Trusted automated electrocardiogram interpretation method, device, equipment and storage medium

Abstract

The present invention relates to the technical field of electrocardiogram interpretation, and discloses a reliable automated electrocardiogram interpretation method, device, equipment and storage medium, including: inputting the electrocardiogram sample to be interpreted into a constructed multi-layer convolutional network to extract the electrocardiogram features; estimating the Dirichlet distribution parameters, determining the loss value and updating the parameters, and predicting the category probability of the electrocardiogram sample; based on the subjective logic framework, calculating the uncertainty estimation result of the category probability prediction result. The present invention constructs a multi-layer convolutional network combined with a shortcut link of maximum pooling, extracts high-level abstract features from ECG data and estimates the parameters of the Dirichlet distribution, introduces the overall uncertainty in the subjective logic evaluation classification, so that the prediction probability obtained using the Dirichlet distribution parameters is not affected by the amplification effect, and considering the overall uncertainty helps to avoid the situation of low confidence but high prediction probability (such as random guessing), reducing the risk of overconfidence.

Description

Technical Field

[0001] The present invention relates to the technical field of electrocardiogram interpretation, and in particular to a reliable automated electrocardiogram interpretation method, device, equipment and storage medium. Background Art

[0002] An electrocardiogram (ECG) records the physiological activity of the heart over time, providing important insights into heart health. Accurate interpretation of ECG data is critical for the diagnosis of certain diseases. Accurate interpretation of ECG signals requires the analytical skills and careful attention to detail of a cardiologist. However, these skills vary widely between cardiologists, and as the volume of ECG readings increases, it becomes difficult for a single physician to fully process them. Automated interpretation of ECG data offers a viable solution to overcome these problems.

[0003] Deep learning models, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) and their variants, have achieved promising results in the application of ECG automated interpretation due to their excellent data learning capabilities and adaptable processing architectures. Currently, deep learning models for ECG interpretation usually consist of one or more feature extraction layers, ending with a softmax layer. The softmax layer assigns probabilities to predefined categories, and the sum of these probabilities is equal to 1. These probabilities are usually interpreted as the confidence of the model's prediction of the corresponding category.

[0004] However, existing research often overlooks a key issue: the credibility of the explanations provided by deep learning models. For human clinicians, a diagnosis will only be made when there is a high degree of confidence in the interpretation of the ECG; otherwise, they may reserve judgment. In contrast, computational models generate predictions for each instance regardless of the confidence level. When faced with ambiguous or challenging cases, these models may occasionally make the right choice, but they are unsure of the accuracy of their decisions. Such results are not only unhelpful to clinicians, but may even introduce interference and further complicate clinical decision-making. In addition, softmax is known to amplify the probability of predicted categories, which means that confidence is exaggerated. This can lead to highly confident but wrong predictions, which can mislead clinicians and even have catastrophic consequences for patients. For such samples amplified by softmax, human doctors tend to give higher uncertainties, but current automated parsing methods have difficulty making such estimates.

[0005] Therefore, how to accurately estimate the uncertainty of the target ECG data and avoid overconfidence in the output interpretation is a technical problem that needs to be solved urgently. Summary of the invention

[0006] The present invention provides a reliable automated electrocardiogram interpretation method, device, equipment and storage medium, aiming to solve at least one of the above-mentioned technical problems.

[0007] To achieve the above object, the present invention provides a reliable automated electrocardiogram interpretation method based on deep learning, comprising the following steps:

[0008] Constructing a multi-layer convolutional network, inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extracting the electrocardiogram features of the electrocardiogram sample;

[0009] Using the electrocardiogram features, estimating Dirichlet distribution parameters;

[0010] According to the Dirichlet distribution parameters, the loss value is determined and the parameters are updated, and the category probability prediction is performed on the electrocardiogram samples to obtain the category probability prediction result;

[0011] Based on the subjective logic framework, the uncertainty estimation results of the category probability prediction results are calculated.

[0012] Optionally, construct a multi-layer convolutional network, including:

[0013] Constructing a first convolutional layer module and m second convolutional layer modules, and connecting the first convolutional layer module and the m second convolutional layer modules end to end in sequence;

[0014] A Conv-BN-LeakyReLU structure is set before the first convolutional layer module, and a BN-LeakyReLU structure is set after the m second convolutional layer modules to form a multi-layer convolutional network.

