Chronic disease detection method and device
By constructing a chronic disease feature matrix and matching it with a case database using an improved neural network model, the problems of insufficient data utilization and low efficiency in traditional detection are solved, thereby improving the accuracy and efficiency of early detection of chronic diseases.
Patent Information
- Application Number
- CN202511733757.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-02-17
AI Technical Summary
Traditional chronic disease detection relies on manual analysis by doctors, which results in insufficient data utilization, low detection efficiency, and difficulty in meeting the needs of early intervention.
By acquiring multi-dimensional medical testing data, a chronic disease feature matrix is constructed. An improved neural network model is used for preliminary detection, which is then matched with a case database. Finally, multiple related test results are combined for collaborative analysis to generate the final test results.
This improves the accuracy and efficiency of early detection of chronic diseases, ensuring the reliability and timeliness of test results.
Smart Images
Figure FT_1 
Figure FT_2 
Figure SMS_1
Abstract
Description
Technical Field
[0001] This invention relates to the fields of intelligent medical care and artificial intelligence technology, and in particular to a method and device for detecting chronic diseases. Background Technology
[0002] Chronic diseases are characterized by long course, insidious early symptoms, and complex pathogenesis. Early and accurate detection is crucial for disease intervention and prognosis improvement. Traditional chronic disease detection relies on doctors' manual analysis of single or limited modalities of medical data (such as imaging reports and blood test results), which has the following limitations: Insufficient data utilization: Doctors struggle to simultaneously integrate multi-source patient information, easily leading to missed or misdiagnosed cases due to missing information; Low detection efficiency: In primary healthcare settings, specialized medical resources are scarce, and manual testing is slow when facing large-scale population screening, failing to meet the needs of early intervention.
[0003] Based on this, the present invention proposes a method and apparatus for detecting chronic diseases to solve the above-mentioned technical problems. Summary of the Invention
[0004] This invention describes a method and apparatus for detecting chronic diseases, which can improve the accuracy of early detection of chronic diseases.
[0005] According to a first aspect, the present invention provides a method for detecting chronic diseases, comprising: Acquire medical testing data; wherein, the medical testing data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data; Based on the aforementioned medical testing data, a feature matrix for chronic diseases is determined; The feature matrix of the chronic disease is input into a preset chronic disease detection model to obtain preliminary detection results of the chronic disease; The feature matrix of the chronic disease is input into a preset case database to obtain multiple related detection results; wherein the overall similarity between the related detection results and the preliminary detection results is greater than a first preset threshold. Based on the preliminary test results and multiple related test results, the final test result is determined.
[0006] According to a second aspect, the present invention provides a device for detecting chronic diseases, comprising: The acquisition unit is configured to acquire medical testing data; wherein the medical testing data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data. The first data processing unit is configured to determine a feature matrix of chronic diseases based on the medical test data; The second data processing unit is configured to input the feature matrix of the chronic disease into a preset chronic disease detection model to obtain preliminary detection results of the chronic disease. The third data processing unit is configured to input the feature matrix of the chronic disease into a preset case database to obtain multiple related detection results; wherein the overall similarity between the related detection results and the preliminary detection results is greater than a first preset threshold. The fourth data processing unit is configured to determine the final detection result based on the preliminary detection result and multiple related detection results.
[0007] Thirdly, embodiments of this specification also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of this specification.
[0008] Fourthly, embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.
