Electrocardiogram waveform evaluation method, device, system, electronic device, and program product

By using a target recognition model and an artificial intelligence system to identify electrocardiogram (ECG) waveforms, generate bounding boxes, and obtain feature points, the system solves the problems of subjectivity and lack of unified standards in traditional ECG analysis, and realizes automated and personalized ECG assessment.

CN123058748APending Publication Date: 2026-11-10MITSEL CORP +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510585142.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Traditional electrocardiogram (ECG) analysis methods rely on doctors' subjective interpretation and lack unified quantitative standards. Existing technologies struggle to achieve efficient and accurate automated analysis under noise interference and complex waveforms, especially when identifying diseases such as myocardial infarction and atrial fibrillation.

Method used

The target recognition model identifies the waveform components of multiple cardiac cycles in an electrocardiogram (ECG) waveform image, generates bounding boxes, obtains feature points, determines waveform features, and combines with an artificial intelligence text analysis system to generate ECG waveform evaluation results.

Benefits of technology

It achieves a high degree of automation and objective and accurate assessment of electrocardiogram analysis, provides standardized waveform feature quantification, improves the accuracy and efficiency of diagnosis, and provides personalized diagnostic suggestions by combining user consultation information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123058748A_ABST
    Figure CN123058748A_ABST
Patent Text Reader

Abstract

The present application provides an electrocardiogram waveform evaluation method, device, system, electronic equipment and program product. The electrocardiogram waveform evaluation method can include: identifying each wave component of a plurality of cardiac cycles in an electrocardiogram waveform image through a target identification model, and generating a bounding box corresponding to each wave component of the plurality of cardiac cycles; obtaining a feature point of each wave component according to the bounding box of each wave component; determining a waveform feature of each wave component according to the feature point of each wave component; and generating an electrocardiogram waveform evaluation result based on the waveform feature of each wave component, and the evaluation result reflects the corresponding cardiac function state of the electrocardiogram.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of medicine, health, or sports, and in particular to electrocardiogram waveform assessment methods, devices, systems, electronic devices, and program products. Background Technology

[0002] Heart disease is one of the major threats to the health of the global population. Electrocardiography (ECG), as an important tool for diagnosing heart disease, reflects the state of cardiac function by recording the heart's electrical activity, providing crucial information for clinical diagnosis.

[0003] Traditionally, doctors or professionals assess heart health by observing the waveforms of an electrocardiogram (ECG). This method requires doctors to possess extensive professional knowledge and clinical experience. Therefore, traditional ECG analysis methods rely heavily on the doctor's subjective interpretation. Different doctors may interpret the same ECG differently, and the lack of a unified quantitative standard makes it difficult to objectively compare diagnostic results.

[0004] With the development of computer technology, various signal processing methods have been applied to electrocardiogram (ECG) analysis, including filtering, denoising, waveform detection, and feature extraction. Commonly used techniques include Fourier transform, wavelet transform, and spectral analysis. However, existing technologies generally suffer from insufficient analytical accuracy and poor intuitiveness, limiting their clinical application. Especially when dealing with complex waveform changes, current technologies struggle to accurately identify and quantify the characteristics of key components, failing to effectively capture the typical ECG manifestations of diseases such as myocardial infarction, atrial fibrillation, and premature ventricular contractions.

[0005] Meanwhile, existing technologies struggle to achieve efficient, accurate, and automated electrocardiogram signal analysis in the presence of noise interference, baseline drift, or arrhythmia.

[0006] Due to the aforementioned limitations, automated analysis and objective quantitative assessment of electrocardiograms remain technical challenges in the field.

[0007] The background description is provided for the purpose of understanding the relevant technologies in this field and is not intended as an admission of prior art. Summary of the Invention

[0008] The present invention aims to provide an electrocardiogram waveform evaluation scheme that at least partially alleviates or solves the above-mentioned technical problems.

[0009] In a first aspect, embodiments of the present invention provide a method for evaluating electrocardiogram waveforms, which may include: The target recognition model identifies the waveform components of multiple cardiac cycles in the electrocardiogram waveform image and generates bounding boxes corresponding to each waveform component of the multiple cardiac cycles. Based on the bounding boxes of each waveform component, the feature points of each waveform component are obtained; Based on the feature points of each waveform component, the waveform characteristics of each waveform component are determined; Based on the waveform characteristics of each waveform component, an electrocardiogram waveform evaluation result is generated, which reflects the cardiac function status corresponding to the electrocardiogram.

[0010] In some embodiments, the electrocardiogram waveform evaluation method further includes: The time-series electrocardiogram (ECG) data is converted into the ECG waveform image.

[0011] In some embodiments, the waveform components of the cardiac cycle include one or more of the P wave, QRS complex, ST segment, T wave, and U wave; The characteristic points of each waveform component include: the start point, peak point, and end point of the P wave; the start point and trough point of the Q wave, the peak point of the R wave, and the trough point and end point of the S wave of the QRS wave; the start point and end point of the ST segment; the start point, peak point, and end point of the T wave; and some or all of the characteristic points of the start point, peak point, and end point of the U wave.

[0012] In some embodiments, obtaining the feature points of each waveform component based on the bounding boxes of each waveform component includes: The coordinates of the feature points of the corresponding waveform components are determined based on the intersection of the bounding boxes of each waveform component with the electrocardiogram waveform.

[0013] In some embodiments, the waveform features include at least one of time features, amplitude features, morphological features, and integrity features.

[0014] In some embodiments, determining the waveform characteristics of each waveform component based on the feature points of each waveform component includes: determining the time characteristics of each waveform component, wherein the time characteristics include one or more of the following: P wave duration, PQ interval, QRS complex duration, QT interval, RR interval, heart rate, and corrected QT interval (QTc), specifically including: The P-wave duration is determined based on the time difference between the start and end of the P-wave. The PQ interval is determined based on the time difference between the start of the P wave and the start of the Q wave. The QRS group duration is determined based on the time difference between the Q wave initiation and the S wave termination. The QT interval is determined based on the time difference between the start of the Q wave and the end of the T wave. The RR interval is determined based on the time difference between two adjacent R-wave peaks. The heart rate is determined by a preset heart rate calculation formula based on the RR interval; and / or The corrected QT interval QTc is determined by a preset correction formula based on the QT interval and the RR interval.

[0015] In some embodiments, determining the waveform features of each waveform component based on the feature points of each waveform component includes: The integrity status of each cardiac cycle is determined based on the boundary frames of the P wave, QRS complex, and T wave, or the presence of characteristic points of the P wave, QRS complex, and T wave in each cardiac cycle.

[0016] In some embodiments, determining the integrity status of each cardiac cycle based on the presence of characteristic points of the P wave, QRS complex, and T wave in each cardiac cycle includes: Determine whether there are missing coordinate sequences of characteristic points of P wave, QRS complex and T wave in each cardiac cycle in the electrocardiogram waveform image; When there are no missing beats, the corresponding cardiac cycle is determined to be a complete rhythm; When a missing component exists, the corresponding missing waveform component of the cardiac cycle is determined based on the coordinates of the missing feature point.

[0017] In some embodiments, the morphological features include the tilt characteristics of the ST segment of the electrocardiogram waveform; Wherein, determining the waveform features of each waveform component based on the feature points of each waveform component includes: Obtain the time and amplitude values ​​of the S-wave endpoint and the T-wave initiation; Calculate the slope of the line connecting the end of the S wave and the beginning of the T wave; and The shape of the ST segment is determined based on the slope of the connecting line. When the slope is zero, the ST segment is determined to be straight; when the slope is positive, the ST segment is determined to be upward sloping; and when the slope is negative, the ST segment is determined to be downward sloping.

[0018] In some embodiments, the morphological features include the directional features of the T wave in an electrocardiogram waveform; Wherein, determining the waveform features of each waveform component based on the feature points of each waveform component includes: Obtain the amplitude values ​​of the T-wave's initiation, peak, and end points; Compare the magnitude of the peak point amplitude with the magnitudes of the start and end points; and The T-wave pattern is determined based on the magnitude relationship described above. When the peak amplitude is greater than the amplitudes at the start and end points, it is determined to be an upright T-wave; when the peak amplitude is less than the amplitudes at the start and end points, it is determined to be an inverted T-wave.

[0019] In some embodiments, generating an electrocardiogram waveform evaluation result based on the waveform characteristics of each waveform component includes: The waveform features are then converted into a first structured text description containing standard medical terminology. The first structured text description is input into the artificial intelligence text analysis system through a preset interface format; and The artificial intelligence text analysis system analyzes the first structured text description and outputs the electrocardiogram waveform evaluation result.

[0020] In some embodiments, generating an electrocardiogram waveform evaluation result based on the waveform characteristics of each waveform component includes: The waveform features are then converted into a first structured text description containing standard medical terminology. Obtain user consultation information and convert it into a second structured text description; The first structured text description and the second structured text description are input into the artificial intelligence text analysis system through a preset interface format.

[0021] The artificial intelligence text analysis system analyzes the first structured text description and the second structured text to output the electrocardiogram waveform evaluation result.

[0022] In some embodiments, the electrocardiogram waveform evaluation method further includes: training the target recognition model, specifically including: Obtain multiple electrocardiogram training samples; Preprocessing the electrocardiogram training samples to obtain preprocessed training samples includes converting time-series electrocardiogram data into electrocardiogram waveform images; The waveform components in the electrocardiogram waveform images of the preprocessed training samples are labeled to generate a training dataset, wherein the labeling includes labeling the location regions of the P wave, QRS complex, ST segment, T wave and U wave; The training dataset is input into the target recognition model for training, and the model parameters are optimized using the backpropagation algorithm until a preset condition is met. In a second aspect, embodiments of the present invention provide an electrocardiogram waveform evaluation device, which may include: The recognition unit is configured to recognize each waveform component of multiple cardiac cycles in an electrocardiogram waveform image through a target recognition model, and generate bounding boxes corresponding to each waveform component of the multiple cardiac cycles. The acquisition unit is configured to acquire feature points of each waveform component based on the bounding boxes of each waveform component; The determining unit is configured to determine the waveform features of each waveform component based on the feature points of each waveform component; and The evaluation unit is configured to generate an electrocardiogram waveform evaluation result based on the waveform characteristics of each waveform component, and the evaluation result reflects the cardiac function status corresponding to the electrocardiogram.

