ECG data processing method and system based on microvolt T-wave alternans

Abstract

This application discloses a method and system for processing electrocardiogram (ECG) data based on microvolt-level T-wave alternation, relating to the field of ECG signal analysis. The method includes: using QRS to locate the heartbeat time anchor point; adaptively determining the T-wave analysis interval according to the heartbeat and dynamically segmenting it to obtain beat-by-beat T-wave segments; forming paired units from the T-wave segments according to odd and even heartbeats, and performing time alignment on the paired segments to suppress interference from time axis phase deviation on the differential waveform; constructing a weighting function that considers both the peak region and the slope change region of the T-wave, and weighting the differential waveform to enhance sensitivity to weak morphological differences; and fusing the weighted alternation amplitudes of each lead to generate a global alternation detection index. Therefore, the output global alternation detection index can more effectively carry weak and early morphological difference information and remain relatively stable under multi-source perturbation conditions, thus supporting more reliable output of ECG abnormality risk warnings.

Description

Technical Field

[0001] This application relates to the field of electrocardiogram (ECG) signal analysis technology, and in particular to an ECG data processing method and system based on microvolt-level T-wave alternation. Background Technology

[0002] Myocardial ischemia may remain in an "early / hidden" state for a considerable period. Current ECG monitoring and data analysis practices often rely on relatively macroscopic waveform changes to indicate abnormalities. However, these macroscopic changes are usually more likely to stabilize and appear after the relevant physiological state has progressed to a certain extent. Therefore, there is an objective possibility that early or hidden risks may be missed or insufficiently indicated under routine monitoring conditions. Against this backdrop, how to extract earlier, weaker, and more subtle electrophysiological clues from ECG signals and output stable risk indications in a quantifiable manner has become an important direction for improving ECG data analysis capabilities.

[0003] Microvolt T-wave alternation (MTWA) reflects the minute alternating fluctuations during repolarization between adjacent heartbeats, possessing potential early indicative value. However, existing MTWA testing practices have long primarily served arrhythmia risk stratification: on the one hand, its negative predictive value is high, but its positive predictive value is low; on the other hand, detection often depends on specific acquisition / testing conditions and is easily affected by factors such as heart rate fluctuations, medication effects, and lead noise, leading to a high proportion of "non-deterministic" results, thus limiting its stable application in routine conditions and daily monitoring scenarios. In recent years, some studies have explored using microvolt T-wave alternation for risk indication related to reversible myocardial ischemia, but the overall sensitivity remains low, and a considerable proportion of related risks are still not effectively indicated. Summary of the Invention

[0004] This application provides a method, system, storage medium, computer program product, and electronic device for processing electrocardiogram data based on microvolt-level T-wave alternation, which at least solves the problem in current related technologies that the extraction of weak electrocardiogram pathological features is difficult to balance high sensitivity and anti-interference stability, thus failing to provide reliable quantitative risk assessment.

[0005] In a first aspect, embodiments of this application provide a method for processing electrocardiogram (ECG) data based on microvolt-level T-wave alternation. The method includes: acquiring multi-lead ECG data to be processed, and performing preprocessing and QRS complex localization on the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat time point information; adaptively determining T-wave analysis intervals for each heartbeat in the ECG waveform data based on the heartbeat time point information, and obtaining T-wave curve segments corresponding to each heartbeat through dynamic T-wave segmentation, thereby generating a T-wave segment set; dividing the T-wave curve segments in the T-wave segment set into odd-numbered sequences and even-numbered sequences according to the heartbeat sequence number, and separating the even-numbered sequence T-waves... T-wave segments are formed by combining curve segments with odd-numbered sequence T-wave curve segments. For each paired T-wave segment, time alignment processing is performed to eliminate phase deviation on the time axis, resulting in aligned paired T-wave curve segments. A weighting function that considers the sensitivity of both the T-wave peak region and the T-wave slope change region is constructed, and the weighting function is used to perform weighted calculations on the differential waveforms between the aligned paired T-wave curve segments to obtain weighted alternating amplitudes that reflect the microvolt-level morphological differences of adjacent heartbeat T-waves. Based on the weighted alternating amplitudes corresponding to each lead, multi-lead fusion processing is performed to generate a global alternating detection index, and ECG abnormality risk warning information for indicating the risk of myocardial ischemia is output according to the global alternating detection index.

[0006] Secondly, embodiments of this application provide an electrocardiogram (ECG) data processing system based on microvolt-level T-wave alternation. The system includes: an ECG data acquisition and preprocessing unit, used to acquire multi-lead ECG data to be processed, and perform preprocessing and QRS complex localization on the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat time point information; a T-wave dynamic segmentation unit, used to adaptively determine the T-wave analysis interval for each heartbeat in the ECG waveform data based on the heartbeat time point information, and obtain T-wave curve segments corresponding to each heartbeat through dynamic T-wave segmentation, thereby generating a T-wave segment set; and a pairing construction unit, used to divide the T-wave curve segments in the T-wave segment set into odd-numbered sequences and even-numbered sequences according to the heartbeat sequence number, and pair the even-numbered sequence T-wave curve segments with the corresponding time sequence. Linear segments and odd-numbered sequence T-wave curve segments form paired T-wave segments; a timing alignment unit performs timing alignment processing on each paired T-wave segment to eliminate phase deviation on the time axis, resulting in aligned paired T-wave curve segments; a weighted alternating amplitude calculation unit constructs a weighting function that considers the sensitivity of both the T-wave peak region and the T-wave slope change region, and uses the weighting function to perform weighted calculation on the differential waveforms between the aligned paired T-wave curve segments to obtain a weighted alternating amplitude reflecting the microvolt-level morphological differences of adjacent heartbeat T-waves; an alternating detection and prompt output unit performs multi-lead fusion processing based on the weighted alternating amplitude corresponding to each lead to generate a global alternating detection index, and outputs ECG abnormality risk prompt information to indicate the risk of myocardial ischemia based on the global alternating detection index.

[0007] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the electrocardiogram data processing method based on microvolt-level T-wave alternation of any embodiment of the present application.

[0008] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the electrocardiogram data processing method based on microvolt-level T-wave alternation of any embodiment of this application.

[0009] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the electrocardiogram data processing method based on microvolt-level T-wave alternation of any embodiment of this application.

[0010] The electrocardiogram data processing method and system based on microvolt-level T-wave alternation provided in this application can achieve at least the following technical effects: (1) A stable measurement link for subtle differences in repolarization is formed by constructing T-wave analysis segments with beat-adaptive characteristics and merging them with consistent alignment before paired differential analysis. After the ECG waveform has undergone basic preprocessing and the heartbeat anchor point has been determined, the T-wave segments are dynamically extracted and incorporated into a unified comparison framework with the heartbeat timing as a reference, so that the comparison objects between adjacent heartbeats remain comparable in physiological meaning and temporal coverage. Furthermore, by introducing timing alignment on the odd-even paired comparison units, the main contribution of differential operation can be shifted from "time axis misalignment" to "morphological ontological difference", reducing the amplification or cancellation effect of small drifts and phase deviations on alternating quantization, so that the subsequent alternating amplitudes can more centrally represent the real subtle changes in the repolarization process between adjacent heartbeats, and improve the stability and consistency of the indicators.

[0011] (2) A weighted mechanism that comprehensively considers the peak region and the slope change region is used as a key measurement enhancement method. By constructing a weighting function to weight the differential waveform, the measurement can be more focused on the key segments that are more sensitive to subtle morphological differences and have more concentrated information, while weakening the unnecessary influence of low-information or more easily disturbed segments on the results. Subsequently, multi-lead fusion based on the alternating amplitude of each lead can establish the abnormal indication on the overall consistency of the cross-leads, so that the final global detection index is not easily dominated by the occasional disturbance of a single lead, thereby making the output risk indication information more reproducible and usable under normal monitoring conditions.

[0012] This technical solution organically couples the construction of comparable T-wave segments, the alignment mechanism for suppressing phase deviation, the weighted measurement for key morphological regions, and translead consistency to form a robust quantitative path for microvolt-level repolarization alternation characteristics. As a result, the output global alternation detection index can more effectively carry weak and early morphological difference information and remain relatively stable under multi-source perturbation conditions, thus supporting more reliable output of ECG abnormality risk warnings. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart is shown as an example of an electrocardiogram data processing method based on microvolt-level T-wave alternation according to an embodiment of this application; Figure 2 A schematic diagram illustrating the operational mechanism of an example of an electrocardiogram data processing method based on microvolt-level T-wave alternation according to an embodiment of this application is shown. Figure 3 A schematic diagram showing the comparison of the average alternating detection index under different background noise levels is presented. Figure 4 A heatmap comparison of detection accuracy under different noise and signal amplitude conditions is shown. Figure 5 A structural block diagram of an example of an electrocardiogram data processing system based on microvolt-level T-wave alternation according to an embodiment of this application is shown. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] It should be noted that among the current related technologies, there are two main approaches to the detection of microvolt-level T-wave alternation: frequency domain analysis and time domain analysis. However, both approaches have obvious technical limitations when dealing with the refined screening of early or occult myocardial ischemia.

