Deep learning-based case anomaly detection method and system

Abstract

The invention discloses a case anomaly detection method and system based on deep learning, and relates to the technical field of medical images. Comprising a data collection and preprocessing module, a multi-modal feature extraction module, a feature fusion module, a context information integration module, an anomaly detection model module, a model training and optimization module, a model iteration training and verification module and a real-time monitoring and feedback module. The method further comprises the following steps: step 1, collecting medical image data from different modalities, and carrying out denoising and standardization processing on the medical image data; 2, extracting features of different modal images by using a deep convolutional neural network, and fusing the features to form a comprehensive feature vector; 3, combining the clinical information of the patient with the fusion features to generate a comprehensive feature vector; 4, analyzing the fused feature vector through the trained deep learning model, and outputting an anomaly detection result; 5, the system monitors a new case in real time, automatically prompts abnormity and continuously updates the model according to feedback of a doctor.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a case abnormality detection method and system based on deep learning. Background Art

[0002] Medical image processing combined with artificial intelligence technology is revolutionizing traditional medical practice and improving the accuracy and efficiency of diagnosis. With the advancement of image acquisition technology (such as X-ray, CT and MRI), massive amounts of image data are emerging. These data contain rich clinical information. By applying deep learning and image processing algorithms, AI can quickly analyze and interpret images, identify subtle lesions, and thus assist doctors in making more accurate diagnoses. With the rapid development of medical imaging technology, the use of imaging equipment such as CT, MRI, and X-ray is becoming increasingly popular, and the amount of medical imaging data generated is also increasing dramatically. However, the existing case anomaly detection technology has the following main disadvantages:

[0003] 1. Doctors rely on personal experience and judgment, which results in heavy workload and is easily influenced by subjective factors, leading to misdiagnosis and missed diagnosis;

[0004] 2. Many existing automated detection systems are based on traditional machine learning algorithms, which make it difficult to fully mine the complex information in images, resulting in low accuracy in anomaly detection;

[0005] 3. Existing technologies usually only process single-modality imaging data, ignoring the complementarity between different modalities and unable to effectively handle the analysis of complex cases;

[0006] 4. Existing technologies mostly focus on imaging data, but ignore the patient's clinical history and other pathological information, resulting in insufficient comprehensiveness of diagnostic results.

[0007] Therefore, in response to the above problems, there is an urgent need for a case anomaly detection system based on deep learning to improve data processing efficiency, protect data privacy, and achieve more flexible system integration. Summary of the invention

[0008] The purpose of the present invention is to provide a case anomaly detection method and system based on deep learning to solve the problems raised in the above background technology.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0010] A case abnormality detection system based on deep learning comprises a data collection and preprocessing module, a multimodal feature extraction module, a feature fusion module, a context information integration module, an abnormality detection model module and a model iteration training and verification module; the data collection and preprocessing module receives medical image data of different modalities, and performs denoising and standardization processing on the images through the data preprocessing module; the multimodal feature extraction module respectively sends the preprocessed images into a multi-branch convolutional neural network in the multimodal feature extraction module for feature extraction; the feature fusion module fuses the feature extraction data of different modalities; the context information integration module integrates the clinical information of the patient with the fusion feature, and combines the information from different sources together by vector splicing to form a more comprehensive feature vector; the abnormality detection and classification module sends the integrated feature vector into a deep learning model for abnormality detection, the model training and optimization model uses a training data set for multiple iterations, and continuously adjusts the model parameters through a back propagation algorithm to minimize the loss function; the model iteration training and verification module adopts a cyclic training and verification step, optimizes the model parameters through a training set, and uses a verification set to test and evaluate the model performance until the model converges.

