN having M inputs corresponding to the leads of the M-inductance system.

[0069] In Example 9, the method of Example 8 may further include an ANN having at least one additional input corresponding to at least one of the following: the angle of the cardiac vector, the magnitude of the cardiac vector, and information about a human subject.

[0070] In Example 10, any of the methods from Examples 1 to 9 may further include a machine learning model that includes an expert committee.

[0071] In Example 11, any of the methods from Examples 1 to 10 may further include an M-guidance system comprising multiple M-guidance systems, the method further including evaluating the accuracy of each M-guidance system and ranking the M-guidance systems in order of their accuracy.

[0072] In Example 12, the method of Example 11 may further include the evaluation of the accuracy of each M-guided system being performed by referring to the Y figures of merit (FoM) of each M-guided system.

[0073] In Example 13, the method of Example 11 may further include selecting one of the M-lead systems for use when monitoring the ECG of a human subject, based on the ranking of one of the M-lead systems.

[0074] In Example 14, the method of Example 11 may further include selecting several M-lead systems for use in monitoring the ECG of a human subject, based on the ranking of the selected M-lead systems.

[0075] In Example 15, the method of Example 13 may further include assessing the accuracy of a reconstruction produced by a selected M-lead system by determining whether the reconstruction satisfies the inherent characteristics of a standard 12-lead ECG signal, and assigning a confidence value to the reconstruction based on the results of the assessment.

[0076] In Example 16, the method of Example 15 may further include performing a calibration of one of the selected M-inductance systems based on the assessment results.

[0077] In Example 17, the method of Example 13 may further include assessing the reliability of the reconstruction generated by one of the selected M-induction systems based on at least one of the external sensor data and contact impedance data.

[0078] Example 18 may further include the implementation of any of the methods from Examples 1 to 17, where the machine learning model is implemented using fuzzy C-means (FCM) with a regressor.

[0079] In Example 19, any of the methods from Examples 1 to 18 may further include the case where M is equal to 3.

[0080] Example 20 is an electrocardiogram (ECG) reconstruction system for reconstructing a 12-lead standard ECG system signal using an M-lead system, comprising: a plurality of electrodes constituting a 12-lead standard ECG system, which are applied to the skin of a human subject; a training module for training a machine learning model for reconstructing a 12-lead standard ECG system signal from a signal acquired by an M-lead system using signals acquired by a 12-lead standard ECG system and signals acquired by an M-lead system; and a reconstruction module for reconstructing a 12-lead standard ECG system signal using an M-lead system with the machine learning model.

[0081] In Example 21, the ECG reconstruction system of Example 20 may further include an M-lead system containing M leads of an enhanced ECG system, which includes a 12-lead standard ECG system.

[0082] In Example 22, the ECG reconstruction system of Example 21 may further include an enhanced ECG system that includes at least one additional electrode.

[0083] In Example 23, any of the ECG reconstruction systems from Examples 20 to 23 may further include an evaluation module for evaluating the accuracy of a machine learning model.

[0084] In Example 24, the ECG reconstruction system of claim 23 may further include an evaluation module that evaluates the accuracy of a machine learning model by comparing the signal acquired by a 12-lead standard ECG system with the reconstructed signal.

[0085] In Example 25, any of the ECG reconstruction systems from Examples 20-24 may further include a machine learning model that includes an artificial neural network (ANN), where each portion of the recorded signal is used to train the coefficients of the ANN.

[0086] In Example 26, the ECG reconstruction system of Example 25 may further include an ANN that includes multiple output ANNs.

[0087] In Example 27, the ECG reconstruction system of Example 25 may further include an ANN containing multiple single-output ANNs.

[0088] In Example 28, any of the ECG reconstruction systems from Examples 20-27 may further include the implementation of a machine learning model using an expert committee.

[0089] In Example 29, any of the ECG reconstruction systems from Examples 20 to 28 may further include the implementation of a machine learning model using fuzzy C-means (FCM) with a regressor.

