Alzheimer disease prediction method and system based on deep learning and electrocardiosignals

Abstract

The invention discloses an Alzheimer disease prediction method and system based on deep learning and electrocardiosignals. Firstly, electrocardiosignals are collected and preprocessed; then, constructing a deep learning prediction model which comprises a feature extraction module and a classification prediction module; the feature extraction module comprises a plurality of residual blocks, and each residual block comprises a convolution layer, a batch normalization layer, a nonlinear activation layer and a jump connection; the classification prediction module realizes a complete mapping process from a feature space to probability prediction through a multi-layer perceptron structure and a softmax function, and obtains a prediction probability of each electrocardiosignal fragment belonging to each category; and fusing the prediction results of the plurality of electrocardiosignal segments of the same sample by adopting a soft voting mechanism to obtain a final prediction result. The system comprises a signal acquisition module, a signal preprocessing module, a deep learning prediction module and a result output module. According to the method, Alzheimer's disease prediction is realized based on the electrocardiosignals, and the method is simple and convenient in acquisition mode, non-invasive, low in cost and suitable for large-scale screening.

Description

Technical Field

[0001] This invention belongs to the field of computer science and technology, specifically a method and system for predicting Alzheimer's disease based on deep learning and electrocardiogram signals. Background Technology

[0002] Alzheimer's disease is a common neurodegenerative disease, mainly characterized by memory decline, cognitive loss, and reduced daily living abilities. With the aging population, the incidence of this disease is increasing year by year, severely impacting patients' quality of life. Currently, clinical diagnosis mainly relies on imaging examinations, cerebrospinal fluid biomarkers, and cognitive function scales. While these methods can provide disease-related information, they generally suffer from high costs, invasiveness, and dependence on specialized equipment and personnel, making it difficult to meet the needs of large-scale early screening.

[0003] In recent years, deep learning has been widely used in Alzheimer's disease research, especially in medical imaging and speech analysis, where it has shown certain advantages. However, these methods are limited by high testing costs, complex hardware requirements, and difficulties in data acquisition, making them difficult to promote in routine clinical screening. In contrast, electrocardiogram (ECG) signal acquisition is simple, non-invasive, and low-cost, and can reflect the function of the autonomic nervous system. Since autonomic dysfunction has been proven to be closely related to the occurrence and development of Alzheimer's disease, ECG-based Alzheimer's disease identification has potential application value.

[0004] However, existing research based on electrocardiogram (ECG) signals still has limitations. On the one hand, traditional analyses rely heavily on manual feature extraction, making it difficult to comprehensively reflect potential pathological patterns. On the other hand, ECG signals are susceptible to noise and artifacts, leading to insufficient model prediction accuracy and stability. Furthermore, research on segmented modeling and result fusion mechanisms for long-term signals is insufficient, preventing the full utilization of ECG data in Alzheimer's disease screening. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the technical problem this invention aims to solve is to provide a method and system for predicting Alzheimer's disease based on deep learning and electrocardiogram signals.

[0006] The present invention solves the aforementioned technical problem by adopting the following technical solution:

[0007] On the one hand, the present invention provides a method for predicting Alzheimer's disease based on deep learning and electrocardiogram signals, characterized by comprising the following steps:

[0008] Step 1: Acquire electrocardiogram (ECG) signals and perform preprocessing;

[0009] Step 2: Construct a deep learning prediction model, including a preliminary feature mapping module, a residual feature extraction module, and a prediction output module;

[0010] The preliminary feature mapping module includes convolution, batch normalization, ReLU activation function, and pooling operations. The feature mapping process is represented as follows:

[0011] F1 = Pool(ReLU(BN(Conv(x))))

[0012] Where F1 represents the output features of the initial feature mapping module, Pool(·) represents the pooling operation, ReLU(·) represents the ReLU activation function, BN(·) represents the batch normalization operation, Conv(·) represents the one-dimensional convolution operation, and x is an electrocardiogram signal segment;

[0013] The residual feature extraction module consists of multiple residual blocks, and the feature extraction process for the residual blocks is as follows:

[0014] F=ReLU(BN(Conv(ReLU(BN(Conv(x))))))+x)

[0015] Where F represents the output feature of the residual block;

[0016] The prediction output module uses a multilayer perceptron to perform probability mapping on the deep features output by the residual feature extraction module to obtain the predicted probability of ECG signal segments belonging to various categories.