[0011] A further improvement of the technical solution of the present invention is that: the data collection and preprocessing module includes a data collection unit and a data processing unit, the data collection unit includes a storage device, the storage device first receives medical image data of different modalities, and represents the collected image data information coordinates as (x, y), the data processing unit includes a processor, and the processor performs denoising and standardization preprocessing on the medical image data collected by the data collection unit, and the denoising processing formula is:

[0012]

[0013] Among them I denoised (x,y) represents the pixel value of the denoised image at the coordinate (x,y), I noisey (x,y) is the pixel value of the noisy input image at coordinate (x,y), λ is the denoising parameter, is the second-order gradient of the image, and x and y are used to represent the pixel coordinates in the image;

[0014] Medical image data is standardized to a specific range (such as 0 to 1). The formula is:

[0015]

[0016] Where μ is the mean value of the image, σ is the standard deviation, and the processed image is:

[0017] A further improvement of the technical solution of the present invention is that the multimodal feature extraction module sends the preprocessed images to the multi-branch convolutional neural network in the multimodal feature extraction module for feature extraction. The specific formula is as follows:

[0018] F modality =CNN modality (I modality )

[0019] Among them I modality Represents a medical image of a specific modality (such as CT, MRI, or X-ray), CNN modality represents the convolutional neural network designed for a specific modality, F modality represents the extracted feature vector.

[0020] A further improvement of the technical solution of the present invention is that the feature fusion module fuses the extracted features of different modalities to form a comprehensive feature vector, and the fusion method adopts weighted average or splicing. The specific formula is as follows:

[0021] F fused =α1F CT +α2F MRI +α3F X-ray

[0022] Among them, α1, α2, and α3 represent the weights of each modal feature, and F CT 、F MRI 、F X-ray are the feature vectors extracted from CT, MRI and X-ray images respectively, and F fused Represents the fused feature vector.

[0023] A further improvement of the technical solution of the present invention is that the context information integration module combines the patient's clinical history, laboratory test results and other context information with the feature fusion module to extract different modal feature data to form a comprehensive feature vector. The formula is as follows:

[0024]

[0025] Where C represents the patient's contextual information, including medical history and laboratory test results. Indicates that the fusion feature is concatenated with the context information to form a more comprehensive feature vector. total The final feature vector containing all relevant features.

[0026] A further improvement of the technical solution of the present invention is that the anomaly detection model module inputs the context information integration module into the deep learning model for anomaly detection based on the integrated feature vector, and the anomaly detection formula is:

[0027] in It represents the probability that the input case is abnormal. RNN is a recursive neural network or a long short-term memory network, which is used to process sequence data or time series features.

[0028] A further improvement of the technical solution of the present invention is that the model training and optimization module is based on the detection data of the anomaly detection model module and the historical case data, the cross entropy loss function is used to train the model, and L2 regularization is introduced to prevent the model from overfitting. The specific formula of the loss function is as follows:

[0029]

[0030] Where L represents the loss function, y i , Represent the true label and the predicted value of the model, λ represents the regularization coefficient, ||θ j || 2 represents the L2 norm of the model parameters, N is the number of samples, and is the model parameter.

[0031] A further improvement of the technical solution of the present invention is that the model iterative training and verification module is based on the results of the preliminary training of the model training and optimization module and the system continuously performs iterative training and verification through new case data, adopts a cyclic training and verification step, optimizes the model parameters through the training set, and uses the verification set for testing, so that the model is continuously optimized in actual application.

[0032] A further improvement of the technical solution of the present invention is that the real-time monitoring and feedback module receives new case data in real time based on the model optimized by the model iteration training and verification module, performs abnormality detection on the data according to the new model of the model iteration training and verification module, and feeds back the detection results to the doctor, so as to continuously optimize the model to adapt to new data characteristics and clinical environment.