[0023] In a third aspect, embodiments of the present invention provide an electrocardiogram waveform evaluation system, characterized in that it includes: a user terminal, an AI feature extraction model, and an AI analysis model, wherein the user terminal is configured to interact with the AI ​​feature extraction model and the AI ​​analysis model to present evaluation results; The user terminal includes an electrocardiogram (ECG) acquisition module, which is configured to acquire ECG data. The AI ​​feature extraction model is configured to generate an electrocardiogram waveform image based on the electrocardiogram data, identify each waveform component of multiple cardiac cycles in the electrocardiogram waveform image, generate corresponding bounding boxes, extract feature points of each waveform component from the bounding boxes, determine the waveform features of each waveform component, and convert the waveform features into a first structured text description. The AI ​​analysis model is configured to receive a first structured text description, analyze and process it, generate an assessment result reflecting the cardiac function status corresponding to the electrocardiogram, and return the assessment result to the user terminal.

[0024] In some embodiments, the user terminal further includes a consultation module configured to receive user consultation information and convert it into a second structured text description; The AI ​​analysis model is configured to receive a first structured text description and a second structured text description, analyze and process them, generate an assessment result reflecting the cardiac function status corresponding to the electrocardiogram, and return the assessment result to the user terminal.

[0025] In a fourth aspect, embodiments of the present invention provide an electronic device, including: a processor and a memory storing a computer program, the processor being configured to implement the method as described in the first aspect when executing the computer program.

[0026] In a fifth aspect, embodiments of the present invention provide a program product including a computer program, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0027] In a sixth aspect, embodiments of the present invention provide a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0028] The electrocardiogram (ECG) waveform evaluation method provided in this invention identifies waveform components of multiple cardiac cycles in an ECG waveform image using a target recognition model, generates bounding boxes for each waveform component, obtains feature points of each waveform component based on the bounding boxes, determines the waveform features of each waveform component, and generates an ECG waveform evaluation result based on the waveform features. This invention addresses the technical problems of strong subjectivity and lack of unified quantitative standards in traditional ECG analysis by providing an objective and accurate ECG waveform evaluation scheme. Specifically, this invention identifies waveform components of multiple cardiac cycles in an ECG waveform image using a target recognition model and generates bounding boxes, achieving accurate segmentation of the ECG waveform; furthermore, based on the feature points obtained from the bounding boxes, the temporal and morphological characteristics of each waveform component can be objectively determined, establishing a standardized waveform feature quantification system, thus solving the problems of strong subjectivity and lack of unified standards in traditional ECG analysis.

[0029] Furthermore, this ECG waveform assessment method uses a target recognition model to accurately identify and segment various waveform components in the ECG waveform image, such as the P wave, QRS complex, ST segment, T wave, and U wave. Based on the intersection of the bounding box of each waveform component with the ECG waveform, the coordinates of the feature points of the corresponding waveform component are determined. This allows for the accurate calculation of time characteristics, including P wave duration, PQ interval, QRS complex duration, QT interval, RR interval, heart rate, and corrected QT interval, as well as morphological characteristics such as ST segment tilt and T wave direction, and the determination of the integrity of the cardiac cycle.

[0030] Furthermore, this electrocardiogram (ECG) waveform assessment method also includes converting the acquired waveform features into a structured text description containing standard medical terminology, performing comprehensive analysis through an artificial intelligence text analysis system, and outputting cardiac function status assessment results to provide objective evidence for cardiac function status assessment. This method can also combine user consultation information to provide a more comprehensive and personalized cardiac function assessment through an artificial intelligence text analysis system, further improving the accuracy and specificity of diagnosis. Therefore, the ECG waveform assessment scheme provided in this embodiment of the invention not only achieves a high degree of automation in ECG analysis but also provides personalized diagnostic suggestions based on user consultation information, significantly improving the efficiency and accuracy of ECG analysis.

[0031] Other optional features and technical effects of the embodiments of the present invention are partly described below and partly apparent from reading this document. Attached Figure Description

[0032] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The elements shown are not limited to the scale shown in the drawings, and the same or similar reference numerals in the drawings denote the same or similar elements, wherein: Figure 1This is an exemplary flowchart of an electrocardiogram waveform evaluation method according to an embodiment of the present invention; Figure 2 An exemplary schematic diagram is shown, illustrating the identification of multiple waveforms in an electrocardiogram using an electrocardiogram waveform evaluation method according to a specific embodiment of the present invention; Figure 3 A YOLO target detection model applicable to an electrocardiogram waveform evaluation method according to a specific embodiment of the present invention is shown; Figure 4 This is an exemplary flowchart of an electrocardiogram waveform evaluation method according to an embodiment of the present invention; Figure 5 A schematic diagram is shown illustrating the determination of feature points using an electrocardiogram waveform evaluation method according to a specific embodiment of the present invention. Figure 6 This is an exemplary flowchart of an electrocardiogram waveform evaluation method according to an embodiment of the present invention; Figure 7 A schematic diagram illustrating the determination of ST segment slope using an electrocardiogram waveform evaluation method according to a specific embodiment of the present invention is shown. Figure 7 (a) Involving the ST straight segment, as shown Figure 7 (b) ST sloping upwards, as shown Figure 7 (c) Involves the downward sloping segment of ST; Figure 8 This is an exemplary judgment flowchart of an electrocardiogram waveform evaluation method according to an embodiment of the present invention; Figure 9 A schematic diagram illustrating the determination of T-wave direction using an electrocardiogram waveform evaluation method according to a specific embodiment of the present invention is shown. Figure 9 (a) Involves the identification of positive T waves, as shown Figure 9 (b) Involving the identification of inverted T waves; Figure 10 This is an exemplary flowchart of an electrocardiogram waveform evaluation method according to an embodiment of the present invention; Figure 11 This is an exemplary flowchart of the training process of an electrocardiogram waveform recognition model according to an embodiment of the present invention; Figure 12 This is a schematic structural diagram of an electrocardiogram waveform evaluation device according to an embodiment of the present invention; Figure 13 This is a schematic structural diagram of an electrocardiogram waveform evaluation system according to an embodiment of the present invention; Figure 14 This is an exemplary structural diagram of an electronic device capable of implementing the methods according to embodiments of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0034] The embodiments of the present invention relate to the field of electrocardiogram (ECG) processing, and more particularly to ECG waveform evaluation schemes.

[0035] As mentioned earlier, traditional electrocardiogram (ECG) analysis relies primarily on the subjective interpretation of physicians. While computer-aided ECG signal analysis methods have been developing for many years, they still have significant shortcomings in terms of analytical complexity and clinical applicability. In current medical practice, the objective quantitative assessment and automated analysis of ECG waveform characteristics remains a technical challenge.

[0036] In some known techniques, traditional signal processing methods such as filtering and threshold detection can be used to identify waveform components in electrocardiograms (ECGs). However, these methods are sensitive to noise and struggle to accurately identify waveform components in complex or irregular ECGs, especially when there are morphological variations or arrhythmias.

[0037] In some known techniques, time-frequency analysis methods such as wavelet transform can be used to process electrocardiogram (ECG) signals and extract feature points. However, these methods have complex parameter settings, insufficient adaptability to the ECG characteristics of different patients, and difficulty in establishing a unified quantitative standard.

[0038] In some known technologies, machine learning methods can be used to directly classify electrocardiograms, but these methods often lack interpretability of intermediate processes, are difficult to provide specific waveform feature parameters, are not conducive to doctors' understanding of diagnostic criteria, and limit clinical application.

[0039] Therefore, there is currently a lack of an integrated solution capable of accurately identifying the boundaries of various waveforms in an electrocardiogram (ECG), objectively quantifying waveform characteristics, and generating evaluation results. This is particularly true in the analysis of complex ECGs, where it is difficult to simultaneously guarantee identification accuracy and provide interpretable, objective analytical results.

[0040] To address the aforementioned technical problems, this invention proposes an innovative method for evaluating electrocardiogram (ECG) waveforms. This method uses a target recognition model to identify and segment waveform components across multiple cardiac cycles in an ECG waveform image, acquires and analyzes the feature points and waveform characteristics of each component, and ultimately generates an evaluation result reflecting the state of cardiac function.

[0041] The electrocardiogram waveform evaluation method provided in this invention can be applied to a variety of scenarios, including but not limited to: electrocardiogram signal analysis in medical auxiliary devices, remote medical monitoring systems, portable health management devices, medical research and teaching, and auxiliary screening in primary medical institutions. It also includes various application scenarios in the health or sports fields, including but not limited to scenarios applied to health apps or sports apps that are compatible with electrocardiogram detection devices.

[0042] As an explanation, all steps of the electrocardiogram waveform evaluation method described in the embodiments of the present invention can be performed by a computer or other device, and thus belong to the category of information processing methods. More specifically, these methods can automate the processing and analysis of acquired electrocardiogram signals by a computer, including waveform recognition, bounding box generation, feature point extraction, waveform feature calculation, and functional status evaluation, etc., and all processing steps can be automatically completed by the computer system.

[0043] The technical solutions and implementation processes of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0044] In such Figure 1 The illustrated embodiment provides an electrocardiogram waveform evaluation method according to an embodiment of the present invention, which may include steps 110 to 140.

[0045] In an optional embodiment, before performing steps 110 to 130 above, the electrocardiogram waveform evaluation method may include one or more optional steps, such as, but not limited to, a data acquisition step, a conversion step, and one or more preprocessing steps. Figure 1 (Not shown in the image).

[0046] Optionally, the electrocardiogram waveform evaluation method according to an embodiment of the present invention may include step 101 (not shown): acquiring electrocardiogram data in time series representation.

[0047] An electrocardiogram (ECG or EKG) is a medical examination method that records and analyzes the electrical activity of the heart. ECG data reflects the time series of the heart's electrophysiological activity.