[0090] In Example 30, any of the ECG reconstruction systems from Examples 20 to 28 may further include an M-inductance system comprising multiple M-inductance systems, the evaluation module further evaluates the accuracy of each M-inductance system, and the reconstruction system further includes a ranking module for ranking the M-inductance systems based on the accuracy of each M-inductance system.

[0091] In Example 31, the ECG reconstruction system of Example 30 may further include the evaluation of the accuracy of each M-inductance system being performed by referring to the Y figures of merit (FoM) of each M-inductance system.

[0092] In Example 32, the ECG reconstruction system of Example 30 may further include an assessment module for assessing whether the intrinsic characteristics of a standard 12-lead ECG signal are satisfied by the reconstruction generated by one of several M-lead systems, and for assigning confidence values ​​to a machine learning model based on the assessment results.

[0093] In Example 33, the ECG reconstruction system of Example 32 may further include a calibration module for calibrating one of several M-lead models based on the assessment results.

[0094] In Example 34, any of the ECG reconstruction systems from Examples 20 to 33 may further include the condition that M is equal to 3.

[0095] Example 35 is a method for reconstructing a 12-lead standard electrocardiogram (ECG) system signal using an M-lead system, comprising: recording a first signal generated by a 12-lead standard ECG system; recording a second signal generated by a set of M leads; training a machine learning model using a first portion of the first recorded signal and the second recorded signal; generating a reconstructed signal by applying the machine learning model to a second portion of the second recorded signal; and evaluating the accuracy of the machine learning model by comparing the first signal with the reconstructed signal.

[0096] In Example 36, the method of Example 35 may further include a machine learning model that includes an artificial neural network (ANN), where each portion of the recorded signal is used to train the coefficients of the ANN.

[0097] In Example 37, the method of Example 36 may further include an ANN with multiple output ANNs.

[0098] In Example 38, the method of Example 36 may further include an ANN comprising multiple single-output ANNs.

[0099] In Example 39, any of the methods from Examples 35 to 38 may further include a machine learning model that includes an expert committee.

[0100] In Example 40, any of the methods from Examples 35 to 39 may further include assessing the accuracy of a machine learning model by determining whether the intrinsic characteristics of the first signal are satisfied by the reconstructed signal, and assigning a confidence value to the machine learning model based on the assessment results.

[0101] In Example 41, the method of Example 40 may further include adjusting the regressor weights of the machine learning model based on confidence levels.

[0102] In Example 42, any of the methods from Examples 35 to 41 may further include an M-induction system comprising multiple M-induction systems, the method further including evaluating the accuracy of each M-induction system and ranking a unique set based on the accuracy of each M-induction system compared to others.

[0103] In Example 43, the method of Example 42 may further include the evaluation of the accuracy of each of the M-guided systems being performed with reference to the Y figures of merit (FoM) of the M-guided systems.

[0104] In Example 44, the method of Example 42 or 43 may further include selecting one of the M-lead systems for use when monitoring the ECG of a human subject, based on the ranking of one of the M-lead systems.

[0105] In Example 45, the method of Example 44 may further include assessing the accuracy of a reconstruction produced by a selected M-lead system by determining whether the reconstruction satisfies the inherent characteristics of a standard 12-lead ECG signal, and assigning a confidence value to the reconstruction based on the results of the assessment.

[0106] In Example 46, the method of Example 45 may further include performing a calibration of one of the selected M-inductance systems based on the assessment results.

[0107] In Example 47, the method of Example 42 or 43 may further include selecting several M-lead systems for use in monitoring the ECG of a human subject, based on the ranking of the selected M-lead systems.

[0108] Example 48 may further include the implementation of any of the methods from Examples 35 to 47 using a fuzzy C-means (FCM) algorithm with a regressor.

[0109] In Example 49, any of the methods from Examples 35 to 48 may further include M equal to 3.

