Medical auxiliary diagnosis method and system based on convolutional neural network

Abstract

The invention discloses a medical auxiliary diagnosis method and system based on a convolutional neural network, and relates to the field of medical big data analysis, and the method comprises the steps: collecting various types of medical data, which comprise medical image data, electronic medical record text data and vital sign monitoring data; constructing a convolutional neural network model with a multi-branch structure; preparing a training data set containing labeled samples, wherein labeled information comprises disease diagnosis results and disease severity grades; training the convolutional neural network model by using the training data set, dynamically adjusting the learning rate in the training process by adopting a stochastic gradient descent optimization algorithm and combining a learning rate attenuation strategy, and introducing a regularization method; inputting to-be-diagnosed medical data into the trained convolutional neural network model, and outputting disease diagnosis probability distribution and disease severity prediction results through forward propagation calculation of the model; and generating a medical auxiliary diagnosis report according to the output prediction result.

Description

Technical Field

[0001] The present invention relates to the technical field of medical big data analysis, and particularly to a medical auxiliary diagnosis method and system based on a convolutional neural network. Background Art

[0002] With the continuous development of medical technology, various medical devices such as X-ray machines, CT scanners, MRI, etc. are widely used, generating a large amount of medical image data. At the same time, the popularization of electronic medical record systems has also led to a sharp increase in medical data in text form. These rich data provide a basis for the application of data-driven convolutional neural network technology in medical auxiliary diagnosis. The workload of medical diagnosis is large. Therefore, an auxiliary diagnosis technology is needed to improve the diagnosis efficiency and accuracy and reduce the workload of doctors.

[0003] As an important branch of deep learning, convolutional neural networks have achieved great success in fields such as image recognition and speech recognition. It has the ability to automatically extract data features and can effectively process image and sequence data, which makes it have great potential in medical image analysis and medical data processing.

[0004] The cost of annotating and organizing medical data is high, and the data volume is limited, resulting in insufficient model training, affecting the diagnostic accuracy. The diversity and representativeness of the data are insufficient, which may lead to deviations in the actual application of the model. Deep learning models such as CNNs are usually regarded as "black boxes" and it is difficult to explain their decision-making processes, which may affect doctors' trust in the diagnostic results in the medical field; there are multiple words describing the same symptom or disease in medical text data, resulting in a decrease in the matching accuracy of knowledge base-based systems. Although CNNs have made significant progress in the diagnosis of certain diseases, misdiagnosis or missed diagnosis may still occur in complex cases or rare diseases. Integrating deep learning technology with existing medical information systems and clinical workflows is difficult, and problems such as data sharing and privacy protection need to be solved. Summary of the Invention

[0005] To solve the above technical problems, a medical auxiliary diagnosis method and system based on a convolutional neural network are provided. To achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] A medical auxiliary diagnosis method and system based on a convolutional neural network, comprising:

[0007] Collect various types of medical data, where the medical data includes medical image data, electronic medical record text data, and vital sign monitoring data;

[0008] Construct a convolutional neural network model with a multi-branch structure, where the model includes an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch;

[0009] Prepare a training dataset containing labeled samples, where the labeling information includes disease diagnosis results and severity grading of the condition.

[0010] Use the training dataset to train the constructed convolutional neural network model. Adopt the stochastic gradient descent optimization algorithm, combine with the learning rate decay strategy, dynamically adjust the learning rate during the training process, and at the same time introduce the regularization method.

[0011] Input the medical data to be diagnosed into the trained convolutional neural network model. Through the forward propagation calculation of the model, output the disease diagnosis probability distribution and the prediction result of the severity of the condition.

[0012] Generate a medical auxiliary diagnosis report according to the output prediction result. The content of the report includes disease diagnosis, diagnostic confidence, and corresponding condition analysis and suggestions.

[0013] Preferably, the collecting various types of medical data, where the medical data includes medical image data, electronic medical record text data, and vital sign monitoring data specifically includes:

[0014] Perform denoising processing on the collected medical image data. Use the adaptive median filtering algorithm to remove salt-and-pepper noise in the image, and at the same time enhance the contrast of the image through histogram equalization.

[0015] Perform word segmentation, stop word removal, and stemming operations on the electronic medical record text data.