[0015] Optionally, the first convolutional layer module specifically includes: a Conv-BN-LeakyReLU structure, a Dropout layer, a Conv layer connected in sequence end to end, and a Max-pool layer connecting the input end of the Conv-BN-LeakyReLU structure and the output end of the Conv layer.

[0016] Optionally, the second convolutional layer module specifically includes: a BN-ReLu structure, a Conv-BN-LeakyReLU structure, a Dropout layer, a Conv layer, and a Max-pool layer connecting the input end of the BN-ReLu structure and the output end of the Conv layer.

[0017] Optionally, in a preferred embodiment, the step of inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network to extract the electrocardiogram features of the electrocardiogram sample specifically includes:

[0018] The ECG sample to be interpreted Input the multi-layer convolutional network constructed by double downsampling the input every other Max-pool layer;

[0019] Based on the multi-layer convolutional network Max-pool layers and Conv layers to extract ECG samples ECG features in ;

[0020] ; Where w represents the filter width of the Conv layer, and n represents the number of filters in the Conv layer.

[0021] Optionally, the step of estimating Dirichlet distribution parameters using the electrocardiogram features specifically includes:

[0022] ECG characteristics Perform shape transformation and activate, the expression is:

[0023] ; ; where σ(·) represents the activation function LeakyReLU, Indicates the number of preset target heart disease types, represents the shape transformation parameters;

[0024] Using activated ECG features Estimate the Dirichlet distribution parameters, expressed as:

[0025] ;

[0026] in, is for Dirichlet distribution parameters without misleading estimates, express The corresponding data labels are represents the Hadamard product.

[0027] Optionally, according to the Dirichlet distribution parameter, determining the loss value and performing a parameter updating step specifically includes:

[0028] According to the Dirichlet distribution parameters, the loss value is determined, and the expression is:

[0029] ;

[0030] ;

[0031] ;

[0032] in, (·)express Regularization, Indicates an updateable parameter. represents the loss value, and Represents the balance between KL divergence loss and The regularization hyperparameters, represents the digamma function, represents a vector containing K 1s, represents the Dirichlet uniform distribution, represents the gamma function, Indicates that the i-th sample belongs to the j-th category, represents the unit simplex corresponding to the i-th sample, represents the Dirichlet parameter estimate of the i-th sample in category j, represents the probability distribution of sample i, represents the Dirichlet parameter estimate for sample i in class k without misleading estimates.

[0033] Optionally, according to the Dirichlet distribution parameters, a loss value is determined and a parameter updating step is performed, specifically including: using the Adam algorithm to update the parameters with a learning rate of 0.001.

[0034] Optionally, based on the subjective logic framework, the step of calculating the uncertainty estimation result of the category probability prediction result specifically includes:

[0035] Based on the calculation formula of the subjective logic framework, the uncertainty estimation result of the category probability prediction result is calculated, and the expression is:

[0036] ;

[0037] in, Represents the uncertainty estimation result.

[0038] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a reliable automated electrocardiogram interpretation device, comprising:

[0039] A feature extraction module, used to construct a multi-layer convolutional network, input the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extract the electrocardiogram features of the electrocardiogram sample;

[0040] A parameter estimation module, used to estimate the parameters of the Dirichlet distribution using the electrocardiogram features;

[0041] A loss value determination module, used to determine the loss value and update the parameters according to the Dirichlet distribution parameters, perform category probability prediction on the electrocardiogram samples, and obtain category probability prediction results;

[0042] The uncertainty estimation module is used to calculate the uncertainty estimation results of the category probability prediction results based on the subjective logic framework.

[0043] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a trusted automated ECG interpretation device, which includes: a memory, a processor, and a trusted automated ECG interpretation program stored on the memory and executable on the processor, and the trusted automated ECG interpretation program, when executed by the processor, implements the steps of the trusted automated ECG interpretation method as described above.

[0044] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a storage medium, on which a trusted automated ECG interpretation program is stored, and when the trusted automated ECG interpretation program is executed by a processor, the steps of the above-mentioned trusted automated ECG interpretation method are implemented.

