An early warning system based on AI sensing technology

Abstract

The present application relates to the medical technical field, especially to an early warning system based on AI sensing technology, which comprises an AI electrocardio sensing and collecting module, an AI preprocessing module, an electrocardio trend modeling module, an AI deduction module, a signal verification module and a hierarchical early warning pushing module. The present application aims to provide an early warning system based on AI sensing technology to solve the problems of early electrocardio signal confusion and misjudgment and early warning lag.

Description

Technical Field

[0001] This invention relates to the field of medical technology, and in particular to an early warning system based on AI sensing technology. Background Technology

[0002] Among all physiological signals monitored at home, electrocardiogram (ECG) signals are the most easily acquired, most widely available, have the lowest operational threshold, can be continuously collected, and possess the highest information density. Currently, devices such as smartwatches, smart bracelets, flexible ECG chest patches, and portable home ECG monitors are widely available. Users can independently complete long-term, continuous, and non-invasive ECG signal collection without the guidance of professional medical personnel. No invasive operation or complex calibration is required, making them suitable for the home use needs of ordinary users of all ages. Furthermore, as a core indicator reflecting the functional state of the cardiovascular system, the waveform morphology, rhythm characteristics, and dynamic evolution trends of ECG signals are not only directly related to early pathological changes in cardiovascular diseases but can also indirectly reflect early changes in various diseases such as systemic infections, metabolic disorders, organ dysfunction, and autonomic nervous system dysfunction. Therefore, it is the most feasible and universally applicable monitoring indicator for early disease warning in home settings.

[0003] However, at the same time, electrocardiogram (ECG) signals are also the physiological signals most prone to waveform confusion, misdiagnosis, and misjudgment in the early stages of disease. Early pathological changes in the disease only cause weak, non-specific changes in ECG signals. These changes highly overlap with ECG waveform changes caused by non-pathological factors such as physiological interference, changes in body position, exercise, emotional fluctuations, sleep state, and environmental factors. In addition, many different etiologies, such as early myocardial ischemia, early heart failure, paroxysmal arrhythmia, electrolyte imbalance, autonomic nervous system dysfunction, and lung disease, can all exhibit highly similar ECG waveform characteristics in the early stages of the disease.

[0004] Existing AI ECG analysis algorithms mostly focus on the post-hoc diagnosis of existing organic lesions and clear arrhythmias, lacking the ability to model the evolution of ECG history over a long period of time, and are unable to predict future pathological trends. At the same time, existing algorithms cannot solve the problems of easy confusion and high misjudgment rate of early ECG signals through cross-validation of multi-source signals.

[0005] Therefore, how to design an early warning system based on AI sensing technology that can solve the above-mentioned technical problems is a technical problem that needs to be solved. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to provide an early warning system based on AI sensing technology to solve the issues of easy confusion and misjudgment of early electrocardiogram signals and delayed early warning.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: including an AI ECG sensing and acquisition module, an AI preprocessing module, an ECG trend modeling module, an AI inference module, a signal verification module, and a graded early warning push module; The AI ​​ECG sensor acquisition module is used to continuously acquire raw ECG signals from the human body surface, synchronously acquire timestamps, and mark the acquisition status. The AI ​​preprocessing module is used to preprocess the raw electrocardiogram signal; The ECG trend modeling module is used to extract the time domain, frequency domain, and nonlinear core features of ECG signals, and construct an ECG historical trend feature matrix covering short-term, medium-term, and long-term time scales. The AI ​​prediction module, combined with the historical trend feature matrix, predicts the trend of electrocardiogram within a preset time window in the future, and simultaneously outputs the probability of occurrence of each trend and the corresponding pathological risk indication. The signal verification module is used to access the auxiliary lift signal that is synchronously acquired from the same source as the electrocardiogram signal, and to cross-verify the output results of the AI ​​inference module. The tiered early warning push module is used to trigger corresponding level of early warning operation based on the risk level and trend probability after cross-validation.

[0008] Furthermore, the AI ​​preprocessing module specifically includes the following steps: Step S11: Perform sliding frame processing on the raw ECG signal; Step S12: Input the framed signal into the lightweight TinyCNN noise reduction network to filter out power line interference, electromyographic noise and high-frequency noise in the original signal; Step S13: Adaptive LMS filtering algorithm is adopted, using the acquired synchronous volume signal as reference noise, and the filtering weights are iteratively updated through the least mean square algorithm. The specific formula is as follows: Where e(n) is the final preprocessed ECG signal, d(n) is the ECG signal after preliminary noise reduction, x(n) is the reference body motion noise signal, and w(n) is the iterative weight coefficient; Step S14: The amplitude of the preprocessed ECG signal is normalized using a standardization method to output a standard ECG signal.