[0048] In this embodiment of the invention, the time-series electrocardiogram (ECG) data can be acquired using various ECG acquisition devices, such as, but not limited to, single-lead or multi-lead ECG devices. The ECG data can be in various forms, such as continuously acquired time series (e.g., real-time acquisition), multiple consecutive buffered data segments ordered by time, or complete time-series data. The sampling frequency of the ECG data can be selected according to application requirements, for example, a sampling value from 100Hz to 1000Hz.

[0049] However, those skilled in the art will realize that the acquired time-series electrocardiogram data can be stored data, which falls within the scope of this invention.

[0050] Optionally, the electrocardiogram waveform evaluation method according to an embodiment of the present invention may include step 101' (not shown): performing data preprocessing on the electrocardiogram data (which may be referred to as first preprocessing).

[0051] In embodiments of the present invention, one or more data preprocessing steps may be provided, including but not limited to filtering (such as bandpass filtering), baseline correction, and noise reduction.

[0052] Optionally, the electrocardiogram waveform evaluation method according to an embodiment of the present invention may include step 102 (not shown): converting time-series electrocardiogram data into an electrocardiogram waveform image.

[0053] In this embodiment of the invention, converting time-series electrocardiogram (ECG) data into ECG waveform images involves transforming one-dimensional time-series data into a two-dimensional image representation.

[0054] In some embodiments, converting time-series electrocardiogram (ECG) data into a two-dimensional ECG waveform image can be achieved by plotting a time-voltage curve. Specifically, time can be used as the horizontal axis and voltage amplitude as the vertical axis, connecting the sampling points to form a continuous waveform curve. During the conversion process, the ratio of the time axis and amplitude axis can be adjusted as needed to optimally display the morphological characteristics of the ECG. In some embodiments, the ECG waveform image can be adapted to different resolutions as required.

[0055] Optionally, the electrocardiogram waveform evaluation method according to an embodiment of the present invention may include step 102' (not shown): performing image preprocessing on the electrocardiogram waveform image (which may be referred to as second preprocessing).

[0056] In this embodiment of the invention, during the conversion process, the electrocardiogram waveform image can be preprocessed as needed, such as adjusting the image resolution, contrast and other parameters, and denoising.

[0057] In some embodiments, the electrocardiogram waveform image can be normalized to a predetermined size (e.g., 448×448 pixels) for subsequent processing by the target recognition model.

[0058] In some embodiments, pixel value normalization may be performed to adjust the pixel values ​​to a standard range, including grayscale processing.

[0059] In some embodiments, preprocessing may also include operations such as contrast enhancement, noise removal, and edge sharpening.

[0060] As described above, the electrocardiogram waveform evaluation method according to embodiments of the present invention may include: Step 110: Identify the waveform components of multiple cardiac cycles in the electrocardiogram waveform image using the target recognition model, and generate bounding boxes corresponding to the waveform components of multiple cardiac cycles.

[0061] In this paper, the cardiac cycle has the meaning known in the art, referring to a complete contraction and relaxation of the heart, typically represented on an electrocardiogram as a sequence of waveforms from one R wave to the next. Multiple consecutive cardiac cycles constitute the rhythm of an electrocardiogram, and by analyzing the characteristics of multiple cardiac cycles (rhythms), the functional state of the heart can be assessed.

[0062] In embodiments of the invention, the waveform components of the identified cardiac cycle include one or more of the P wave, QRS complex, ST segment, T wave, and U wave, preferably all of them. As an illustrative explanation and not a limitation of the invention, the P wave may represent the atrial depolarization process; the QRS complex may represent the ventricular depolarization process, including the Q wave, R wave, and S wave; the ST segment, located between the end of the QRS complex and the beginning of the T wave, may reflect the early repolarization phase of the ventricle; the T wave may represent the ventricular repolarization process; and the U wave, a small wave following the T wave, may be related to the late repolarization of the ventricle. Figure 2 The electrocardiogram waveform shows that it contains various waveform components, such as P wave, QRS complex, ST segment, T wave and U wave.

[0063] In this embodiment of the invention, the target recognition model includes the YOLO (You Only Look Once) target detection model, which is trained, for example, to recognize components of an electrocardiogram (ECG) waveform. The YOLO model can complete target localization and classification tasks in a single neural network and can directly predict the bounding boxes and class probabilities of ECG waveform components, making it suitable for real-time recognition and localization of each waveform component in an ECG waveform.

[0064] like Figure 3As shown, the structure of the YOLO model may include a backbone network 31, a feature aggregation network (PAN) 32, and segment heads 33 and 34. In this embodiment, the segment heads adopt a one-to-many head structure. The backbone network 31 is responsible for extracting feature information from the electrocardiogram waveform image; the feature aggregation network 32 is responsible for fusing features at different scales; and the segment heads, such as the one-to-many heads 33 and 34, are responsible for generating prediction results. In this embodiment, the segment heads, such as the one-to-many heads 33 and 34, include intersection-to-union regression (IoU) submodules 331 and 341 and classification submodules 332 and 342, wherein the IoU regression submodule can be used to predict bounding boxes, and the classification submodule can be used to generate labels for the bounding boxes. Thus, the model can simultaneously detect electrocardiogram waveform components of different scales and morphologies.

[0065] In an exemplary embodiment, the workflow of the YOLO model may include: adjusting an electrocardiogram waveform image to a fixed size (e.g., 448×448 pixels), optionally maintaining the original aspect ratio and adding appropriate padding; then inputting it into a backbone network for feature extraction, capturing different frequency and temporal features of the electrocardiogram signal through structures such as multiple convolutional layers, pooling layers, and residual connections; then fusing the extracted features through a feature aggregation network; and finally generating prediction results from a multi-head network, including the bounding box coordinates and class labels of each waveform component.

[0066] For illustrative purposes, and not as a limitation, a bounding box refers to a rectangular box that encloses each waveform component, defined by four coordinate values ​​(x, y, width, height), where (x, y) represents the coordinates of the top-left corner of the bounding box, and width and height represent the width and height of the bounding box, respectively. These bounding boxes allow for precise location and extent of each waveform component within the electrocardiogram, providing a foundation for subsequent feature point extraction and waveform feature analysis.

[0067] In an optional embodiment, the output of the YOLO model also includes a confidence score, representing the model's degree of confidence in the detection result. A confidence threshold (e.g., 0.5) can be set to retain only detection results with a confidence score higher than the threshold, thereby improving the accuracy of recognition. Those skilled in the art will conceive of using any feasible optional features, which fall within the scope of this invention.

[0068] Preferably, the YOLO model used in the embodiments of the present invention is end-to-end, thereby directly outputting the position, bounding box and category of each waveform component without the need for complex intermediate processing steps, making the entire model more efficient and easier to deploy.

[0069] As mentioned above, Figure 2 The diagram schematically illustrates that an electrocardiogram (ECG) waveform includes various waveform components, such as the P wave, QRS complex, ST segment, T wave, and U wave. Furthermore, Figure 2 The diagram also shows exemplary waveform components (such as P-waves, QRS complexes, and T-waves) and their corresponding bounding boxes identified using a target recognition model according to embodiments of the present invention. The identification of these bounding boxes can serve as the basis for subsequent determination of feature points. In some embodiments, bounding boxes for P-waves, QRS complexes, and T-waves can be generated. In some embodiments, bounding boxes for P-waves, QRS complexes, T-waves, and U-waves can be generated.

[0070] Although the exemplary embodiment describes the target recognition model as the YOLO model, those skilled in the art will understand that any suitable target recognition model capable of performing target recognition and forming detection boxes can also be used to implement the technical solution of the present invention, including but not limited to models such as SSD (Single Shot Detector) and Faster R-CNN.

[0071] Through this step, the embodiments of the present invention achieve accurate identification and boundary localization of electrocardiogram waveform components, providing a foundation for subsequent feature point extraction and waveform feature analysis.

[0072] Step 120: Obtain the feature points of each waveform component based on the bounding boxes of each waveform component.

[0073] In this embodiment of the invention, based on the bounding box generated by the target recognition model in step 110, feature points of each waveform component can be extracted in step 120.

[0074] As mentioned above, the identified waveform components of the cardiac cycle include one or more of the following: P wave, QRS complex, ST segment, T wave, and U wave. This allows for the generation of corresponding bounding boxes, such as bounding boxes for the P wave, QRS complex, and T wave, or bounding boxes for the P wave, QRS complex, T wave, and U wave. In some embodiments, the feature points of each waveform component include: the start, peak, and end points of the P wave; the start and trough points of the Q wave, the peak point of the R wave, and the trough and end points of the S wave of the QRS complex; the start and end points of the ST segment; the start, peak, and end points of the T wave; and some, preferably all, of the start, peak, and end points of the U wave.

[0075] In this embodiment of the invention, the characteristic points of each waveform component in each cardiac cycle (rhythm) can be represented by a sequence of characteristic points. For example, the P wave characteristic point sequence consists of P... i {P 1i ,P 2i ,P 3i} represents the QRS group characteristic point sequence composed of QRS i {q1i ,q 2i ,r i ,s 1i ,s 2i} indicates that the ST segment feature point sequence is derived from ST. i {st 1i ,st 2i} represents the T-wave characteristic point sequence composed of T i {t 1i ,t 2i ,t 3i} and U-wave characteristic point sequences are derived from U i {u 1i ,u 2i ,u 3i} indicates that, where p 1i ,p 2i ,p 3i The starting point, peak point, and ending point of the P wave; q 1i ,q 2i ,r i ,s 1i ,s 2i For the QRS complex, the Q wave start and trough, the R wave peak, and the S wave trough and end; st 1i ,st 2i For the start and end points of segment ST; t 1i ,t 2i ,t 3i These represent the start, peak, and end points of the T wave, respectively; u 1i ,u 2i ,u 3i These represent the start point, peak point, and end point of the U wave, respectively. Here, the electrocardiogram waveform may show, for example, N cardiac cycles (rhythms), i=1,⋯,N.

[0076] In some embodiments, obtaining the feature points of each waveform component based on the bounding box of each waveform component includes: step 120' (not shown): determining the coordinates of the feature points of the corresponding waveform component based on the intersection of the bounding box of each waveform component and the electrocardiogram waveform.