[0016] Perform normalization processing on the vital sign monitoring data, mapping vital sign data in different ranges to a unified numerical interval.

[0017] Preferably, the constructing a convolutional neural network model with a multi-branch structure, where the model includes an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch specifically includes:

[0018] The image feature extraction branch adopts an improved ResNet architecture, and introduces an attention mechanism module on the basis of the original residual block to enhance the attention to key features in medical images.

[0019] The text feature extraction branch adopts a structure based on a bidirectional long short-term memory network, combines with a pre-trained word vector model, and extracts features from the electronic medical record text.

[0020] The vital sign feature extraction branch adopts a multi-layer perceptron structure to perform feature transformation on the normalized vital sign data.

[0021] Fuse the features extracted from the three branches. Adopt a method that combines feature concatenation and weighted summation to perform weighted combination of the features of different branches through learnable weight parameters.

[0022] Preferably, the preparation includes training data sets with labeled samples. The annotation information includes disease diagnosis results and severity grading of the condition, specifically including:

[0023] For different types of medical images, clearly label whether a specific disease exists and accurately label the specific location of the disease in the body part;

[0024] When there are multiple diseases in the image, each disease needs to be accurately labeled;

[0025] Grade according to the size, quantity, and morphological characteristics of the lesions in the image; Consider the degree of influence of the disease on organ function for grading;

[0026] Extract the clear disease diagnosis names given by doctors from the text descriptions in the electronic medical records. For some undetermined diagnoses and suspected diseases, make corresponding labels; Record the complications that the patient has at the same time.

[0027] Preferably, use the training data set to train the constructed convolutional neural network model. Adopt the stochastic gradient descent optimization algorithm, combined with the learning rate decay strategy, dynamically adjust the learning rate during the training process, and at the same time introduce regularization methods, specifically including:

[0028] Assign initial values to all learnable parameters of the convolutional neural network model, determine the hyperparameters required for training, and divide the prepared training data set into a training set and a validation set;

[0029] Randomly extract a batch of sample data from the training data set and input it into the convolutional neural network model;

[0030] The sample data is sequentially calculated through the convolutional layer, pooling layer, and fully connected layer of the model to obtain the predicted output of the model;

[0031] Compare the predicted output of the model with the true label of the sample, calculate the loss value, and according to the loss function, use the backpropagation algorithm to calculate the gradient of the loss function with respect to all learnable parameters of the model; At the same time, also take the derivative of the regularization term and add it to the gradient calculation;

[0032] According to the calculated gradient, use the stochastic gradient descent algorithm to update the parameters of the model; Average the gradients calculated for the current sample data and use this average gradient to update the model parameters to reduce the variance of the gradients;

[0033] According to the preset learning rate decay strategy, the learning rate is dynamically adjusted during the training process. When the decay condition is met, the current learning rate is updated and the new learning rate is used in subsequent parameter updates;

[0034] After each epoch ends, the model is evaluated using the validation set, and the loss and accuracy on the validation set are calculated; the above steps are repeated until the preset number of training epochs is reached.

[0035] Preferably, the combination of the learning rate decay strategy, dynamically adjusting the learning rate during the training process, and introducing the regularization method specifically includes:

[0036] The loss function after adding the regularization term:

[0037]

[0038] In the formula, L(θ) is the loss function after adding the regularization term, θ is all the learnable parameters of the model, L original is the average cross-entropy loss, λ is the weight of the regularization term, used to control the influence degree of the regularization term in the total loss, w j is the j-th parameter of the model, is the sum of the squares of all parameters.

[0039] Preferably, the combination of the learning rate decay strategy, dynamically adjusting the learning rate during the training process, and introducing the regularization method specifically includes:

[0040] Learning rate decay strategy, fixed-step decay, every certain number of training epochs step_size, multiply the learning rate by a fixed decay factor γ:

[0041]

[0042] In the formula, α is the learning rate, α0 is the initial learning rate, α old is the learning rate of the previous round, and epoch is the current number of training epochs.