[0017] Step 3: Train the deep learning prediction model and use the trained deep learning prediction model to predict Alzheimer's disease.

[0018] Furthermore, a soft voting mechanism is used to fuse the prediction results of multiple ECG signal segments from the same sample. The fusion process is represented as follows:

[0019]

[0020] in, Let P(y) be the final predicted category of the sample, N be the total number of ECG signal segments in the same sample, and P(y) be the final predicted category of the sample. i |x j ) represents the j-th electrocardiogram signal segment x j The predicted probability for category i.

[0021] Furthermore, the preprocessing includes artifact removal and high-frequency noise suppression.

[0022] On the other hand, the present invention also provides an Alzheimer's disease prediction system, comprising:

[0023] Signal acquisition module: used to acquire electrocardiogram (ECG) signals;

[0024] Signal preprocessing module: used to preprocess ECG signals;

[0025] Deep learning prediction module: used to extract features from electrocardiogram signals, classify them according to the features, and generate prediction probabilities for each category;

[0026] The results output module is used to fuse the prediction results of multiple ECG signal segments and output the final prediction result.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] Alzheimer's disease prediction based on electrocardiogram signals is simple, non-invasive, and low-cost, making it suitable for large-scale population screening.

[0029] By using deep learning models to achieve automatic feature extraction, the reliance on human experience in traditional methods is avoided, and the feature representation capability is improved.

[0030] By combining sliding window segmentation with soft voting results, the interference of ECG signal noise was effectively mitigated, and the accuracy and stability of the prediction results were improved.

[0031] The overall solution is non-invasive, portable, and easy to promote, providing an efficient and reliable technical means for the early auxiliary diagnosis of Alzheimer's disease. Attached Figure Description

[0032] Figure 1 This is a comparison chart of the data before and after preprocessing in this invention;

[0033] Figure 2 This is a schematic diagram of the structure of the deep learning prediction model of the present invention;

[0034] Figure 3 This is a schematic diagram of the system structure of the present invention;

[0035] Figure 4 This is the confusion matrix and ROC curve of the present invention. Detailed Implementation

[0036] Specific embodiments are given below with reference to the accompanying drawings. These specific embodiments are only used to describe the technical solution of the present invention in detail, and are not intended to limit the scope of protection of this application.

[0037] This invention provides a method for predicting Alzheimer's disease based on deep learning and electrocardiogram signals, comprising the following steps:

[0038] Step 1: Acquire electrocardiogram (ECG) signals and perform preprocessing;

[0039] ECG signals can be derived from clinically collected ECG monitoring data or from publicly available databases for model training and validation. They can be collected using portable devices, wearable devices, or medical monitoring systems.

[0040] To improve signal quality and remove artifacts, preprocessing includes:

[0041] Multiple statistical signal quality indicators, such as mean, variance, and kurtosis, are extracted, and an adaptive baseline reference is established in the stable signal region based on a standardization method. A multi-scale sliding window approach is used for automatic detection of artifact regions, and precise localization is achieved through iterative steps. For the signal after artifact removal, the R-peak is preferentially used as the time reference point during reconstruction to reconnect normal signal segments and maintain the physiological continuity of the signal.

[0042] To further suppress high-frequency noise, a discrete wavelet transform thresholding method can be used. This thresholding method can be based on the MinMax thresholding method, selectively thresholding the coefficients of high-frequency components to achieve noise suppression and signal reconstruction. Alternatively, other time-frequency analysis methods such as empirical mode decomposition can be used to achieve similar results.

[0043] like Figure 1 As shown, Figure 1 (a) shows the raw, unprocessed electrocardiogram signal. Figure 1 (b) shows the ECG signal after preprocessing according to this embodiment; the preprocessing process of the present invention can remove artifacts and high-frequency noise in the signal, thereby obtaining ECG signal data that is more suitable for subsequent feature extraction.

[0044] After completing the above preprocessing, the long-term ECG signal is further divided into fixed-length segments using the sliding window technique, so that it can be input into the deep learning model for training and prediction.