[0077] In this embodiment of the invention, the following waveform coordinates can be determined by the intersection points of the bounding boxes of each waveform component and the electrocardiogram waveform lines: In some embodiments, the P-wave feature point can be determined by identifying the intersection of the left boundary of the P-wave bounding box and the electrocardiogram waveform. 1i V p1i (p 1i The peak point of the P-wave is determined by searching for local maxima within the P-wave bounding box. 2i V p2i (p 2iThe endpoint of the P-wave is determined by the intersection of the right boundary and the waveform. 3i V p3i (p 3i )).

[0078] In some embodiments, determining the characteristic points of a QRS group involves determining five characteristic points: the Q-wave initiation and trough, the R-wave peak, and the S-wave trough and end. The Q-wave initiation (q) can be determined by the intersection of the left boundary of the QRS group bounding box and the waveform. 1i V q1i (q 1i Then search for the first local minimum point to determine the Q-wave trough (q). 2i V q2i (q 2i Then, within the QRS group bounding box, search for the global maximum point to determine the R peak point (r). i V ri (r i Then, search for the second local minimum point to determine the S-wave trough point (s). 1i V s1i (s 1i Then, the S-wave endpoint is determined by the intersection of the right boundary of the QRS group bounding box and the waveform. 2i V s2i (s 2i )).

[0079] In some embodiments, the ST segment is a transition segment connecting the QRS complex and the T wave, and its characteristic points share coordinates with adjacent waveform components; that is, the ST segment start point (st 1i V s1i (s 1i )) and the S-wave endpoint (s 2i V s2i (s 2i They share the same coordinates, and the ST segment ends at (st 2i V s2i (s 2i )) and T-wave onset (t 1i V t1i (t 1i They share the same coordinates.

[0080] In some embodiments, the method for determining T-wave feature points is similar to that for P-waves, but the possible directional variations of the T-wave can be considered. The T-wave initiation point (t) can be determined by the intersection of the left boundary of the T-wave bounding box and the waveform. 1i V t1i (t 1i Within the T-wave bounding box, search for local extreme points (such as the maximum point (positive T-wave) or the local minimum point (inverted T-wave)) to determine the T-wave peak point (t). 2i Vt2i (t 2i The endpoint of the T-wave is determined by the intersection of the right boundary and the waveform. 3i V t3i (t 3i )).

[0081] In some embodiments, the existence of the U-wave can be confirmed first (e.g., determined based on the bounding box label), and then the starting point of the U-wave can be determined by the intersection of the left boundary of the U-wave bounding box and the waveform. 1i V u1i (u 1i The peak point of the U-wave is determined by searching for local maximum points within the U-wave bounding box. 2i V u2i (u 2i The endpoint of the U-wave is determined by the intersection of the right boundary and the waveform. 3i V u3i (u 3i )).

[0082] In this embodiment of the invention, the feature points in the form of extreme points (maximum or minimum points) can be analyzed by combining the sliding window technique to determine the changing trend of the window maximum value sequence and the turning point of the difference from positive to negative. More specifically, the feature points in the form of extreme points (maximum or minimum points) can be determined by combining the sliding window with the gradient method and analyzing the changing trend of the signal maximum value in adjacent windows.

[0083] By using this bounding box and waveform intersection analysis, the embodiments of the present invention can accurately obtain the feature point coordinates of each waveform component of the electrocardiogram, providing a basis for determining the waveform features in the subsequent step 130.

[0084] Step 130: Determine the waveform characteristics of each waveform component based on the characteristic points of each waveform component.

[0085] In this embodiment of the invention, based on the waveform component feature points obtained in step 120, step 130 can determine the waveform features.

[0086] In some embodiments of the present invention, the waveform features include at least one of time features, amplitude features, morphological features, and integrity features.

[0087] In some embodiments of the present invention, in conjunction with reference to the reference Figure 5 The step 130, which determines the waveform characteristics of each waveform component based on the feature points of each waveform component, may also include step 131 (not shown): determining the time characteristics of each waveform component.

[0088] The time characteristics include one or more of the following: P wave duration, PQ interval, QRS complex duration, QT interval, RR interval, heart rate, and corrected QT interval (QTc).

[0089] In a further embodiment of the present invention, the step of determining the time characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the P-wave duration based on the time difference between the P-wave start point and the P-wave end point. Specifically, the P-wave duration P... i =p 3i -p 1i ; , where P 3i It is the time coordinate of the end of the P-wave, P 1i This is the time coordinate of the P wave initiation. For illustrative purposes, and not restrictive, the P wave duration reflects the duration of atrial depolarization, with a normal range of less than 0.1 seconds (100 milliseconds). A P wave duration exceeding 0.1 seconds may indicate atrial enlargement or other atrial abnormalities.

[0090] In a further embodiment of the present invention, the step of determining the temporal characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the PQ interval based on the time difference between the P-wave initiation point and the Q-wave initiation point. Specifically, the PQ interval is PQ i =q 1i - p 1i ; , where q 1i It is the time coordinate of the Q-wave initiation point, p 1i It is the time coordinate of the P wave initiation. For the purpose of interpretation and not limitation, the PQ interval reflects the time it takes for cardiac electrical impulses to conduct from the atria to the ventricles.

[0091] In a further embodiment of the present invention, the step of determining the time characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the QRS group duration based on the time difference between the Q wave start point and the S wave end point. Specifically, the QRS group duration is QRS... i =s 2i -q 1i ; , where s 2i It is the time coordinate of the S-wave endpoint, q 1iIt is the time coordinate of the Q wave initiation. For the purpose of interpretation and not limitation, the QRS duration reflects the time of ventricular depolarization. A normal QRS duration is less than 0.12 seconds; a QRS duration exceeding 0.12 seconds suggests intraventricular conduction delay, such as left bundle branch block (LBBB) or right bundle branch block (RBBB).

[0092] In a further embodiment of the present invention, the step of determining the time characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the QT interval based on the time difference between the Q wave initiation point and the T wave termination point. Specifically, the QT interval is QT... i =q 1i - t 3i ; , where t 3i It is the time coordinate of the T-wave endpoint, q 1i It is the time coordinate of the Q wave initiation. For the purpose of interpretation and not limitation, the QT interval reflects the total electrical activity time of the ventricle, including the entire process of depolarization and repolarization. QT interval abnormalities may be associated with serious arrhythmias.

[0093] In a further embodiment of the present invention, the step of determining the time characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the RR interval based on the time difference between two adjacent R-wave peak points. Specifically, the RR interval RR... i = r (i+1) -r i ; ,in r (i+1) It is the time coordinate of the R-wave peak point in the (i+1)th cardiac cycle, r i This is the time coordinate of the peak R-wave point in the i-th cardiac cycle. For illustrative purposes and not as a limitation, the RR interval reflects the length of the cardiac cycle. The RR interval in normal sinus rhythm is typically between 0.6 and 1.0 seconds and is relatively constant. Abnormal RR intervals suggest possible arrhythmias or other cardiac problems. For example, a prolonged RR interval (greater than 1.0 seconds) suggests sinus bradycardia or atrioventricular block; a shortened RR interval (less than 0.6 seconds) suggests sinus tachycardia or rapid arrhythmias; and an irregular RR interval suggests possible arrhythmias such as atrial fibrillation, premature contractions, or ventricular tachycardia.

[0094] In a further embodiment of the present invention, the determination of the temporal characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the heart rate according to the RR interval using a preset heart rate calculation formula. Specifically, the heart rate is HR=60×1000 / RR, which converts the RR interval into the number of heartbeats per minute, intuitively reflecting the frequency of cardiac activity.

[0095] In a further embodiment of the present invention, the step of determining the time characteristics of each waveform component, i.e., step 131 (not shown), may specifically include: determining the corrected QT interval QTc using a preset correction formula based on the QT interval and the RR interval. In an embodiment of the present invention, the Framingham formula can be used to calculate the QT interval QTc: By way of explanation and not limitation, the corrected QT interval (QTc) excludes the influence of heart rate variations on the QT interval, facilitating comparisons of QT intervals at different heart rates and can be used to assess ventricular repolarization abnormalities. However, those skilled in the art will recognize that other formulas can be used to calculate the QT interval (QTc), including but not limited to the Bazett formula, the Fridericia formula, or the Hodges formula.

[0096] In some embodiments, the time feature may include one or more of the time features described above or otherwise suitable, which falls within the scope of this invention.

[0097] In some embodiments of the present invention, the step of determining the waveform characteristics of each waveform component based on the characteristic points of each waveform component, i.e., step 130, may include step 132 (not shown): determining the integrity status of each cardiac cycle based on the boundary boxes of the P wave, QRS complex, and T wave, or the presence of characteristic points of the P wave, QRS complex, and T wave in each cardiac cycle. By detecting whether the integrity characteristics (including their presence or absence) of each key waveform component in the cardiac cycle are present, specific types of arrhythmias, such as atrial fibrillation and premature beats, can be identified. As an explanation, and not a limitation, a normal electrocardiogram rhythm should consist of a P wave, QRS complex, T wave, and U wave; the absence of certain waveform components may indicate the presence of a specific type of heart disease.

[0098] In some embodiments of the present invention, such as Figure 4 Combination Figure 5 As shown, determining the integrity status of each cardiac cycle based on the presence of characteristic points of the P wave, QRS complex, and T wave in each cardiac cycle, i.e., step 132 (not shown), may include: Step 410: Determine whether there are missing coordinate sequences of characteristic points of P wave, QRS complex and T wave in each cardiac cycle in the electrocardiogram waveform image.

[0099] In this embodiment of the invention, in step 410, it is possible to detect whether the P wave, QRS complex and T wave of each cardiac cycle are completely present; in some embodiments, it is possible to detect whether the P wave, QRS complex, ST segment, T wave and U wave of each cardiac cycle are completely present; it is also conceivable to detect more or fewer waveform components.