[0045] The beneficial effects of the present invention are: a reliable automated electrocardiogram interpretation method, device, equipment and storage medium are proposed, the method comprising: constructing a multi-layer convolutional network, inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, extracting the electrocardiogram features of the electrocardiogram sample; using the electrocardiogram features, estimating the Dirichlet distribution parameters; determining the loss value and updating the parameters according to the Dirichlet distribution parameters, predicting the category probability of the electrocardiogram sample, and obtaining the category probability prediction result; based on the subjective logic framework, calculating the uncertainty estimation result of the category probability prediction result. The present invention extracts high-level abstract features from the original ECG data by constructing a multi-layer convolutional network combined with a shortcut link of maximum pooling, and estimates the parameters of the Dirichlet distribution according to the extracted features. Finally, subjective logic is introduced to evaluate the overall uncertainty in classification, which has the following two significant advantages: first, the prediction probability obtained based on the Dirichlet distribution parameters is not affected by the amplification effect; second, the overall uncertainty helps to avoid the situation of low confidence but high prediction probability (such as random guessing). Therefore, the risk of overconfidence is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention;

[0047] Figure 2 A flowchart of a reliable automated electrocardiogram interpretation method is provided;

[0048] Figure 3 The algorithm framework diagram of the trusted automated ECG interpretation method;

[0049] Figure 4 Schematic diagram of the prediction performance on experimental data under different uncertainty thresholds

[0050] Figure 5 It is a schematic diagram of the structure of a reliable automated electrocardiogram interpretation device.

[0051] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0054] like Figure 1 As shown, Figure 1 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0055] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0056] Those skilled in the art will understand that Figure 1 The structure of the device shown in the figure does not constitute a limitation of the device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0057] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a trusted automated electrocardiogram interpretation program.

[0058] exist Figure 1In the terminal shown, the network interface 1004 is mainly used to connect to the backend server and perform data communication with the backend server; the user interface 1003 is mainly used to connect to the client (user end) and perform data communication with the client; and the processor 1001 can be used to call the trusted automated electrocardiogram interpretation program stored in the memory 1005 and perform the following operations:

[0059] Constructing a multi-layer convolutional network, inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extracting the electrocardiogram features of the electrocardiogram sample;

[0060] Using the electrocardiogram features, estimating Dirichlet distribution parameters;

[0061] According to the Dirichlet distribution parameters, the loss value is determined and the parameters are updated, and the category probability prediction is performed on the electrocardiogram samples to obtain the category probability prediction result;

[0062] Based on the subjective logic framework, the uncertainty estimation results of the category probability prediction results are calculated.

[0063] The specific embodiments of the present invention applied to the device are basically the same as the embodiments of the following application of the reliable automated electrocardiogram interpretation method, and will not be described in detail here.

[0064] The embodiment of the present invention provides a reliable automated electrocardiogram interpretation method, referring to Figure 2 , Figure 2 Schematic diagram of the process of an embodiment of a reliable automated electrocardiogram interpretation method of the present invention.

[0065] In this embodiment, the reliable automated electrocardiogram interpretation method comprises the following steps:

[0066] S100: constructing a multi-layer convolutional network, inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extracting the electrocardiogram features of the electrocardiogram sample;

[0067] S200: using the electrocardiogram features, estimating Dirichlet distribution parameters;

[0068] S300: According to the Dirichlet distribution parameters, determine the loss value and update the parameters, perform category probability prediction on the electrocardiogram samples, and obtain category probability prediction results;

[0069] S400: Based on the subjective logic framework, the uncertainty estimation result of the category probability prediction result is calculated.

[0070] It should be noted that existing studies often overlook a key issue: the credibility of the explanations provided by deep learning models. For human clinicians, a diagnosis will only be made when there is a high degree of confidence in the interpretation of the ECG; otherwise, they may reserve their judgment. In contrast, computational models generate predictions for each instance regardless of the confidence level. When faced with ambiguous or challenging cases, these models may occasionally make the right choice, but they are unsure of the accuracy of their decisions. Such results are not only unhelpful to clinicians, but may even introduce interference and further complicate clinical decision-making. In addition, softmax is known to amplify the probability of predicted categories, which means that confidence is exaggerated. This can lead to highly confident but wrong predictions, misleading clinicians and even having catastrophic consequences for patients.

