A deep learning-based illness warning method and system

Abstract

The application provides a disease early warning method and system based on deep learning, and relates to the technical field of medical data analysis. The method comprises the following steps: acquiring an electrocardiogram signal of a patient; performing denoising processing on the electrocardiogram signal to obtain a denoised electrocardiogram signal; performing data enhancement processing on the denoised electrocardiogram signal through an SMOTE algorithm to obtain an enhanced electrocardiogram signal; constructing a disease prediction classification model based on deep learning; taking the enhanced electrocardiogram signal as input, outputting a category prediction result through the disease prediction classification model based on deep learning; and performing myocardial infarction disease early warning according to the category prediction result. In the application, the denoising processing on the electrocardiogram signal and the construction of the disease prediction classification model based on deep learning can effectively improve the data quality, automatically learn complex features, capture key time sequence correlations, and thus improve the accuracy of the prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data analysis, in particular to a disease warning method and system based on deep learning. BACKGROUND

[0002] With the rapid development of deep learning technology, disease warning methods based on deep learning have become one of the important research directions in the medical field. Through in-depth analysis of a large amount of medical data, deep learning models can discover potential health risks and timely warn of disease, thereby effectively improving the accuracy of early diagnosis.

[0003] Currently, disease warning technologies based on deep learning mainly rely on deep neural network (DNN), convolutional neural network (CNN) and recurrent neural network (RNN) models. These models can process and analyze a large amount of medical data, including electronic health records (EHR), medical images, genomic data, laboratory test data, etc. In medical image analysis, CNN has been widely used in cancer detection, fracture identification and other fields.

[0004] However, the existing methods are not good at data scarcity due to the lack of case data and the difficulty of obtaining labeled data. At the same time, due to the strong heterogeneity, high dimensionality and noise of medical data, the model may not be able to fully capture the key temporal correlation in the feature extraction process, resulting in inaccurate prediction results. SUMMARY

[0005] In view of the above deficiencies of the prior art, the purpose of the embodiments of the present application is to provide a disease warning method based on deep learning, which can solve the technical problems that the existing methods are not good at data scarcity due to the lack of case data and the difficulty of obtaining labeled data. At the same time, due to the strong heterogeneity, high dimensionality and noise of medical data, the model may not be able to fully capture the key temporal correlation in the feature extraction process, resulting in inaccurate prediction results.

[0006] The first aspect of the embodiments of the present application proposes a disease warning method based on deep learning, comprising:

[0007] S1: obtaining an electrocardiogram signal of a patient;

[0008] S2: performing denoising processing on the electrocardiogram signal to obtain a denoised electrocardiogram signal;

[0009] S3: based on the denoised electrocardiogram signal, performing data enhancement processing through SMOTE algorithm to obtain an enhanced electrocardiogram signal;

[0010] S4: constructing a disease prediction classification model based on deep learning;

[0011] S5: outputting a category prediction result through the deep learning-based illness prediction classification model with the enhanced electrocardiogram signal as input;

[0012] S6: performing myocardial infarction illness early warning according to the category prediction result.

[0013] The second aspect of the embodiment of the present application provides a deep learning-based illness early warning system, comprising a processor and a memory.

[0014] The memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the deep learning-based illness early warning method according to the first aspect.

[0015] The third aspect of the embodiment of the present application provides a readable storage medium, and the readable storage medium stores programs or instructions, and the programs or instructions are executed by the processor to implement the steps of the deep learning-based illness early warning method according to the first aspect.

[0016] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects:

[0017] In the embodiment of the present application, the SMOTE algorithm is used for data enhancement to alleviate the problem of insufficient data, thereby improving the performance of the model. At the same time, by denoising the electrocardiogram signal and constructing a deep learning-based illness prediction classification model, the data quality can be effectively improved, complex features can be automatically learned, and key time sequence correlations can be captured, thereby improving the accuracy of myocardial infarction prediction. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings are only for the purpose of illustrating specific embodiments and are not considered as limiting the present application. Throughout the drawings, the same reference symbols indicate the same components. Obviously, the accompanying drawings described below are only some embodiments described in the embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0019] Figure 1 is a flowchart of a deep learning-based illness early warning method provided by the embodiment of the present application;

[0020] Figure 2 is a structural schematic diagram of a deep learning-based illness early warning system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0021] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] The following detailed description, in conjunction with the accompanying drawings, of the deep learning-based disease early warning method provided by the embodiments of the present invention through specific examples and application scenarios, will be provided in detail.