[0043] Preferably, inputting the medical data to be diagnosed into the trained convolutional neural network model, and through the forward propagation calculation of the model, outputting the disease diagnosis probability distribution and the prediction result of the disease severity specifically includes:

[0044] Through the forward propagation calculation of the model, the disease diagnosis probability distribution and the prediction result of the disease severity are obtained;

[0045] Using the softmax function to convert the output into a probability distribution, representing the probabilities of the patient having different diseases; using the mean squared error to output the predicted value of the disease severity; using the softmax function to output the classification probability of the severity;

[0046] For the disease diagnosis probability distribution, find the disease category with the highest probability as the possible diagnosis result. For the prediction result of the disease severity, clarify its corresponding grading and specific value;

[0047] According to the results obtained from the analysis, clearly list the possible disease diagnosis names in the report. If there are multiple possible diagnoses, arrange them in descending order of diagnosis confidence.

[0048] Furthermore, a medical auxiliary diagnosis system based on a convolutional neural network is used to implement a medical auxiliary diagnosis method based on a convolutional neural network as described above, including:

[0049] Data acquisition and preprocessing module: used to collect various medical data, including medical image data, electronic medical record text data, and vital sign monitoring data;

[0050] Convolutional neural network model construction module: construct a convolutional neural network model with multi-scale feature fusion to extract medical data features at different levels;

[0051] Model training module: prepare a training data set containing labeled samples, and the labeling information includes disease diagnosis results and disease severity grading; use the stochastic gradient descent optimization algorithm with an adaptive learning rate to train the convolutional neural network model;

[0052] Medical auxiliary diagnosis module: input the medical data to be diagnosed into the trained convolutional neural network model, and through the forward propagation calculation of the model, output the disease diagnosis probability distribution and the prediction result of the disease severity; generate a medical auxiliary diagnosis report according to the output result.

[0053] Optionally, the medical auxiliary diagnosis module specifically includes:

[0054] The disease diagnosis probability distribution converts the model output into probability values through the Softmax function;

[0055] Compare and analyze the generated medical auxiliary diagnosis report with the actual diagnosis results, and adjust the parameters and structure of the convolutional neural network model according to the comparison results to optimize the model performance

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] The present invention proposes to collect various types of data such as medical image data, electronic medical record text data, and vital sign monitoring data; multi-source data fusion can provide more comprehensive and rich patient information, which helps to more accurately capture disease characteristics and changes in the condition, thereby improving the accuracy and reliability of diagnosis; a convolutional neural network model with a multi-branch structure is constructed, including an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch respectively. Different types of data have different characteristics and structures, and the multi-branch structure can perform specialized feature extraction and processing according to the characteristics of each data type; the stochastic gradient descent optimization algorithm is adopted and combined with a learning rate decay strategy to dynamically adjust the learning rate during training. In the initial stage of training, a larger learning rate can enable the model to quickly converge to near the optimal solution; as training progresses, gradually reducing the learning rate can prevent the model from oscillating near the optimal solution, thus converging to the optimal solution more stably; the model outputs the disease diagnosis probability distribution and the prediction result of the disease severity. The disease diagnosis probability distribution can provide doctors with the confidence of each possible disease, helping doctors to more comprehensively understand the possibilities of various diseases, rather than just giving a single diagnosis result. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flowchart of a medical auxiliary diagnosis method based on a convolutional neural network;

[0059] Figure 2 is an internal framework diagram of a medical auxiliary diagnosis system based on a convolutional neural network. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0061] Referring to Figure 1 as shown, a medical auxiliary diagnosis method based on a convolutional neural network includes:

[0062] Collect various types of medical data, where the medical data includes medical image data, electronic medical record text data, and vital sign monitoring data;

[0063] Construct a convolutional neural network model with a multi-branch structure, and the model includes an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch;

[0064] Prepare a training data set containing labeled samples, and the labeling information includes disease diagnosis results and disease severity grading;

[0065] Use the training dataset to train the constructed convolutional neural network model. Adopt the stochastic gradient descent optimization algorithm, combine the learning rate decay strategy, dynamically adjust the learning rate during the training process, and introduce the regularization method at the same time;

[0066] Input the medical data to be diagnosed into the trained convolutional neural network model. Through the forward propagation calculation of the model, output the disease diagnosis probability distribution and the prediction result of the disease severity;

[0067] Generate a medical auxiliary diagnosis report according to the output prediction result. The report content includes disease diagnosis, diagnostic confidence, and corresponding disease analysis and suggestions.