[0100] In some embodiments, integrity features can be determined by determining whether the feature point coordinate sequence is missing. In a further embodiment, for the i-th cardiac cycle, the feature point coordinates P of the P wave can be checked. i {p 1i ,p 2i ,p 3i}; Coordinates of characteristic points of the QRS group i {q 1i ,q 2i ,r i ,s 1i ,s 2i}; Coordinates of feature points in segment ST i {st 1i ,st 2i The characteristic point coordinates of the T-wave are T. i {t 1i ,t 2i ,t 3i}; Coordinates of the characteristic point of the U-wave U i {u 1i ,u 2i ,u 3i The presence of missing coordinates and / or incorrect sequence order is checked. If all feature point coordinates are present and / or the sequence order is correct, the waveform components of the cardiac cycle are considered complete; if some feature point coordinates are missing and / or the sequence order is incorrect, it indicates that the corresponding waveform components may be missing.

[0101] In a further specific embodiment, the coordinate point sequence of the bounding boxes segmented by the target recognition model can be determined by marking them according to rhythmic sequence numbers, thereby assisting in determining the integrity features. However, those skilled in the art will understand that the marking of the rectangular feature boxes segmented by the target recognition model can be performed in other steps or in combination with other steps, or as a specific implementation of other steps, for example, it can be combined with step 120 or as a partial implementation of step 120, which falls within the scope of this invention.

[0102] In a further specific embodiment, an exemplary scheme for marking the rectangular feature boxes (bounding boxes) segmented by the target recognition model according to the rhythm sequence number to determine the coordinate point sequence of the bounding boxes is as follows: For P-wave feature boxes: PK i {(p 1i,0);(p 1i V p2i (p 2i ));(V p2i (p 2i ),p 3i );(p 3i ,0)}, where PK i The rectangular feature box representing the P wave in the i-th cardiac cycle is defined by four coordinate points, namely the lower left corner (p... 1i ,0), top left corner point (p 1i V p2i (p 2i )), upper right corner (V) p2i (p 2i ),p 3i ) and the bottom right corner point (p 3i ,0).

[0103] For QRS group feature boxes: QRSK i {(q 1i V q1i (q 1i ));(q 1i V ri (r i ));(s 2i V ri (r i ));(s 2i V s2i (s 2i ))}, where QRSK i The rectangular feature box representing the QRS complex in the i-th cardiac cycle is defined by four coordinate points, namely the lower left corner (q... 1i V q1i (q 1i )), top left corner (q) 1i V ri (r i )), top right corner point (s 2i V ri (r i )) and the bottom right corner point (s) 2i V s2i (s 2i Furthermore, if V s2i (s 2i ) <V q1i (q 1i ), then V s2i (s 2i )=V q1i (q 1i If V s2i (s 2i )>Vq1i (q 1i ), then V q1i (q 1i )=V s2i (s 2i ).

[0104] For T-wave feature boxes: TK i {(t 1i V t1i (t 1i ));(t 1i V t2i (t 2i ));(t 3i V t2i (t 2i ));(t 3i V t3i (t 3i ))}, where TKi represents the rectangular feature box of the T wave in the i-th cardiac cycle, defined by four coordinate points, namely the lower left corner point (t 1i V t1i (t 1i )), top left corner point (t) 1i V t2i (t 2i )), top right corner point (t) 3i V t2i (t 2i )) and the bottom right corner point (t) 3i V t3i (t 3i Furthermore, if V t1i (t 1i ) <V t3i (t 3i ), then V t1i (t 1i )=V t3i (t 3i If V t1i (t 1i )>V t3i (t 3i ), then V t3i (t 3i )=V t1i (t 1i ).

[0105] If there is no missing data, proceed to step 420: determine if the corresponding cardiac cycle is a complete rhythm.

[0106] In this embodiment of the invention, if the coordinates of all characteristic points of the P wave, QRS complex, and T wave in the cardiac cycle are detected, then the cardiac cycle is marked as a complete rhythm in step 420. Specifically, the system detects P waves sequentially from i=1,⋯N. i QRS i ,T i Check if the electrocardiogram waveform of the i-th rhythm is complete. If all P... i QRS i ,T i If all waveforms are present, the system marks that cardiac cycle as a complete rhythm. For the purposes of interpretation and not limitation, a complete rhythm generally indicates that the heart's excitation, conduction, and repolarization processes are normal.

[0107] When a missing feature point exists, proceed to step 420': determine the missing waveform component of the corresponding cardiac cycle based on the coordinates of the missing feature point.

[0108] In this embodiment of the invention, if the coordinates of all feature points of the P wave, QRS complex, and T wave in the cardiac cycle are missing and / or ordered incorrectly, it is marked as an incomplete rhythm. As an explanation, and not a limitation, when the P wave cannot be extracted... i When the coordinates of the characteristic points are known, it is called a P-wave missing rhythm; when the QRS complex cannot be extracted... i When the coordinates of the characteristic points are known, it is called a missing QRS rhythm; when T cannot be extracted... i When the coordinates of the characteristic points are known, it is called a T-wave missing rhythm.

[0109] While the above embodiments describe waveform integrity determination based on feature point coordinates, in other embodiments, it is conceivable to determine waveform integrity based on whether the bounding box (feature box) is complete or in the correct order. As an explanation, and not a limitation, when P cannot be detected... i When the feature box is visible, it is called a P-wave missing rhythm; when the QRS cannot be detected... i When the feature box is visible, it is called a missing QRS rhythm; when T waves cannot be detected... i When the feature box is drawn, it is called a T-wave missing rhythm.

[0110] As an explanation, and not a limitation, waveform integrity determination based on feature point coordinates provides a more accurate description. In other embodiments, a combination of the two approaches may also be conceived.

[0111] In some embodiments of the present invention, in conjunction with reference to the reference Figure 5 The step 130, which determines the waveform features of each waveform component based on the feature points of each waveform component, may also include step 133 (not shown): determining the morphological features of each waveform component.

[0112] In some embodiments of the present invention, morphological features include the tilt characteristics of the ST segment of the electrocardiogram waveform. By way of explanation and not limitation, the tilt of the ST segment is an indicator for assessing myocardial ischemia, injury, or other cardiac diseases, and analyzing its morphological changes can provide objective evidence for clinical diagnosis.

[0113] like Figure 6 Combination Figure 7 As shown, in some embodiments of the present invention, step 133 (not shown) may specifically include: Step 610: Obtain the time and amplitude values ​​of the S-wave end point and the T-wave start point.

[0114] In this embodiment of the invention, in step 610, the time and amplitude values ​​(s) of the S-wave endpoint can be extracted based on the feature point coordinates obtained in step 120. 2i V s2i (s 2i The time and amplitude values ​​of the T-wave initiation (t) 1i V t1i (t 1i )) is used to determine the tilt state of the ST segment, such as the slope.

[0115] Step 620: Calculate the slope of the line connecting the end of the S wave and the beginning of the T wave.

[0116] In this embodiment of the invention, in step 620, the slope of the straight line connecting the two points can be calculated based on the coordinate values ​​of the S-wave endpoint and the T-wave initiation point.

[0117] Specifically, the formula for calculating the slope is:

[0118] Step 630: Determine the shape of the ST segment based on the slope of the connecting line.

[0119] In this embodiment of the invention, in step 630, the shape of the ST segment can be determined based on the slope determined in step 620, such as... Figure 7 As shown in (a), when the slope is zero, segment ST is determined to be straight; as Figure 7 As shown in (b), when the slope is positive, the ST segment is determined to be upward sloping; Figure 7 As shown in (c), when the slope is negative, the ST segment is determined to be downward sloping.

[0120] As an explanation rather than a limitation, ST segment changes may indicate myocardial ischemia, injury, inflammation, or other cardiac diseases; for example, ST segment elevation (sloping upward) may indicate acute myocardial infarction, pericarditis, or early repolarization syndrome; ST segment depression (sloping downward) may indicate myocardial ischemia, myocarditis, or excessive cardiac workload.

[0121] In some embodiments of the present invention, the morphological features also include the directional features of the T wave in the electrocardiogram waveform. By way of explanation and not limitation, the direction of the T wave is a morphological feature in the electrocardiogram representing ventricular repolarization, and its abnormalities may be associated with various cardiac pathological states.

[0122] like Figure 8 Combination Figure 9 As shown, in some embodiments of the present invention, step 133 (not shown) may specifically include: Step 810: Obtain the amplitude values ​​of the T-wave start point, peak point, and end point.

[0123] In this embodiment of the invention, in step 810, the amplitude value V of the T-wave initiation point can be extracted based on the feature point coordinates obtained in step 120. t1i (t 1i ), the amplitude value V at the peak point t2i (t 2i ) and the amplitude value V at the endpoint t3i (t 3i ), used to determine the direction of the T wave.

[0124] Step 820: Compare the magnitude of the peak point amplitude with the magnitude of the starting and ending points.

[0125] In this embodiment of the invention, in step 820, the directional characteristics of the T-wave can be determined by comparing the relative magnitudes of the peak amplitude of the T-wave with the amplitudes of the start and end points.

[0126] Step 830: Determine the T-wave morphology based on the size relationship.

[0127] In embodiments of the present invention, such as Figure 9 As shown, in step 830, the direction of the T-wave can be determined based on the comparison results, such as... Figure 9 As shown in (a), when the peak amplitude is greater than the amplitudes at the start and end points, it is determined to be a positive T-wave; Figure 9 As shown in (b), when the peak amplitude is smaller than the amplitudes at the start and end points, it is defined as an inverted T wave. As an explanation and not a limitation, different T wave morphologies correspond to different clinical meanings. For example, a flat or inverted T wave may indicate myocardial ischemia, myocardial infarction, hypokalemia, or drug effects, while a tall and peaked T wave may indicate hyperkalemia, early acute myocardial infarction, or apical hypertrophic cardiomyopathy.

[0128] In this embodiment of the invention, by determining the ST segment tilt characteristics and T wave direction characteristics, the morphological characteristics of the electrocardiogram waveform can be objectively and quantitatively analyzed, providing morphological information for assessing cardiac function. Although not shown, the waveform features of embodiments of the present invention may also include amplitude features, particularly the amplitude features of the peaks and valleys of each waveform component (such as any combination of P-wave, QRS complex, ST segment, T-wave and U-wave).