[0023] Reference manual attached Figure 1 The diagram shows a flowchart of a disease early warning method based on deep learning provided by an embodiment of the present invention.

[0024] This invention provides a deep learning-based disease early warning method, which may include the following steps:

[0025] S1: Acquire the patient's electrocardiogram signal.

[0026] An electrocardiogram (ECG) is a technique used to record the electrical activity of the heart. It involves placing electrodes on the skin to capture the electrical signals generated with each heartbeat. ECG signals are commonly used to diagnose heart disease and monitor heart health.

[0027] It should be noted that ST segment elevation, Q waves, and T wave inversion on an electrocardiogram (ECG) are important indicators for early warning of myocardial infarction. Therefore, during signal acquisition, it is necessary to ensure proper electrode placement and that the acquired ECG signal covers all critical information about the heart.

[0028] Specifically, by using an electrocardiogram (ECG) monitoring device, electrodes are placed on the patient's skin to record the electrical activity signals of the heart in real time, thus obtaining the patient's electrocardiogram signal.

[0029] In this embodiment of the invention, the electrocardiogram (ECG) monitoring device can record electrical signals in real time during the patient's cardiac activity. This is crucial for detecting important features related to myocardial infarction, such as ST-segment elevation, Q-wave formation, and T-wave inversion. Simultaneously, proper electrode placement allows for the capture of comprehensive and accurate cardiac electrical activity, ensuring that the monitored signals reflect the true state of the heart.

[0030] S2: Denoise the electrocardiogram signal to obtain a denoised electrocardiogram signal.

[0031] In one possible implementation, S2 specifically includes:

[0032] S201: decompose the electrocardiogram signal into multiple sub-signals of different frequency ranges using discrete wavelet transform:

[0033]

[0034] where X(t) represents the electrocardiogram signal at time t, ψ j (t) represents the value of the wavelet basis function at time t at the jth level scale, and W j represents the wavelet coefficient at the jth level scale.

[0035] where discrete wavelet transform (DWT) is a mathematical tool for analyzing and processing signals, which can decompose signals into different frequency components for analysis at multiple scales. DWT is a form of wavelet transform that decomposes signals recursively while preserving their time and frequency domain information. DWT has wide applications in signal denoising, data compression, feature extraction, and other fields, especially in electrocardiogram signal processing and image processing.

[0036] In the embodiments of the present application, DWT helps to preserve the characteristics of the signal during denoising. Important waveforms in electrocardiogram signals are usually located in the low-frequency part, while noise (such as electromyographic noise and baseline drift) is usually located in the high-frequency part. Therefore, by decomposing the wavelet coefficients, the signal and noise can be processed separately, achieving better denoising effect.

[0037] S202: denoising the decomposed wavelet coefficients at each level using soft threshold denoising processing:

[0038]

[0039] where, represents the wavelet coefficient at the jth level scale after denoising, sign represents the sign function, λ represents the soft threshold, and max represents the maximum value.

[0040] where the basic idea of soft threshold denoising processing is to remove noise by applying a threshold to the wavelet transform coefficients (or other transform coefficients). The denoising process is to modify the transform coefficients so that the coefficients less than the threshold are suppressed to zero, and the coefficients greater than the threshold are reduced in amplitude but not completely eliminated.

[0041] In the embodiments of the present application, the coefficients after wavelet transform can be divided into low-frequency and high-frequency parts, and noise is usually located in the high-frequency part. Through soft threshold denoising processing, high-frequency noise can be effectively removed, while the key components in the electrocardiogram signal, especially the useful information in the low-frequency part, are preserved.

[0042] S203: Reconstruct the ECG signal based on the denoised wavelet coefficients to obtain the denoised ECG signal:

[0043]

[0044] in, This represents the denoised electrocardiogram signal at time t. This represents the wavelet coefficients after denoising.

[0045] In this embodiment of the invention, the accuracy of the electrocardiogram (ECG) signal is improved through reconstruction, reducing errors introduced by noise. For clinical applications, accurate ECG signals can improve the reliability and accuracy of diagnosis.