[0009] Furthermore, the ECG trend modeling module specifically includes the following steps: Step S21: Using the Pan-Tompkins algorithm combined with a lightweight binary classification CNN network, the start and end points of the QRS complex, P wave, T wave, and ST segment of the standardized ECG signal are located, wherein the lightweight binary classification network is used to distinguish between real QRS complexes and artifacts. Step S22: Based on the localized feature points, extract the core ECG features and input them into a two-layer Bi-GRU network. The forward GRU captures the positive dependencies of the ECG time series, and the backward GRU captures the negative dependencies of the ECG time series. Output the deep time series feature vector. Step S23: Construct a three-level time sliding window, namely a short-term window, a medium-term window, and a long-term window. Input the deep temporal feature vector within each window into a TCN temporal convolutional network with an attention mechanism. Through causal convolution, capture the trend evolution patterns at different time scales. The attention layer automatically weights the trend features with high contribution, and finally outputs a historical trend feature matrix covering multiple time scales. , where T is the time step and F is the feature dimension.

[0010] Furthermore, the AI ​​inference module specifically includes the following steps: Step S31: Input the historical trend feature matrix E into the Bi-GRU encoder of the Seq2Seq architecture to generate the context vector C. Step S32: Set up 4 parallel GRU decoder branches, each branch corresponding to a future ECG trend. Using the context vector C as input, deduce the ECG trend features at the next three time points. The trend directions corresponding to the 4 branches are as follows: Branch 1: Normal and stable trend, ECG characteristics remain within the normal physiological range, with no pathological abnormality tendency; Branch 2: Slow deterioration trend, with progressive pathological deviations in electrocardiogram characteristics; Branch 3: Rapid deterioration trend, with sudden abnormal fluctuations in electrocardiogram characteristics; Branch 4: Non-specific fluctuation trend, transient abnormalities in electrocardiogram characteristics, with no clear tendency for pathological progression; Step S33: Using the Softmax activation function, output the probability P of the occurrence of the four trends at the next three time points. , where P1+P2+P3+P4=1; Step S34: Based on trend probability and pathological indication, complete the initial risk classification.

[0011] Furthermore, the auxiliary signals accessed by the signal verification module specifically include respiratory wave signals, blood oxygen saturation signals, triaxial body motion signals, and skin conductance signals that are synchronously acquired from the same source as the electrocardiogram signals.

[0012] Furthermore, the signal verification module specifically includes: fusing the features of each auxiliary signal with the ECG historical trend feature matrix, inputting it into a pre-trained XGBoost classifier, cross-validating the initial risk classification and trend pathological indication, correcting the trend probability, and filtering false positive warnings caused by non-pathological interference.

[0013] The present invention has the following beneficial effects: 1- This invention uses an electrocardiogram trend modeling module and an AI inference module to fully capture the progressive pathological evolution of electrocardiogram signals through multi-timescale historical trend modeling. At the same time, it uses a multi-branch Seq2Seq architecture to complete the inference of future multi-path trends, which can identify the early pathological tendencies of diseases in advance.

[0014] 2- This invention accurately distinguishes between organic pathological changes, secondary changes, and physiological interference in electrocardiogram (ECG) abnormalities through an auxiliary signal verification module, effectively solving the problems of high overlap, easy confusion, and multiple types of misjudgments in ECG signal waveforms in the early stages of disease. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: See Figure 1 As shown, the solution includes an AI ECG sensor acquisition module, an AI preprocessing module, an ECG trend modeling module, an AI inference module, a signal verification module, and a graded early warning push module. The AI ​​ECG sensor acquisition module is used to continuously acquire raw ECG signals from the human body surface, synchronously acquire timestamps, and mark the acquisition status. The AI ​​preprocessing module is used to preprocess the raw electrocardiogram signal; The ECG trend modeling module is used to extract the time domain, frequency domain, and nonlinear core features of ECG signals, and construct an ECG historical trend feature matrix covering short-term, medium-term, and long-term time scales. The AI ​​prediction module, combined with the historical trend feature matrix, predicts the trend of electrocardiogram within a preset time window in the future, and simultaneously outputs the probability of occurrence of each trend and the corresponding pathological risk indication. The signal verification module is used to access the auxiliary lift signal that is synchronously acquired from the same source as the electrocardiogram signal, and to cross-verify the output results of the AI ​​inference module. The tiered early warning push module is used to trigger corresponding level of early warning operation based on the risk level and trend probability after cross-validation.