[0045] Step 2: Construct a deep learning prediction model;

[0046] like Figure 3 As shown, the deep learning prediction model includes a preliminary feature mapping module, a residual feature extraction module, and a prediction output module. The preliminary feature mapping module is used to convert the preprocessed ECG signal segments into features required by the neural network. The residual feature extraction module performs deep feature extraction by stacking multiple residual blocks. The prediction output module generates prediction probabilities for each category based on the deep features.

[0047] The preliminary feature mapping module includes convolution, batch normalization, ReLU activation function, and pooling operations. The feature mapping process is represented as follows:

[0048] F1 = Pool(ReLU(BN(Conv(x))))

[0049] Where F1 represents the output features of the initial feature mapping module, Pool(·) represents the pooling operation, ReLU(·) represents the ReLU activation function, BN(·) represents the batch normalization operation, Conv(·) represents the one-dimensional convolution operation, and x is an electrocardiogram signal segment.

[0050] The residual feature extraction module models and analyzes the temporal features of the ECG signal, including multiple residual blocks. Within each residual block, the input features are sequentially processed through convolution, batch normalization, ReLU activation, convolution, and batch normalization before being connected to the block via skip connections. The features obtained from these skip connections are then processed by the ReLU activation function to obtain the output features of the residual block. The feature extraction process for the residual block is represented as follows:

[0051] F=ReLU(BN(Conv(ReLU(BN(Conv(x))))))+x)

[0052] Where F represents the output characteristics of the residual block.

[0053] The prediction output module uses a multilayer perceptron to map the deep features output by the residual feature extraction module from the feature space to probability prediction, obtaining the predicted probability of each ECG signal segment belonging to a specific category. The processing of the prediction output module can be represented as follows:

[0054]

[0055] Wherein, P(y i |x) represents the predicted probability that ECG signal segment x belongs to category i, W i b i The weights and bias parameters of the output layer are represented by MLP(·), where MLP(·) represents a multilayer perceptron, C is the total number of categories, and F2 represents the deep features output by the residual feature extraction module.

[0056] A soft voting mechanism is employed to fuse the prediction results of multiple ECG signal segments from the same sample, thereby improving the stability and accuracy of the prediction. Since ECG signals are typically segmented into time windows, each sample generates multiple predicted probabilities for each segment. A reasonable fusion strategy can effectively reduce the prediction error of a single segment and improve overall prediction performance. For multiple ECG signal segments from the same sample, the deep learning model outputs the predicted probability of each segment separately. The soft voting fusion mechanism determines the final prediction category by calculating the arithmetic mean of the predicted probabilities of all ECG signal segments. The entire fusion and prediction process can be represented as follows:

[0057]

[0058] in, Let P(y) be the final predicted category of the sample, N be the total number of ECG signal segments in the same sample, and P(y) be the final predicted category of the sample. i |x j ) represents the j-th electrocardiogram signal segment x j The predicted probability for category i.

[0059] The core advantage of the soft voting mechanism lies in preserving the detailed features of probabilistic information, avoiding the information loss problem of hard voting. By using the average probability distribution, this method not only determines the final predicted category but also provides relative probability information for each category, offering clinicians a more comprehensive reference for diagnostic decisions. Simultaneously, the maximum probability value after fusion can serve as a quantitative indicator of prediction confidence, helping to assess the reliability of the prediction results. In the output phase, the system provides complete prediction results for each sample, including the final predicted category, the fused probability distribution, the prediction confidence, and the number of fragments involved in the fusion. This detailed output information not only provides quantitative prediction results for clinical diagnosis but also offers transparency support for doctors to understand the model's decision-making process, enhancing the system's interpretability and clinical acceptability.

[0060] Through the aforementioned soft voting mechanism, this invention can fully utilize the temporal characteristics and multi-segment information of electrocardiogram signals, effectively integrate prediction results from different time windows, improve the stability and reliability of Alzheimer's disease prediction results, and thus provide auxiliary reference for clinical diagnosis.

[0061] Step 3: Train the deep learning prediction model and use the trained deep learning prediction model to predict Alzheimer's disease.

[0062] During model training, the cross-entropy loss function is used for parameter optimization, and an adaptive weight initialization strategy is employed to ensure training stability. To improve the model's generalization ability, dropout regularization and layer normalization techniques are introduced into the classifier. An end-to-end learning approach is used to extract ECG signal features, thereby reducing reliance on manual feature design, and outputting the predicted probability distribution for each ECG signal segment.