[0046] S3: Based on the denoised electrocardiogram signal, data augmentation processing is performed using the SMOTE algorithm to obtain an enhanced electrocardiogram signal.

[0047] SMOTE (Synthetic Minority Over-sampling Technique) is a data augmentation method commonly used to address class imbalance. It balances the class distribution by synthesizing samples from the minority class, thereby improving the predictive ability of machine learning models for the minority class.

[0048] In one possible implementation, based on the denoised electrocardiogram (ECG) signal, data augmentation processing is performed using the following formula to obtain the enhanced ECG signal:

[0049] X new =X i +rand(0,1)×(X i,j -X i )

[0050] Among them, X new X represents the generated minority class sample. i Let X represent the i-th heartbeat sample in the minority class. i,j Let represent the neighboring samples of the i-th heartbeat sample, and rand(0,1) represent a random number between 0 and 1. It's worth noting that data augmentation using the SMOTE algorithm helps generate more samples when there is a shortage of minority class samples (such as myocardial infarction). This is extremely useful, especially for rare events like myocardial infarction, where the SMOTE algorithm can synthesize new samples, thereby balancing the dataset and helping deep learning models better learn the features of minority class samples.

[0051] In this embodiment of the invention, by applying the SMOTE algorithm to the denoised electrocardiogram (ECG) signal, class imbalance can be effectively balanced, minority class samples can be enhanced, and the model's classification ability can be improved, especially in medical signal analysis such as abnormal heartbeat detection. SMOTE enhances the model's generalization ability, robustness, and minority class recognition ability by generating synthetic minority class samples, thereby providing more reliable results for ECG signal classification tasks.

[0052] S4: Construct a disease prediction and classification model based on deep learning.

[0053] Among them, disease prediction and classification models based on deep learning are a type of system that uses deep neural networks to predict and classify diseases. Deep learning technology automatically learns features from large amounts of data, enabling it to extract useful information from complex biomedical signals (such as electrocardiograms, electroencephalograms, and medical images) to help doctors make disease diagnoses and select treatment plans.

[0054] S5: Using enhanced electrocardiogram signals as input, the system outputs category prediction results through a deep learning-based disease prediction classification model.

[0055] In one possible implementation, S5 specifically includes:

[0056] S501: Using dilated causal convolution to capture spatial features of enhanced ECG signals:

[0057]

[0058] in, Let W represent the spatial features at time step t, ReLU represent the activation function, and W represent the spatial features at time step t. dil The weight matrix represents the dilated convolution, * denotes the symbol for the convolution operation, and x t-k:t b represents the signal segment from time step tk to time step t. dil This represents the bias term of the dilated convolution.

[0059] It should be noted that dilated causal convolution is used to extract spatial features related to myocardial infarction, such as ST segment elevation, Q wave changes, and T wave inversion.

[0060] In this embodiment of the invention, dilated convolution can effectively capture important local spatial features in electrocardiogram signals.

[0061] S502: Introduces a gated cyclic unit to capture the temporal characteristics of enhanced electrocardiogram signals:

[0062]

[0063] in, Let x represent the temporal characteristics at time step t.t This represents the electrocardiogram signal at time step t. W represents the temporal hidden feature at time step t-1. gru This represents the learnable weight matrix in the GRU network.

[0064] It should be noted that the GRU is used to capture temporal features in electrocardiogram (ECG) signals. Here, the GRU specifically focuses on temporal changes associated with myocardial infarction, such as persistent ST segment elevation and T wave inversion. Through long-term temporal dependence, the model can learn the temporal changes of these features, thereby determining the occurrence of myocardial infarction.

[0065] In this embodiment of the invention, the electrocardiogram (ECG) signal is a typical time series signal. The GRU can effectively capture time series information, learn the long-term time dependence in the signal, and thus identify important time series features in the ECG signal.

[0066] S503: Spatial and temporal features are initially fused using a feature cross-attention mechanism to obtain preliminary fused features:

[0067]

[0068] in, Let represent the initial fusion feature at time step t, A represent the cross-attention weight, Softmax represent the activation function, Q represent the query vector obtained after linear transformation of the spatial features, K represent the key vector obtained after linear transformation of the temporal features, T represent the transpose operation, and d represent the scaling factor.