[0009] According to the chronic disease detection method and apparatus provided by the present invention, firstly, multi-dimensional medical testing data of the patient is acquired through a standardized data interface, wherein the medical testing data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data. Based on the medical testing data, a feature matrix of chronic diseases is determined. On the one hand, the feature matrix is input into a preset chronic disease detection model to obtain preliminary detection results of chronic diseases, including disease type (such as diabetes, coronary heart disease), probability of disease (such as 92.5%), and stage of disease (such as early stage); on the other hand, the same feature matrix is input into a preset case database (containing 150,000+ desensitized clinical cases), and multiple related detection results with a comprehensive similarity greater than a first preset threshold (92%) are selected to ensure that the related cases are highly consistent with the current patient in terms of core features (such as imaging lesion type, gene site mutation). Finally, a collaborative analysis is conducted based on the preliminary test results and related test results: if the preliminary results are consistent with the diagnosis of most related cases (e.g., all 3 related cases are "early stage of type 2 diabetes"), the final report is generated directly based on the preliminary results; if there are differences (e.g., 1 related case is "prediabetes"), cross-validation is performed by combining the specific clinical background of related cases (e.g., patient age, comorbidities) with the current patient condition, and after correcting for deviations, the final test results are generated, which include the diagnosed disease, stage, key evidence, and a summary of reference cases. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 A flowchart illustrating a method for detecting chronic diseases according to one embodiment is shown; Figure 2 A schematic block diagram of a chronic disease detection device according to one embodiment is shown. Detailed Implementation
[0012] The solution provided by the present invention will now be described with reference to the accompanying drawings.
[0013] Figure 1 This diagram illustrates a flowchart of a method for detecting chronic diseases according to one embodiment. It is understood that this method can be executed by any device, apparatus, platform, or cluster of devices with computing and processing capabilities. Figure 1 As shown, the method includes: Step 100: Obtain medical testing data; including imaging data, physiological indicator data, electronic medical record data, and gene sequencing data. Step 102: Based on medical testing data, determine the feature matrix of chronic diseases; Step 104: Input the feature matrix of chronic diseases into the preset chronic disease detection model to obtain the preliminary detection results of chronic diseases; Step 106: Input the feature matrix of chronic diseases into the preset case database to obtain multiple related test results; among them, the comprehensive similarity between the related test results and the preliminary test results is greater than the first preset threshold. Step 108: Based on the preliminary test results and multiple related test results, determine the final test results.
[0014] In this embodiment, firstly, multi-dimensional medical testing data of the patient is acquired through a standardized data interface. This medical testing data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data. Based on this medical testing data, a feature matrix for chronic diseases is determined. On one hand, the feature matrix is input into a preset chronic disease detection model to obtain preliminary detection results for chronic diseases, including disease type (e.g., diabetes, coronary heart disease), probability of disease (e.g., 92.5%), and disease stage (e.g., early stage). On the other hand, the same feature matrix is input into a preset case database (containing 150,000+ desensitized clinical cases), and multiple relevant detection results with a comprehensive similarity greater than a first preset threshold (92%) are selected to ensure that the relevant cases highly match the current patient in core features (e.g., imaging lesion type, gene locus mutation). Finally, a collaborative analysis is conducted based on the preliminary test results and related test results: if the preliminary results are consistent with the diagnosis of most related cases (e.g., all 3 related cases are "early stage of type 2 diabetes"), the final report is generated directly based on the preliminary results; if there are differences (e.g., 1 related case is "prediabetes"), cross-validation is performed by combining the specific clinical background of related cases (e.g., patient age, comorbidities) with the current patient condition, and after correcting for deviations, the final test results are generated, which include the diagnosed disease, stage, key evidence, and a summary of reference cases.
[0015] In one embodiment of the present invention, determining a feature matrix of chronic diseases based on medical testing data includes: Based on imaging data, physiological index data, electronic medical record data, and gene sequencing data, the image feature vector, physiological index feature vector, electronic medical record feature vector, and gene sequencing feature vector are determined sequentially. Based on image feature vectors, physiological indicator feature vectors, electronic medical record feature vectors, and gene sequencing feature vectors, a feature matrix for chronic diseases is constructed. The imaging data includes CT images, fundus optical coherence tomography images, and electrocardiogram images; the physiological indicator data includes blood pressure, blood glucose, and blood lipids; and the electronic medical record data includes medical history records, symptom descriptions, medication history, and family medical history.