[0017] In frequency domain analysis, relevant methods typically utilize Fourier transform to extract the power spectrum at specific frequencies to estimate alternating amplitudes. However, these methods place extremely high demands on signal stability, usually requiring subjects to reach a high target heart rate under exercise load or specific drug effects and maintain a stable state for a prolonged period. This is often difficult to achieve for elderly individuals or patients with underlying diseases. Furthermore, some studies indicate that due to the difficulty in maintaining strict heart rate stability over long periods or limitations caused by noise interference, a significant proportion of test results are classified as "non-deterministic," failing to provide effective clinical guidance. Additionally, frequency domain analysis often uses fixed time windows to extract T-wave signals. When subjects experience QT interval prolongation or dynamic changes in T-wave morphology, the fixed window is prone to missing crucial information such as the T-wave termination, leading to false negative interpretations.

[0018] In time-domain analysis, some improved moving average techniques calculate alternating amplitudes by constructing odd-even heartbeat templates. While this reduces the stringent requirements for heart rate stability to some extent, its core logic still has shortcomings. On the one hand, these methods are often limited to calculating the difference in the absolute amplitude of the T-wave peak, ignoring the subtle changes in the slope, symmetry, and microscopic morphology of the T-wave, which are often important indicators of early myocardial ischemia. On the other hand, currently used thresholds are mostly based on risk assessments of malignant arrhythmias, which are significantly higher in magnitude than the tiny signal fluctuations at the microvolt level during early ischemia. Directly using such thresholds can easily lead to missed detections of early ischemia risks.

[0019] Furthermore, current technologies for signal interception and feature extraction mostly employ static strategies. Studies have shown that in individuals with prolonged QT intervals, traditional fixed T-wave intercept windows often fail to fully cover the entire T-wave duration, leading to incomplete extraction of alternating signals. Although some researchers have attempted to introduce machine learning models for classification, existing models are largely based on traditional energy or peak features, lacking in-depth modeling of ischemic pathological mechanisms, and are highly dependent on microvolt-level labeled data, making it difficult to achieve stable generalization across different individuals.

[0020] Therefore, how to overcome the dependence of current related technologies on high heart rate conditions, overcome noise interference under normal conditions, and accurately capture microvolt-level electrophysiological cues highly related to ischemia from a multidimensional morphological perspective is a technical problem that urgently needs to be solved.

[0021] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0022] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0023] Figure 1 A flowchart illustrating an example of an electrocardiogram data processing method based on microvolt-level T-wave alternation according to an embodiment of this application is shown.

[0024] Regarding the execution entity of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as an electrocardiogram (ECG) data analysis / monitoring platform. It executes ECG data processing program instructions (e.g., algorithm modules) stored in memory to process multi-lead ECG data and generate corresponding risk warning outputs. In some examples, it can be integrated into a computer device through software, hardware, or a combination of both. This computer device can be various electronic devices or terminals with computing capabilities.

[0025] like Figure 1 As shown, in step S110, multi-lead ECG data to be processed is acquired, and preprocessing and QRS complex localization are performed on the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat time point information.

[0026] In some implementations, multi-lead ECG data may be derived from 12-lead static ECG, Holter monitoring, or wearable multi-lead acquisition devices; the sampling rate may be 250Hz, 500Hz, or higher, and may carry lead identifiers, timestamps, and device gain information, which are not limited here.

[0027] Preprocessing of ECG data can be varied and may include: baseline drift suppression (e.g., using high-pass or morphological filtering to suppress low-frequency drift components), power line interference suppression (e.g., 50 / 60Hz notch filtering or adaptive suppression), EMG and high-frequency noise suppression (e.g., low-pass or band-pass filtering to limit bandwidth), and may include lead quality assessment (e.g., calculating short-time energy, saturation ratio, spike interference ratio, signal-to-noise ratio), marking substandard leads as "low-confidence leads" for deweighting or removal during subsequent fusion. Furthermore, to ensure cross-lead consistency, preprocessing can employ uniform filter parameters and maintain phase consistency (e.g., using zero-phase filtering or linear-phase filtering) to avoid introducing additional "spurious differences" between leads due to phase distortion.

[0028] In QRS complex localization, R-peak detection can be performed independently for each lead, with consistency correction at the multi-lead level. For example, candidate R-peak sets are first output for each lead, then the global R-peak time point is determined based on the principle of "simultaneous occurrence in multiple leads," and abnormal detection points in isolated leads are corrected or discarded. Furthermore, QRS onset / offset points can be located for more precise constraints on subsequent T-wave intervals, and the time point information of each heartbeat is organized into a structured heartbeat index (e.g., including heartbeat number, R-peak time, RR interval, QRS onset / offset times, heartbeat validity markers, etc.). Thus, while ensuring the authenticity of waveform timing, major noise sources are suppressed, and unified, traceable heartbeat anchor point information is output.

[0029] In step S120, based on the heartbeat time point information, the T-wave analysis interval is adaptively determined for each heartbeat in the electrocardiogram waveform data, and the T-wave curve segment corresponding to each heartbeat is obtained by dynamic T-wave segmentation, thereby generating a set of T-wave segments.

[0030] In some implementations, for each heartbeat, a candidate T-wave window that varies with heart rate is given by taking the R-peak time as a reference and combining the adjacent RR intervals. For example, the starting point of the analysis window is set in the delayed region after the end of the QRS (to avoid QRS remnants and early ST segment disturbances), and the ending point is set in the safe region before the next QRS (to avoid overlap with the P wave / QRS of the next heartbeat). In addition, to enhance adaptability, the boundary of this window can be determined by combining proportional constraints and physiological boundary constraints: the proportional constraint refers to the boundary scaling with RR (e.g., the starting point and ending point are respectively taken within a certain proportional range of RR), and the physiological boundary constraint refers to the minimum delay that must not be earlier than the QRS-offset and the minimum advance that must not be later than the next R-peak. When an abnormal RR is encountered (e.g., significantly too short or too long), the heartbeat can be marked as an "unstable heartbeat" by RR outlier detection (e.g., the relative median RR deviation exceeds a threshold), and removed or treated separately in subsequent alternation analysis to avoid T-wave window shift caused by sudden heart rate changes.

[0031] After determining the candidate window, dynamic T-wave segmentation is performed to generate T-wave curve segments. This segmentation can refine the T-wave start and end points by fusing morphological constraints and slope / curvature constraints. For example, the T-peak position (such as the point of maximum amplitude or maximum energy) is searched within the candidate window, and then backtracking from the T-peak to both sides to find the boundary points that satisfy the condition that the absolute value of the slope decreases to a threshold, the curvature change tends to stabilize, or the point returns to the vicinity of the baseline as the T-wave start and end points. If there are bimodal T-waves or complex morphological cases, the main T-wave segment can be preferably determined with the peak with the highest energy proportion of the main peak as the center, or the bimodal information can be retained as segmentation features for compatibility with subsequent weighting functions. Finally, a set of beat-by-beat T-wave segments is output for each lead, and the segments can be normalized and aligned to a baseline (e.g., using the mean of the short window at the beginning of the segment as the zero baseline) to reduce residual baseline drift. This ensures that the T-wave segments of each heartbeat are comparable in terms of time coverage and morphology, reduces the "truncation inconsistency" error caused by fixed window truncation, and makes microvolt-level morphological differences less likely to be masked or amplified by window boundary drift.

[0032] In step S130, the T-wave curve segments in the T-wave segment set are divided into odd-numbered sequences and even-numbered sequences according to the heartbeat sequence number, and the even-numbered sequence T-wave curve segments and the odd-numbered sequence T-wave curve segments corresponding to the time sequence are combined to form paired T-wave segments.

[0033] In some implementations, alternating analysis can be performed within a sliding time window (e.g., taking N consecutive valid heartbeats to form an analysis window), and only heartbeats that satisfy the condition of "heartbeat validity being marked as true" are included in the odd-even sequence. When a heartbeat is eliminated (e.g., noisy heartbeats, premature beats, or abnormal RR caused by missed / false detections), a skipping strategy that maintains temporal consistency can be adopted. That is, invalid heartbeats are skipped and the sequence number of the next valid heartbeat is redefined during sequence construction to maintain the continuity of the odd-even alternation relationship and avoid the pairing logic being disrupted by a single abnormal heartbeat.

[0034] For example, a pairing relationship can be defined as: the first even number in a sequence k The segment and the first segment in the odd sequence k Each segment constitutes a pair; in the case of multiple leads, this pair shares the same set of heartbeat indices across all leads to ensure synchronous comparison of the same pair of heartbeats across different leads. By fixing the comparison unit of "adjacent heartbeat alternation difference" into a standardized paired structure, the alternation detection is constrained from "any two-beat difference" to "strictly odd-even paired difference," giving the subsequent weighted alternation amplitudes a clear alternation semantics.