[0129] Step 140: Based on the waveform characteristics of each waveform component, generate an electrocardiogram waveform evaluation result, which reflects the cardiac function status corresponding to the electrocardiogram.

[0130] In this embodiment of the invention, in step 140, the waveform features determined in step 130 can be further transformed into intuitive and understandable cardiac function status assessment results. By introducing an artificial intelligence text analysis system, the quantified waveform feature data is transformed into clinically interpretable assessment conclusions, providing support for a comprehensive evaluation of cardiac function status.

[0131] like Figure 10 As shown, based on the waveform characteristics of each waveform component, an electrocardiogram waveform evaluation result is generated, i.e., step 140, which may include the following steps: Step 1010: Transform the waveform features into a first structured text description containing standard medical terminology.

[0132] In step 1010, the waveform features (including time features, morphological features, integrity features, etc.) obtained in step 130 can be converted into standardized medical text descriptions. This conversion process involves mapping quantified waveform feature parameters to corresponding medical terminology and descriptive language to conform to the professional expression of clinical electrocardiogram diagnostic reports. In some embodiments, it may include only a direct description of the waveform features of the electrocardiogram, or it may include a description of the standard corresponding to the waveform features; in some embodiments, it may also include a quantitative or qualitative description of the judgment value and / or extended value obtained by comparing the waveform features of the electrocardiogram with the standard; it may also include a reasonable combination of the above.

[0133] As an example, and not a limitation, the first structured text description may include multidimensional features such as P wave description (e.g., "P wave duration is XX ms", "P wave duration is XX ms, standard P wave duration is XX ms", "P wave duration is normal / prolonged"), PQ interval description (e.g., "PQ interval is XX ms, slightly prolonged"), QRS complex description (e.g., "QRS complex width is XX ms, slightly widened"), ST segment description (e.g., "ST segment horizontal depression XX mm"), T wave description (e.g., "T wave inversion", "Signal T wave inversion"), QT interval description (e.g., "QTc=XX ms, at the upper limit of normal"), and heart rate description (e.g., "Sinus rhythm, heart rate XX beats / min"). These multidimensional waveform feature descriptions together constitute the complete electrocardiogram waveform feature text.

[0134] In this embodiment of the invention, the first structured text description can adopt various data formats, such as JSON, XML, YAML, or key-value pairs, to facilitate subsequent system processing and analysis.

[0135] Step 1030: Input the first structured text description into the artificial intelligence text analysis system through a preset interface format.

[0136] In step 1030, the first structured text description can be input into the artificial intelligence text analysis system through a preset interface format. The interface format is designed to meet the input requirements of the artificial intelligence text analysis system and typically includes components such as task description, contextual information, waveform feature description, and indication of the expected output.

[0137] In some embodiments of the present invention, interface input and output reception can be implemented through API protocols. The types of API protocols include, but are not limited to, standard interface protocols such as REST API and OpenAPI. Other interfaces or calling methods suitable for large language models can also be used, including but not limited to model context protocols and function call interface protocols.

[0138] In a preferred embodiment, the artificial intelligence text analysis system is based on one or more large language models, which are trained with large-scale medical texts and cardiology knowledge, enabling them to understand complex electrocardiogram terms and concepts and perform in-depth analysis and reasoning based on input waveform feature descriptions.

[0139] In some embodiments of the present invention, various parameters of the artificial intelligence text analysis system, such as temperature parameters, sampling strategies, and maximum output length, can be configured to optimize analysis performance. The artificial intelligence text analysis system can also employ advanced technologies such as thought chain reasoning, retrieval-enhanced generation, or hybrid expert models to improve the accuracy and reliability of electrocardiogram analysis.

[0140] Step 1040: Analyze the first structured text description using an artificial intelligence text analysis system and output the electrocardiogram waveform evaluation result.

[0141] In step 1040, the AI ​​text analysis system processes the input first structured text description and generates an electrocardiogram waveform assessment result based on built-in cardiology knowledge. The assessment result may include cardiac rhythm status assessment (such as sinus rhythm, atrial fibrillation, tachycardia, etc.), conduction system assessment (such as atrioventricular block, bundle branch block), myocardial status assessment (such as myocardial ischemia, signs of myocardial infarction), and comprehensive risk assessment.

[0142] In some embodiments, AI text analysis systems can use domain-specific knowledge to guide the reasoning process, ensuring the clinical relevance and accuracy of the results. AI text analysis systems can provide confidence metrics for the assessment results and explain the reasoning basis, enhancing the credibility and interpretability of the results and facilitating physician understanding and adoption.

[0143] Optionally, the ECG waveform assessment results are generated based on the waveform characteristics of each waveform component. This process can also incorporate user consultation information to provide more comprehensive and personalized assessment results. Specifically, step 140 may include: Step 1010' (not shown): Convert the waveform features into a first structured text description containing standard medical terminology; Similar to step 1010, this step transforms the waveform features into a standardized medical text description. This first structured text description includes specialized descriptions of the temporal, morphological, and integrity characteristics of each waveform component, conforming to the expression style of clinical electrocardiogram diagnostic reports.

[0144] Step 1020' (not shown): Obtain the user's consultation information and convert it into a second structured text description.

[0145] In this step, relevant health information provided by the user during the electrocardiogram (ECG) examination can be obtained. This user consultation information includes symptom descriptions, medical history, medication use, and lifestyle habits, with symptom descriptions and medical history being particularly important for assessing cardiac function. Artificial intelligence text analysis acquires this information through a pre-defined consultation interface or natural language interaction, and converts unstructured user input into standardized, second-structured text descriptions through semantic analysis, facilitating subsequent integration and analysis with waveform feature information.

[0146] Step 1030' (not shown): Input the first structured text description and the second structured text description into the artificial intelligence text analysis system through a preset interface format.

[0147] Similar to step 1030, this step also inputs the AI ​​text analysis system through a preset interface format. The difference from step 1030 is that this step simultaneously inputs a first structured text description representing the characteristics of the electrocardiogram waveform and a second structured text description representing the user's consultation information. This allows the AI ​​text analysis system to process two types of relevant information at the same time, thereby generating more personalized and comprehensive cardiac function assessment results.

[0148] Step 1040' (not shown): Analyze the first structured text description and the second structured text using the artificial intelligence text analysis system, and output the electrocardiogram waveform evaluation result.

[0149] Similar to step 1040, this step analyzes the input data and outputs an electrocardiogram (ECG) waveform assessment result through an AI-powered text analysis system. The difference lies in that this step combines waveform characteristics with comprehensive data from the user's medical history, enabling a more personalized assessment tailored to the patient's specific situation. For example, when the ECG shows a prolonged QT interval and the user reports taking a specific medication, the AI-powered text analysis system can consider the potential impact of the medication in the assessment result, providing a more accurate risk assessment and recommendations.

[0150] Through step 140 and its sub-steps, the present invention realizes the transformation from objective waveform characteristics to clinical assessment results, provides intelligent analysis capabilities for electrocardiogram waveform assessment, and effectively supports the assessment of cardiac function status and the medical decision-making process.

[0151] In this embodiment of the invention, prior to the identification step, the electrocardiogram waveform evaluation method may further include step 100': training a target identification model.

[0152] In some embodiments, target recognition model training is a preliminary step in electrocardiogram (ECG) waveform recognition, aiming to enable the model to accurately identify various waveform components such as the P wave, QRS complex, ST segment, T wave, and U wave in ECG waveform images. Through supervised learning, the model learns the feature representations of ECG waveform components, thereby enabling it to automatically identify various waveform components in new, unlabeled ECGs.

[0153] like Figure 11 The flowchart shown is a training process for an electrocardiogram waveform recognition model according to an embodiment of the present invention. The training target recognition model specifically includes steps 101' to 104': Step 101': Obtain multiple electrocardiogram training samples.

[0154] In this embodiment of the invention, step 101' may involve collecting and preparing electrocardiogram (ECG) data for training the target recognition model. The ECG training samples involved in this embodiment can be training data from any legal source, including but not limited to data from hospital electronic medical record systems, clinical ECG databases, records from cardiology hospitals, and data collected by medical devices or apps and / or health or fitness devices or apps uploaded with the user's legal authorization. In a preferred embodiment, the training samples can cover different types of ECG waveforms, including normal ECGs and various abnormal ECGs related to heart diseases. Preferably, various abnormal ECGs related to heart diseases can account for a given proportion.

[0155] Step 102': Preprocess the ECG training samples to obtain preprocessed training samples, including converting time-series ECG data into ECG waveform images.

[0156] In this embodiment of the invention, the preprocessing step may include converting the original time-series electrocardiogram data into a two-dimensional waveform image, where the horizontal axis represents time and the vertical axis represents voltage amplitude. This conversion makes the electrocardiogram data suitable as an image input into the target recognition model.

[0157] In some embodiments, the preprocessing of the training dataset further includes: filtering the electrocardiogram training samples using a bandpass filter to remove DC interference and environmental noise interference; enhancing the electrocardiogram training samples, the enhancement process including adding random noise, adjusting waveform amplitude, stretching or compressing the time axis at least one; and uniformly adjusting the processed electrocardiogram training samples into standardized images of a predetermined size.

[0158] In this embodiment of the invention, the preprocessing during the training process can refer to the relevant image preprocessing or its features during the recognition process, and vice versa.

[0159] In one embodiment, the processed electrocardiogram samples can be uniformly adjusted to standardized images of a predetermined size, such as the 448×448 pixel size typically used by the YOLO model, which is similar to the size used in the recognition process.

[0160] Step 103': Label each waveform component in the electrocardiogram waveform image of the preprocessed training sample to generate a training dataset.

[0161] In this embodiment of the invention, the marking includes marking the location areas of the P wave, QRS complex, ST segment, T wave, and U wave.

[0162] In this embodiment of the invention, step 103' involves electrocardiogram (ECG) Figure 2 In the 3D waveform dataset, each waveform component, such as the P wave, QRS complex, ST segment, T wave, and U wave, is labeled with a rectangular feature box. Figure 2 As shown in (b) above. In some embodiments, the annotation work can be performed by medical professionals, such as electrocardiogram (ECG) professionals, who, based on their expertise in ECG interpretation, mark the precise location regions of each waveform component on each ECG waveform image. Thus, each annotated rectangular feature box defines the region of the corresponding waveform component in the image, including information such as the upper-left corner coordinates, width, and height. Accordingly, the annotation of this training data will guide the model to learn to recognize the features of each waveform component.