[0071] In order to solve the above problems, this embodiment extracts high-level abstract features from the original ECG data by constructing a multi-layer convolutional network combined with a shortcut link of maximum pooling, and estimates the parameters of the Dirichlet distribution based on the extracted features. Finally, the introduction of subjective logic to evaluate the overall uncertainty in classification has the following two significant advantages: first, the prediction probability based on the parameters of the Dirichlet distribution is not affected by the amplification effect; second, the overall uncertainty helps to avoid situations with low confidence but high prediction probability (such as random guessing). Therefore, the risk of overconfidence is reduced.

[0072] In a preferred embodiment, the step of constructing a multi-layer convolutional network specifically includes:

[0073] S110: constructing a first convolutional layer module and m second convolutional layer modules, and connecting the first convolutional layer module and the m second convolutional layer modules end to end in sequence;

[0074] S120: Setting a Conv-BN-LeakyReLU structure before the first convolutional layer module, and setting a BN-LeakyReLU structure after the m second convolutional layer modules to form a multi-layer convolutional network.

[0075] Furthermore, the first convolutional layer module specifically includes: a Conv-BN-LeakyReLU structure, a Dropout layer, a Conv layer connected end to end, and a Max-pool layer connecting the input end of the Conv-BN-LeakyReLU structure and the output end of the Conv layer.

[0076] Furthermore, the second convolutional layer module specifically includes: a BN-ReLu structure, a Conv-BN-LeakyReLU structure, a Dropout layer, a Conv layer, and a Max-pool layer connecting the input end of the BN-ReLu structure and the output end of the Conv layer. In a preferred embodiment, the electrocardiogram sample to be interpreted is input into the constructed multi-layer convolutional network, and the electrocardiogram feature step of the electrocardiogram sample is extracted, which specifically includes:

[0077] S130: The electrocardiogram sample to be interpreted Input the multi-layer convolutional network constructed by double downsampling the input every other Max-pool layer;

[0078] S140: Based on the multi-layer convolutional network Max-pool layers and Conv layers to extract ECG samples ECG features in ;

[0079] ;

[0080] Among them, w represents the filter width of the Conv layer, and n represents the number of filters in the Conv layer.

[0081] In a preferred embodiment, the step of estimating the parameters of the Dirichlet distribution using the electrocardiogram features specifically includes:

[0082] S210: ECG characteristics Perform shape transformation and activate, the expression is:

[0083] ;

[0084] ;

[0085] Among them, σ(·) represents the activation function LeakyReLU, Indicates the number of preset target heart disease types, represents the shape transformation parameters;

[0086] S220: Using activated ECG features Estimate the Dirichlet distribution parameters, expressed as:

[0087] ;

[0088] in, is for Dirichlet distribution parameters without misleading estimates, express The corresponding data labels are represents the Hadamard product.

[0089] In a preferred embodiment, according to the Dirichlet distribution parameters, determining the loss value and performing the parameter updating step specifically includes:

[0090] S310: Determine the loss value according to the Dirichlet distribution parameter, expressed as:

[0091] ;

[0092] ;

[0093] ;

[0094] in, (·)express Regularization, Indicates an updateable parameter. represents the loss value, and Represents the balance between KL divergence loss and The regularization hyperparameters, represents the digamma function, represents a vector containing K 1s, represents the Dirichlet uniform distribution, represents the gamma function, Indicates that the i-th sample belongs to the j-th category, represents the unit simplex corresponding to the i-th sample, represents the Dirichlet parameter estimate of the i-th sample in category j, represents the probability distribution of sample i, represents the Dirichlet parameter estimate for sample i in class k without misleading estimates.

[0095] In a preferred embodiment, according to the Dirichlet distribution parameters, the loss value is determined and the parameter updating step is performed, which specifically includes: using the Adam algorithm to update the parameters with a learning rate of 0.001.

[0096] In a preferred embodiment, based on the subjective logic framework, the step of calculating the uncertainty estimation result of the category probability prediction result specifically includes:

[0097] S410: Based on the calculation formula of the subjective logic framework, the uncertainty estimation result of the category probability prediction result is calculated and obtained, and the expression is:

[0098] ;

[0099] in, Represents the uncertainty estimation result.

