An AI-based ECG diagnostic model system

Abstract

The present invention discloses an AI-based ECG diagnostic model system, comprising: an ECG data uploading module for uploading first ECG data to a data preprocessing module; a data preprocessing module for preprocessing the first ECG data; a first identification and detection module for extracting and detecting features from the first ECG data and outputting a first anomaly detection result; a sampling frequency acquisition module for outputting the first ECG data to a signal reconstruction module when the first ECG data has no anomaly and the sampling frequency is not greater than the standard sampling frequency; a signal reconstruction module for performing signal reconstruction based on the first ECG data and restoring its waveform details; a second identification and detection module for extracting and detecting features from the reconstructed first ECG data and outputting a second anomaly detection result; and a conclusion uploading module for sending an anomaly detection reference report to a doctor client. The present invention improves the recognition accuracy of low-sampling-frequency ECGs.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence technology, and in particular to an AI-based electrocardiogram (ECG) diagnostic model system. Background Art

[0002] ECG diagnosis is an important medical method for assessing cardiac function and diagnosing heart disease by recording and analyzing the heart's electrical activity. The core tool for ECG diagnosis is the electrocardiogram (ECG / EKG), which uses electrodes on the body's surface to capture the weak electrical signals generated by the heartbeat and converts them into visual waveforms. These waveforms reflect the rhythm, frequency, and conduction pathways of the heart's electrical activity.

[0003] Early approaches relied on physician experience, which led to efficiency and subjectivity issues. With the rise of artificial intelligence (AI), researchers began exploring its application in ECG analysis. With the advancement of AI, particularly the application of convolutional neural networks (CNNs), AI can automatically extract features from massive amounts of data, build fully automated diagnostic models, and significantly improve diagnostic accuracy. AI ECG diagnosis can be applied to various ECG devices, but some devices, due to their low sampling frequency, can suffer from the following issues: 1. Loss of high-frequency features: blurred P wave and QRS complex details, and distortion of ST segment slopes. 2. Insufficient temporal accuracy: increased errors in QT and RR interval measurement. Loss of high-frequency features during AI diagnosis can easily affect the diagnosis of conditions such as myocardial ischemia and atrial fibrillation, while insufficient temporal accuracy can complicate the analysis of long QT syndrome and heart rate variability. Consequently, AI ECG diagnosis can experience reduced recognition accuracy in ECGs with low sampling frequencies. Summary of the Invention

[0004] In view of some of the above-mentioned defects in the prior art, the technical problem to be solved by the present invention is to provide an AI-based ECG diagnosis model system, aiming to improve the recognition accuracy of low sampling frequency ECGs.

[0005] To achieve the above objectives, the present invention provides an AI-based ECG diagnostic model system, which includes: an ECG data uploading module, a data preprocessing module, a first recognition and detection module, a sampling frequency acquisition module, a signal reconstruction module, a second recognition and detection module, and a conclusion uploading module;

[0006] The ECG data uploading module is configured to upload the first ECG data to the data preprocessing module; wherein the first ECG data at least includes a first electrocardiogram;

[0007] The data preprocessing module is configured to perform denoising filtering and signal segmentation on the first ECG data to eliminate baseline drift and power frequency interference, and to standardize the first ECG data;

[0008] The first recognition and detection module is configured to extract features from the preprocessed first electrocardiogram data, perform anomaly detection based on the extracted features, and output a first anomaly detection result; wherein the first anomaly detection result includes no anomaly and various anomaly conclusions and their corresponding confidence levels;

[0009] The sampling frequency obtaining module is configured to obtain a first sampling frequency corresponding to the first ECG data in response to the first abnormality detection result being no abnormality; determine whether the first sampling frequency is greater than a standard sampling frequency, and if so, output the first ECG data as normal; if not, output the first ECG data to the signal reconstruction module;

[0010] The signal reconstruction module is configured to perform signal reconstruction on the first ECG data based on the first ECG data, increase the sampling frequency of the first ECG data, and restore waveform details corresponding to the first ECG data; wherein the signal reconstruction module is trained using multiple sets of corresponding ECG data, the corresponding ECG data including ECG data collected by the same user at a second sampling frequency and a third sampling frequency during the same time period, respectively, under ECG equipment with the second sampling frequency being greater than the standard sampling frequency, and the third sampling frequency being less than the standard sampling frequency;

[0011] a second recognition and detection module, configured to extract features from the first electrocardiogram data after signal reconstruction, perform anomaly detection based on the extracted features, and output a second anomaly detection result;

[0012] The conclusion uploading module is used to send the abnormality detection reference report to the doctor client and collect and save the doctor's opinions and treatment suggestions on the abnormality detection reference report; wherein, the abnormality detection reference report includes the first abnormality detection result or the second abnormality detection result.

