Deadly ventricular tachycardia detection system based on independent weight update

Abstract

The invention provides a fatal ventricular tachycardia detection system based on independent weight updating, and relates to the technical field of fatal arrhythmia detection.The fatal ventricular tachycardia detection system comprises a data acquisition module used for acquiring original heart sound data and preprocessing the original heart sound data to obtain signal fragments; the model reasoning module is used for inputting the signal frequency band into a multi-outlet neural network reasoning model, calculating the confidence coefficient of a prediction result of a current network outlet, judging whether the current network outlet is a final outlet or not according to the confidence coefficient, and if not, continuing to go to the next network outlet to obtain a prediction result with higher precision; wherein multiple layers of backbone networks are arranged in the multi-outlet neural network inference model, each layer of backbone network corresponds to one network outlet, the weight of each layer of backbone network is optimized and finely adjusted through an independent weight updating algorithm, and efficient personalized tachycardia detection of the multi-outlet neural network inference model is achieved. According to the system disclosed by the invention, the detection accuracy of the fatal ventricular tachycardia is improved.

Description

Technical Field

[0001] This disclosure relates to the field of lethal arrhythmia detection technology, specifically to a lethal ventricular tachycardia detection system based on independent weight updates. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] An implantable cardioverter-defibrillator (ICD) is a small medical device implanted in the body that improves the survival rate of patients with sudden cardiac arrest by detecting fatal ventricular arrhythmias (VA) and automatically delivering a defibrillation shock. Traditional VA detection algorithms rely on multiple criteria such as heart rate, detection intervals, rhythm stability, and initiation characteristics, but are prone to misjudgment and inappropriate discharge due to individual differences in intracardiac signal morphology. With the development of deep neural networks (DNNs), the accuracy of ventricular arrhythmia detection has been significantly improved. However, deep neural networks typically require a large amount of computational resources and storage space during inference, which conflicts with the low power consumption, low memory, and low computational requirements of implantable devices, making it difficult to directly deploy traditional deep models.

[0004] To address this issue, multi-exit neural networks (MENs) have enabled efficient inference for neural networks on implantable devices. This architecture allows models to have multiple classification exits at varying depths, adaptively selecting appropriate inference levels based on task complexity or input features, thereby significantly reducing computational overhead while maintaining accuracy. Most VA (Visual Analog Detection) tasks can be effectively handled by branches with fewer layers and smaller parameter sizes; only a few samples with complex morphologies or special features require deeper networks for analysis. Therefore, to balance computational efficiency and generalization ability across scenarios of varying complexity, MENs offer an effective way to achieve a dynamic balance between performance and resource consumption.

[0005] While multi-exit neural networks can significantly improve the inference efficiency of models and effectively adapt to the resource constraints of implantable devices, they still face the following limitations in practical applications: (1) For specific samples with special morphological characteristics and rare rhythmic patterns, it is often difficult to achieve accurate detection.

[0006] (2) When performing personalized fine-tuning on pre-trained models, existing solutions often rely on a small number of clinical samples to complete parameter calibration through backpropagation. However, different exits of multi-exit neural networks correspond to different layers of feature representations, which often leads to inconsistent gradient update directions, causing gradient conflict in the backbone network. This conflict can cause disordered parameter updates, which may not only reduce the effect of personalized fine-tuning, but also increase computational overhead. Summary of the Invention

[0007] To address the aforementioned issues, this disclosure proposes a lethal ventricular tachycardia detection system based on independent weight updates. Specifically, for multi-outlet neural networks in lethal ventricular tachycardia detection systems, the system improves prediction accuracy by controlling the independent weight fine-tuning mechanism of the neural network weights. This mitigates the impact of insufficient performance of pre-trained models on specific samples, resulting in a high-precision, low-latency model suitable for implantable devices.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions: A lethal ventricular tachycardia detection system based on independent weight updates includes: The data acquisition module is used to acquire raw heart sound data, preprocess it, and segment it into signal segments; The model inference module is used to input the signal frequency band into the multi-exit neural network inference model, calculate the confidence level of the prediction result of the current network exit, and determine whether the current network exit is the final exit based on the confidence level. Otherwise, it will continue to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0009] According to some embodiments, the present disclosure adopts the following technical solutions: An implantable device includes the aforementioned lethal ventricular tachycardia detection system based on independent weight updates.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program that, when executed by a processor, implements a detection method for the lethal ventricular tachycardia detection system based on independent weight updates, comprising the following steps: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the detection method of the lethal ventricular tachycardia detection system based on independent weight updates, comprising the following steps: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute a detection method for implementing the aforementioned lethal ventricular tachycardia detection system based on independent weight updates, the steps of which include: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0013] Compared with the prior art, the beneficial effects of this disclosure are as follows: This disclosed system for detecting lethal ventricular tachycardia based on independent weight updates addresses the inference process of a multi-exit neural network inference model within the system. By controlling the independent weight fine-tuning mechanism of the neural network weights, it improves the model's prediction accuracy, thereby mitigating the impact of insufficient performance of the pre-trained model on specific samples. This results in a high-precision, low-latency model suitable for implantable devices. While reducing computational complexity, it efficiently resolves gradient conflicts during multi-exit neural network training, achieving accurate and stable parameter updates.