[0016] In this embodiment, for image data, it is necessary to first distinguish the characteristics of CT images, fundus optical coherence tomography (OCT) images, and electrocardiogram (ECG) images. CT images are processed by an improved ResNet-34 network to extract spatial features such as ground-glass opacities in the lungs and abnormal organ structures, outputting a 2048-dimensional vector. Fundus OCT images are processed by a lightweight U-Net network to segment the macular region, and then an attention mechanism is used to focus on lesion features such as microaneurysms and hard exudates, generating a 1024-dimensional vector. ECG images are processed by a 1D-CNN network to capture temporal features such as ST-T segment depression and arrhythmia, converting them into a 512-dimensional vector. Finally, the three types of image feature vectors are concatenated to form a 3584-dimensional image feature vector. For physiological indicator data (blood pressure, blood glucose, blood lipids), Z-score standardization is first used to eliminate dimensional differences (e.g., mapping systolic blood pressure mmHg and blood glucose mmol / L to the [0,1] interval), and then feature encoding is performed through a 3-layer fully connected network to highlight abnormal correlations between indicators (e.g., the synergistic effect of hyperglycemia and hyperlipidemia), generating a 256-dimensional physiological indicator feature vector. For electronic medical record data (medical history records, symptom descriptions, medication history, family medical history), semantic word segmentation is performed based on a pre-trained BERT model in the medical field, combined with a Bi-LSTM network to capture semantic associations such as "chronic cough - lung disease" and "polydipsia and polyuria - diabetes", outputting a 512-dimensional electronic medical record feature vector. For gene sequencing data, SNP sites related to chronic diseases (e.g., rs7903146 associated with diabetes) are screened through 1D-CNN, and then dimensionality reduction is performed through an autoencoder to generate a 256-dimensional gene sequencing feature vector. Finally, following the order of image feature vector, physiological indicator feature vector, electronic medical record feature vector, and gene sequencing feature vector, the four types of vectors are stacked in rows to construct a chronic disease feature matrix with a dimension of 4×4608 (each row corresponds to one type of feature vector, and each column corresponds to a single feature dimension). Each element in the matrix is labeled with the data source and feature meaning, which ensures the integrity of multimodal information.
[0017] In one embodiment of the present invention, the preset chronic disease detection model is an improved neural network model. The improved neural network model includes an input layer, a feature fusion subnetwork, a feature optimization subnetwork, and an output layer connected in sequence. The feature fusion subnetwork includes a feature mapping layer, an attention calculation layer, a Transformer encoding layer, and a modality weight layer connected in sequence. The feature optimization subnetwork includes a residual network and a batch normalization layer. The output layer includes at least two fully connected networks and a Sigmoid activation layer.
[0018] In this embodiment, the preset chronic disease detection model is a customized improved neural network model, consisting of an input layer, a feature fusion subnetwork, a feature optimization subnetwork, and an output layer connected in sequence, ensuring efficient processing and detection output of multimodal features. The input layer receives the chronic disease feature matrix and standardizes it into a tensor format suitable for the model through a data interface. The feature fusion subnetwork, as the core of the model, operates according to the process of "feature mapping - attention calculation - encoding integration - weight allocation." The feature mapping layer uses four parallel fully connected branches to uniformly map four types of single-modal features to a 512-dimensional space to eliminate dimensional differences. The attention calculation layer uses an 8-head self-attention mechanism to calculate the intermodal correlation weights (such as the detection correlation between images and physiological indicators). The Transformer encoding layer captures cross-modal global dependencies through a 6-layer encoder. The modality weight layer dynamically adjusts the contribution of each modality using a Sigmoid gating mechanism to enhance the role of high-value features. The feature optimization subnetwork integrates a residual network (3 residual blocks) and a batch normalization layer. The residual connections solve the gradient vanishing problem, and the batch normalization layer stabilizes the training process, together filtering feature redundancy and noise. The output layer contains two parallel fully connected networks (corresponding to disease classification and disease staging tasks, respectively), paired with a sigmoid activation layer, and finally outputs the disease type probability result. In one embodiment of the present invention, the preset chronic disease detection model is trained in the following manner: For each iteration of training the pre-defined chronic disease detection model, the feature matrix of labeled chronic diseases is input into the improved neural network model to obtain the prediction result of the current iteration. Based on the prediction results and the labels of the feature matrix of chronic diseases, obtain the loss for the current iteration of training; Backpropagation is performed based on the loss from the current iteration of training to update the parameters of the improved neural network model, thus completing the current iteration of training.