[0013] Optionally, the signal reconstruction module is specifically used to: perform frequency interpolation on the first ECG data based on the first ECG data so as to increase the corresponding sampling frequency of the first ECG data; the training process of the signal reconstruction module includes: inputting multiple groups of the corresponding ECG data into the first training model, so that the first training model obtains the correspondence between the ECG data less than the standard sampling frequency and the ECG data greater than the standard sampling frequency in the same group of the corresponding ECG data, and then obtains the signal reconstruction module, and the signal reconstruction module performs frequency interpolation on the input ECG data less than the standard sampling frequency according to the correspondence, and outputs the corresponding ECG data greater than the standard sampling frequency.

[0014] Optionally, the system further comprises: a report generating module;

[0015] The report generating module is configured to generate a comprehensive report based on the abnormality detection reference report and the doctor's opinion and treatment suggestion on the abnormality detection reference report.

[0016] Optionally, the data preprocessing module is specifically used to:

[0017] The first electrocardiogram data is subjected to denoising filtering by using wavelet transform or adaptive filter to eliminate baseline drift and power frequency interference; the heart beats are segmented according to the peak value of the R wave, and the first electrocardiogram data is standardized to a uniform length.

[0018] Optionally, the first identification and detection module is specifically configured to:

[0019] A convolutional neural network is used to extract the time-frequency domain features of the first electrocardiogram data; a long short-term memory network model or a Transformer model is used to obtain the first abnormality detection result based on the time-frequency domain features.

[0020] Optionally, the first identification and detection module and the second identification and detection module are a common module.

[0021] Optionally, the training process of the data preprocessing module includes:

[0022] Noise characteristics of different ECG devices are mixed to generate first training data, and the data preprocessing module is trained using the first training data to improve the robustness of the data preprocessing module in preprocessing the first ECG data.

[0023] The beneficial effects of the present invention are as follows: 1. The signal reconstruction module of the present invention is trained with multiple sets of corresponding ECG data and can reconstruct the low-sampling frequency ECG data based on the corresponding relationship between low-sampling frequency ECG data and high-sampling frequency data. Therefore, when an anomaly in the low-sampling frequency ECG data cannot be detected, the signal reconstruction module is used to reconstruct the ECG data to increase its sampling frequency and restore the corresponding waveform details for further detection. This, with the help of artificial intelligence, avoids the problems of high-frequency feature loss and insufficient time precision caused by the low sampling frequency, effectively improving the accuracy of anomaly detection. 2. The present invention performs signal reconstruction only when an anomaly in the low-sampling frequency ECG data cannot be detected. Compared to performing signal reconstruction on the low-sampling frequency ECG data from the outset, the present invention can effectively save computing resources. 3. The present invention sends the anomaly detection reference report to the doctor client to obtain the doctor's opinion and treatment recommendations on the anomaly detection reference report, forming an effective interaction with the expert, and then analyzing the first ECG data from multiple perspectives, avoiding the problem of insufficient detection accuracy caused by single-level analysis.

[0024] In summary, the present invention can effectively improve the recognition accuracy of low sampling frequency electrocardiograms. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a structural diagram of an AI-based electrocardiogram diagnostic model system provided by a specific embodiment of the present invention.

[0026] Figure 2 This is a business flow chart of an AI-based electrocardiogram diagnostic model system during application, provided by a specific embodiment of the present invention. DETAILED DESCRIPTION

[0027] The present invention discloses an AI-based electrocardiogram diagnostic model system. Those skilled in the art can refer to the content of this article and appropriately improve the technical details. It should be noted in particular that all similar replacements and modifications are obvious to those skilled in the art and are considered to be included in the present invention. The device and application of the present invention have been described through preferred embodiments. Relevant personnel can obviously modify or appropriately change and combine the device and application described herein without departing from the content, spirit and scope of the present invention to implement and apply the technology of the present invention.