[0069] In this embodiment of the invention, the cross-attention mechanism helps the model focus on changes in ST segment elevation and Q waves, thereby more accurately fusing information from both and improving prediction accuracy. Simultaneously, the cross-attention mechanism helps capture the nonlinear relationship between spatial and temporal features, enhancing the model's understanding of complex patterns in electrocardiograms.

[0070] S504: Use a dynamic gating mechanism to perform fine-grained fusion on the initial fused features to obtain fine-grained fused features:

[0071]

[0072] in, G represents the fine-grained fusion feature at time step t. t ⊙ denotes the weighting factor, σ denotes element-wise multiplication, and W denotes the sigmoid activation function. g Let b represent the weight matrix. g This represents the bias term used to adjust the calculation of the gating value.

[0073] In this embodiment of the invention, the dynamic gating mechanism automatically adjusts the fusion of spatial and temporal features, enabling the model to prioritize features of myocardial infarction and improve prediction accuracy.

[0074] S505: Based on refined fusion features, the final fusion features are obtained through multi-layer convolution and skip connections.

[0075]

[0076] in, Let α represent the final fused feature at time step t. l This represents the weighting coefficients used to weight the convolution results of each layer, l = 1, 2, ..., L, where L represents the number of skip connections, and Conv1D represents a one-dimensional convolution operation. This represents the finely fused feature after the l-th convolutional operation at the t-th time step.

[0077] In this embodiment of the invention, by using multi-layer convolution and skip connections, the model can fuse features from different levels, thereby enhancing the model's ability to learn features at different scales.

[0078] S506: Input the final fused features into the fully connected layer for classification and output the category prediction results.

[0079] In one possible implementation, S506 specifically includes:

[0080] S5061: The final fused features are input into the fully connected layer for classification to obtain scores for each category:

[0081]

[0082] Among them, y m W represents the score for the m-th category. cls Let b represent the weight matrix of the fully connected layer. cls This represents the bias term of the fully connected layer.

[0083] S5062: Convert the scores of each category into category prediction results using the Softmax function:

[0084]

[0085] Among them, P m Let represent the predicted probability of the m-th class, e represent the exponential function, n represent the class index, and y represent the predicted probability of the m-th class. n This represents the raw score value for the nth category.

[0086] In summary, by employing dilated convolution, GRU, attention mechanisms, and gating mechanisms, the model can effectively extract spatiotemporal features from electrocardiogram signals, enhancing the understanding of the signals and significantly improving the accuracy of the classification model, especially for the identification of minority classes.

[0087] In one possible implementation, the category prediction results specifically include: the predicted probability of a normal heartbeat and the predicted probability of myocardial infarction.

[0088] Optionally, the deep learning-based disease prediction classification model can be optimized using an improved adam optimizer.

[0089] Specifically, the loss function for the disease prediction classification model is set as the mean squared error loss function:

[0090]

[0091] Where MSE represents the mean squared error loss function, and k represents the total number of samples. Let y represent the predicted value of the b-th sample. b This represents the true value of the b-th sample.

[0092] Initialize parameters, including model parameters θ0, learning rate η, and batch size N. B The attenuation rates β1, β2 and constant ε.

[0093] Initialize the first-order moment estimator m0 = 0, the second-order moment estimator v0 = 0, and

[0094] Obtain the training dataset, which includes normal samples and heartbeat samples from myocardial infarction, and label each sample.

[0095] Extract a batch of training samples from the training dataset and calculate the current parameter θ. t-1 Gradient of the objective function:

[0096]

[0097] Among them, G t Let x represent the gradient at time t. ik Let y represent the input data for the k-th sample. ik Let θ represent the label of the k-th sample. t-1 Let L((x) represent the model parameters in the previous iteration, I represent the cross-entropy loss function, and L((x) represent the model parameters in the previous iteration. ik ,y ik ),θ t-1 ) represents the output of the model, N B This represents the total number of samples.

[0098] Update the first-order moment estimate using the current gradient:

[0099] m t =β1·m t-1 +(1-β1)G t

[0100] Where, m t m represents the first moment estimate at time t. t-1 G represents the first moment estimate at time t-1, β1 represents the decay rate of the first moment, and G... t This represents the gradient at time t.