[0035] In step S140, timing alignment processing is performed on each paired T-wave segment to eliminate phase deviation on the time axis, resulting in aligned paired T-wave curve segments.

[0036] Specifically, the two T-wave segments can be resampled to a uniform length first. L (For example, using spline interpolation or equidistant resampling) to ensure that subsequent alignment and differencing are performed on a uniform sampling grid; then, two levels of alignment are performed: the first level is global translation alignment, which searches for the offset that maximizes the similarity between two segments within a limited time offset range (similarity can be characterized by normalized cross-correlation or minimization of mean square error) to correct the overall peak position shift; the second level is local fine-tuning alignment, which allows slight local scaling (e.g., piecewise linear time mapping or restricted local alignment) while maintaining the monotonicity of the time mapping, but imposes constraints on the scaling amplitude to avoid excessive deformation that fits noise into the "alignment result".

[0037] Furthermore, if anomalies are detected after alignment (such as requiring an offset beyond the allowable range for alignment, or the residual structure exhibiting a clearly non-physiological morphology after alignment), the pairing can be marked as a "low-confidence pairing." This decouples the differential calculation from "temporal misalignment differences," allowing the differential waveform to primarily reflect true morphological differences rather than peak drift. Especially at the microvolt scale, slight phase deviations are often sufficient to significantly increase the differential amplitude or cancel out the true alternation. Constrained alignment can significantly reduce the dominant role of this structural error in alternation quantization, thereby improving the reproducibility of alternation amplitudes across different heartbeats and time periods.

[0038] In step S150, a weighting function that takes into account the sensitivity of the T wave peak region and the T wave slope change region is constructed, and the weighting function is used to perform weighted calculation on the differential waveforms between the aligned paired T wave curve segments to obtain a weighted alternating amplitude that reflects the microvolt level morphological differences of adjacent heartbeats' T waves.

[0039] Here, the weighting function can be designed to cover two key regions simultaneously: one is the region near the peak of the T wave, which is most sensitive to amplitude alternation; the other is the region of slope change of the rising / falling branch of the T wave, which is more sensitive to subtle changes in shape (such as slope, inflection point drift).

[0040] Specifically, the position of the T-peak (e.g., the point of maximum amplitude) can be determined first on the aligned segment, and the derivative sequence of the segment can be calculated to characterize the intensity of slope change. Then, weights are constructed: peak weights can adopt a concentrated weighting centered on the T-peak (e.g., higher weights near the peak and decreased weights further away), and slope weights can be positively correlated with the absolute value of the slope or the rate of change of the slope, giving higher attention to slope abrupt change segments. Finally, the two weights are merged and normalized according to a preset ratio, so that the weight sum is 1, thereby avoiding the drift of the total weight due to different segment lengths or amplitude scales. Furthermore, to suppress the misleading effect of peak noise on the slope weights, the derivative sequence can be smoothed or truncated first (e.g., limiting the amplitude of slope values ​​exceeding a reasonable range) to ensure that the weight function reflects stable morphological information (rather than instantaneous noise).

[0041] In some implementations, a differential waveform can be constructed first in the weighted alternating amplitude calculation. (Or the opposite definition, but it must remain consistent throughout), then calculate the weighted alternating amplitude. ,in The absolute value can be chosen to characterize the intensity of amplitude difference, or the squared term can be chosen to enhance the contribution of larger differences. When robustness to noise is required, a robust function that suppresses large-amplitude anomalous differences (such as piecewise linear suppression) can be used to avoid single-point anomalous differences dominating the amplitude results. This allows the alternation measure to focus more on key segments that are more sensitive to "weak morphological differences" at the numerical level, while reducing the influence of low-information segments and occasional interference on the results, thereby improving the detectability and stability of microvolt alternation characteristics without changing the physical meaning of the original signal.

[0042] In step S160, multi-lead fusion processing is performed based on the weighted alternating amplitude corresponding to each lead to generate a global alternating detection index, and ECG abnormality risk warning information for indicating the risk of myocardial ischemia is output according to the global alternating detection index.

[0043] More specifically, a lead quality factor can be established for each lead (e.g., obtained by combining lead quality assessment indicators, alignment residuals, low confidence pairing ratios, etc.), and the alternation amplitude of that lead can be weighted accordingly. Subsequently, robust aggregation of the alternation amplitudes of multiple leads can be performed, for example, by using weighted mean, weighted median, or energy-based fusion (e.g., taking the square root after weighted summation of the squares of the amplitudes of each lead) to obtain a global alternation detection index. This allows the alternation trend consistently supported by most leads to contribute more to the global index, while occasional abnormalities in a single lead are unlikely to significantly improve the global results.

[0044] Furthermore, when outputting risk warning information based on global alternating detection indicators, the warning information can be organized into machine-readable structured results (e.g., including time period, global indicator value, risk level, confidence level, set of leads involved in fusion, proportion of excluded heartbeats, etc.) for record-keeping and backtracking in daily monitoring systems. Thus, utilizing multi-lead redundancy information improves the robustness and reproducibility of the detection, ensuring that the final output not only provides risk warning conclusions but also a credibility representation coupled with data quality and lead consistency, thereby better adapting to the stable application needs of routine ECG data monitoring environments.

[0045] Regarding the implementation details of data preprocessing and noise reduction in step S110, in some examples of embodiments of this application, baseline drift correction and power frequency interference suppression are performed on multi-lead ECG data, and empirical mode decomposition algorithm is used to decompose the corrected signal.

[0046] In some implementations, the original multi-lead ECG data is first linearly processed using a bidirectional zero-phase high-pass filter (e.g., with a cutoff frequency of 0.5Hz) and a notch filter to eliminate baseline drift and power line interference, preventing phase distortion. Subsequently, considering the non-stationary and nonlinear characteristics of ECG signals, an Empirical Mode Decomposition (EMD) algorithm is used to adaptively decompose the linearly filtered signal, breaking down the complex ECG waveform into several Intrinsic Mode Function (IMF) components arranged in descending order of frequency. This allows for the rapid removal of deterministic interference through linear means, and the use of EMD to separate the signal at local characteristic scales, thereby physically separating broadband random noise from the effective ECG components at different modes.

[0047] Then, based on the spectral characteristics and energy distribution of each intrinsic mode component, the high-frequency noise-dominated component is identified and removed, and the remaining components are reconstructed to generate preprocessed electrocardiogram waveform data.

[0048] In some implementations, after obtaining several IMF components, the spectral energy distribution characteristics (such as the centroid of power spectral density or energy percentage) of each IMF component are calculated. Typically, electromyographic interference and high-frequency random noise are mainly concentrated in the first few high-frequency modes (such as IMF1 to IMF2), exhibiting high frequency and relatively dispersed energy; while effective electrophysiological signals such as T waves are mainly distributed in the mid-to-low frequency modes.

[0049] Based on this principle, the system automatically identifies and removes modal components determined to be dominated by high-frequency noise, and then performs linear superposition and reconstruction of the remaining effective signal-dominant modes to generate preprocessed ECG waveform data. Thus, the adaptive screening and reconstruction mechanism effectively suppresses microvolt-level background noise while preserving the subtle morphological features of the T wave (such as peak roundness and fine notches), avoiding waveform smoothing distortion that may be caused by traditional low-pass filtering.

[0050] Subsequently, based on the preprocessed ECG waveform data, candidate R wave positions for each lead were extracted.

[0051] In some implementations, differential thresholding or wavelet transform modulus maxima methods can be used to dynamically adaptively threshold and lock local maxima in the signals of each lead, marking them as candidate R-wave locations for that lead. This process does not rely on the inter-lead relationships but rather maximizes the capture of all possible excitation moments from a single channel. Thus, by independently detecting each lead, it is ensured that even in cases of poor contact or weak signals in some leads, other leads can still provide effective cardiac cues, thereby establishing a comprehensive and highly sensitive initial candidate cardiac pool.

[0052] Furthermore, the candidate R wave positions are screened using a multi-lead collaborative verification strategy. Candidate R waves that appear simultaneously in at least a preset number of leads within a preset time tolerance range are identified as valid QRS complexes, thereby obtaining the heartbeat time point information for each heartbeat.

[0053] In some implementations, the system sets a very short time tolerance window (e.g., ±10 ms) and counts the number of leads in which candidate R waves are detected within this window. Only when candidate R waves appear simultaneously in at least a preset number of leads (e.g., more than half the total number of leads or a fixed three) within the same time window is the location determined as a true valid QRS complex, and the weighted average of the R wave times in each valid lead is used as the final heartbeat timing information. Thus, by utilizing the spatial synchronicity of cardiac electrical activity projected onto the body surface, false positives caused by single electrode loosening or local electromyographic artifacts are effectively eliminated, significantly improving the specificity and robustness of heartbeat localization.