[0017] Furthermore, the AI ​​preprocessing module specifically includes the following steps: Step S11: Perform sliding frame processing on the raw ECG signal; Step S12: Input the framed signal into the lightweight TinyCNN noise reduction network to filter out power line interference, electromyographic noise and high-frequency noise in the original signal; Step S13: Adaptive LMS filtering algorithm is adopted, using the acquired synchronous volume signal as reference noise, and the filtering weights are iteratively updated through the least mean square algorithm. The specific formula is as follows: Where e(n) is the final preprocessed ECG signal, d(n) is the ECG signal after preliminary noise reduction, x(n) is the reference body motion noise signal, and w(n) is the iterative weight coefficient; Step S14: The amplitude of the preprocessed ECG signal is normalized using a standardization method to output a standard ECG signal.

[0018] Furthermore, the ECG trend modeling module specifically includes the following steps: Step S21: Using the Pan-Tompkins algorithm combined with a lightweight binary classification CNN network, the start and end points of the QRS complex, P wave, T wave, and ST segment of the standardized ECG signal are located, wherein the lightweight binary classification network is used to distinguish between real QRS complexes and artifacts. Step S22: Based on the localized feature points, extract the core ECG features and input them into a two-layer Bi-GRU network. The forward GRU captures the positive dependencies of the ECG time series, and the backward GRU captures the negative dependencies of the ECG time series. Output the deep time series feature vector. Step S23: Construct a three-level time sliding window, namely a short-term window, a medium-term window, and a long-term window. Input the deep temporal feature vector within each window into a TCN temporal convolutional network with an attention mechanism. Through causal convolution, capture the trend evolution patterns at different time scales. The attention layer automatically weights the trend features with high contribution, and finally outputs a historical trend feature matrix covering multiple time scales. , where T is the time step and F is the feature dimension.

[0019] Furthermore, the AI ​​inference module specifically includes the following steps: Step S31: Input the historical trend feature matrix E into the Bi-GRU encoder of the Seq2Seq architecture to generate the context vector C. Step S32: Set up 4 parallel GRU decoder branches, each branch corresponding to a future ECG trend. Using the context vector C as input, deduce the ECG trend features at the next three time points. The trend directions corresponding to the 4 branches are as follows: Branch 1: Normal and stable trend, ECG characteristics remain within the normal physiological range, with no pathological abnormality tendency; Branch 2: Slow deterioration trend, with progressive pathological deviations in electrocardiogram characteristics; Branch 3: Rapid deterioration trend, with sudden abnormal fluctuations in electrocardiogram characteristics; Branch 4: Non-specific fluctuation trend, transient abnormalities in electrocardiogram characteristics, with no clear tendency for pathological progression; Step S33: Using the Softmax activation function, output the probability P of the occurrence of the four trends at the next three time points. , where P1+P2+P3+P4=1; Step S34: Based on trend probability and pathological indications, complete the initial risk classification. Specifically, low risk: only generate a daily health report on the user's app, indicating the ECG trend status, without proactive warning push; medium risk: the app pushes real-time warning prompts, explaining the abnormal ECG trend and possible causes, and recommends completing an ECG re-examination and rest adjustment within 24 hours; high risk: simultaneously push warnings to the user's app, preset emergency contacts, and contracted family doctor, simultaneously push continuous ECG data and trend analysis reports, and recommend seeking medical verification within 4 hours; very high risk: automatically trigger an emergency medical call, simultaneously push the user's location, continuous ECG data, and risk analysis report to the nearest medical institution's chest pain center, and simultaneously notify the emergency contact.

[0020] Furthermore, the auxiliary signals accessed by the signal verification module specifically include respiratory wave signals, blood oxygen saturation signals, triaxial body motion signals, and skin conductance signals that are synchronously acquired from the same source as the electrocardiogram signals.

[0021] Furthermore, the signal verification module specifically includes: fusing the features of each auxiliary signal with the ECG historical trend feature matrix, inputting it into a pre-trained XGBoost classifier, cross-validating the initial risk classification and trend pathological indication, correcting the trend probability, and filtering false positive warnings caused by non-pathological interference.