[0100] To further explain the present invention, a specific example of the reliable automated electrocardiogram interpretation method based on deep learning of the present invention is provided below.

[0101] This specific example proposes a reliable automated ECG interpretation method based on deep learning. The method first extracts high-level abstract features from the raw ECG data through a multi-layer convolutional network combined with max-pooling shortcut links. Then, it estimates the parameters of the Dirichlet Distribution based on the extracted features. Finally, Subjective Logic is introduced to evaluate the overall uncertainty in the classification. This method brings two significant advantages: first, the prediction probability based on the parameters of the Dirichlet distribution is not affected by the amplification effect; second, the overall uncertainty helps to avoid situations with low confidence but high prediction probability (such as random guessing). Therefore, the risk of overconfidence is reduced. Figure 3 It is the overall architecture diagram of the present invention.

[0102] like Figure 3 As shown in the figure, each convolutional layer (Conv) is followed by batch normalization (BN), leaky rectified linear unit (LeakyReLU) activation function and dropout (the first and last ones have no dropout). Each max-pooling shortcut link (Max-pool) connects the computational modules before the two convolutional layers to the modules after them. With this structure, the depth of the model can be adjusted according to the computing power and data complexity. However, the ratio between the two computational modules of the max-pooling shortcut link and the convolutional layer should be kept at ( ≥1). When When >2, part of the structure will be repeated.

[0103] For clarity, in Figure 4 The structure is represented by a "repeating module", the number of which is ,and ≥2. Assume that the filter width of the convolutional layer is , the number of filters is , every other shortcut link downsamples the input by a factor of two. is a hyperparameter, and the padding is set to valid. Given the input and has 2 +1 convolutional layer and A network of shortcut links, we will obtain the corresponding representation ,in Then, the present invention is Perform shape transformation and activate σ(·) represents the activation function LeakyReLU, , is the number of types of preset target heart diseases. Assume that the current electrocardiogram sample Corresponding data labels is a one-hot vector, i.e. , there are N data samples in total, and the present invention calculates the loss value of the interpretation algorithm according to the following formula: For parameter update:

[0104] ;

[0105] (·)yes Regularization, Θ is an updateable parameter, and Is to balance the KL divergence loss and The regularization hyperparameters. , is the digamma function; , is for Dirichlet prediction parameters without misleading estimates, is a vector containing K 1s, is the Dirichlet uniform distribution, is the gamma function, is the digamma function, represents the Hadamard product, Indicates that the i-th sample belongs to the j-th category, represents the unit simplex corresponding to the i-th sample, represents the Dirichlet parameter estimate of the i-th sample in category j, represents the probability distribution of sample i, represents the Dirichlet parameter estimate for sample i in class k without misleading estimates.

[0106] In addition, in order to avoid the model focusing too much on the KL divergence loss in the early stage of training, which may lead to insufficient exploration of the parameter space and produce an almost flat uniform distribution, the present invention gradually increases The parameter update is performed using the Adam algorithm, and the learning rate is set to 0.001. Then, according to the calculation formula of subjective logic:

[0107] Calculate for Uncertainty estimate .

[0108] In order to verify the beneficial effects of the present invention, the experimental verification results of the reliable automated electrocardiogram interpretation method based on deep learning of the present invention are provided below.

[0109] The present invention is experimentally verified on real electrocardiogram data PhysioNet Challenge 2017 databases. The verification is evaluated by five indicators: accuracy (ACC), F1 score (F1), PR curve area (PR-AUC), ROC curve area (ROC-AUC) and overconfidence (OC). The verification data is unified as samples with a duration of 30 seconds and a sampling rate of 300Hz (the input data points are 9000) as input, and the hyperparameters of the algorithm in the present invention are , , , , They are set to 7, 15, [12, 12, 12, 12, 24, 24, 24, 24, 48, 48, 48, 96, 96], 1, and 0.001 respectively. When the other parameters are randomly initialized, the model is trained for 300 epochs. The above calculation process will obtain the verification results shown in Table 1.