[0054] Regarding the implementation details of generating the T-wave fragment set in step S120, in some examples of embodiments of this application, the first T-wave fragment is determined based on the heartbeat time point information. The peak moment of the R wave of a heartbeat The current RR interval is calculated based on the peak time of the R wave in adjacent heartbeats. .

[0055] Specifically, the system searches for the voltage maximum point within a small window near the heartbeat time point as the R-wave peak, and calculates... get This established the time reference anchor point for subsequent T-wave analysis and incorporated heart rate variability (HRV) into consideration; since the duration of the T wave is highly correlated with heart rate, real-time acquisition... This provides a physiological basis for the subsequent dynamic adjustment of the search window, ensuring that the algorithm can adapt to different heart rate states such as bradycardia or tachycardia.

[0056] Then, based on Construct a dynamic search interval that can cover long QT interval features, and set the time range of the dynamic search interval to be [missing information]. .

[0057] It should be noted that the coefficient of 0.15 is used to skip the initial isoelectric period of the ST segment and potential interference near the J point, while the coefficient of 0.6 is significantly wider than the end position of the conventional physiological T wave, realizing a "high tolerance" search strategy. It is specifically optimized for the phenomenon of QT interval prolongation (Long QT) caused by myocardial ischemia or drug effects. Compared with a fixed duration search window, this dynamic range can effectively prevent waveform truncation (i.e., "tailing") caused by delayed T wave appearance or prolonged duration, thereby ensuring the integrity of ischemic information.

[0058] Then, within the dynamic search interval, indexed by sampling points. Representing the discrete-time position within the dynamic search interval, calculate the first derivative sequence of the electrocardiogram waveform data. (Reflecting slope changes) and second derivative sequence (Reflecting curvature changes), and statistically analyzing the maximum modulus of the first derivative within the dynamic search interval. ,Will Determined as an adaptive slope threshold, where This is the preset scaling factor.

[0059] More specifically, An empirical value between 0.1 and 0.3 (e.g., 0.2) can be used to dynamically adjust the threshold based on the steepness of the current heartbeat T wave. This eliminates the need for a fixed voltage threshold and instead uses a relative slope threshold related to the waveform morphology, making the algorithm highly robust to differences in signal amplitude across different leads and baseline drift. Even low-amplitude, inverted ischemic T waves can achieve suitable detection sensitivity.

[0060] Then, a forward search is performed within the dynamic search interval to calculate the first derivative. The magnitude of the curve exceeded the adaptive slope threshold for the first time. The time marker is the T-wave initiation point. This allows us to find the "takeoff point" where the waveform rapidly climbs (or falls) from the relatively flat state of the ST segment to the main peak of the T wave. It can accurately separate the equipotential ST segment from the main repolarization process, avoiding misjudgment of the starting point caused by ST segment offset.

[0061] Then, in determining the T-wave initiation point The search continues until a location is found that satisfies the first derivative. The modulus value falls below the adaptive slope threshold. And at the same time, the second derivative The moment of zero crossover is marked as the end of the T wave. .

[0062] It should be noted that for the T-wave endpoint For localization, the algorithm employs a dual-constraint strategy: after determining the starting point, it continues searching to find a location that simultaneously satisfies the condition that "the magnitude of the first derivative falls back to below..." "and the second derivative" The moment of "zero crossover" is when the second derivative zero crossover point corresponds to the inflection point of the waveform curve, which usually corresponds to the transition position where the tail of the T wave descends from steep to gentle, that is, near the tangent point where it merges with the baseline.

[0063] This solves the technical problem that the T-wave ends often have severe tailing and are difficult to define; compared with the simple regression baseline method, the introduction of second derivative constraints can more accurately identify the fine boundary between the end of the T-wave and the beginning of the U-wave or the noise baseline, preventing non-T-wave components from being included in the analysis.

[0064] Furthermore, based on the determined T-wave initiation point With the end of the T wave The T-wave curve segment of the current heartbeat is completely extracted from the electrocardiogram waveform data, and the T-wave curve segments of each heartbeat are collected to generate a T-wave segment set.

[0065] More specifically, the system stores the extracted data segments as vectors and retains their original amplitude information for subsequent microvolt-level difference analysis, thereby generating a set of pure T-wave datasets that have been rigorously morphologically defined, eliminating phase noise introduced by heart rate fluctuations and positioning errors.

[0066] Regarding the implementation details of pairing construction in step S130, before combining the even-numbered sequence T-wave curve segments and the odd-numbered sequence T-wave curve segments corresponding to the time series into paired T-wave segments, the system can first perform a template learning step based on historical data. Specifically, based on the T-wave curve segments corresponding to a preset number of historical heartbeats before the current heartbeat, a reference T-wave template is constructed through statistical aggregation after time axis alignment and amplitude normalization.

[0067] Specifically, the system maintains a first-in-first-out (FIFO) buffer queue containing a preset number (e.g., the past 20 to 50) valid heartbeats preceding the current heartbeat. The T-wave curve segments in the queue are centroidally or peak-aligned, and amplitude normalization is performed to eliminate the effects of baseline drift and respiratory amplitude modulation. Subsequently, a robust statistical aggregation method (such as pointwise median calculation or trimmed mean) is used to generate the current reference T-wave template. This constructs a baseline waveform that represents the "standard" T-wave morphology under the current physiological state, effectively filtering out random noise in a single heartbeat using statistical averaging, providing a reliable morphological benchmark for assessing the quality of a single heartbeat.

[0068] Then, the morphological cross-correlation coefficient between each T-wave curve segment in the T-wave segment set and the reference T-wave template is calculated, and the residual noise energy characterizing the degree of waveform difference is calculated. Based on the morphological cross-correlation coefficient and the residual noise energy, a segment quality score is generated.

[0069] More specifically, on the one hand, the morphological cross-correlation coefficient between the current T-wave segment and the reference template is calculated to quantify the similarity of the overall waveform trend; on the other hand, the differential residual energy or high-frequency component energy of the two is calculated as a residual noise energy index. For example, the system weights high cross-correlation coefficients with low residual noise energy and maps them to a higher segment quality score. Thus, an objective signal quality evaluation system is established, capable of sensitively identifying waveform distortions caused by electromyographic interference, poor electrode contact, or sudden motion artifacts, thereby distinguishing non-pathological signal distortion from true pathological microvolt alternation.

[0070] Then, T-wave curve segments with a quality score below a preset validity threshold are marked as abnormal artifact segments, and abnormal artifact segments are prohibited from directly participating in the pairing of even-numbered and odd-numbered sequences.

[0071] In some implementations, T-wave curve segments with scores below the threshold are marked as anomalous artifact segments. In the subsequent odd-even sequence pairing logic, the system forcibly prevents these marked artifact segments from directly participating in pairing combinations as "even" or "odd" elements. Thus, by employing a proactive "circuit breaker" protection mechanism, low-quality data is prevented from entering the subsequent high-sensitivity microvolt alternation analysis stage, preventing situations where large artifacts mask weak alternation signals or generate false positive alternation readings.

[0072] Furthermore, for the missing pairing positions caused by the marking of abnormal artifact segments, the effective T-wave curve segments adjacent to the abnormal artifact segment on the time axis are retrieved, and the abnormal artifact segment is replaced and filled by the effective T-wave curve segments to ensure that the generated paired T-wave segments meet the sequence continuity requirements.

[0073] Here, for index gaps (i.e. missing pairing positions) left in odd or even sequences due to the removal of abnormal artifact fragments, the system executes a signal compensation strategy based on the principle of temporal nearest neighbor.

[0074] Specifically, the system searches backward or forward for the nearest valid T-wave curve segment on the time axis to the missing position. If valid segments exist before and after, linear weighted interpolation is used to generate a replacement segment. If a valid segment exists only on one side, an amplitude-adapted replication strategy is used. After filling the gap with these generated replacement segments, odd-even pairing is performed. Thus, while ensuring data purity, the integrity of the data stream is automatically restored, ensuring that paired T-wave segments meet the sequence continuity requirements, effectively maintaining the rhythmic stability required for microvolt-level T-wave alternation (ABAB mode) analysis, and avoiding analysis window resets or spectral leakage caused by data interruption.

[0075] Regarding the implementation details of step S140, in some examples of embodiments of this application, the even-numbered sequence T-wave curve segment in the paired T-wave segment is set as the reference sequence. The odd-numbered T-wave segments in the paired T-wave segments are set as the sequences to be aligned. ,in and These are discrete time series, and their number of sampling points are respectively and .

[0076] Here, the paired T-wave segments are assigned roles. The system designates the even-numbered T-wave segment from the paired sequence as the reference sequence. (Usually, even-numbered heartbeats are physiologically more stable), so odd-numbered T-wave curve segments are designated as sequences to be aligned. At this point, due to the influence of micro-HRV (micro-heart rate fluctuations), the sampling point lengths of both are... and They may not be equal.