[0022] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0023] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0024] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0025] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0026] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An early warning system based on AI sensing technology, characterized in that: It includes an AI ECG sensor acquisition module, an AI preprocessing module, an ECG trend modeling module, an AI inference module, a signal verification module, and a graded early warning push module; The AI ​​ECG sensor acquisition module is used to continuously acquire raw ECG signals from the human body surface, synchronously acquire timestamps, and mark the acquisition status. The AI ​​preprocessing module is used to preprocess the raw electrocardiogram signal; The ECG trend modeling module is used to extract the time domain, frequency domain, and nonlinear core features of ECG signals, and construct an ECG historical trend feature matrix covering short-term, medium-term, and long-term time scales. The AI ​​prediction module, combined with the historical trend feature matrix, predicts the trend of electrocardiogram within a preset time window in the future, and simultaneously outputs the probability of occurrence of each trend and the corresponding pathological risk indication. The signal verification module is used to access the auxiliary lift signal that is synchronously acquired from the same source as the electrocardiogram signal, and to cross-verify the output results of the AI ​​inference module. The tiered early warning push module is used to trigger corresponding level of early warning operation based on the risk level and trend probability after cross-validation.

2. The early warning system based on AI sensing technology according to claim 1, characterized in that: The AI ​​preprocessing module specifically includes the following steps: Step S11: Perform sliding frame processing on the raw ECG signal; Step S12: Input the framed signal into the lightweight TinyCNN noise reduction network to filter out power line interference, electromyographic noise and high-frequency noise in the original signal; Step S13: Adaptive LMS filtering algorithm is adopted, using the acquired synchronous volume signal as reference noise, and the filtering weights are iteratively updated through the least mean square algorithm. The specific formula is as follows: ； Where e(n) is the final preprocessed ECG signal, d(n) is the ECG signal after preliminary noise reduction, x(n) is the reference body motion noise signal, and w(n) is the iterative weight coefficient; Step S14: The amplitude of the preprocessed ECG signal is normalized using a standardization method to output a standard ECG signal.

3. The early warning system based on AI sensing technology according to claim 1, characterized in that: The electrocardiogram trend modeling module specifically includes the following steps: Step S21: Using the Pan-Tompkins algorithm combined with a lightweight binary classification CNN network, the start and end points of the QRS complex, P wave, T wave, and ST segment of the standardized ECG signal are located, wherein the lightweight binary classification network is used to distinguish between real QRS complexes and artifacts. Step S22: Based on the localized feature points, extract the core ECG features and input them into a two-layer Bi-GRU network. The forward GRU captures the positive dependencies of the ECG time series, and the backward GRU captures the negative dependencies of the ECG time series. Output the deep time series feature vector. Step S23: Construct a three-level time sliding window, namely a short-term window, a medium-term window, and a long-term window. Input the deep temporal feature vector within each window into a TCN temporal convolutional network with an attention mechanism. Through causal convolution, capture the trend evolution patterns at different time scales. The attention layer automatically weights the trend features with high contribution, and finally outputs a historical trend feature matrix covering multiple time scales. , where T is the time step and F is the feature dimension.

4. The early warning system based on AI sensing technology according to claim 1, characterized in that: The AI ​​inference module specifically includes the following steps: Step S31: Input the historical trend feature matrix E into the Bi-GRU encoder of the Seq2Seq architecture to generate the context vector C. Step S32: Set up 4 parallel GRU decoder branches, each branch corresponding to a future ECG trend. Using the context vector C as input, deduce the ECG trend features at the next three time points. The trend directions corresponding to the 4 branches are as follows: Branch 1: Normal and stable trend, ECG characteristics remain within the normal physiological range, with no pathological abnormality tendency; Branch 2: Slow deterioration trend, with progressive pathological deviations in electrocardiogram characteristics; Branch 3: Rapid deterioration trend, with sudden abnormal fluctuations in electrocardiogram characteristics; Branch 4: Non-specific fluctuation trend, transient abnormalities in electrocardiogram characteristics, with no clear tendency for pathological progression; Step S33: Using the Softmax activation function, output the probability P of the occurrence of the four trends at the next three time points. , where P1+P2+P3+P4=1; Step S34: Based on trend probability and pathological indication, complete the initial risk classification.

5. The early warning system based on AI sensing technology according to claim 1, characterized in that: The signal verification module receives auxiliary signals including respiratory wave signals, blood oxygen saturation signals, triaxial body motion signals, and skin conductance signals, which are collected synchronously with the electrocardiogram signals.

6. The early warning system based on AI sensing technology according to claim 5, characterized in that: The signal verification module specifically includes: fusing the features of each auxiliary signal with the ECG historical trend feature matrix, inputting it into a pre-trained XGBoost classifier, cross-validating the initial risk classification and trend pathological indication, correcting the trend probability, and filtering false positive warnings caused by non-pathological interference.

Images