[0110] It should be noted that all specifications, dimensions, and relationships outlined herein (e.g., elements, operations, number of steps, etc.) are provided for illustrative and teaching purposes only. Such information may change materially without departing from the spirit of this disclosure or the scope of the appended claims. Specifications apply to only one non-limiting embodiment and should therefore be interpreted as such. In the foregoing description, exemplary embodiments have been described with reference to the configuration of specific components. Various modifications and changes may be made to such embodiments without departing from the scope of the appended claims. Accordingly, the description and drawings should be considered illustrative rather than limiting.

[0111] It should be noted that in some of the embodiments provided herein, interactions are described in terms of two, three, four, or more electrical components and / or modules. However, this is done solely for the purpose of clarity and illustration. It should be understood that the system can be integrated in any preferred manner. According to similar design alternatives, any of the components, modules, and elements shown in the figures can be combined into a variety of possible configurations, all of which are clearly within the broad scope of this specification. In particular, it may be easier to describe one or more of the functionality of a given flowset by referring to only a limited number of electrical elements. It should be understood that the electrical circuits in the figures and their teachings are readily expandable and can accommodate many components, as well as more complex / advanced arrangements and configurations. Thus, the embodiments provided do not limit the scope of electrical circuits or hinder their broad teachings when potentially applied to a wide variety of other architectures.

[0112] Furthermore, any references in this specification to various features (e.g., elements, structures, modules, components, steps, operations, characteristics, etc.) included in “one embodiment,” “exemplary embodiment,” “another embodiment,” “several embodiments,” “various embodiments,” “other embodiments,” or “alternative embodiments” are intended to mean that any such features are included in one or more embodiments of this disclosure, but may be combined in the same embodiment or not necessarily.

[0113] Furthermore, it should be noted that the functions related to the circuit architecture shown in the figures represent only some of the possible circuit architecture functions that may be performed by or within the system shown. Some of these operations may be deleted or removed as appropriate, or may be significantly modified or altered without departing from the scope of this disclosure. In addition, the timing of these operations may be significantly altered. The above operation flows are provided for the purposes of examples and discussion. Substantial flexibility is provided by the embodiments described herein in that any preferred arrangement, temporal order, configuration, and timing mechanism may be provided without departing from the teachings of this disclosure.

[0114] Numerous other changes, substitutions, modifications, alterations, and modifications may be apparent to those skilled in the art, and this disclosure is intended to encompass all such changes, substitutions, modifications, alterations, and modifications that fall within the scope of the appended claims.

[0115] Furthermore, all optional features of the devices and systems described herein may be implemented in relation to the methods or processes described herein, and details of the embodiments may be used throughout one or more embodiments.

[0116] The “means to do” in these (above) examples may include, but are not limited to, the use of any suitable software, circuitry, hubs, computer code, logic, algorithms, hardware, controllers, interfaces, links, buses, communication paths, and others, along with any suitable components considered herein.

[0117] It should be noted that in the embodiments provided above, and in many other embodiments provided herein, interactions may be described with respect to two, three, or four network elements. However, this is done solely for the purpose of clarity and illustration. In certain cases, it may be easier to describe one or more of the functionalities of a given flowset by referring to only a limited number of network elements. It should be understood that the topologies illustrated and described with reference to the accompanying figures (and their teachings) are readily extensible and can accommodate a large number of components, as well as more complex / advanced arrangements and configurations. Therefore, the embodiments provided do not limit the scope of the shown topologies or hinder their extensive teachings when they are potentially applicable to a wide variety of other architectures.

[0118] It is also important to note that the steps in the flowchart above represent only some of the possible signaling scenarios and patterns that may be performed by or within the communication system shown in the diagram. Some of these steps may be deleted or removed as appropriate, or they may be significantly modified or changed without departing from the scope of this disclosure. In addition, some of these operations are described as being performed simultaneously with or in parallel with one or more additional operations. However, the timing of these operations may be significantly changed. The above operation flow is provided for illustrative and discussion purposes only. Substantial flexibility is provided by the communication system shown in the diagram in that any preferred arrangement, temporal order, configuration, and timing mechanism may be provided without departing from the teachings of this disclosure.