[0077] Then, to eliminate the nonlinear time-domain distortion caused by heart rate fluctuations, a dynamic time warping algorithm is used to construct a local cost matrix, and a dynamic programming strategy is used to search for warping paths in the local cost matrix that satisfy monotonicity and continuity constraints. And determine the optimal bending path that minimizes the cumulative distance from each bending path. The cumulative distance along the path satisfy: Equation (1) In the formula, For curved paths Index point pairs on, and Reference sequences With the sequence to be aligned Time axis index, Represents the reference sequence In the index The amplitude at that point, Represents the sequence to be aligned In the index The amplitude at that point, This represents the absolute value of the amplitude difference.

[0078] Here, to eliminate nonlinear time-domain distortion, the algorithm employs Dynamic Time Warping (DTW) technology to construct a dimension of... The local cost matrix. When searching for the optimal path, the system strictly imposes monotonic constraints (time index). The system employs a constraint that allows only increases, not decreases, to prevent time reversal, and a continuity constraint (index step size limit to prevent signal breakage). Under these constraints, a dynamic programming strategy is used to traverse the matrix and find the starting point. To the finish line curved path Therefore, through the elastic matching mechanism, the waveform is allowed to be locally warped on the time axis, thereby simulating the non-uniform scaling characteristics of the cardiac repolarization process at different heart rates.

[0079] Specifically, in the path search process, the core optimization objective is to minimize the cumulative distance. Equation (1) specifically expresses the path along a curved path. The total cost of morphological differences, of which As a coupled index pair, it represents the... The points and The Align the points, and The instantaneous amplitude difference between the two points under aligned conditions was then calculated. Through iterative comparison, the factor that makes this cumulative distance... The path that reaches the minimum value is the optimal curved path. Therefore, a time mapping relationship that maximizes morphological similarity is found, rather than a simple overlap of moments, thus ensuring that even if the waveform changes in width, its corresponding anatomical feature points such as peaks and troughs can still be correctly associated.

[0080] Furthermore, based on the optimal bending path The determined index mapping relationship is used for the sequences to be aligned. The time axis is subjected to nonlinear time scaling mapping, and the mapped sampling points are interpolated and resampled to generate aligned odd-numbered sequence T-wave curve fragments. This makes its key morphological features and reference sequence Phase synchronization is achieved in the time dimension.

[0081] In some implementations, since the mapped time points may no longer fall on the original sampling grid, the system further performs interpolation resampling (e.g., using cubic spline interpolation or linear interpolation) to recalculate the amplitude based on the mapped time positions, generating a sequence with the same length as the reference sequence (i.e., a length of...). Aligned odd sequence Thus, the phase deviation caused by heart rate fluctuations is truly eliminated mathematically, and phase synchronization of the two sequences is achieved at key morphological feature points (such as the T wave peak and the maximum slope point).

[0082] Finally, the reference sequence Aligned odd-numbered sequence T-wave curve fragments This forms aligned, paired T-wave curve segments. At this point, and Not only are they of equal length, but they have also been calibrated in the time dimension, eliminating phase errors. This ensures that the differential waveforms calculated subsequently will purely reflect the microvolt-level pathological alternation of the T wave in amplitude and morphology (i.e., the difference in the vertical direction), without being mixed with false high-frequency differential noise caused by time misalignment, which significantly improves the signal-to-noise ratio of alternating detection.

[0083] Regarding the implementation details of step S150, in some examples of embodiments of this application, the reference sequence determined in the timing alignment process is extracted. The morphological features, including the reference sequence Peak position index and the discrete first derivative at each sampling point .

[0084] In this embodiment, the weighting function is constructed based on a reference sequence determined during the temporal alignment process. Specifically, system traversal Locate the index corresponding to the point with the maximum amplitude among all sampling points. This point represents the main energy center of ventricular repolarization; simultaneously, the discrete first derivative at each point in the sequence is calculated using a difference algorithm. This derivative sequence reflects the rate of rise and fall and the steepness of the T-wave waveform. Thus, the "skeleton" features of the T-wave were extracted, and key time regions (such as peaks and points of maximum slope) were identified.

[0085] Then, construct a combined weight function based on morphological feature sensitivity. The combined weighting function is composed of a Gaussian decay term focused on the peak region and a derivative enhancement term focused on morphological distortion: Equation (2) In the formula, For sampling point index, To control the standard deviation parameter of the range of interest in the peak region, These are weighting coefficients used to adjust the sensitivity to changes in T-wave slope and waveform broadening characteristics. This indicates taking the absolute value.

[0086] Based on the extracted morphological features, the system constructs a combined weight function. Its design logic employs a "dual concern mechanism" as shown in equation (2). Specifically, It consists of two superimposed parts: the first part is based on Gaussian decay term It exhibits a bell-shaped distribution, used to assign the highest weight to the peak region where T-wave energy is most concentrated, and through the standard deviation parameter The second part is the derivative enhancement term, which controls the scope of focus. It is proportional to the absolute value of the waveform's slope and is used to enhance sensitivity to regions with significant slope changes, such as the rising and falling limbs of the T-wave. Therefore, by adjusting the weighting coefficient... , It has the ability to flexibly capture both "T-wave amplitude alternation" and "T-wave morphological distortion (such as widening or slowing down)".

[0087] Furthermore, by utilizing the combined weight function For the reference sequence Aligned odd-numbered sequence T-wave curve fragments A weighted average is calculated for the absolute difference sequences of amplitudes between them to obtain weighted alternating amplitudes. : Equation (3) In the formula, The sampling point length of the aligned paired T-wave curve segment. Characterizes the microvolt-level morphological differences of adjacent heartbeat T waves after phase deviation has been eliminated.

[0088] After determining the weight distribution, the system uses formula (3) to calculate the weighted alternating amplitude. It employs a weighted average absolute error (WMAE) calculation process. In equation (3), the numerator represents the summation of the microvolt-level differences at each sampling point multiplied by their corresponding morphological importance weights; the denominator is a normalization factor, thus supporting simultaneous signal enhancement and noise suppression. Specifically, in regions rich in pathological information, such as T-wave peaks or steep slopes, minute differences are amplified by weights; while in regions with flat waveforms or near the baseline (typically with low signal-to-noise ratios), differences are suppressed by weights. Therefore, the final output... It is not just an amplitude difference, but also a high signal-to-noise ratio alternation index that has been morphologically weighted and corrected.

[0089] Compared to traditional, simple "peak subtraction" or "full-band averaging" methods, the calculation method based on equations (2) and (3) in this application embodiment can keenly capture subtle morphological changes caused by early myocardial ischemia (such as local differences caused by T wave flattening or asymmetric changes), effectively solving the problem that microvolt-level T wave alternation is easily submerged by background noise. As a result, the detection sensitivity of the algorithm in the low alternation amplitude range (such as 1~5 microvolts) is significantly improved, which can provide more discriminative hidden ischemia risk warning information for clinical use.

[0090] Regarding the implementation details of generating the global alternating detection index in step S160, in some examples of the embodiments of this application, the various leads of the multi-lead ECG data are traversed. Extract the lead reference sequence determined during the timing alignment process for this lead. And calculate the signal energy of the reference sequence for that lead. As initial fusion weights: Equation (4) In the formula, For lead traversal variables, , Total number of leads; For lead reference sequence, indicating In the Specific examples on each lead; Represents the lead reference sequence In the The amplitude of each sampling point.

[0091] It should be noted that the first step in multi-lead fusion is to assess the basic reliability of the signals from each lead. Specifically, the system traverses all leads and extracts the reference sequence determined during the timing alignment phase for each lead. The signal energy is calculated using equation (4), which physically represents the total power integral of the T wave in the time domain. Since the projection of the electrocardiogram vector onto the body surface varies depending on the lead location, leads with higher T wave amplitudes typically contain richer electrophysiological details and are relatively less affected by quantization noise; therefore, the signal energy... As an initial weight, it ensures that the dominant leads with full morphology have a greater say in the final result, while low or weak leads contribute less.

[0092] Then, obtain the first Background noise level of each lead The background noise level is determined based on the amplitude standard deviation within a preset baseline window outside the T-wave analysis interval of the lead reference sequence, and / or the standard deviation of the high-frequency residuals after performing a preset high-pass filtering on the lead reference sequence.

[0093] Here, while calculating the initial weights, the system needs to acquire the "cleanliness" index of each lead, i.e., the background noise level, in parallel. In practice, an isoelectric segment outside the T-wave analysis interval (such as the TP segment) can be selected as the baseline window, and its amplitude standard deviation can be calculated; or a high-pass filter (e.g., above 20Hz) can be applied to the T-wave sequence, and the standard deviation of the high-frequency residuals can be calculated. This allows for real-time quantification of the degree to which the current lead is affected by poor contact or baseline drift, helping to prevent misleading analysis results from high-energy but also high-noise inferior leads.

[0094] Then, combined with a preset noise threshold With attenuation coefficient Calculate the first Corrected fusion weights for each lead : Equation (5) In the formula, Used to control the penalty level for high-noise leads.