[0101] Find the maximum value of the first moment at the previous time step and the current gradient.

[0102] Calculate the target ratio based on the updated first-order moment estimate, the maximum value of the first-order moment from the previous time step, and the current gradient:

[0103]

[0104] Where Λ(t) represents the target ratio.

[0105] In this embodiment of the invention, by dynamically judging the target ratio, abnormal situations that occur during training (such as a sudden surge in gradient) can be identified, and optimization strategies can be automatically switched, thereby improving the robustness and fault tolerance of model training.

[0106] The second-order moment estimate is calculated by controlling the magnitude of the update using the target ratio:

[0107] v t =β2·v t-1 +(1-β2)·|G t | Λ(t)

[0108] Among them, v t Let v represent the second moment estimate at time t. t-1 Let β2 represent the second moment estimate at time t-1, and let β2 represent the decay rate of the second moment.

[0109] If the target ratio is less than 2, switch to the AMSGrad optimizer, set amsgrad=True, and update the parameter θ. t :

[0110]

[0111] Where, θ t Let θ represent the model parameters at time t. t-1 Let m represent the model parameters at time t-1, η represent the learning rate, and m represent the model parameters at time t-1. t This represents the first moment estimate at time t, where ε is a constant. This represents the corrected second-order moment estimate in the AMSGrad optimizer. This represents the corrected second-order moment estimate from the previous time step.

[0112] If the target ratio is greater than or equal to 2, continue using the Adam optimizer and update the parameter θ. t :

[0113]

[0114] When the preset number of iterations is reached, the iteration stops and the final optimized parameters are output.

[0115] In this embodiment of the invention, by introducing a target ratio and dynamically switching between the Adam and AMSGrad optimization strategies, both rapid convergence and optimization stability can be achieved during model training. When gradient fluctuations are large, Adam is used to maintain efficient updates; when training tends to stabilize or faces local optima, AMSGrad is switched to suppress overshoot, effectively improving the robustness and accuracy of training. Furthermore, this method has adaptive adjustment capabilities, automatically controlling the parameter update amplitude to avoid oscillations and divergences, reducing the burden of manual parameter tuning. This enables the trained model to meet the high stability requirements of medical scenarios.

[0116] S6: Provide early warning of myocardial infarction based on category prediction results.

[0117] In one possible implementation, S6 specifically includes:

[0118] S601: Sort the category probabilities of each category in the category prediction results in descending order, and select the target category with the highest prediction probability.

[0119] S602: Continue monitoring if the target category is normal heartbeat. Issue an alarm if the target category is myocardial infarction.

[0120] In this embodiment of the invention, by setting different response mechanisms for different categories, the system can precisely control when to trigger an alert. For example, if a normal heartbeat is predicted, the system continues monitoring without interfering with normal operation. However, if a myocardial infarction is predicted, the system will immediately issue an alert. This approach avoids excessive alerts and improves the system's efficiency. Furthermore, the system will not easily trigger an alert in the absence of a myocardial infarction. An alert will only be issued when the model predicts a myocardial infarction with a high probability, which effectively reduces interference or unnecessary emergency responses caused by excessive alerts.

[0121] In one possible implementation, issuing an alarm specifically includes:

[0122] An alarm is triggered by a warning light.

[0123] And / or issue an alarm via a buzzer.

[0124] In this embodiment of the invention, the warning light and the buzzer alarm each provide different sensing methods: the warning light provides a visual signal, while the buzzer alarm provides an audible signal. Using both together ensures that alarm information is not missed due to environmental noise, equipment malfunction, or personnel failing to notice a particular signal source.

[0125] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0126] In this embodiment of the invention, the SMOTE algorithm is used for data augmentation to alleviate the problem of insufficient data, thereby improving the model's performance. Simultaneously, by denoising the electrocardiogram signal and constructing a deep learning-based disease prediction classification model, data quality can be effectively improved, complex features can be automatically learned, and key temporal correlations can be captured, thus enhancing the accuracy of myocardial infarction prediction.

[0127] Reference manual attached Figure 2 The diagram shows a schematic representation of a disease early warning system based on deep learning provided in an embodiment of the present invention.