[0068] It should be noted that when training the convolutional neural network (CNN) model, the learning rate decay strategy is a commonly used technique for dynamically adjusting the learning rate during the training process. Among them, meeting the decay conditions specifically includes:

[0069] Based on the number of training epochs:

[0070] Fixed interval: Every time a certain number of epochs pass, the learning rate decays once. For example, multiply the learning rate by a decay factor every 10 epochs;

[0071] Exponential decay: The learning rate decays exponentially over time, and the learning rate can be dynamically adjusted according to a preset time constant or decay rate;

[0072] Based on validation performance

[0073] Validation loss stagnation: When the loss value on the validation set does not decrease significantly for several consecutive epochs, it indicates that the model may enter a convergence plateau. At this time, the learning rate can be reduced to make the optimization process more refined, which helps the model to further approach the optimal solution;

[0074] Validation accuracy stagnation: Similarly, if the accuracy on the validation set does not increase significantly for multiple epochs, it can also be used as a condition to trigger the learning rate decay.

[0075] Refer to Figure 2 As shown, a medical auxiliary diagnosis system based on a convolutional neural network includes:

[0076] Data acquisition and preprocessing module: Used to collect various medical data, including medical image data, electronic medical record text data, and vital sign monitoring data;

[0077] Convolutional neural network model construction module: Construct a convolutional neural network model with multi-scale feature fusion to extract medical data features at different levels;

[0078] Model training module: Prepare a training data set containing labeled samples, where the labeling information includes disease diagnosis results and severity grading of the condition; Use the stochastic gradient descent optimization algorithm with an adaptive learning rate to train the convolutional neural network model;

[0079] Medical auxiliary diagnosis module: Input the medical data to be diagnosed into the trained convolutional neural network model. Through the forward propagation calculation of the model, output the disease diagnosis probability distribution and the prediction result of the disease severity; Generate a medical auxiliary diagnosis report according to the output result.

[0080] Example: Step 1: Collect various types of medical data

[0081] Medical image data: Cooperate with the imaging departments of major hospitals to collect X-ray, CT, and MRI images of different parts (such as lungs, brains, bones, etc.). For example, collect 5000 cases of lung CT images, covering cases of normal lungs, pneumonia, lung cancer, etc. Classify and store these image data to prepare for subsequent processing and analysis.

[0082] Electronic medical record text data: Extract the medical record information of patients from the hospital's electronic medical record system, including symptom descriptions, past medical histories, examination results, etc.; Collect about 8000 electronic medical records, covering various disease types. Conduct a preliminary screening and sorting of the electronic medical records to remove duplicate, invalid, or incomplete records.

[0083] Vital sign monitoring data: Obtain the vital sign data of patients, such as heart rate, blood pressure, body temperature, respiratory rate, etc.; Data can be collected from the hospital's monitoring equipment or wearable devices. A total of 6000 groups of vital sign data are collected, and the data is sorted in a time series to ensure the continuity and accuracy of the data.

[0084] Step 2: Build a convolutional neural network model with a multi-branch structure

[0085] Image feature extraction branch: Use an improved ResNet architecture as the basis for image feature extraction. On the basis of ResNet, add an attention mechanism module, such as the SE (Squeeze-and-Excitation) module, to enhance the model's attention to key features in the image. For example, for lung CT images, the model can focus more on the lesion areas of the lungs. The input of this branch is medical image data, and after multiple layers of convolution, pooling, and activation function processing, the feature vector of the image is output.

[0086] Text feature extraction branch: Use a pre-trained model based on BERT to extract features from the electronic medical record text. After tokenizing the text data, input it into the BERT model to obtain the semantic features of the text, and then reduce the dimensionality of the feature vector through a fully connected layer to obtain text features suitable for subsequent fusion.

[0087] Vital sign feature extraction branch: Construct a simple multi-layer perceptron (MLP) network to process vital sign monitoring data. After normalizing the vital sign data, input it into the MLP. Through the calculation of multiple layers of neurons, output the feature representation of the vital signs.

[0088] Feature fusion: Concatenate the feature vectors extracted from the above three branches, and then perform feature fusion through a fully connected layer to obtain the final comprehensive feature vector. Then input the comprehensive feature vector into the output layer to output the disease diagnosis probability distribution and the prediction result of the disease severity.