[0095] In equation (5), a piecewise nonlinear penalty logic is employed. When the lead noise is below a preset safety threshold... At that time, the signal is considered pure, and its initial energy weight is maintained. The noise remains unchanged; once the noise exceeds the threshold, an exponential decay term is introduced. The weights are drastically compressed. Among them, the attenuation coefficient... This determines the severity of the punishment. The larger the threshold, the lower the tolerance for excessive noise. Therefore, a "soft threshold" filtering mechanism was constructed. Compared with directly discarding noisy leads, this mechanism retains some of the effective information contained therein, but greatly reduces its interference weight on the global result through mathematical means, thus significantly improving the robustness of the algorithm in complex clinical environments.

[0096] Furthermore, by utilizing the modified fusion weights The weighted alternating amplitudes calculated for each lead Perform a weighted average calculation to obtain the global alternating detection index. : Equation (6) In equation (6), the numerator sums the weighted alternating amplitudes of each lead (i.e., the result of a single lead). Multiply by the adjusted weight The denominator is the sum of all adjusted weights, used for normalization. Thus, the output... It is no longer a single-view observation, but a robust indicator that integrates comprehensive cardiac electrical activity information and has undergone signal-to-noise ratio weighting correction. As a result, it can effectively avoid the problem of "single-lead blind zone" (i.e., the ischemic area projects as zero in a specific lead) and suppress the influence of local artifacts.

[0097] Regarding the implementation details of outputting ECG abnormality risk warning information in step S160, in some examples of embodiments of this application, within a preset analysis window, morphological feature parameters of each cardiac T-wave curve segment are extracted from the T-wave segment set, and statistical representative values ​​of the morphological feature parameters within the analysis window are calculated. The morphological feature parameters include the T-wave width. Slope symmetry ratio and ST segment offset The statistical representative values ​​include the mean and / or median.

[0098] Here, macroscopic features of the T-wave morphology within the analysis window are extracted and statistically reduced. More specifically, although the preceding steps focus on alternating changes in microvolts, the ECG data characteristics of myocardial ischemia often also express macroscopic changes in the T-wave morphology. Within a preset analysis window (e.g., 128 heartbeats), the system calculates the T-wave width beat by beat. (Reflecting the duration of repolarization) and the proportion of slope symmetry (Reflecting waveform asymmetry caused by ischemia) and ST segment offset (Reflecting the injury current). Subsequently, statistically representative values ​​of these parameters within this window are calculated (preferably the median to combat individual abnormal heartbeats, or a truncated mean can be used). This resolves the issues of "large fluctuations in single-beat characteristics" and "window-level alternation indicators." The problem of "dimensional mismatch" is addressed by compressing high-frequency, frame-by-frame morphological features into stable window-level statistical features, laying the data foundation for multidimensional fusion.

[0099] Then, construct a system containing global alternation detection metrics. Multidimensional eigenvectors of statistical representative values ​​of morphological feature parameters ;in , This represents the vector transpose operation.

[0100] Here, based on the calculated statistical representative values, the system constructs a multidimensional feature vector. This vector is essentially a physiological descriptor that combines dynamic and static elements, in which... This represents the dynamic instability of myocardial repolarization at the microscopic level. This represents a static structural distortion (SSD) in the repolarization process at the macroscopic level. Therefore, through feature fusion, the system can capture hidden risks that are difficult to detect with a single indicator. For example, in some early ischemic cases, the ST segment may not yet have shown significant shift (negative conventional ECG), but... A slight increase has been observed, combined with Slight changes, feature vector This allows for a keen characterization of the multi-source pathological distribution.

[0101] Next, the multidimensional feature vectors Input a pre-trained logistic regression model and use the Sigmoid activation function to calculate the risk score of ischemia-related electrocardiographic abnormalities. : Equation (7) In the formula, For the intercept term, These are the model regression coefficients for the corresponding feature dimensions.

[0102] In equation (7), the linear part Weighted summation of features was implemented, where the regression coefficients The magnitude and sign of the values ​​were derived from training with a large amount of clinical labeled data, quantifying the contribution of each feature to ischemic events (e.g., Generally, a positive value indicates that the higher the alternating amplitude, the greater the risk. (Outer Sigmoid function) It then acts as a nonlinear activation function, converting linear results whose values ​​can range from negative infinity to positive infinity. The parameters are forcibly mapped to the (0, 1) interval. This transforms the complex combination of physiological parameters into an intuitive probability value. Compared to a simple hard threshold determination, the probabilistic output can smoothly reflect the continuous changes in risk level.

[0103] Furthermore, based on the risk score of ischemia-related electrocardiographic abnormalities... It outputs ECG abnormality risk warning information; the ECG abnormality risk warning information includes risk score value and risk level obtained according to preset risk threshold classification.

[0104] In some implementations, the system not only outputs The specific value (e.g., 0.85) is also used to classify the current status into "low risk / observation / high risk" levels based on preset clinical grading thresholds (e.g., 0.3 for low risk and 0.7 for high risk). If the high-risk threshold is exceeded, the system can trigger a visual or auditory alarm and highlight the main feature contributing to the high score (such as prompting "alternating indicators are significantly abnormal").

[0105] Therefore, the ECG data analysis platform provides users with tiered auxiliary diagnostic information, which can quantitatively display the level of ischemia and quickly screen out patient data that requires emergency intervention through a tiered mechanism, significantly improving the clinical early warning efficiency of ECG monitoring.

[0106] It should be noted that the ECG abnormality risk warning information, risk score, and risk level output in this application embodiment are quantitative reference data calculated based on ECG signal characteristics. They are only used as auxiliary analysis results and should not be directly used as the final clinical diagnostic conclusion. The final diagnosis still needs to be made by a professional physician based on the patient's clinical symptoms, medical history, and other imaging or biochemical test results.

[0107] Figure 2 A schematic diagram illustrating the operational mechanism of an example of an electrocardiogram data processing method based on microvolt-level T-wave alternation according to an embodiment of this application is shown.

[0108] like Figure 2As shown, this application embodiment constructs a closed-loop process from raw signal acquisition to final risk grading. First, the system performs cascaded denoising (linear filtering + EMD) and multi-lead collaborative QRS localization on the multi-lead ECG data acquired via a high-resolution sensor to establish a high signal-to-noise ratio waveform and accurate time reference. Subsequently, T-wave segments are dynamically segmented using an adaptive window covering long QT features, and nonlinear time alignment of paired sequences is performed using dynamic time warping (DTW) technology, effectively eliminating phase deviations before microvolt-level analysis. Next, the data stream enters the deep analysis stage, sequentially performing morphology-sensitive weighted alternation amplitude calculation (Multi-Point Interbeat Magnitude Alternans, MPIMA) and multi-lead signal-to-noise ratio fusion based on a noise penalty mechanism to generate a robust global alternation detection index. Finally, this global index is fused with multi-dimensional morphological features such as ST segment shift and slope symmetry, and input into a logistic regression model to calculate the ischemic risk score. It will output corresponding risk classification prompts.

[0109] To verify the effectiveness of the algorithm proposed in this application, an embodiment was constructed by simulation to build a microvolt-level T-wave alternation model and compared with traditional detection methods. The experiment first constructed a baseline T-wave, using a Gaussian function to simulate a normal T-wave with a peak amplitude of 100 µV and a width of 0.1 s, and set the heart rate to 100 bpm (corresponding to a heartbeat length of approximately 0.6 s). Then, 200 consecutive heartbeats were generated, and the peak values ​​of the T-waves of odd and even heartbeats were alternately increased and decreased by ±2 µV to simulate a microvolt-level alternation signal. Simultaneously, white noise with a mean of 0 and a standard deviation varying between 1 µV and 10 µV was superimposed, thereby constructing test data at different signal-to-noise ratio levels.

[0110] Based on this, the traditional peak difference method was applied as a baseline, and the detection method in this embodiment, which includes time alignment and morphological weighting calculation, was used to calculate the alternation amplitude of each heartbeat pair. By statistically analyzing the average detection index under different noise conditions and comparing the calculated alternation amplitude with a preset threshold (e.g., 1 µV), the correct detection rate of each method under different noise levels was calculated, and a two-dimensional accuracy heatmap was generated to intuitively evaluate and verify the accuracy and robustness of the algorithm in extracting microvolt-level features in a strong noise background.

[0111] Figure 3 A comparative diagram of the average alternation detection index under different background noise levels is shown. The horizontal axis represents the standard deviation of the superimposed white noise, and the vertical axis represents the calculated average alternation detection index. This comparative data is based on a constructed microvolt-level T-wave alternation simulation model, where a preset alternation amplitude is set. The noise standard deviation range is set to to .