[0028] The applicant's research found that traditional electrocardiogram diagnosis mainly relies on the experience of doctors, which has problems such as low efficiency, strong subjective judgment, and easy misdiagnosis in complex cases. Most of the current artificial intelligence-assisted systems are limited to single-level analysis and lack effective interaction with expert consultation. In addition, AI electrocardiogram diagnosis can be applied to different electrocardiogram devices, but some devices have the following problems due to low sampling frequency: 1. Loss of high-frequency features: blurred details of P waves and QRS waves, and distortion of ST segment slope. 2. Insufficient time accuracy: increased measurement errors of QT interval and RR interval. In the AI ​​diagnosis process, the loss of high-frequency features can easily affect the diagnosis of myocardial ischemia, atrial fibrillation, etc., while insufficient time accuracy can easily lead to difficulties in analyzing long QT syndrome and heart rate variability. Therefore, the recognition accuracy of AI electrocardiogram diagnosis will decrease in electrocardiograms corresponding to low sampling frequencies.

[0029] Therefore, the embodiment of the present invention provides an AI-based ECG diagnostic model system, such as Figure 1 The ECG diagnosis model system shown in the figure includes: an ECG data uploading module 101, a data preprocessing module 102, a first recognition and detection module 103, a sampling frequency acquisition module 104, a signal reconstruction module 105, a second recognition and detection module 106 and a conclusion uploading module 107;

[0030] The ECG data uploading module 101 is configured to upload first ECG data to the data preprocessing module 102 ; wherein the first ECG data at least includes a first electrocardiogram;

[0031] The data preprocessing module 102 is configured to perform denoising filtering and signal segmentation on the first ECG data to eliminate baseline drift and power frequency interference, and to standardize the first ECG data;

[0032] The first recognition and detection module 103 is configured to extract features from the preprocessed first ECG data, perform anomaly detection based on the extracted features, and output a first anomaly detection result; wherein the first anomaly detection result includes no anomaly and various anomaly conclusions and their corresponding confidence levels;

[0033] The sampling frequency obtaining module 104 is configured to obtain a first sampling frequency corresponding to the first ECG data in response to the first abnormality detection result being no abnormality; determine whether the first sampling frequency is greater than a standard sampling frequency, and if so, output the first ECG data as normal; otherwise, output the first ECG data to the signal reconstruction module 105;

[0034] a signal reconstruction module 105 for performing signal reconstruction on the first ECG data based on the first ECG data, increasing the sampling frequency of the first ECG data, and restoring waveform details corresponding to the first ECG data; wherein the signal reconstruction module 105 is trained using multiple sets of corresponding ECG data, the corresponding ECG data including ECG data collected from the same user at a second sampling frequency and a third sampling frequency during a same time period, respectively, using ECG equipment with the second sampling frequency being greater than a standard sampling frequency, and the third sampling frequency being less than the standard sampling frequency;

[0035] The second recognition and detection module 106 is used to extract features from the first ECG data after signal reconstruction, perform anomaly detection based on the extracted features, and output a second anomaly detection result;

[0036] The conclusion uploading module 107 is used to send the abnormality detection reference report to the doctor client and collect and save the doctor's opinions and treatment suggestions on the abnormality detection reference report; wherein, the abnormality detection reference report includes the first abnormality detection result or the second abnormality detection result.

[0037] It should be noted that the embodiments of the present invention utilize a recognition-reconstruction-rerecognition model to effectively identify low-sampling-rate ECG data, ensuring accurate recognition while conserving computing power. Compared to existing technologies, the embodiments of the present invention can identify more cardiac issues from low-sampling-rate ECG data through signal reconstruction, improving detection accuracy and avoiding missed or misdiagnoses.

[0038] In this specific embodiment, multiple devices are compatible (such as a 12-lead electrocardiograph). The system of the embodiment of the present invention is connected to each electrocardiograph device, and the electrocardiograph device automatically uploads the collected data to the cloud platform to ensure data security.

[0039] In this specific embodiment, the signal reconstruction module 105 is specifically used to: perform frequency interpolation on the first ECG data based on the first ECG data so as to increase the corresponding sampling frequency of the first ECG data; the training process of the signal reconstruction module 105 includes: inputting multiple groups of corresponding ECG data into the first training model so that the first training model obtains the correspondence between the ECG data with a frequency less than the standard sampling frequency and the ECG data with a frequency greater than the standard sampling frequency in the same group of corresponding ECG data, and then obtains the signal reconstruction module 105. The signal reconstruction module 105 performs frequency interpolation on the input ECG data with a frequency less than the standard sampling frequency according to the correspondence, and outputs the corresponding ECG data with a frequency greater than the standard sampling frequency.

[0040] It should be noted that machine learning effectively captures the relationship between high- and low-sampling-frequency ECG data under the same cardiac conditions. This model then enables the reconstruction of low-sampling-frequency ECG data into high-sampling-frequency ECG data. Furthermore, ECG data collected by the same user at the second and third sampling frequencies during the same time period was used to minimize variation, avoid interference from other factors, and maximize the relationship.