[0128] This invention provides a disease early warning system 20 based on deep learning, including: a processor 201 and a memory 202;

[0129] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-mentioned deep learning-based disease early warning method and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0130] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0131] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0132] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0133] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0134] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0135] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0137] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0139] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the above-described deep learning-based disease early warning method and achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A disease early warning method based on deep learning, characterized in that, include: S1: Acquire the patient's electrocardiogram signal; S2: Denoise the electrocardiogram signal to obtain a denoised electrocardiogram signal; S3: Based on the denoised electrocardiogram signal, data enhancement processing is performed using the SMOTE algorithm to obtain an enhanced electrocardiogram signal; S4: Construct a disease prediction and classification model based on deep learning; S5: Using the enhanced electrocardiogram signal as input, the category prediction result is output through the deep learning-based disease prediction classification model; S6: Provide early warning of myocardial infarction based on the predicted category results; Specifically, S3 is: Based on the denoised electrocardiogram (ECG) signal, data augmentation processing is performed using the following formula to obtain the enhanced ECG signal: X new =X i +rand(0,1)×(X i,j -X i ); Among them, X new X represents the generated minority class sample. i Let X represent the i-th heartbeat sample in the minority class. i,j This represents the neighboring samples of the i-th heartbeat sample, and rand(0,1) represents a random number between 0 and 1; Specifically, S5 includes: S501: Use dilated causal convolution to capture the spatial features of the enhanced electrocardiogram signal; S502: Introducing a gated loop unit to capture the temporal characteristics of the enhanced electrocardiogram signal; S503: The spatial features and the temporal features are initially fused using a feature cross-attention mechanism to obtain preliminary fused features; S504: Use a dynamic gating mechanism to perform fine-grained fusion on the preliminary fusion features to obtain fine-grained fusion features: in, G represents the fine-grained fusion feature at time step t. t The circle (⊙) represents the weighting factor, and the circle (⊙) represents element-wise multiplication. Represents the spatial characteristics at time step t. Let W represent the temporal characteristics at time step t, σ represent the Sigmoid activation function, and W represent the temporal features at time step t. g Represents the weight matrix. b represents the initial fusion feature at time step t. g This represents the bias term used to adjust the calculation of the gate value; S505: Based on the aforementioned refined fusion features, the final fusion features are obtained through multi-layer convolution and skip connections: in, Let α represent the final fused feature at time step t. l This represents the weighting coefficients used to weight the convolution results of each layer, l = 1, 2, ..., L, where L represents the number of skip connections, and Conv1D represents a one-dimensional convolution operation. This represents the finely fused features after the l-th convolutional operation at the t-th time step; S506: Input the final fused features into the fully connected layer for classification, and output the category prediction result.

2. The disease early warning method based on deep learning according to claim 1, characterized in that, S2 specifically includes: S201: Use discrete wavelet transform to decompose the electrocardiogram signal into multiple sub-signals of different frequency ranges; S202: Use soft thresholding to denoise the wavelet coefficients of each decomposed layer. S203: Reconstruct the electrocardiogram signal based on the denoised wavelet coefficients to obtain the denoised electrocardiogram signal.

3. The disease early warning method based on deep learning according to claim 1, characterized in that, Specifically, S506 includes: S5061: Input the final fused features into the fully connected layer for classification to obtain the score value of each category; S5062: The scores for each category are converted into category prediction results using the Softmax function.

4. The disease early warning method based on deep learning according to claim 1, characterized in that, The specific prediction results include: the predicted probability of a normal heartbeat and the predicted probability of myocardial infarction.

5. The disease early warning method based on deep learning according to claim 4, characterized in that, S6 specifically includes: S601: Sort the category probabilities of each category in the category prediction results in descending order, and select the target category with the highest prediction probability; S602: If the target category is normal heartbeat, continue monitoring; if the target category is myocardial infarction, issue an alarm.

6. The disease early warning method based on deep learning according to claim 5, characterized in that, The issuance of the alarm specifically includes: An alarm is triggered by warning lights; And / or issue an alarm via a buzzer.

7. A disease early warning system based on deep learning, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the deep learning-based disease early warning method as described in any one of claims 1 to 6.

8. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the deep learning-based disease early warning method as described in any one of claims 1 to 6.

Images