[0019] In this embodiment, for each training iteration, samples are first extracted from the fully labeled training dataset (which contains over 100,000 desensitized clinical samples covering target chronic diseases such as diabetes and cardiovascular diseases), with each batch consisting of 32 samples. These samples are all labeled, with two categories of labels: one is "disease type labels," using a multi-label annotation method (e.g., a sample corresponding to diabetes is labeled as 1, and non-diabetic samples as 0); the other is "disease stage labels," using a one-hot encoding annotation method (e.g., labels corresponding to different stages such as early and middle stages). Simultaneously, a 46,008-dimensional chronic disease feature matrix corresponding to these samples is extracted, standardized into tensor format by the model input layer, and then input into the improved neural network model. The model first completes feature processing through a feature fusion subnetwork and a feature optimization subnetwork, and then outputs the prediction result of the current iteration from the output layer. This result includes the probability of each disease type and the probability of each disease stage. Subsequently, combining the prediction results and sample labels, a hybrid loss function is used to calculate the loss value for the current iteration of training: the disease classification loss is calculated using the binary cross-entropy algorithm, and the disease staging loss is calculated using the focus loss algorithm. The two losses are superimposed with a weight ratio of 0.6 to 0.4 to quantify the degree of deviation between the prediction results and the true labels. Finally, based on the calculated loss value, a backpropagation operation is performed using an adaptive momentum weight decay optimizer (with an initial learning rate set to 1 / 100,000 and a weight decay coefficient set to 5 / 100,000): starting from the model output layer, the gradients of each parameter (including convolutional kernel weights, fully connected layer weights, etc.) are calculated backward along each layer of the network, and the parameters are updated in the direction of gradient descent; at the same time, gradient clipping (with a clipping threshold set to 1.0) is used to avoid the gradient explosion problem. In one embodiment of the present invention, after each iteration of training, it is determined whether the improved neural network model has converged; If so, end the model training and save the trained improved neural network model as a preset chronic disease detection model; If not, continue with the next iteration of training.
[0020] In this embodiment, after each iteration of training, a dual metric is used to determine whether the improved neural network model has converged, ensuring stable model performance and satisfactory generalization ability. First, the average loss value of the validation dataset (independent of the training dataset, containing 20,000 anonymized samples) is calculated. If, within five consecutive iterations, the decrease in the average loss value of the validation set is less than one ten-thousandth (e.g., from 0.1230 to 0.1229), and there is no significant rebound; simultaneously, the model's detection accuracy on the validation set (e.g., accuracy in diabetes detection, accuracy in cardiovascular disease staging) fluctuates by no more than one percent within five consecutive iterations, then the model is considered converged. If so, model training is immediately terminated, and all parameters of the current model (including weights and biases of each sub-network) are saved in binary format to a dedicated storage directory, generating a "preset chronic disease detection model" file, and a training log (including iteration count, final loss value, and validation set accuracy) is recorded. If not, the learning rate is adjusted according to a cosine annealing strategy (e.g., from one in ten thousand to one in twelve thousand), and the next iteration of training continues until the convergence condition is met.
[0021] In one embodiment of the present invention, the loss of the current iteration of training is determined by the following formula:
[0022] In the formula, The loss for the current iteration of training, The number of target chronic disease categories, Labels for the feature matrix of chronic diseases. This is the predicted result.
[0023] In one embodiment of the present invention, the overall similarity is determined by the following formula:
[0024] In the formula, To assess overall similarity, A feature matrix for chronic diseases. For the first case in the case database Example: Feature matrix of historical cases For divergence, This represents the joint probability distribution of the current patient's multimodal features. For the first For example, the joint probability distribution of multimodal features of historical cases. The total number of categories for multimodal features. Assign modal index number, For the first Modal weighting coefficients, For the current patient The eigenvectors of a mode, For the first Example of historical case number The eigenvectors of a mode.