[0110] ACC F1 PR-AUC ROC-AUC OC 86.12 83.14 85.27 92.87 0.59 Table 1. Performance of the present invention on experimental data

[0111] This result shows that the confidence of the present invention is significantly reliable (ideally, OC will be 0). In addition, it is worth noting that when the present invention refuses to interpret samples that exceed a certain uncertainty, the absolute interpretation performance indicators (ACC, F1, PR-AUC and ROC-AUC) will be improved, reaching up to ACC 93.59%, F1 93.22%, PR-AUC 89.17% and ROC-AUC95.1%. Figure 4 Shown is how the absolute prediction performance (measured by ACC, F1, PR-AUC, and ROC-AUC) changes when the present invention rejects explanations exceeding different uncertainty thresholds. Figure 4 In , the horizontal axis represents the uncertainty probability of the judged sample, and the vertical axis represents the prediction accuracy of the sample used for model performance.

[0112] Reference Figure 5 , Figure 5 Schematic diagram of the structure of a reliable automated electrocardiogram interpretation device in an embodiment of the present invention.

[0113] like Figure 5As shown, the reliable automated electrocardiogram interpretation device proposed in the embodiment of the present invention includes:

[0114] The feature extraction module 10 is used to construct a multi-layer convolutional network, input the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extract the electrocardiogram features of the electrocardiogram sample;

[0115] A parameter estimation module 20, for estimating Dirichlet distribution parameters using the electrocardiogram features;

[0116] A loss value determination module 30, for determining a loss value and updating the parameters according to the Dirichlet distribution parameters, performing a category probability prediction on the electrocardiogram sample, and obtaining a category probability prediction result;

[0117] The uncertainty estimation module 40 is used to calculate the uncertainty estimation result of the category probability prediction result based on the subjective logic framework.

[0118] Other embodiments or specific implementations of the reliable automated electrocardiogram interpretation device of the present invention can refer to the above-mentioned method embodiments, which will not be described in detail here.

[0119] In addition, the present invention also proposes a trusted automated ECG interpretation device, which includes: a memory, a processor, and a trusted automated ECG interpretation program stored on the memory and executable on the processor, wherein the trusted automated ECG interpretation program implements the steps of the trusted automated ECG interpretation method as described above when executed by the processor.

[0120] The specific implementation of the trusted automated ECG interpretation device of the present application is basically the same as the above-mentioned trusted automated ECG interpretation method embodiments, and will not be repeated here.

[0121] In addition, the present invention also proposes a readable storage medium, the readable storage medium includes a computer readable storage medium, on which a reliable automated electrocardiogram interpretation program is stored. The readable storage medium may be Figure 1 The memory 1005 in the terminal may also be at least one of a ROM (Read-Only Memory) / RAM (Random Access Memory), a magnetic disk, and an optical disk. The readable storage medium includes a number of instructions for enabling a trusted automated electrocardiogram interpretation device having a processor to execute the trusted automated electrocardiogram interpretation method described in each embodiment of the present invention.

[0122] The specific implementation methods in the readable storage medium of the present application are basically the same as the embodiments of the above-mentioned trusted automated electrocardiogram interpretation method, and will not be repeated here.

[0123] It is understood that, in the description of this specification, the description with reference to the terms "one embodiment", "another embodiment", "other embodiments", or "first to Nth embodiments" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0124] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0125] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0126] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0127] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A reliable automated electrocardiogram interpretation method, characterized in that: The following steps are involved: Constructing a multi-layer convolutional network, inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extracting the electrocardiogram features of the electrocardiogram sample; Using the electrocardiogram features, estimating Dirichlet distribution parameters; specifically including: ECG characteristics i ∈R d×1 Perform shape transformation and activate, the expression is: a i =σ(W·r i ); W∈R K×d ; Where σ(·) represents the activation function LeakyReLU, K represents the number of preset target heart diseases, and W represents the shape transformation parameter; Using the activated ECG feature r i ∈R d×1 Estimate the Dirichlet distribution parameters, expressed as: in, For x i Dirichlet distribution parameter without misleading estimates, y i Represents x i The corresponding data labels,⊙,represent the Hadamard product; According to the Dirichlet distribution parameters, the loss value is determined and the parameters are updated, and the category probability prediction is performed on the electrocardiogram samples to obtain the category probability prediction result; According to the Dirichlet distribution parameters, the loss value is determined, and the expression is: Where L2(·) represents L2 regularization, Θ represents an updateable parameter, represents the loss value, λ t and λ represent the hyperparameters for balancing KL divergence loss and L2 regularization, ψ(·) represents the double gamma function, 1 represents a vector containing K 1s, Dir(p i |1) represents the Dirichlet uniform distribution, γ(·) represents the gamma function, and y ij Indicates that the i-th sample belongs to the j-th category, S i represents the unit simplex corresponding to the i-th sample, α ij represents the Dirichlet parameter estimate of the i-th sample in category j, p i represents the probability distribution of sample i, represents the Dirichlet parameter estimate for sample i in category k after removing misleading estimates; Based on the subjective logic framework, the uncertainty estimation results of the category probability prediction results are calculated.