[0089] Step 3: Prepare a training data set containing labeled samples

[0090] Disease diagnosis result annotation: Organize professional medical experts to annotate the collected medical data for disease diagnosis; for medical image data, experts judge whether the patient has a certain disease according to the image features, such as pneumonia, lung cancer, etc.; for electronic medical record text data, experts give a diagnosis conclusion according to the symptom descriptions and examination results in the medical records; for vital sign monitoring data, combine other data and clinical experience for diagnosis annotation.

[0091] Disease severity grading: Similarly, medical experts grade the disease severity according to the disease diagnosis criteria and clinical experience. For example, for pneumonia patients, it is divided into three grades: mild, moderate, and severe; for cancer patients, it is graded according to the stage of the tumor; finally, a training data set containing 10,000 labeled samples is formed.

[0092] Step 4: Use the training data set to train the constructed convolutional neural network model

[0093] Hyperparameter setting: Set the batch size to 32, the number of training epochs to 50, and the initial learning rate to 0.001; adopt the stochastic gradient descent (SGD) optimization algorithm, combined with the learning rate strategy of exponential decay, and decay the learning rate to 0.1 times the original every 10 training epochs. At the same time, introduce the L2 regularization method, and set the regularization coefficient to 0.0001.

[0094] Training process: Divide the training data set into a training set and a validation set with a ratio of 8:2. In each training epoch, randomly extract a batch of sample data from the training set and input it into the model for forward propagation calculation to obtain the predicted output of the model. Compare the predicted output with the true label of the sample to calculate the loss value. Use the backpropagation algorithm to calculate the gradient of the loss function with respect to all learnable parameters of the model, and update the parameters by combining the gradient of the regularization term. After each training epoch, use the validation set to evaluate the model, and record the loss and accuracy on the validation set.

[0095] Step 5: Input the medical data to be diagnosed into the trained convolutional neural network model

[0096] Data preprocessing: Perform the same preprocessing operations on the medical data to be diagnosed as on the training data. For medical image data, perform denoising, enhancement, and size adjustment; for electronic medical record text data, perform word segmentation, stop word removal, etc.; for vital sign monitoring data, perform normalization processing.

[0097] Model prediction: Input the preprocessed data to be diagnosed into the trained convolutional neural network model. Through the forward propagation calculation of the model, obtain the disease diagnosis probability distribution and the prediction result of the disease severity. For example, for a lung CT image to be diagnosed and the corresponding electronic medical record and vital sign data, the model outputs that the probability of the patient having pneumonia is 0.8 and the disease severity is moderate.

[0098] Step 6: Generate a medical assistant diagnosis report based on the output prediction result

[0099] Disease diagnosis: According to the disease diagnosis probability distribution output by the model, select the disease with the highest probability as the main diagnosis result. For example, if the probability of pneumonia output by the model is the highest, clearly diagnose pneumonia in the report.

[0100] Diagnostic confidence: Write the diagnosis probability of the disease as the diagnostic confidence into the report. In the above example, the diagnostic confidence is 80%.

[0101] Disease analysis: Combine the medical knowledge base and clinical experience to analyze the disease condition. For example, for pneumonia patients, analyze the possible causes, symptoms, and impacts on the body.

[0102] Suggestions: Based on the disease analysis, give corresponding suggestions. For example, for mild pneumonia patients, suggest rest, drink more water, and appropriately use antibiotics for treatment; for moderate pneumonia patients, suggest hospitalization for observation and treatment, etc. Finally, organize these contents into a detailed medical assistant diagnosis report for doctors' reference.

[0103] The usage process of the present invention is as follows:

[0104] Step 1: Collect various types of medical data, where the medical data includes medical image data, electronic medical record text data, and vital sign monitoring data;

[0105] Step 2: Perform denoising on the collected medical image data. Use the adaptive median filtering algorithm to remove the salt-and-pepper noise in the image, and at the same time enhance the contrast of the image through histogram equalization;

[0106] Step 3: Perform word segmentation, stop word removal, and stemming on the electronic medical record text data;

[0107] Step 4: Normalize the vital sign monitoring data to map vital sign data in different ranges to a unified numerical interval;

[0108] Step 5: The image feature extraction branch adopts an improved ResNet architecture, and introduces an attention mechanism module on the basis of the original residual block to enhance the attention to key features in medical images;