[0025] In this embodiment, the comprehensive similarity is used to determine the matching degree between the current patient and historical cases, and the divergence is used to quantify the difference in feature distribution between the current patient and historical cases. It is a calculation operator for distribution calibration. The weight coefficient of the m-th modality is used to reflect the contribution of different modalities to the diagnosis (e.g., 0.42 in the diagnosis of diabetes, i.e., the image modality weight is 42%). This invention introduces KL divergence into the case matching model for the first time. This divergence quantifies the difference in feature distribution between the current patient and historical cases, and then maps it to a calibration factor in the (0,1] interval through the natural exponential function. If the feature distributions of the two cases are highly consistent (e.g., both are "fasting blood glucose consistently 8~10 mmol / L"), the KL divergence approaches 0, the calibration factor approaches 1, and the weighted similarity is not corrected; if the distribution differences are large (e.g., one case is "constantly high fasting blood glucose", and the other is "temporarily high postprandial blood glucose"), the KL divergence increases, the calibration factor drops below 0.6, and the final similarity value is reduced.
[0026] In one embodiment of the present invention, when the detection duration of the preset chronic disease detection model is greater than a second preset threshold, the image feature vector, physiological indicator feature vector, electronic medical record feature vector, and gene sequencing feature vector are arranged from high to low according to a preset priority to obtain the sorted feature matrix of chronic diseases. The feature matrix of the sorted chronic diseases is reduced in dimensionality to obtain the feature matrix of the chronic diseases after dimensionality reduction; and the feature matrix of the chronic diseases after dimensionality reduction is used as the input of the preset chronic disease detection model.
[0027] In this embodiment, when the detection time of the preset chronic disease detection model exceeds the second preset threshold (this threshold is set based on the timeliness requirements of clinical diagnosis and treatment, with a default value of 30 seconds, which can be dynamically adjusted according to hardware computing power), the feature vector priority sorting process is initiated first. The priority division here is determined based on the clinical value and information density of chronic disease diagnosis: the first priority is physiological indicator feature vectors, as they directly reflect the abnormal metabolic state of the body and are a key basis for preliminary screening of chronic diseases, with low data dimensionality and low computational cost; the second priority is image feature vectors, which can provide intuitive evidence of organ structural and functional lesions and play an important supporting role in disease classification; the third priority is electronic medical record feature vectors, which can supplement clinical background information and help eliminate interference from "same symptoms, different diseases"; the fourth priority is gene sequencing feature vectors, which, although valuable for disease risk prediction, have high data dimensionality and are computationally time-consuming, and are therefore listed as the lowest priority. The four types of feature vectors are stacked sequentially according to this order to generate a sorted chronic disease feature matrix, ensuring that high-value, low-time-consuming features participate in model calculation first. Subsequently, the sorted feature matrix undergoes targeted dimensionality reduction. First, principal components with a contribution rate ≥95% are extracted using PCA (preserving core information while reducing dimensionality redundancy). This results in a dimensionality-reduced chronic disease feature matrix. For example, the original matrix has a dimension of 4×4608, which can be compressed to approximately 4×1200 after dimensionality reduction, a reduction of 74%. Finally, the dimensionality-reduced chronic disease feature matrix is used as the input to a pre-defined chronic disease detection model. This approach improves the model's computational efficiency while controlling accuracy loss.
[0028] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0029] According to another embodiment, the present invention provides a detection device for chronic diseases. Figure 2 A schematic block diagram of a chronic disease detection device according to one embodiment is shown. It will be understood that this device can be implemented by any apparatus, device, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the device includes: an acquisition unit 200, a first data processing unit 202, a second data processing unit 204, a third data processing unit 206, and a fourth data processing unit 208. The main functions of each component are as follows: The acquisition unit 200 is configured to acquire medical test data; wherein the medical test data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data. The first data processing unit 202 is configured to determine a feature matrix of chronic diseases based on the medical test data; The second data processing unit 204 is configured to input the feature matrix of the chronic disease into a preset chronic disease detection model to obtain preliminary detection results of the chronic disease. The third data processing unit 206 is configured to input the feature matrix of the chronic disease into a preset case database to obtain multiple related detection results; wherein the overall similarity between the related detection results and the preliminary detection results is greater than a first preset threshold. The fourth data processing unit 208 is configured to determine the final detection result based on the preliminary detection result and multiple related detection results.