[0163] Step 104': Input the training dataset into the target recognition model for training, and optimize the model parameters through the backpropagation algorithm until the preset conditions are met.

[0164] In this embodiment of the invention, step 104' is the training step of the model.

[0165] In some embodiments of the present invention, the YOLO object detection model is selected. As previously described and as... Figure 3 As shown, the YOLO model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, a detection layer, and an output layer. These components together constitute an end-to-end target recognition model, which can directly identify and locate each waveform component from an electrocardiogram waveform image.

[0166] In some embodiments, the input layer may integrate the image preprocessing functions described above (or may not), which falls within the scope of this invention. When integrating the above functions, the ECG waveform image size can be cropped to 448×448 pixels for the YOLO model. Regardless of whether the above image resizing function is integrated, the input layer 31 can input training data into subsequent functional modules, such as subsequent convolutional layers.

[0167] In some embodiments, convolutional layers may include multiple convolutional layer structures for extracting image features. Each layer typically includes convolution, batch normalization, and an activation function (such as ReLU). For example, Conv2D(64, (3, 3), padding='same') -> BatchNorm -> ReLU (i.e., a sequentially connected two-dimensional convolutional layer (outputting 64 feature maps, using 3×3 convolutional kernels with same-dimensional padding), a batch normalization layer, and a ReLU activation function layer that introduces a non-linear transformation).

[0168] In some embodiments, pooling layers can be used to reduce the size of the feature map, thereby reducing computation and extracting higher-level features. This embodiment employs max pooling. For example, MaxPooling2D(pool size=(2, 2), strides=(2, 2)) (i.e., using a 2×2 pooling window, sliding across the feature map with a stride of 2 pixels, selecting the maximum value within the window as the output each time, thereby reducing the height and width of the feature map by half, preserving salient features while reducing computational complexity).

[0169] In some embodiments, fully connected layers can be used to transform feature maps into the final detection result. For example, Conv2D(1024, (3, 3), padding='same') -> BatchNorm -> ReLU (i.e., a convolutional layer with 1024 output channels, a 3×3 kernel size, and padding of the same dimension, followed by, for example, the batch normalization and ReLU activation function described above). As an explanation, and not a limitation, in a preferred embodiment of the invention, convolutional layers (as described above) are used instead of "traditional" fully connected layers. By replacing traditional fully connected layers, spatial information is preserved while reducing the number of model parameters, making it more suitable for processing the spatial features of electrocardiogram waveforms. However, those skilled in the art will understand that in embodiments of the invention, this feature can still be referred to as a fully connected layer and encompasses schemes of traditional fully connected layers or non-traditional fully connected layers, such as fully connected layers presenting as convolutional layers.

[0170] In some embodiments, detection layers can be used to output detection results. The detection layer is the last layer and can be constructed as a convolutional layer. For example, the detection layer can output a feature map containing bounding box coordinates, class probabilities, and confidence. For example, Conv2D(num_classes + 5, (1, 1)) (i.e., a two-dimensional convolutional layer using a 1×1 convolutional kernel, whose output channel number is the number of classes plus n (n=5), where n can correspond to the N (e.g., five) basic parameters of the bounding box: center coordinates x, y, width w, height h, and confidence), thereby enabling the model to simultaneously predict the bounding box location information and class probability at each detection location.

[0171] In some embodiments, output layer 36 is used to output object detection results. In some embodiments, the output feature map is divided into an S×S grid, with each grid predicting B bounding boxes (B is typically 2 or 3). Each bounding box contains 5 values: (x, y, w, h, confidence) and the probabilities of C classes. For example, the output shape is (S, S, B * (5 + C)), where S is the grid size, B is the number of bounding boxes per grid, and C is the number of classes.

[0172] In this embodiment of the invention, during the training process, the training data can be divided into a training set and a test set according to a given ratio, such as 8:2. The training set is used for model training, and the test set is used for model optimization. The model receives preprocessed and labeled electrocardiogram waveform images as input and outputs prediction results for each waveform component. By comparing the prediction results with the true labels, the loss value is calculated, and the model parameters are updated through the backpropagation algorithm, continuously iterating and optimizing until the preset conditions are met.

[0173] In this embodiment of the invention, the preset conditions include, but are not limited to: when the performance of the target recognition model on the validation dataset meets the preset conditions, the training is determined to be complete, enabling the target recognition model to recognize each waveform component in the electrocardiogram waveform image and generate the corresponding bounding box. These preset conditions may include performance indicators such as the average intersection-over-union (IoU) on the validation set reaching a predetermined threshold and the classification accuracy reaching a predetermined value.

[0174] After training, the trained model is deployed and can be used to complete step 110 in the electrocardiogram waveform evaluation method, that is, to identify the waveform components of multiple cardiac cycles in the electrocardiogram waveform image through the target recognition model and generate bounding boxes corresponding to the waveform components of the multiple cardiac cycles.

[0175] In some respects, as mentioned above, the training process described can be used as part of the electrocardiogram (ECG) waveform evaluation method, for example, as a preliminary step. In this implementation, the ECG waveform evaluation method comprises two parts: model training and waveform evaluation.

[0176] On the other hand, some embodiments of the electrocardiogram waveform evaluation method of the present invention may not include a training process, whereby the model used is already trained.

[0177] In another aspect, some embodiments of the present invention may relate to related training methods, which, for example, include or incorporate the training process described above. Those skilled in the art will understand that the training methods may be independent of the electrocardiogram waveform evaluation methods of the embodiments of the present invention.

[0178] like Figure 12 As shown, this embodiment of the invention also provides an electrocardiogram waveform evaluation device 1200, which may include: The recognition unit 1210 is configured to recognize each waveform component of multiple cardiac cycles in an electrocardiogram waveform image through a target recognition model, and generate bounding boxes corresponding to each waveform component of multiple cardiac cycles. The acquisition unit 1220 is configured to acquire feature points of each waveform component based on the bounding box of each waveform component; The determining unit 1230 is configured to determine the waveform characteristics of each waveform component based on the characteristic points of each waveform component; and The evaluation unit 1240 is configured to generate an electrocardiogram waveform evaluation result based on the waveform characteristics of each waveform component. The evaluation result reflects the cardiac function status corresponding to the electrocardiogram.

[0179] Under the guidance of this invention, features or steps (sub-steps) of the method embodiments of this invention can be combined with device embodiments in a non-contradictory manner to obtain new embodiments, and features or components (sub-components) of device embodiments can also be combined with method embodiments in a non-contradictory manner to obtain new embodiments, which falls within the scope of this invention.

[0180] exist Figure 13 The illustrated embodiment provides an electrocardiogram (ECG) waveform evaluation system. This ECG waveform evaluation system may include: a user terminal 131, an AI feature extraction model 132, and an AI analysis model 133.

[0181] In some embodiments of the present invention, the user terminal is configured to interact with the AI ​​feature extraction model and the AI ​​analysis model to present evaluation results.

[0182] In some embodiments of the present invention, the user terminal includes an electrocardiogram (ECG) acquisition module 1311, which is configured to acquire ECG data.

[0183] In some embodiments of the present invention, the AI ​​feature extraction model is configured to generate an electrocardiogram waveform image based on the electrocardiogram data, identify each waveform component of multiple cardiac cycles in the electrocardiogram waveform image, generate corresponding bounding boxes, extract feature points of each waveform component from the bounding boxes, determine the waveform features of each waveform component, and convert the waveform features into a first structured text description.

[0184] In some embodiments of the present invention, the AI ​​analysis model is configured to receive a first structured text description, analyze and process it, generate an assessment result reflecting the cardiac function status corresponding to the electrocardiogram, and return the assessment result to the user terminal.

[0185] In some embodiments of the present invention, the user terminal further includes a consultation module 1312, configured to receive user consultation information and convert it into a second structured text description.

[0186] In some embodiments of the present invention, the AI ​​analysis model is configured to receive a first structured text description and a second structured text description, analyze and process them, generate an assessment result reflecting the cardiac function status corresponding to the electrocardiogram, and return the assessment result to the user terminal.

[0187] Under the guidance of this invention, features or steps (sub-steps) of the method embodiments of this invention can be combined with system embodiments in a non-contradictory manner to obtain new embodiments, and features or components (sub-components) of system embodiments can also be combined with method embodiments in a non-contradictory manner to obtain new embodiments, which falls within the scope of this invention.

[0188] In some embodiments of the present invention, an electronic device is provided, including a processor and a memory storing a computer program, the processor being configured to implement the method of any embodiment of the present invention when running the computer program. Figure 14 A schematic diagram is shown of a method or electronic device 1400 that can be used to implement embodiments of the present invention. In some embodiments, the number of electronic devices may be more or less than the number shown. In some embodiments, it can be implemented using a single or multiple electronic devices. It can also be implemented using cloud or distributed electronic devices in some embodiments.

[0189] like Figure 14 As shown, the electronic device 1400 includes a processor 1401 and a memory 1402. The processor executes programs stored in the memory, which, when executed by a computer, can implement the methods, steps, or functions described in the above embodiments. The processor 1401 may include various types of processors, such as a central processing unit (CPU), a graphics processing unit (GPU), a neural network processor (NPU), a digital signal processor (DSP), etc. The processor 1401 and the memory 1402 are interconnected via a bus 1403. Input / output (I / O) interfaces may also be connected to the bus 1403. The systems, devices, modules, or units described in the above embodiments can be implemented by a computer or its associated components. The computer may be, for example, a mobile terminal, a smartphone, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a personal digital assistant, a media player, a navigation device, a game console, a tablet computer, a wearable device, a smart TV, an Internet of Things system, a smart home, an industrial computer, a server, or a combination thereof.

[0190] Although not shown, in this embodiment of the invention, a program product is provided, including a computer program that, when executed by a processor, implements the method of any embodiment of the invention.