[0119] While this disclosure has been described in detail with reference to specific configurations and arrangements, these exemplary configurations and arrangements can be substantially modified without departing from the scope of this disclosure. For example, while this disclosure has been described with reference to specific communication exchanges, embodiments described herein may be applicable to other architectures.

[0120] Numerous other changes, substitutions, modifications, alterations, and modifications may be apparent to those skilled in the art, and this disclosure is intended to encompass all such changes, substitutions, modifications, alterations, and modifications as those found in the attached claims. To assist the United States Patent and Trademark Office (USPTO) and, in addition, all readers of any patents issued relating to this application in interpreting the attached claims herein, the applicant would like to note that (a) unless the terms “means for” or “steps for” are used specifically in a particular claim, none of the attached claims are intended to exercise Section 142, paragraph 6(6) of the United States Patent Act, since it exists as of the filing date, and (b) nothing in the specification is intended to limit this disclosure in any way that is not otherwise reflected in the attached claims. [Explanation of Symbols]

[0121] 100 Systems 102 Enhanced ECG System 103A electrode 103B Electrode 104 Subjects 106 Guidance System 108 Guidance System 110 Training Modules 112 trained models 114 Reconstruction Module 116 Evaluation Modules 118 Ranking Module 120 Individual Rankings 130 M guidance system 132 On-the-fly verification module 136 Self-Calibration Module 200 Artificial Neural Networks (ANNs) 202 Input Layer 204 layers 220 ANN configuration 222 Input Layers 224 layers 300 Networks 302A~302F Graph 400 System 500 Systems 700 FoM tensor 708 score 1100 Processing System 1102 Processor 1104 memory elements 1106 System Bus 1108 Local memory 1110 Mass Storage Devices 1112 Input Devices 1114 Output Device 1116 Network Adapter 1120 ECG Read Reconstruction Module

Claims

1. A computer-based method for reconstructing electrocardiogram (ECG) system signals for a human subject, To record the first signal acquired by the 12-guidance standard ECG system, To record the second signal acquired by the M-guidance system, This includes training a machine learning model to generate a reconstructed 12-lead standard ECG system signal using the M-lead system, by inputting the recorded first signal and the recorded second signal. During training, the machine learning model clusters the second signal input to the machine learning model using fuzzy C-means (FCM) to generate multiple clusters, and then trains a specific linear regressor for each of the multiple clusters. A method comprising: each data sample of the second signal input to the machine learning model is assigned a degree to which it belongs to each of the clusters, and generating the reconstructed 12-lead standard ECG system signal involves combining the specific linear regressor with the degree to which the data sample belongs.

2. The method according to claim 1, wherein the M-lead system comprises a plurality of leads, including a subset of the leads of the enhanced ECG system, and the enhanced ECG system comprises the 12-lead standard ECG system and at least one additional electrode.

3. The method according to claim 1 or 2, further comprising evaluating the performance of the machine learning model by comparing the recorded first signal with the reconstructed 12-lead standard ECG system signal.

4. To record the third signal generated by the aforementioned M-induction system, The method according to any one of claims 1 to 3, further comprising applying the machine learning model to the recorded third signal to generate the reconstructed signal.

5. The method according to any one of claims 1 to 4, wherein the machine learning model comprises an artificial neural network (ANN) including a multi-output ANN or a multi-single-output ANN, wherein portions of the recorded first signal and the recorded second signal are used to train the coefficients of the ANN, and the ANN comprises a plurality of inputs corresponding to each lead of the M-guidance system.

6. The method according to claim 5, wherein the ANN comprises at least one additional input corresponding to at least one of the angles of the cardiac vector, the magnitude of the cardiac vector, and information relating to the human subject.