[0030] In one embodiment of the present invention, the first data processing unit 202 is configured to perform the following operations: Based on the image data, the physiological index data, the electronic medical record data, and the gene sequencing data, the image feature vector, the physiological index feature vector, the electronic medical record feature vector, and the gene sequencing feature vector are determined sequentially. Based on the image feature vector, the physiological indicator feature vector, the electronic medical record feature vector, and the gene sequencing feature vector, a feature matrix of the chronic disease is constructed. The image data includes CT images, fundus optical coherence tomography images, and electrocardiogram images; the physiological indicator data includes blood pressure, blood glucose, and blood lipids; and the electronic medical record data includes medical history records, symptom descriptions, medication history, and family medical history.
[0031] In one embodiment of the present invention, the preset chronic disease detection model is an improved neural network model, which includes an input layer, a feature fusion subnetwork, a feature optimization subnetwork, and an output layer connected in sequence; the feature fusion subnetwork includes a feature mapping layer, an attention calculation layer, a Transformer encoding layer, and a modality weight layer connected in sequence; the feature optimization subnetwork includes a residual network and a batch normalization layer; and the output layer includes at least two fully connected networks and a Sigmoid activation layer.
[0032] In one embodiment of the present invention, the apparatus further includes a fourth data processing unit, the fourth data processing unit being configured to perform the following operations: For each iteration of training the preset chronic disease detection model, the feature matrix of the labeled chronic disease is input into the improved neural network model to obtain the prediction result of the current iteration. Based on the prediction results and the labels of the feature matrix of chronic diseases, obtain the loss for the current iteration of training; Backpropagation is performed based on the loss of the current iteration of training to update the parameters of the improved neural network model in order to complete the current iteration of training; After each iteration of training, it is determined whether the improved neural network model has converged; If so, end the model training and save the trained improved neural network model as the preset chronic disease detection model; If not, continue with the next iteration of training.
[0033] In one embodiment of the present invention, the loss of the current iteration of training is determined by the following formula:
[0034] In the formula, The loss for the current iteration of training, The number of target chronic disease categories, The labels are the feature matrix of the chronic disease. The prediction result is given.
[0035] In one embodiment of the present invention, the comprehensive similarity is determined by the following formula:
[0036] In the formula, To assess overall similarity, This is the feature matrix of the chronic disease. For the first case in the case database Example: Feature matrix of historical cases For divergence, This represents the joint probability distribution of the current patient's multimodal features. For the first For example, the joint probability distribution of multimodal features of historical cases. The total number of categories for multimodal features. Assign modal index number, For the first Modal weighting coefficients, For the current patient The eigenvectors of a mode, For the first Example of historical case number The eigenvectors of a mode.
[0037] In one embodiment of the present invention, the apparatus further includes a fifth data processing unit, the fifth data processing unit being configured to perform the following operations: When the detection duration of the preset chronic disease detection model exceeds the second preset threshold, the image feature vector, the physiological indicator feature vector, the electronic medical record feature vector, and the gene sequencing feature vector are arranged from high to low according to the preset priority to obtain the sorted chronic disease feature matrix. The feature matrix of the sorted chronic diseases is reduced in dimensionality to obtain the feature matrix of the chronic diseases after dimensionality reduction; and the feature matrix of the chronic diseases after dimensionality reduction is used as the input of the preset chronic disease detection model.
[0038] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.
[0039] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.
[0040] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0041] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.
[0042] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for detecting chronic diseases, characterized in that, include: Acquire medical testing data; wherein, the medical testing data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data; Based on the aforementioned medical testing data, a feature matrix for chronic diseases is determined; The feature matrix of the chronic disease is input into a preset chronic disease detection model to obtain preliminary detection results of the chronic disease; The feature matrix of the chronic disease is input into a preset case database to obtain multiple related detection results; wherein the overall similarity between the related detection results and the preliminary detection results is greater than a first preset threshold. Based on the preliminary test results and multiple related test results, the final test result is determined.