[0041] In this specific embodiment, the system further includes: a report generation module;

[0042] The report generation module is used to generate a comprehensive report based on the abnormality detection reference report and the doctor's opinions and treatment suggestions on the abnormality detection reference report.

[0043] It should be noted that the integrated AI diagnosis, expert opinions and treatment recommendations support PDF format.

[0044] In this specific embodiment, the data preprocessing module 102 is specifically configured to:

[0045] The first ECG data is subjected to denoising filtering by using wavelet transform or adaptive filter to eliminate baseline drift and power frequency interference; the heart beats are segmented according to the peak value of the R wave, and the first ECG data is standardized to a uniform length.

[0046] Furthermore, the embodiment of the present invention first identifies which ECG device the ECG data comes from, and then pre-processes the data in a corresponding manner according to the type of ECG device to ensure that the pre-processed data is accurate and reliable.

[0047] It should be noted that the newly collected ECG data generally contains noise. In order to prevent the noise from interfering with subsequent detection, data preprocessing effectively solves this problem.

[0048] In this specific embodiment, the first identification and detection module is specifically configured to:

[0049] A convolutional neural network is used to extract the time-frequency domain features of the first electrocardiogram data; a long short-term memory network model or a Transformer model is used to obtain a first anomaly detection result based on the time-frequency domain features.

[0050] It should be noted that the AI ​​diagnostic analysis corresponding to the first recognition and detection module includes:

[0051] Feature extraction: Convolutional neural network (CNN) is used to extract time-frequency domain features.

[0052] Anomaly detection: Identify arrhythmias (such as atrial fibrillation and premature ventricular contractions) based on long short-term memory (LSTM) or Transformer models.

[0053] Confidence score: Outputs anomaly detection reference report and confidence level (threshold is adjustable, default ≥90% for high confidence).

[0054] In this specific embodiment, the first recognition and detection module 103 and the second recognition and detection module 106 are common modules.

[0055] It should be noted that identical modules can reduce system redundancy.

[0056] In this specific embodiment, the training process of the data preprocessing module 102 includes:

[0057] Noise characteristics of different ECG devices are mixed to generate first training data, and the data preprocessing module 102 is trained using the first training data to improve the robustness of the data preprocessing module 102 in preprocessing the first ECG data.

[0058] It should be noted that the above training method can improve the performance of data preprocessing and obtain more accurate ECG data.

[0059] In specific applications, the business process diagram can be Figure 2 As shown, it includes: patient side, hospital side, doctor side, and AI side.

[0060] Figure 2The flowchart shows in detail each link from data collection to report generation, ensuring that each step is accurate and correct. First, the patient's ECG data is collected by professional equipment, and then the data is automatically uploaded to the cloud platform for secure storage. In the cloud, the data preprocessing step begins, including denoising filtering and signal segmentation to optimize data quality and lay a solid foundation for subsequent analysis. Next, the corresponding module of AI diagnostic analysis intervenes to accurately identify ECG abnormalities such as arrhythmias through feature extraction, anomaly detection and confidence scoring. Once the diagnosis is completed, the system automatically generates a comprehensive report containing AI anomaly detection reference report, expert opinions and treatment recommendations and distributes it to the patient. The entire process is scientifically designed and efficient, which fully reflects the application value of the embodiment of the present invention in the ECG consultation scenario.

[0061] The signal reconstruction module 105 of the embodiment of the present invention is trained with multiple sets of corresponding ECG data and can reconstruct the low-sampling-frequency ECG data based on the corresponding relationship between the low-sampling-frequency ECG data and the high-sampling-frequency ECG data. Therefore, when an anomaly in the low-sampling-frequency ECG data cannot be detected, the signal reconstruction module 105 is used to reconstruct the ECG data to increase its sampling frequency and restore the corresponding waveform details for further detection. This utilizes artificial intelligence to avoid problems such as loss of high-frequency features and insufficient time precision caused by low sampling frequencies, effectively improving the accuracy of anomaly detection.

[0062] By first detecting and then reconstructing the low-sampling-frequency ECG data if no abnormality is detected, the embodiment of the present invention can effectively save computing resources compared to reconstructing the low-sampling-frequency ECG data from the outset.