7. The M-guidance system includes a plurality of M-guidance systems, and the method is To evaluate the accuracy of each of the aforementioned multiple M-guidance systems, The method according to any one of claims 1 to 6, comprising ranking the plurality of M-guidance systems in order of the accuracy of each of the plurality of M-guidance systems.

8. The method according to claim 7, wherein the evaluation of the accuracy of each of the plurality of M-guidance systems is performed by referring to the Y figures of merit (FoM) of each M-guidance system.

9. The method according to claim 7 or 8, further comprising selecting a first M-lead system from the plurality of M-lead systems based on the ranking of the first M-lead systems for use in monitoring the ECG of the human subject, and / or further comprising selecting a plurality of first M-lead systems from the plurality of M-lead systems based on the ranking of each of the plurality of first M-lead systems for use in monitoring the ECG of the human subject.

10. The method according to claim 9, further comprising: assessing the accuracy of the reconstruction generated by the selected first M-lead system by determining whether the intrinsic characteristics of a 12-lead standard ECG signal are satisfied by the reconstruction; and assigning a confidence value to the reconstruction based on the results of the assessment.

11. The method according to claim 10, further comprising adjusting the regressor weights of the machine learning model based on the confidence value.

12. The method according to claim 10, further comprising performing calibration of the selected first M-inductance system based on the results of the assessment.

13. The method according to claim 10 or 12, further comprising assessing the reliability of the reconstruction generated by the selected first M-induction system based on at least one of external sensor data or contact impedance data.

14. The method according to any one of claims 1 to 13, wherein the reconstruction reliability is assigned to the reconstructed 12-lead standard ECG system signal by determining whether the inherent characteristics of the 12-lead standard ECG system signal are satisfied by the reconstructed 12-lead standard ECG system signal.

15. An electrocardiogram (ECG) reconstruction system, Multiple electrodes constituting a 12-lead standard ECG system, which are applied to a human subject, A training module for training a machine learning model to reconstruct the 12-lead standard ECG system signal from the second signal acquired by the M-lead system, by inputting the first signal acquired by the 12-lead standard ECG system and the second signal acquired by the M-lead system into the machine learning model. The system comprises a reconstruction module for reconstructing the 12-lead standard ECG system signal using the M-lead system with the machine learning model described above, During training, the machine learning model uses fuzzy C-means (FCM) to cluster the second signal input to the machine learning model to generate multiple clusters, and then trains a specific linear regressor for each of the multiple clusters. An ECG reconstruction system wherein each data sample of the second signal input to the machine learning model is assigned a degree to which it belongs to each of the clusters, and reconstructing the 12-lead standard ECG system signal includes combining the specific linear regressor with the degree to which the data sample belongs.

16. The ECG reconstruction system according to claim 15, wherein the M-lead system comprises a plurality of leads of an enhanced ECG system including the 12-lead standard ECG system.

17. The ECG reconstruction system according to claim 15 or 16, further comprising an evaluation module for evaluating the accuracy of the machine learning model by comparing the first signal with the reconstructed 12-lead standard ECG system signal.

18. The ECG reconstruction system according to claim 17, wherein the M-guidance system comprises a plurality of M-guidance systems, the evaluation module further evaluates the accuracy of each of the plurality of M-guidance systems, and the ECG reconstruction system further comprises a ranking module for ranking the M-guidance systems based on the accuracy of each of the plurality of M-guidance systems.

19. The ECG reconstruction system according to claim 18, further comprising an assessment module for assessing whether the intrinsic characteristics of the 12-lead standard ECG signal are satisfied by a reconstruction generated by a selected M-lead system from among the plurality of M-lead systems, and assigning confidence values ​​to the machine learning model based on the results of the assessment.

20. The ECG reconstruction system according to claim 19, further comprising a calibration module for calibrating the selected M-guidance system from among the plurality of M-guidance systems based on the results of the assessment.

Images