[0109] Step 6: The text feature extraction branch adopts a structure based on a bidirectional long short-term memory network, and combines a pre-trained word vector model to extract features from the electronic medical record text;

[0110] Step 7: The vital sign feature extraction branch adopts a multi-layer perceptron structure to perform feature transformation on the normalized vital sign data;

[0111] Step 8: Fuse the features extracted by the three branches, adopt a combination of feature splicing and weighted summation, and perform weighted combination of the features of different branches through learnable weight parameters;

[0112] Step 9: Prepare a training data set containing labeled samples, and the labeling information includes disease diagnosis results and disease severity grading;

[0113] Step 10: For different types of medical images, clarify whether a specific disease exists and accurately label the specific location of the disease in the body part;

[0114] Step 11: When there are multiple diseases in the image, each disease needs to be accurately labeled;

[0115] Step 12: Grade according to the size, quantity, and morphological characteristics of the lesions in the image; consider the degree of influence of the disease on organ function for grading;

[0116] Step 13: Extract the clear disease diagnosis names given by doctors from the text descriptions of the electronic medical records, and make corresponding labels for some unclear diagnoses and suspected diseases; record the complications that the patient has at the same time;

[0117] Step 14: Assign initial values to all learnable parameters of the convolutional neural network model, determine the hyperparameters required for training, and divide the prepared training data set into a training set and a validation set;

[0118] Step 15: Randomly extract a batch of sample data from the training data set and input it into the convolutional neural network model;

[0119] Step 16: The sample data is sequentially calculated through the convolutional layer, pooling layer, and fully connected layer of the model to obtain the predicted output of the model;

[0120] Step 17: Compare the predicted output of the model with the true labels of the samples, calculate the loss value, and according to the loss function, use the backpropagation algorithm to calculate the gradients of the loss function with respect to all learnable parameters of the model; at the same time, also take the derivative of the regularization term and add it to the gradient calculation;

[0121] Step 18: According to the calculated gradients, use the stochastic gradient descent algorithm to update the parameters of the model; average the gradients calculated for the current sample data and use this average gradient to update the model parameters to reduce the variance of the gradients;

[0122] Step 19: According to the pre-set learning rate decay strategy, dynamically adjust the learning rate during the training process. When the decay condition is met, update the current learning rate and use the new learning rate in subsequent parameter updates;

[0123] Step 20: After each epoch, evaluate the model using the validation set, calculate the loss and accuracy on the validation set; repeat the above steps until the preset number of training epochs is reached or other stopping conditions are met;

[0124] Step 21: Extract information related to disease diagnosis and disease severity from the results output by the convolutional neural network model;

[0125] Step 22: For the disease diagnosis probability distribution, find the disease category with the highest probability as the possible diagnosis result. For the prediction result of disease severity, clarify its corresponding grading and specific value;

[0126] Step 23: According to the parsed results, clearly list the possible disease diagnosis names in the report. If there are multiple possible diagnoses, arrange them in descending order of diagnostic confidence;

[0127] Step 24: Explanation of the diagnostic basis, briefly explain the basis for making this diagnosis, that is, the medical data characteristics on which the model is based;

[0128] Step 25: Based on medical expertise and clinical experience, conduct an in-depth analysis of the diagnosis results and disease severity;

[0129] Step 26: According to the disease analysis, provide specific treatment recommendations for doctors, including recommended treatment methods, types and dosages of drugs, treatment courses, and precautions, and generate a medical auxiliary diagnosis report;

[0130] Step 27: Have professional medical staff review the generated medical auxiliary diagnosis report to check the accuracy, reasonableness, and completeness of the report content;

[0131] Step 28: After passing the review, deliver the medical auxiliary diagnosis report to relevant doctors and patients to provide reference for medical decision-making.