[0063] like Figure 3 As shown, the present invention also provides an Alzheimer's disease prediction system, comprising:

[0064] The signal acquisition module is used to acquire electrocardiogram (ECG) signals.

[0065] The signal preprocessing module is used for quality assessment, artifact removal, and segmentation of electrocardiogram signals;

[0066] The deep learning prediction module is used to extract features from electrocardiogram signals, classify them based on the features, and generate prediction probabilities for each category.

[0067] The results output module is used to fuse the prediction results of multiple ECG signal segments and output the final prediction result.

[0068] To illustrate the feasibility and applicability of the method of this invention, a sliding window technique was used to segment the electrocardiogram (ECG) signal. The window length was set to 60 seconds, and the overlap rate was 0.5 to ensure the consistency of the input data. During training, the batch size was set to 64, the initial learning rate was set to 1e-4, and a learning rate warm-up phase was set in the early stage of training, gradually increasing the learning rate from a small value to the target value. A learning rate decay strategy was further adopted during training, with a decay factor of 0.1. The optimizer was Adam, and the loss function was the cross-entropy loss function. The maximum number of training epochs was set to 100, and an early stopping mechanism was used to reduce the risk of overfitting. To verify the stability of the model under different conditions, this embodiment independently repeated the experiment under multiple random seeds, and the experimental results are shown in Table 1.

[0069] Table 1 Performance test results of the method of the present invention under different random seeds.

[0070]

[0071] As shown in Table 1, under different random seeds, the method maintains a high level in terms of accuracy, precision, recall, and F1 score, and the results are consistent.

[0072] Figure 4 (a) is the confusion matrix obtained on the test set, used to show the classification of each category; Figure 4 (b) shows the corresponding ROC curve, used to demonstrate the classification performance of the model at different thresholds. (From Table 1 and...) Figure 4 As can be seen, the method in this embodiment can achieve stable classification results under multiple experimental conditions, thus verifying the feasibility and applicability of the method.

[0073] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A method for predicting Alzheimer's disease based on deep learning and electrocardiogram signals, characterized in that, Includes the following steps: Step 1: Acquire electrocardiogram (ECG) signals and perform preprocessing; Step 2: Construct a deep learning prediction model, including a preliminary feature mapping module, a residual feature extraction module, and a prediction output module; The preliminary feature mapping module includes convolution, batch normalization, ReLU activation function, and pooling operations. The feature mapping process is represented as follows: F1 = Pool(ReLU(BN(Conv(x)))) Where F1 represents the output features of the initial feature mapping module, Pool(·) represents the pooling operation, ReLU(·) represents the ReLU activation function, BN(·) represents the batch normalization operation, Conv(·) represents the one-dimensional convolution operation, and x is an electrocardiogram signal segment; The residual feature extraction module consists of multiple residual blocks, and the feature extraction process for the residual blocks is as follows: F=ReLU(BN(Conv(ReLU(BN(Conv(x))))))+x) Where F represents the output feature of the residual block; The prediction output module uses a multilayer perceptron to perform probability mapping on the deep features output by the residual feature extraction module to obtain the predicted probability of ECG signal segments belonging to various categories. Step 3: Train the deep learning prediction model and use the trained deep learning prediction model to predict Alzheimer's disease.

2. The Alzheimer's disease prediction method based on deep learning and electrocardiogram signals according to claim 1, characterized in that, A soft voting mechanism is used to fuse the prediction results of multiple ECG signal segments from the same sample. The fusion process is represented as follows: in, Let P(y) be the final predicted category of the sample, N be the total number of ECG signal segments in the same sample, and P(y) be the final predicted category of the sample. i |x j ) represents the j-th electrocardiogram signal segment x j The predicted probability for category i.

3. The Alzheimer's disease prediction method based on deep learning and electrocardiogram signals according to claim 1, characterized in that, The preprocessing includes artifact removal and high-frequency noise suppression.

4. A system for predicting Alzheimer's disease using the method described in any one of claims 1 to 3, characterized in that, include: Signal acquisition module: used to acquire electrocardiogram (ECG) signals; Signal preprocessing module: used to preprocess ECG signals; Deep learning prediction module: used to extract features from electrocardiogram signals, classify them according to the features, and generate prediction probabilities for each category; The results output module is used to fuse the prediction results of multiple ECG signal segments and output the final prediction result.

Images