[0112] like Figure 3 As shown, with the increase of background noise level, the detection index of the baseline peak difference detection method shows a slow upward trend; in contrast, the detection index of the weighted alternation detection method proposed in this application shows a more significant increase, and the detection value at the same noise level is always higher than that of the baseline method. This result indicates that, thanks to the sensitivity focus of the constructed weighting function on the T-wave peak region and slope change region, the method of this application has a stronger signal response and capture capability for subtle morphological alternations in strong noise environments, and can more effectively highlight alternation features, preventing them from being submerged by noise under low signal-to-noise ratio conditions.

[0113] Figure 4 A heatmap comparison of detection accuracy under different noise and signal amplitude conditions is shown.

[0114] like Figure 4 As shown, the horizontal axis of both heatmaps represents the set alternation amplitude ( The vertical axis represents the standard deviation of background noise. The color bar on the right indicates the proportion of correctly detected alternations, with a value range of 0.5 to 1.0. The closer the color is to yellow, the higher the accuracy; the closer it is to purple, the lower the accuracy.

[0115] contrast Figure 4 The left side of the figure (baseline method) and Figure 4 As can be seen from the heatmap distribution in the right-hand panel (the method of this application), the baseline method only exhibits a high detection rate in regions with low noise and large alternation amplitudes (the bright yellow area in the upper right corner); however, in the low signal-to-noise ratio region in the lower left corner (e.g., when the noise intensity exceeds...), the detection rate is significantly lower. And the alternation amplitude is less than When the detection accuracy drops sharply and turns dark purple, the value is generally below 0.6 (i.e. 60%), indicating that it is difficult to effectively extract weak signals in a strong noise environment.

[0116] In contrast, the heatmap of the method in this application exhibits a generally bright distribution. Thanks to the temporal alignment and weighted fusion strategy, even with background noise levels as high as [missing information], [missing information]. And the alternation amplitude is only Under extreme testing conditions, the detection accuracy can still be maintained at around 0.9 (i.e., 90%) (shown in the light yellow-green area in the figure); and as the alternation amplitude increases, the accuracy rapidly approaches 1.0 (saturated yellow area). This comparison intuitively verifies that the method of this application has a significantly better ability to capture microvolt-level alternating signals than the baseline method in complex noise environments.

[0117] Simulation results show that the proposed method significantly outperforms traditional peak difference detection methods in terms of detection sensitivity, noise immunity, and positive resolution. This is especially true when the alternation amplitude is lower than... In the weak signal range, baseline methods often fail due to low signal-to-noise ratios, while the method in this application can still maintain high detection accuracy (in...). In a noisy environment The detection rate of alternating signals exceeds 90%. This advantage is mainly due to three core improvements in this scheme: First, dynamic T-wave segmentation and the DTW-based phase alignment strategy effectively eliminate the phase deviation caused by the fixed window, preventing the stretching or compression of small alternating signals in the time domain; second, morphology-sensitive weighted alternating amplitude calculation (MPIMA) significantly reduces the interference of unrelated background noise on the overall detection results by focusing on the key regions of T-wave peaks and slopes; finally, the multi-lead energy fusion mechanism based on signal-to-noise ratio perception effectively avoids the risk of misjudgment caused by poor contact or local distortion in a single lead.

[0118] Regarding further work, the logistic regression model in this application exhibits good scalability, allowing for iterative training with large-scale clinical data to further optimize model coefficients. It also supports the development of personalized risk assessment thresholds based on the presence of structural heart disease or a history of medication use. Furthermore, given the algorithm's online computation and high robustness, it is applicable not only to standard in-hospital static or exercise stress ECG monitoring but also to wearable devices and remote medical monitoring systems with limited computing power. This provides valuable quantitative early warning data for early screening of occult myocardial ischemia, long-term data management of coronary artery disease patients, and precise intervention suggestions.

[0119] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0120] Figure 5 A structural block diagram of an example of an electrocardiogram data processing system based on microvolt-level T-wave alternation according to an embodiment of this application is shown.

[0121] like Figure 5As shown, the ECG data processing system 500 based on microvolt-level T-wave alternation includes an ECG data acquisition and preprocessing unit 510, a T-wave dynamic segmentation unit 520, a pairing construction unit 530, a timing alignment unit 540, a weighted alternation amplitude calculation unit 550, and an alternation detection and prompt output unit 560.

[0122] The ECG data acquisition and preprocessing unit 510 is used to acquire multi-lead ECG data to be processed, and to perform preprocessing and QRS complex localization on the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat time point information.

[0123] The T-wave dynamic segmentation unit 520 is used to adaptively determine the T-wave analysis interval for each heartbeat in the electrocardiogram waveform data based on the heartbeat time point information, and to obtain the T-wave curve segment corresponding to each heartbeat through dynamic T-wave segmentation, thereby generating a set of T-wave segments.

[0124] The pairing construction unit 530 is used to divide the T-wave curve segments in the T-wave segment set into odd-numbered sequences and even-numbered sequences according to the heartbeat sequence number, and to form paired T-wave segments by combining the even-numbered sequence T-wave curve segments and the odd-numbered sequence T-wave curve segments corresponding to the time sequence.

[0125] The timing alignment unit 540 is used to perform timing alignment processing on each paired T-wave segment to eliminate phase deviation on the time axis and obtain aligned paired T-wave curve segments.

[0126] The weighted alternating amplitude calculation unit 550 is used to construct a weighting function that takes into account the sensitivity of the T wave peak region and the T wave slope change region, and uses the weighting function to perform weighted calculation on the differential waveforms between the aligned paired T wave curve segments to obtain a weighted alternating amplitude that reflects the microvolt level morphological differences of adjacent heartbeat T waves.

[0127] The alternating detection and prompting output unit 560 is used to perform multi-lead fusion processing based on the weighted alternating amplitude corresponding to each lead to generate a global alternating detection index, and output electrocardiogram abnormality risk prompting information to indicate the risk of myocardial ischemia based on the global alternating detection index.

[0128] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, server, or network device) to perform the steps of any of the above-described electrocardiogram data processing methods based on microvolt-level T-wave alternation.

[0129] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described electrocardiogram data processing methods based on microvolt-level T-wave alternation.

[0130] In some embodiments, this application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform steps of a microvolt-level T-wave alternation-based electrocardiogram data processing method.

[0131] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0132] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0133] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0134] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for processing electrocardiogram data based on microvolt-level T-wave alternation, applied to computer equipment, characterized in that, The method includes: Acquire multi-lead ECG data to be processed, and perform preprocessing and QRS complex localization on the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat time point information for each heartbeat; Based on the heartbeat time point information, the T-wave analysis interval is adaptively determined for each heartbeat in the electrocardiogram waveform data, and the T-wave curve segment corresponding to each heartbeat is obtained through dynamic T-wave segmentation, thereby generating a set of T-wave segments; According to the heartbeat sequence number, the T-wave curve segments in the T-wave segment set are divided into odd-numbered sequences and even-numbered sequences, and the even-numbered sequence T-wave curve segments and the odd-numbered sequence T-wave curve segments corresponding to the time sequence are combined into paired T-wave segments; For each paired T-wave segment, timing alignment processing is performed to eliminate phase deviation on the time axis, resulting in aligned paired T-wave curve segments. A weighting function is constructed that takes into account the sensitivity of both the T-wave peak region and the T-wave slope change region. The weighting function is then used to perform weighted calculations on the differential waveforms between the aligned paired T-wave curve segments to obtain weighted alternating amplitudes that reflect the microvolt-level morphological differences of adjacent heartbeat T-waves. Multi-lead fusion processing is performed based on the weighted alternation amplitude corresponding to each lead to generate a global alternation detection index, and ECG abnormality risk warning information for indicating the risk of myocardial ischemia is output according to the global alternation detection index.

2. The method according to claim 1, characterized in that, Based on the heartbeat time point information, the T-wave analysis interval is adaptively determined for each heartbeat in the electrocardiogram waveform data, and T-wave curve segments corresponding to each heartbeat are obtained through dynamic T-wave segmentation, thereby generating a set of T-wave segments, including: Based on the heartbeat timing information, determine the first The peak moment of the R wave of a heartbeat The current RR interval is calculated based on the peak time of the R wave in adjacent heartbeats. ; Based on the above A dynamic search interval capable of covering long QT interval characteristics is constructed, wherein the time range of the dynamic search interval is set to... ; Within the dynamic search interval, indexed by sampling points Representing the discrete-time position within the dynamic search interval, the first derivative sequence of the electrocardiogram waveform data is calculated. With second derivative sequence And calculate the maximum modulus of the first derivative within the dynamic search interval. ,Will Determined as an adaptive slope threshold, where This is a preset proportional coefficient; Perform a forward search within the dynamic search interval to obtain the first derivative. The magnitude of the curve exceeded the adaptive slope threshold for the first time. The time marker is the T-wave initiation point. ; In determining the T-wave initiation point The search continues until the first derivative is satisfied. The modulus value falls back below the adaptive slope threshold. And at the same time, the second derivative The moment of zero crossover is marked as the end of the T wave. ; Based on the determined T-wave initiation point With the T-wave endpoint The T-wave curve segment of the current heartbeat is completely extracted from the electrocardiogram waveform data, and the T-wave curve segments of each heartbeat are collected to generate the T-wave segment set.