2. The method according to claim 1, characterized in that, The determination of the feature matrix of chronic diseases based on the medical test data includes: Based on the image data, the physiological index data, the electronic medical record data, and the gene sequencing data, the image feature vector, the physiological index feature vector, the electronic medical record feature vector, and the gene sequencing feature vector are determined sequentially. Based on the image feature vector, the physiological indicator feature vector, the electronic medical record feature vector, and the gene sequencing feature vector, a feature matrix of the chronic disease is constructed. The image data includes CT images, fundus optical coherence tomography images, and electrocardiogram images; the physiological indicator data includes blood pressure, blood glucose, and blood lipids; and the electronic medical record data includes medical history records, symptom descriptions, medication history, and family medical history.
3. The method according to claim 2, characterized in that, The preset chronic disease detection model is an improved neural network model, which includes an input layer, a feature fusion subnetwork, a feature optimization subnetwork, and an output layer connected in sequence. The feature fusion subnetwork includes a feature mapping layer, an attention calculation layer, a Transformer encoding layer, and a modality weight layer connected in sequence. The feature optimization subnetwork includes a residual network and a batch normalization layer. The output layer includes at least two fully connected networks and a Sigmoid activation layer.
4. The method according to claim 3, characterized in that, The preset chronic disease detection model is trained in the following way: For each iteration of training the preset chronic disease detection model, the feature matrix of the labeled chronic disease is input into the improved neural network model to obtain the prediction result of the current iteration. Based on the prediction results and the labels of the feature matrix of chronic diseases, obtain the loss for the current iteration of training; Backpropagation is performed based on the loss of the current iteration of training to update the parameters of the improved neural network model in order to complete the current iteration of training; After each iteration of training, it is determined whether the improved neural network model has converged; If so, end the model training and save the trained improved neural network model as the preset chronic disease detection model; If not, continue with the next iteration of training.
5. The method according to claim 4, characterized in that, The loss for the current iteration of training is determined by the following formula: In the formula, The loss for the current iteration of training, The number of target chronic disease categories, The labels are the feature matrix of the chronic disease. The prediction result is given.
6. The method according to claim 1, characterized in that, The overall similarity is determined by the following formula: In the formula, To assess overall similarity, This is the feature matrix of the chronic disease. For the first case in the case database Example: Feature matrix of historical cases For divergence, This represents the joint probability distribution of the current patient's multimodal features. For the first For example, the joint probability distribution of multimodal features of historical cases. The total number of categories for multimodal features. Assign modal index number, For the first Modal weighting coefficients, For the current patient The eigenvectors of a mode, For the first Example of historical case number The eigenvectors of a mode.
7. The method according to claim 2, characterized in that, Also includes: When the detection duration of the preset chronic disease detection model exceeds the second preset threshold, the image feature vector, the physiological indicator feature vector, the electronic medical record feature vector, and the gene sequencing feature vector are arranged from high to low according to the preset priority to obtain the sorted chronic disease feature matrix. The feature matrix of the sorted chronic diseases is reduced in dimensionality to obtain the feature matrix of the chronic diseases after dimensionality reduction; and the feature matrix of the chronic diseases after dimensionality reduction is used as the input of the preset chronic disease detection model.
8. A device for detecting chronic diseases, characterized in that, include: The acquisition unit is configured to acquire medical testing data; wherein the medical testing data includes imaging data, physiological indicator data, electronic medical record data, and gene sequencing data. The first data processing unit is configured to determine a feature matrix of chronic diseases based on the medical test data; The second data processing unit is configured to input the feature matrix of the chronic disease into a preset chronic disease detection model to obtain preliminary detection results of the chronic disease. The third data processing unit is configured to input the feature matrix of the chronic disease into a preset case database to obtain multiple related detection results; wherein the overall similarity between the related detection results and the preliminary detection results is greater than a first preset threshold. The fourth data processing unit is configured to determine the final detection result based on the preliminary detection result and multiple related detection results.
9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer program is stored thereon, which, when executed in the computer, causes the computer to perform the method of any one of claims 1-7.