[0063] The embodiment of the present invention sends the abnormality detection reference report to the doctor client to obtain the doctor's opinion and treatment suggestions on the abnormality detection reference report, form effective interaction with the expert, and then analyze the first electrocardiogram data from multiple aspects to avoid the problem of insufficient detection accuracy caused by single-level analysis.

[0064] In summary, the embodiments of the present invention can effectively improve the recognition accuracy of low sampling frequency electrocardiograms.

[0065] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0066] Each embodiment in this specification is described in a related manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so the description is relatively simple. For related parts, refer to the description of the method embodiment.

[0067] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. An AI-based electrocardiogram diagnostic model system, characterized in that: The ECG diagnosis model system includes: an ECG data uploading module, a data preprocessing module, a first identification and detection module, a sampling frequency acquisition module, a signal reconstruction module, a second identification and detection module, and a conclusion uploading module; The ECG data uploading module is configured to upload the first ECG data to the data preprocessing module; wherein the first ECG data at least includes a first electrocardiogram; The data preprocessing module is configured to perform denoising filtering and signal segmentation on the first ECG data to eliminate baseline drift and power frequency interference, and to standardize the first ECG data; The first recognition and detection module is configured to extract features from the preprocessed first electrocardiogram data, perform anomaly detection based on the extracted features, and output a first anomaly detection result; wherein the first anomaly detection result includes no anomaly and various anomaly conclusions and their corresponding confidence levels; The sampling frequency obtaining module is configured to obtain a first sampling frequency corresponding to the first ECG data in response to the first abnormality detection result being no abnormality; determine whether the first sampling frequency is greater than a standard sampling frequency, and if so, output the first ECG data as normal; if not, output the first ECG data to the signal reconstruction module; The signal reconstruction module is configured to perform signal reconstruction on the first ECG data based on the first ECG data, increase the sampling frequency of the first ECG data, and restore waveform details corresponding to the first ECG data; wherein the signal reconstruction module is trained using multiple sets of corresponding ECG data, the corresponding ECG data including ECG data collected by the same user at a second sampling frequency and a third sampling frequency during the same time period, respectively, under ECG equipment with the second sampling frequency being greater than the standard sampling frequency, and the third sampling frequency being less than the standard sampling frequency; a second recognition and detection module, configured to extract features from the first electrocardiogram data after signal reconstruction, perform anomaly detection based on the extracted features, and output a second anomaly detection result; The conclusion uploading module is used to send the abnormality detection reference report to the doctor client and collect and save the doctor's opinions and treatment suggestions on the abnormality detection reference report; wherein the abnormality detection reference report includes the first abnormality detection result or the second abnormality detection result; Among them, the signal reconstruction module is specifically used to: perform frequency interpolation on the first ECG data according to the first ECG data so as to increase the corresponding sampling frequency of the first ECG data; the training process of the signal reconstruction module includes: inputting multiple groups of the corresponding ECG data into the first training model, so that the first training model obtains the correspondence between the ECG data less than the standard sampling frequency and the ECG data greater than the standard sampling frequency in the same group of the corresponding ECG data, and then obtains the signal reconstruction module. The signal reconstruction module performs frequency interpolation on the input ECG data less than the standard sampling frequency according to the correspondence, and outputs the corresponding ECG data greater than the standard sampling frequency.

2. The AI-based ECG diagnostic model system according to claim 1, characterized in that: The system further comprises: a report generating module; The report generating module is configured to generate a comprehensive report based on the abnormality detection reference report and the doctor's opinion and treatment suggestion on the abnormality detection reference report.

3. The AI-based ECG diagnostic model system according to claim 1, characterized in that: The data preprocessing module is specifically used to: The first electrocardiogram data is subjected to denoising filtering by using wavelet transform or adaptive filter to eliminate baseline drift and power frequency interference; the heart beats are segmented according to the peak value of the R wave, and the first electrocardiogram data is standardized to a uniform length.

4. The AI-based ECG diagnostic model system according to claim 1, characterized in that: The first identification and detection module is specifically configured to: A convolutional neural network is used to extract the time-frequency domain features of the first electrocardiogram data; a long short-term memory network model or a Transformer model is used to obtain the first abnormality detection result based on the time-frequency domain features.

5. The AI-based ECG diagnostic model system according to claim 1, characterized in that: The first identification and detection module and the second identification and detection module are common modules.

6. The AI-based ECG diagnostic model system according to claim 1, characterized in that: The training process of the data preprocessing module includes: Noise characteristics of different ECG devices are mixed to generate first training data, and the data preprocessing module is trained using the first training data to improve the robustness of the data preprocessing module in preprocessing the first ECG data.

Images