3. The method according to claim 2, characterized in that, Before combining even-numbered sequence T-wave curve segments with odd-numbered sequence T-wave curve segments to form paired T-wave segments, the method further includes: Based on a preset number of historical heartbeats preceding the current heartbeat, a reference T-wave template is constructed through statistical aggregation after time axis alignment and amplitude normalization. Calculate the morphological cross-correlation coefficient between each T-wave curve segment in the T-wave segment set and the reference T-wave template, and calculate the residual noise energy characterizing the degree of waveform difference. Generate a segment quality score based on the morphological cross-correlation coefficient and the residual noise energy. T-wave curve segments with a quality score below a preset validity threshold are marked as anomalous artifact segments, and these anomalous artifact segments are prohibited from directly participating in the pairing of even-numbered and odd-numbered sequences. For the missing pairing positions caused by the marking of abnormal artifacts, the effective T-wave curve segments adjacent to the abnormal artifact segment on the time axis are retrieved, and the abnormal artifact segment is replaced and filled by the effective T-wave curve segments to ensure that the generated paired T-wave segments meet the sequence continuity requirements.

4. The method according to claim 1 or 3, characterized in that, The step involves performing time alignment processing on each paired T-wave segment to eliminate phase deviations on the time axis, resulting in aligned paired T-wave curve segments, including: The even-numbered T-wave curve segments in the paired T-wave segments are set as the reference sequences. The odd-numbered T-wave curve segments in the paired T-wave segments are set as the sequences to be aligned. ,in and These are discrete time series, and their number of sampling points are respectively and ; To eliminate nonlinear time-domain distortion caused by heart rate fluctuations, a dynamic time warping algorithm is used to construct a local cost matrix. Then, a dynamic programming strategy is employed to search for warping paths within this local cost matrix that satisfy both monotonicity and continuity constraints. And determine the optimal bending path that minimizes the cumulative distance from each bending path. The cumulative distance along the path satisfy: ， In the formula, For curved paths Index point pairs on, and The reference sequences are respectively With the sequence to be aligned Time axis index, Represents the reference sequence In the index The amplitude at that point, Indicates the sequence to be aligned In the index The amplitude at that point, Represents the absolute value of the amplitude difference; Based on the optimal bending path The determined index mapping relationship applies to the sequence to be aligned. The time axis is subjected to nonlinear time scaling mapping, and the mapped sampling points are interpolated and resampled to generate aligned odd-numbered sequence T-wave curve fragments. So that its key morphological feature points are consistent with the reference sequence. Achieve phase synchronization in the time dimension; and use the reference sequence The aligned odd-numbered sequence T-wave curve segment These are composed of aligned paired T-wave curve segments.

5. The method according to claim 4, characterized in that, The construction of a weighting function that considers the sensitivity of both the T-wave peak region and the T-wave slope change region, and the use of the weighting function to perform weighted calculations on the differential waveforms between the aligned paired T-wave curve segments to obtain weighted alternating amplitudes reflecting the microvolt-level morphological differences of adjacent heartbeat T waves, includes: Extract the reference sequence determined in the time alignment process The morphological features, including the reference sequence Peak position index and the discrete first derivative at each sampling point ; Constructing a combined weight function based on morphological feature sensitivity The combined weighting function is composed of a Gaussian decay term focused on the peak region and a derivative enhancement term focused on morphological distortion: ， In the formula, For sampling point index, To control the standard deviation parameter of the range of interest in the peak region, These are weighting coefficients used to adjust the sensitivity to changes in T-wave slope and waveform broadening characteristics. Indicates taking the absolute value; Using the combined weight function For the reference sequence The aligned odd-numbered sequence T-wave curve segment A weighted average is calculated on the absolute difference sequence of amplitudes between them to obtain the weighted alternating amplitudes. : ， In the formula, The sampling point length of the aligned paired T-wave curve segment. Characterizes the microvolt-level morphological differences of adjacent heartbeat T waves after phase deviation has been eliminated.

6. The method according to claim 5, characterized in that, The process of performing multi-lead fusion processing based on the weighted alternation amplitudes corresponding to each lead to generate a global alternation detection index includes: Traversing each lead of the multi-lead ECG data Extract the lead reference sequence determined during the timing alignment process for this lead. And calculate the signal energy of the reference sequence for that lead. As initial fusion weights: ， In the formula, For lead traversal variables, , Total number of leads; For lead reference sequence, indicating In the Specific examples on each lead; Represents the lead reference sequence In the The amplitude of each sampling point; Get the Background noise level of each lead The background noise level is determined based on the amplitude standard deviation within a preset baseline window outside the T-wave analysis interval corresponding to the lead reference sequence, and / or the high-frequency residual standard deviation after performing a preset high-pass filtering on the lead reference sequence. Combined with preset noise threshold With attenuation coefficient Calculate the first Corrected fusion weights for each lead : ， In the formula, Used to control the penalty level for high-noise leads; Using modified fusion weights The weighted alternating amplitudes calculated for each lead Perform a weighted average calculation to obtain the global alternating detection index. : 。 7. The method according to claim 6, characterized in that, The step of outputting electrocardiogram (ECG) abnormality risk warning information based on the global alternating detection index to indicate the risk of myocardial ischemia includes: Within a preset analysis window, morphological feature parameters of each cardiac T-wave curve segment are extracted from the T-wave segment set, and statistical representative values ​​of the morphological feature parameters within the analysis window are calculated; the morphological feature parameters include T-wave width. Slope symmetry ratio and ST segment offset The statistical representative values ​​include the mean and / or median; Constructing a system that includes the global alternation detection index A multidimensional feature vector of statistical representative values ​​of the morphological feature parameters. ;in , This represents the vector transpose operation; The multidimensional feature vector Input a pre-trained logistic regression model and use the Sigmoid activation function to calculate the risk score of ischemia-related electrocardiographic abnormalities. : ， In the formula, For the intercept term, These are the model regression coefficients for the corresponding feature dimensions; Based on the ischemia-related electrocardiogram abnormality risk score The system outputs ECG abnormality risk warning information; wherein, the ECG abnormality risk warning information includes a risk score value and a risk level obtained according to a preset risk threshold.

8. The method according to claim 1, characterized in that, The preprocessing and QRS complex localization of the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat timing information includes: Baseline drift correction and power line interference suppression were performed on the multi-lead ECG data, and the corrected signal was decomposed using an empirical mode decomposition algorithm. Based on the spectral characteristics and energy distribution of each intrinsic mode component, the high-frequency noise-dominated component is identified and removed, and the remaining components are reconstructed to generate the preprocessed electrocardiogram waveform data. Based on the preprocessed ECG waveform data, candidate R wave positions for each lead are extracted. The candidate R wave positions are screened using a multi-lead collaborative verification strategy. Candidate R waves that appear simultaneously in at least a preset number of leads within a preset time tolerance range are identified as valid QRS complexes, thereby obtaining the heartbeat time point information for each heartbeat.

9. A microvolt-level T-wave alternation-based electrocardiogram data processing system, deployed on a computer device, characterized in that, The system includes: The ECG data acquisition and preprocessing unit is used to acquire multi-lead ECG data to be processed, and to perform preprocessing and QRS complex localization on the multi-lead ECG data to obtain preprocessed ECG waveform data and corresponding heartbeat time point information for each heartbeat. The T-wave dynamic segmentation unit is used to adaptively determine the T-wave analysis interval for each heartbeat in the electrocardiogram waveform data based on the heartbeat time point information, and to obtain the T-wave curve segment corresponding to each heartbeat through dynamic T-wave segmentation, thereby generating a set of T-wave segments. The pairing construction unit is used to divide the T-wave curve segments in the T-wave segment set into odd-numbered sequences and even-numbered sequences according to the heartbeat sequence number, and to form a pair of T-wave segments by combining the even-numbered sequence T-wave curve segments and the odd-numbered sequence T-wave curve segments corresponding to the time sequence. The timing alignment unit is used to perform timing alignment processing on each paired T-wave segment to eliminate phase deviation on the time axis and obtain aligned paired T-wave curve segments. The weighted alternating amplitude calculation unit is used to construct a weight function that takes into account the sensitivity of the T wave peak region and the T wave slope change region, and to use the weight function to perform weighted calculation on the differential waveforms between the aligned paired T wave curve segments to obtain a weighted alternating amplitude that reflects the microvolt level morphological differences of adjacent heartbeat T waves. The alternating detection and prompting output unit is used to perform multi-lead fusion processing based on the weighted alternating amplitude corresponding to each lead to generate a global alternating detection index, and output electrocardiogram abnormality risk prompting information to indicate the risk of myocardial ischemia based on the global alternating detection index.

Images