[0014] This disclosure presents a lethal ventricular tachycardia detection system based on independent weight updates, achieving a lethal ventricular tachycardia detection method through personalized fine-tuning of a multi-exit neural network. In the case of a pre-trained multi-exit neural network, the entropy of the predicted label for each exit is used to measure its confidence level, enabling the model to determine the exit. Furthermore, this disclosure designs an independent weight fine-tuning method for the multi-exit neural network. By segmenting and updating each part of the multi-exit neural network separately to overcome gradient conflicts, an efficient personalized fine-tuning method for the multi-exit neural network is achieved, improving the accuracy of lethal ventricular tachycardia detection and obtaining a high-precision, low-latency multi-exit neural network. Attached Figure Description

[0015] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0016] Figure 1 This is a schematic flowchart of the lethal ventricular tachycardia detection method of the lethal ventricular tachycardia detection system based on independent weight update according to an embodiment of this disclosure; Figure 2 This is a schematic diagram of a lethal ventricular tachycardia detection method using a multi-outlet neural network inference model according to an embodiment of this disclosure; Figure 3 This is a schematic diagram of an independent weight update fine-tuning method for a multi-exit neural network inference model according to an embodiment of this disclosure. Detailed Implementation

[0017] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0018] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0019] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0020] Example 1 One embodiment of this disclosure provides a lethal ventricular tachycardia detection system based on independent weight updates. The system includes a data acquisition module and a model inference module. Specifically, the lethal ventricular tachycardia detection system based on independent weight updates includes: The data acquisition module is used to acquire raw heart sound data, preprocess it, and segment it into signal segments; The model inference module is used to input the signal frequency band into the multi-exit neural network inference model, calculate the confidence level of the prediction result of the current network exit, and determine whether the current network exit is the final exit based on the confidence level. Otherwise, it will continue to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0021] As one embodiment, a lethal ventricular tachycardia detection system based on independent weight updates detects lethal ventricular tachycardia. It achieves tachycardia detection by training a multi-exit neural network inference model and controls the process through an independent weight fine-tuning mechanism to improve the model's prediction accuracy. This mitigates the impact of insufficient performance of pre-trained models on specific samples, resulting in a high-precision, low-latency model suitable for implantable devices. The specific implementation process includes: Step 1: The data acquisition module acquires the raw heart sound data, preprocesses it, and segments it into signal segments; Specifically, the raw heart sound data came from 30 individuals collected using the SonR sensor with a Platinum Sorin ICD. A training dataset was constructed containing myocardial vibration signals reflecting heart sounds, with each individual's data segment ranging from 6 to 10 minutes, a sampling rate of 1000 Hz, and each record containing two labels (i.e., VA and non-VA) to describe its corresponding category.

[0022] First, the sampled raw heart sound data is preprocessed. Since the raw heart sound data contains noise, the low frequency needs to be filtered out. Considering that the high frequency component is related to the murmurs produced by abnormal heart sounds, pre-emphasis is used to highlight the high frequency murmur component information. The filtered data undergoes further processing to remove baseline offset; Finally, the preprocessed heart sound signal data is divided into segments with a set time period (2 seconds each) for use.

[0023] Step 2: Using the model inference module, input the signal frequency band into the multi-exit neural network inference model, calculate the confidence level of the prediction result of the current network exit, and determine whether the current network exit is the final exit based on the confidence level. Otherwise, continue to the next network exit to obtain a more accurate prediction result.