2. The reliable automated electrocardiogram interpretation method according to claim 1, characterized in that: The steps to build a multi-layer convolutional network include: Constructing a first convolutional layer module and m second convolutional layer modules, and connecting the first convolutional layer module and the m second convolutional layer modules end to end in sequence; A Conv-BN-LeakyReLU structure is set before the first convolutional layer module, and a BN-LeakyReLU structure is set after the m second convolutional layer modules to form a multi-layer convolutional network; The first convolutional layer module specifically includes: a Conv-BN-LeakyReLU structure, a Dropout layer, a Conv layer, and a Max-pool layer connecting the input end of the Conv-BN-LeakyReLU structure and the output end of the Conv layer. Among them, the second convolutional layer module specifically includes: a BN-ReLu structure, a Conv-BN-LeakyReLU structure, a Dropout layer, a Conv layer, and a Max-pool layer connecting the input end of the BN-ReLu structure and the output end of the Conv layer.

3. The reliable automated electrocardiogram interpretation method according to claim 1, characterized in that: Inputting the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network to extract the electrocardiogram features of the electrocardiogram sample specifically includes: The ECG sample to be interpreted is x i ∈R n×1 Input the multi-layer convolutional network constructed by double downsampling the input every other Max-pool layer; Based on the l Max-pool layers and 2l+1 Conv layers in the multi-layer convolutional network, the electrocardiogram sample x is extracted. i ∈R n×1 ECG features in r i ∈R d×1 ; Among them, w represents the filter width of the Conv layer, and n represents the number of filters in the Conv layer.

4. The reliable automated electrocardiogram interpretation method according to claim 1, characterized in that: According to the Dirichlet distribution parameters, the loss value is determined and a parameter updating step is performed, specifically including: using the Adam algorithm to update the parameters with a learning rate of 0.

001.

5. The reliable automated electrocardiogram interpretation method according to claim 4, characterized in that: Based on the subjective logic framework, the steps for calculating the uncertainty estimation results of the category probability prediction results include: Based on the calculation formula of the subjective logic framework, the uncertainty estimation result of the category probability prediction result is calculated, and the expression is: Where u represents the uncertainty estimation result.

6. A reliable automated electrocardiogram interpretation device, characterized in that: A reliable automated electrocardiogram interpretation method as claimed in any one of claims 1 to 5, comprising: A feature extraction module, used to construct a multi-layer convolutional network, input the electrocardiogram sample to be interpreted into the constructed multi-layer convolutional network, and extract the electrocardiogram features of the electrocardiogram sample; A parameter estimation module, used to estimate the parameters of the Dirichlet distribution using the electrocardiogram features; A loss value determination module, used to determine the loss value and update the parameters according to the Dirichlet distribution parameters, perform category probability prediction on the electrocardiogram samples, and obtain category probability prediction results; The uncertainty estimation module is used to calculate the uncertainty estimation results of the category probability prediction results based on the subjective logic framework.

7. A reliable automated electrocardiogram interpretation device, characterized in that: The trusted automated ECG interpretation device comprises: a memory, a processor, and a trusted automated ECG interpretation program stored in the memory and executable on the processor. When the trusted automated ECG interpretation program is executed by the processor, the steps of the trusted automated ECG interpretation method as described in any one of claims 1 to 5 are implemented.

8. A storage medium, characterized in that: The storage medium stores a trusted automated electrocardiogram interpretation program, which, when executed by a processor, implements the steps of the trusted automated electrocardiogram interpretation method according to any one of claims 1 to 5.

Images