[0191] Although not shown, in an embodiment of the invention, a storage medium is provided storing a computer program configured to be run to implement the method of any embodiment of the invention.

[0192] Storage media in embodiments of the present invention include articles that are permanent and non-permanent, removable and non-removable, capable of storing information by any method or technology. Examples of storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0193] The methods, programs, systems, apparatuses, etc., in embodiments of the present invention can be executed or implemented in one or more networked computers, or practiced in a distributed computing environment. In the embodiments of this specification, in these distributed computing environments, tasks can be performed by remote processing devices connected via a communication network.

[0194] Those skilled in the art will understand that the embodiments described in this specification can be provided as methods, systems, or computer program products. Therefore, those skilled in the art will realize that the functional modules / units or controllers and related method steps described in the above embodiments can be implemented in software, hardware, or a combination of both.

[0195] Unless explicitly stated otherwise, the actions or steps of the methods and procedures described in the embodiments of the present invention do not necessarily have to be performed in a specific order and can still achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0196] This document describes several embodiments of the present invention; however, for the sake of brevity, the descriptions of the embodiments are not exhaustive, and identical or similar features or parts between the embodiments may be omitted. In this document, "one embodiment," "some embodiments," "example," "specific example," or "some examples" refers to embodiments applicable to at least one, but not all, of the present invention. The above terms do not necessarily refer to the same embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of the different embodiments or examples.

[0197] The exemplary systems and methods of the present invention have been specifically shown and described with reference to the above embodiments, which are merely examples of the best mode for implementing the systems and methods. Those skilled in the art will understand that various changes can be made to the embodiments of the systems and methods described herein without departing from the spirit and scope of the invention as defined in the appended claims when implementing the systems and / or methods.

Claims

1. A method for evaluating electrocardiogram waveforms, characterized in that, include: The target recognition model identifies the waveform components of multiple cardiac cycles in the electrocardiogram waveform image and generates bounding boxes corresponding to each waveform component of the multiple cardiac cycles. Based on the bounding boxes of each waveform component, the feature points of each waveform component are obtained; Based on the feature points of each waveform component, the waveform characteristics of each waveform component are determined; as well as Based on the waveform characteristics of each waveform component, an electrocardiogram waveform evaluation result is generated, which reflects the cardiac function status corresponding to the electrocardiogram.

2. The electrocardiogram waveform evaluation method according to claim 1, characterized in that, The method further includes: The time-series electrocardiogram (ECG) data is converted into the ECG waveform image.

3. The electrocardiogram waveform evaluation method according to claim 1, characterized in that, The waveform components of the cardiac cycle include one or more of the following: P wave, QRS complex, ST segment, T wave, and U wave; The characteristic points of each waveform component include: the start point, peak point, and end point of the P wave; the start point and trough point of the Q wave, the peak point of the R wave, and the trough point and end point of the S wave of the QRS wave; the start point and end point of the ST segment; the start point, peak point, and end point of the T wave; and some or all of the characteristic points of the start point, peak point, and end point of the U wave.

4. The electrocardiogram waveform evaluation method according to claim 1, characterized in that, The step of obtaining feature points of each waveform component based on the bounding boxes of each waveform component includes: The coordinates of the feature points of the corresponding waveform components are determined based on the intersection of the bounding boxes of each waveform component with the electrocardiogram waveform.

5. The electrocardiogram waveform evaluation method according to claim 1, characterized in that, The waveform features include at least one of time features, amplitude features, morphological features, and integrity features.

6. The electrocardiogram waveform evaluation method according to claim 5, characterized in that, The step of determining the waveform characteristics of each waveform component based on its characteristic points includes: determining the time characteristics of each waveform component, wherein the time characteristics include one or more of the following: P wave duration, PQ interval, QRS complex duration, QT interval, RR interval, heart rate, and corrected QT interval (QTc), specifically including: The P-wave duration is determined based on the time difference between the start and end of the P-wave. The PQ interval is determined based on the time difference between the onset of the P wave and the onset of the Q wave. The QRS group duration is determined based on the time difference between the Q wave initiation and the S wave termination. The QT interval is determined based on the time difference between the start of the Q wave and the end of the T wave. The RR interval is determined based on the time difference between two adjacent R-wave peaks. The heart rate is determined by a preset heart rate calculation formula based on the RR interval; and / or The corrected QT interval QTc is determined by a preset correction formula based on the QT interval and the RR interval.

7. The electrocardiogram waveform evaluation method according to claim 5, characterized in that, The step of determining the waveform features of each waveform component based on the feature points of each waveform component includes: The integrity status of each cardiac cycle is determined based on the boundary frames of the P wave, QRS complex, and T wave, or the presence of characteristic points of the P wave, QRS complex, and T wave in each cardiac cycle.

8. The electrocardiogram waveform evaluation method according to claim 7, characterized in that, The determination of the integrity status of each cardiac cycle based on the presence of characteristic points of the P wave, QRS complex, and T wave in each cardiac cycle includes: Determine whether there are missing coordinate sequences of characteristic points of P wave, QRS complex and T wave in each cardiac cycle in the electrocardiogram waveform image; When there are no missing beats, the corresponding cardiac cycle is determined to be a complete rhythm; When a missing component exists, the corresponding missing waveform component of the cardiac cycle is determined based on the coordinates of the missing feature point.

9. The electrocardiogram waveform evaluation method according to claim 5, characterized in that, The morphological features include the tilt characteristics of the ST segment of the electrocardiogram waveform; Wherein, determining the waveform features of each waveform component based on the feature points of each waveform component includes: Obtain the time and amplitude values ​​of the S-wave endpoint and the T-wave initiation; Calculate the slope of the line connecting the end of the S wave and the beginning of the T wave; and The shape of the ST segment is determined based on the slope of the connecting line. When the slope is zero, the ST segment is determined to be straight; when the slope is positive, the ST segment is determined to be upward sloping; and when the slope is negative, the ST segment is determined to be downward sloping.

10. The electrocardiogram waveform evaluation method according to claim 5, characterized in that, The morphological features include the directional features of the T wave in the electrocardiogram waveform; Wherein, determining the waveform features of each waveform component based on the feature points of each waveform component includes: Obtain the amplitude values ​​of the T-wave's initiation, peak, and end points; Compare the magnitude of the peak point amplitude with the magnitudes of the start and end points; and The T-wave pattern is determined based on the magnitude relationship described above. When the peak amplitude is greater than the amplitudes at the start and end points, it is determined to be an upright T-wave; when the peak amplitude is less than the amplitudes at the start and end points, it is determined to be an inverted T-wave.

11. The electrocardiogram waveform evaluation method according to claim 1, characterized in that, The generation of electrocardiogram waveform evaluation results based on the waveform characteristics of each waveform component includes: The waveform features are then converted into a first structured text description containing standard medical terminology. The first structured text description is input into the artificial intelligence text analysis system through a preset interface format; and The artificial intelligence text analysis system analyzes the first structured text description and outputs the electrocardiogram waveform evaluation result.

12. The electrocardiogram waveform evaluation method according to claim 1, characterized in that, The generation of electrocardiogram waveform evaluation results based on the waveform characteristics of each waveform component includes: The waveform features are then converted into a first structured text description containing standard medical terminology. Obtain user consultation information and convert it into a second structured text description; The first structured text description and the second structured text description are input into the artificial intelligence text analysis system through a preset interface format; The artificial intelligence text analysis system analyzes the first structured text description and the second structured text to output the electrocardiogram waveform evaluation result.

13. The electrocardiogram waveform evaluation method according to any one of claims 1 to 12, characterized in that, Also includes: Training the target recognition model specifically includes: Obtain multiple electrocardiogram training samples; Preprocessing the electrocardiogram training samples to obtain preprocessed training samples includes converting time-series electrocardiogram data into electrocardiogram waveform images; The waveform components in the electrocardiogram waveform images of the preprocessed training samples are labeled to generate a training dataset, wherein the labeling includes labeling the location regions of the P wave, QRS complex, ST segment, T wave and U wave; The training dataset is input into the target recognition model for training, and the model parameters are optimized through the backpropagation algorithm until the preset conditions are met.

14. An electrocardiogram waveform evaluation device, characterized in that, include: The recognition unit is configured to recognize each waveform component of multiple cardiac cycles in an electrocardiogram waveform image through a target recognition model, and generate bounding boxes corresponding to each waveform component of the multiple cardiac cycles. The acquisition unit is configured to acquire feature points of each waveform component based on the bounding boxes of each waveform component; The determining unit is configured to determine the waveform characteristics of each waveform component based on the feature points of each waveform component; as well as The evaluation unit is configured to generate an electrocardiogram waveform evaluation result based on the waveform characteristics of each waveform component, and the evaluation result reflects the cardiac function status corresponding to the electrocardiogram.

15. An electrocardiogram waveform evaluation system, characterized in that, include: The user terminal, AI feature extraction model, and AI analysis model are configured to interact with the AI ​​feature extraction model and AI analysis model and present evaluation results. The user terminal includes an electrocardiogram (ECG) acquisition module, which is configured to acquire ECG data. The AI ​​feature extraction model is configured to generate an electrocardiogram waveform image based on the electrocardiogram data, identify each waveform component of multiple cardiac cycles in the electrocardiogram waveform image, generate corresponding bounding boxes, extract feature points of each waveform component from the bounding boxes, determine the waveform features of each waveform component, and convert the waveform features into a first structured text description. The AI ​​analysis model is configured to receive a first structured text description, analyze and process it, generate an assessment result reflecting the cardiac function status corresponding to the electrocardiogram, and return the assessment result to the user terminal.

16. The electrocardiogram waveform evaluation system according to claim 15, characterized in that, The user terminal also includes a consultation module, configured to receive user consultation information and convert it into a second structured text description; The AI ​​analysis model is configured to receive a first structured text description and a second structured text description, analyze and process them, generate an assessment result reflecting the cardiac function status corresponding to the electrocardiogram, and return the assessment result to the user terminal.

17. An electronic device, characterized in that, include: A processor and a memory storing a computer program, the processor being configured to implement the method of any one of claims 1-13 when executing the computer program.

18. A program product, characterized in that, Includes a computer program, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1-13.