[0024] Specifically, a multi-exit neural network with multiple classification exits, trained using electrocardiogram signals for ventricular tachycardia detection, is selected as the pre-trained inference model. The backbone network of the multi-exit neural network inference model has L convolutional layers, and each backbone network corresponds to a fully connected layer exit block. Each exit block yields the probability of predicting the classification label, and the confidence score is calculated to determine whether to exit at the current exit and output the prediction result.

[0025] Furthermore, during the inference process, for any given input signal segment, the multi-exit neural network model measures its confidence at each exit based on the entropy of the predicted label. If the confidence exceeds the preset threshold of the current exit, the inference process terminates prematurely at that exit; otherwise, it continues to the next exit to obtain a more accurate prediction result.

[0026] As one example, for any given 2s input signal segment X, the multi-exit neural network model measures its confidence at each exit based on the entropy of the predicted label Y. The entropy is calculated as follows:

[0027] in, It is a vector of prediction results, containing the predicted probabilities of all possible class labels; It is any label for the prediction result. It represents the set of all possible labels.

[0028] If the confidence level exceeds the preset threshold of the current exit, the inference process terminates prematurely at that exit; otherwise, the model will continue to the next exit to obtain higher prediction accuracy.

[0029] In the training process of this multi-exit neural network inference model, to improve the model's adaptation performance to edge samples and thus enhance individual prediction accuracy, the specific implementation method is as follows: Assume the backbone of this neural network has L layers (i.e., ... Each layer of the backbone network is assigned a weight, and the weights of all backbone networks are denoted as follows: These correspond to L exits (i.e., The export weight of each export is denoted as . When performing independent weight updates on a multi-exit neural network inference model, the weights of the convolutional layers connected to each exit are decoupled into two independent parts: weights shared by the exits and weights specific to the exits. These two parts constitute the parameters of the backbone neural network corresponding to the current exit. The ratio coefficients of the two independent weights are set and the convolutional layers and their corresponding input feature maps are separated so that they can be used by the next layer and the corresponding exits respectively.

[0030] Each individual's data-label pair is divided into a training set and a test set, where the training set is used for personalized fine-tuning of the neural network, achieving efficient multi-exit model personalization while maintaining the cross-subject representation ability of the pre-trained model; Specifically, firstly, each exit The weights of the connected convolutional layers Decoupled into two independent parts:

[0031] in, These are the parameters of the backbone neural network corresponding to the current export. This is the number of output channels of the current convolutional layer in the backbone network. This is the number of input channels in the current convolutional layer of the backbone network. This is the kernel matrix of the convolutional layers of the current backbone network. Indicates export Shared weights, and Indicates export Specific weights. Parameters To determine the ratio, it can be preset based on experience. Usually, it is necessary to conduct repeated trials to obtain the optimal ratio for the current dataset. This ratio is used to segment between the convolutional layer and its corresponding input feature map, so that they can be used by the next layer and the corresponding output respectively.

[0032] Then, the gradient of the parameters of each layer of the backbone neural network is calculated:

[0033] in, The final exit on the training set The loss function for export shared weights The partial derivatives, For specific exits on the training set The loss function with export-specific weights Partial derivatives of exports, shared weights Based solely on the last exit of the multi-exit network Update the loss; while export-specific weights gradient Then it is updated only based on the inference loss of the target exit.

[0034] The loss function is calculated based on data-label pairs in the training set, using the cross-entropy loss function. In the classification problem of fatal ventricular tachycardia disclosed in this paper, the training data labels are: y Forecast results for each export Cross-entropy loss function Repeat the above steps for all data in the training set to obtain a model that has been individually fine-tuned.

[0035] Once the fine-tuning phase is complete, for the input heart sound signal segment X, the prediction result for each exit of the ventricular tachycardia (VA) detection based on the personalized multi-outlet network can be expressed as:

[0036] in, for and The parameters of the spliced ​​backbone neural network are used to predict the detection and classification results, including fatal ventricular arrhythmias and non-fatal ventricular arrhythmias.

[0037] Example 2 One embodiment of this disclosure provides a detection method for a lethal ventricular tachycardia detection system based on independent weight updates, comprising: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0038] Example 3 One embodiment of this disclosure provides an implantable device including the aforementioned lethal ventricular tachycardia detection system based on independent weight updates, and performs a method for detecting lethal ventricular tachycardia based on independent weight updates.