[0132] In summary, the advantages of the present invention are as follows: collecting various types of data such as medical image data, electronic medical record text data, and vital sign monitoring data; multi-source data fusion can provide more comprehensive and rich patient information, which helps to more accurately capture disease characteristics and changes in the condition, thereby improving the accuracy and reliability of diagnosis; constructing a convolutional neural network model with a multi-branch structure, respectively setting an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch. Different types of data have different characteristics and structures, and the multi-branch structure can perform specialized feature extraction and processing for the characteristics of each data type; adopting the stochastic gradient descent optimization algorithm combined with the learning rate decay strategy to dynamically adjust the learning rate during training. In the initial stage of training, a larger learning rate can make the model quickly converge near the optimal solution; as training progresses, gradually reducing the learning rate can prevent the model from oscillating near the optimal solution, thereby converging to the optimal solution more stably; the model outputs the disease diagnosis probability distribution and the prediction result of the disease severity. The disease diagnosis probability distribution can provide doctors with the confidence of each possible disease, helping doctors to more comprehensively understand the possibilities of various diseases, rather than just giving a single diagnosis result.

[0133] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A medical auxiliary diagnosis method based on a convolutional neural network, characterized in that, Including: Collecting various types of medical data, where the medical data includes medical image data, electronic medical record text data, and vital sign monitoring data; Constructing a convolutional neural network model with a multi-branch structure, where the model includes an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch; Preparing a training data set containing labeled samples, where the labeling information includes disease diagnosis results and disease severity grading; Using the training data set to train the constructed convolutional neural network model, adopting the stochastic gradient descent optimization algorithm, combining with the learning rate decay strategy, dynamically adjusting the learning rate during the training process, and at the same time introducing regularization methods; Inputting the medical data to be diagnosed into the trained convolutional neural network model, through the forward propagation calculation of the model, outputting the disease diagnosis probability distribution and the disease severity prediction result; Generating a medical auxiliary diagnosis report according to the output prediction result, where the report content includes disease diagnosis, diagnostic confidence, and corresponding disease analysis and suggestions.

2. The medical auxiliary diagnosis method based on a convolutional neural network according to claim 1, wherein The step of collecting various types of medical data, where the medical data includes medical image data, electronic medical record text data, and vital sign monitoring data specifically includes: Performing denoising processing on the collected medical image data, using the adaptive median filtering algorithm to remove salt-and-pepper noise in the image, and at the same time enhancing the contrast of the image through histogram equalization; Performing word segmentation, stop word removal, and stemming extraction operations on the electronic medical record text data; Performing normalization processing on the vital sign monitoring data, mapping vital sign data in different ranges to a unified numerical interval.

3. The medical auxiliary diagnosis method based on a convolutional neural network according to claim 2, wherein The step of constructing a convolutional neural network model with a multi-branch structure, where the model includes an image feature extraction branch, a text feature extraction branch, and a vital sign feature extraction branch specifically includes: The image feature extraction branch adopts an improved ResNet architecture, and an attention mechanism module is introduced on the basis of the original residual block to enhance the attention to key features in medical images; The text feature extraction branch adopts a structure based on a bidirectional long short-term memory network, combines with a pre-trained word vector model, and extracts features from the electronic medical record text; The vital sign feature extraction branch adopts a multi-layer perceptron structure to perform feature transformation on the normalized vital sign data; Fusing the features extracted by the three branches, adopting a combination of feature concatenation and weighted summation, and performing weighted combination of the features of different branches through learnable weight parameters.

4. A medical auxiliary diagnosis method based on a convolutional neural network according to claim 3, characterized in that, The step of preparing a training data set containing labeled samples, where the labeling information includes disease diagnosis results and disease severity grading specifically includes: For different types of medical images, clearly label whether a specific disease exists, and accurately label the specific location of the disease in the body part; When there are multiple diseases in the image, each disease needs to be accurately labeled; Grading according to the size, quantity, and morphological characteristics of the lesions in the image; grading considering the degree of influence of the disease on organ function; Extracting the clear disease diagnosis names given by doctors from the text descriptions of the electronic medical records, making corresponding labels for some unclear diagnoses and suspected disease situations; recording the complications existing in the patient at the same time.

5. The medical auxiliary diagnosis method based on a convolutional neural network according to claim 4, wherein The constructed convolutional neural network model is trained using the training dataset. The stochastic gradient descent optimization algorithm is adopted, combined with a learning rate decay strategy to dynamically adjust the learning rate during the training process. Meanwhile, the regularization method is introduced, which specifically includes: Assign initial values to all learnable parameters of the convolutional neural network model, determine the hyperparameters required for training, and divide the prepared training dataset into a training set and a validation set; Randomly extract a batch of sample data from the training dataset and input it into the convolutional neural network model; The sample data is sequentially calculated through the convolutional layer, pooling layer, and fully connected layer of the model to obtain the predicted output of the model; Compare the predicted output of the model with the true label of the sample, calculate the loss value, and according to the loss function, use the backpropagation algorithm to calculate the gradient of the loss function with respect to all learnable parameters of the model; at the same time, also take the derivative of the regularization term and add it to the gradient calculation; Update the parameters of the model using the stochastic gradient descent algorithm; average the gradients calculated for the current sample data and use the average gradient to update the model parameters to reduce the variance of the gradients; According to the pre-set learning rate decay strategy, dynamically adjust the learning rate during the training process. When the decay condition is met, update the current learning rate and use the new learning rate in subsequent parameter updates; After each epoch ends, use the validation set to evaluate the model and calculate the loss and accuracy on the validation set; repeat the above steps until the preset number of training rounds is reached.

6. A medical auxiliary diagnosis method based on a convolutional neural network according to claim 5, characterized in that, The combination of the learning rate decay strategy to dynamically adjust the learning rate during the training process and the introduction of the regularization method specifically includes: The loss function after adding the regularization term: Where \(L(\theta)\) is the loss function after adding the regularization term, \(\theta\) is all the learnable parameters of the model, \(L\) original is the average cross-entropy loss, \(\lambda\) is the weight of the regularization term, which is used to control the influence degree of the regularization term in the total loss, \(w\) j is the \(j\)-th parameter of the model, is the sum of the squares of all parameters.

7. The medical auxiliary diagnosis method based on a convolutional neural network according to claim 5, wherein The combination of the learning rate decay strategy to dynamically adjust the learning rate during the training process and the introduction of the regularization method specifically includes: Learning rate decay strategy, fixed-step decay. Every certain number of training rounds step_size, multiply the learning rate by a fixed decay factor γ: where α is the learning rate, α0 is the initial learning rate, and α old is the learning rate of the previous round, and epoch is the current training epoch.

8. A medical auxiliary diagnosis method based on a convolutional neural network according to claim 7, characterized in that, Input the medical data to be diagnosed into the trained convolutional neural network model, and through the forward propagation calculation of the model, output the disease diagnosis probability distribution and the prediction result of the disease severity specifically includes: Through the forward propagation calculation of the model, obtain the disease diagnosis probability distribution and the prediction result of the disease severity; Use the softmax function to convert the output into a probability distribution, representing the probabilities of the patient having different diseases; use the mean squared error to output the predicted value of the disease severity; use the softmax function to output the classification probability of the severity; For the disease diagnosis probability distribution, find the disease category with the highest probability as the possible diagnosis result. For the disease severity prediction result, clarify its corresponding grading and specific value; According to the parsed results, clearly list the possible disease diagnosis names in the report. If there are multiple possible diagnoses, arrange them in descending order of diagnosis confidence.

9. A medical auxiliary diagnosis system based on a convolutional neural network, characterized in that, Used to implement a medical auxiliary diagnosis method according to any one of claims 1-8, including: Data acquisition and preprocessing module: used to collect various medical data, including medical image data, electronic medical record text data, and vital sign monitoring data; Convolutional neural network model construction module: constructs a convolutional neural network model with multi-scale feature fusion to extract medical data features at different levels; Model training module: prepares a training data set containing labeled samples, and the labeling information includes disease diagnosis results and disease severity grading; uses the stochastic gradient descent optimization algorithm with an adaptive learning rate to train the convolutional neural network model; Medical assistant diagnosis module: inputs the medical data to be diagnosed into the trained convolutional neural network model, and through the forward propagation calculation of the model, outputs the disease diagnosis probability distribution and the predicted result of the disease severity; generates a medical assistant diagnosis report based on the output result.

10. A medical auxiliary diagnosis system based on a convolutional neural network according to claim 9, characterized in that, The medical assistant diagnosis module specifically includes: The disease diagnosis probability distribution converts the model output into probability values through the Softmax function; Compares and analyzes the generated medical assistant diagnosis report with the actual diagnosis result, and adjusts the parameters and structure of the convolutional neural network model according to the comparison result to optimize the model performance.

Images