[0039] Example 4 One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the detection method of the lethal ventricular tachycardia detection system based on independent weight updates, comprising the following steps: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0040] Example 5 One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When executed by a processor, the computer instructions implement the detection method of the lethal ventricular tachycardia detection system based on independent weight updates, comprising the following steps: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0041] Example 6 One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute a detection method for implementing the aforementioned lethal ventricular tachycardia detection system based on independent weight updates, the steps of which include: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

[0042] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. 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, create a machine 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.

[0043] 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.

[0044] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A lethal ventricular tachycardia detection system based on independent weight updates, characterized in that, include: The data acquisition module is used to acquire raw heart sound data, preprocess it, and segment it into signal segments; The model inference module is used to input the signal frequency band into the multi-exit neural network inference model, calculate the confidence level of the prediction result of the current network exit, and determine whether the current network exit is the final exit based on the confidence level. Otherwise, it will continue to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

2. The lethal ventricular tachycardia detection system based on independent weight updates as described in claim 1, characterized in that, The raw heart sound data is collected using sensors and includes myocardial vibration signals with a sampling rate of 1000Hz. The raw heart sound data is preprocessed by filtering low frequencies, using pre-emphasis to highlight high-frequency noise components, removing baseline offset after filtering, and dividing the preprocessed heart sound signal data into signal segments for a set time period.

3. The lethal ventricular tachycardia detection system based on independent weight updates as described in claim 1, characterized in that, During the inference process of the multi-exit neural network inference model, for any given input signal segment, the multi-exit neural network model measures its confidence level at each exit based on the entropy of the predicted label. If the confidence level exceeds the preset threshold of the current exit, the inference process terminates prematurely at that exit; otherwise, it continues to the next exit to obtain a more accurate prediction result.

4. The lethal ventricular tachycardia detection system based on independent weight updates as described in claim 1, characterized in that, The backbone network of the multi-exit neural network inference model has L layers. Weights are set for each backbone layer, corresponding to L network exits. Each exit has its own exit weight. When the multi-exit neural network inference model performs independent weight updates, the weights of the convolutional layers connected to each exit are decoupled into two independent parts: weights shared by the exits and weights specific to the exits. These two parts constitute the parameters of the backbone neural network corresponding to the current exit. The ratio coefficients of the two independent weights are set and the convolutional layers and their corresponding input feature maps are separated so that they can be used by the next layer and the corresponding exit.

5. The lethal ventricular tachycardia detection system based on independent weight updates as described in claim 4, characterized in that, The parameter gradients of each layer of the backbone neural network are calculated as follows: in, The final exit on the training set The loss function shares weights with exports. The partial derivatives, For specific exits on the training set The loss function with export-specific weights Partial derivatives of exports, shared weights Based solely on the last exit of the multi-exit network Update the loss; while export-specific weights gradient Then it is updated only based on the inference loss of the target exit.

6. The lethal ventricular tachycardia detection system based on independent weight updates as described in claim 1, characterized in that, Once the fine-tuning phase is complete, for the input heart sound signal segment, the prediction result for each exit of the ventricular tachycardia detection based on the multi-exit neural network inference model can be expressed as: in, for and The parameters of the spliced ​​backbone neural network are used to predict the detection and classification results, including fatal ventricular arrhythmias and non-fatal ventricular arrhythmias.

7. An implantable device comprising a lethal ventricular tachycardia detection system based on independent weight updates as described in any one of claims 1-6.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the detection method of the lethal ventricular tachycardia detection system based on independent weight updates as described in any one of claims 1-6, the steps of which include: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the detection method of the lethal ventricular tachycardia detection system based on independent weight updates as described in any one of claims 1-6, comprising the following steps: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

10. An electronic device, characterized in that, include: The system comprises a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to perform a detection method for a lethal ventricular tachycardia detection system based on independent weight updates as described in any one of claims 1-6, the steps of which include: Raw heart sound data is acquired, preprocessed, and segmented to obtain signal segments; The signal frequency band is input into the multi-exit neural network inference model, and the confidence level of the prediction result of the current network exit is calculated. Based on the confidence level, it is determined whether the current network exit is the final exit. Otherwise, it continues to the next network exit to obtain a more accurate prediction result. The multi-exit neural network inference model is configured with multiple backbone networks, each corresponding to one network exit. The weights of each backbone network are optimized and fine-tuned through an independent weight update algorithm, thereby achieving efficient personalized tachycardia detection using the multi-exit neural network inference model.

Images