[0033] A further improvement of the technical solution of the present invention is: a case anomaly detection method based on deep learning, used to implement a case anomaly detection system based on deep learning as described in any one of claims 1-9, comprising the following steps:

[0034] Step 1: Collect medical imaging data from different modalities and improve image quality through denoising and standardization steps;

[0035] Step 2: Use a deep convolutional neural network to extract features from images of different modalities and fuse them to form a comprehensive feature vector;

[0036] Step 3: Combine the patient’s clinical information with the fusion features to generate a comprehensive feature vector that considers more information for anomaly detection;

[0037] Step 4: Analyze the fused feature vector through the trained deep learning model and output the anomaly detection results;

[0038] Step 5: The system can monitor new cases in real time, automatically prompt abnormalities and continuously update the model based on doctor feedback to improve accuracy.

[0039] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:

[0040] 1. The present invention provides an innovative case abnormality detection system and method based on deep learning. The present invention can comprehensively capture potential abnormal information in the case from multiple angles by integrating multimodal medical imaging data (such as CT, MRI, X-ray, etc.) and combining contextual information such as the patient's medical history and laboratory tests. This fusion of multiple features greatly improves the accuracy of detection and effectively reduces the risk of missed diagnosis and misdiagnosis;

[0041] 2. The case anomaly detection system and method based on deep learning provided by the present invention realizes automatic anomaly detection of medical images by adopting deep learning technology and recursive neural network, and can quickly and accurately analyze abnormal features in cases, which greatly reduces the time and error of manual analysis, reduces the workload of doctors, shortens the diagnosis time, and improves the overall efficiency of the medical process;

[0042] 3. The case abnormality detection system and method based on deep learning provided by the present invention can realize real-time detection during medical operations, quickly judge new cases, and immediately feedback the results to doctors, avoiding the lag of traditional manual analysis and ensuring the timeliness and effectiveness of clinical decision-making;

[0043] 4. The case anomaly detection system and method based on deep learning provided by the present invention have the ability of continuous learning through the system, and can continuously optimize the model parameters through the feedback mechanism. As the amount of data increases, the generalization ability and detection accuracy of the model are gradually improved. By combining the latest case data and expert feedback, the system can show high adaptability in different clinical environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0045] Figure 1 A flowchart of a case abnormality detection method and system based on deep learning of the present invention; DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Examples, such as Figure 1 As shown, a case anomaly detection system based on deep learning includes a data collection and preprocessing module, a multimodal feature extraction module, a feature fusion module, a context information integration module, an anomaly detection model module and a model iteration training and verification module; the data collection and preprocessing module receives medical image data of different modalities, and performs denoising and standardization processing on these images through the data preprocessing module; the multimodal feature extraction module sends the preprocessed images to the multi-branch convolutional neural network in the multimodal feature extraction module for feature extraction; the feature fusion module fuses the feature extraction data of different modalities; the context information integration module integrates the patient's clinical information with the fusion features, and combines information from different sources by vector splicing to form a more comprehensive feature vector; the anomaly detection and classification module sends the integrated feature vector to the deep learning model for anomaly detection, and the model training and optimization model uses the training data set for multiple iterations, and continuously adjusts the model parameters through the back propagation algorithm to minimize the loss function; the model iteration training and verification module adopts a cyclic training and verification step, optimizes the model parameters through the training set, and uses the verification set to test and evaluate the model performance until the model converges.

[0048] The data collection and preprocessing module includes a data collection unit and a data processing unit. The data collection unit includes a storage device. The storage device first receives medical image data of different modalities and represents the collected image data information coordinates as (x, y). The data processing unit includes a processor. The processor performs denoising and standardization preprocessing on the medical image data collected by the data collection unit. The denoising processing formula is:

[0049]

[0050] Among them I denoised (x,y) represents the pixel value of the denoised image at the coordinate (x,y), I noisey (x, y) is the pixel value of the noisy input image at the coordinate (x, y), λ is the denoising parameter, which controls the denoising intensity, is the second-order gradient of the image, which represents the rate of change of the image at (x, y), where x and y are used to represent the pixel coordinates in the image;

[0051] The storage device receives medical image data of different modalities including multimodal image data of patients such as CT, MRI, and X-ray;

[0052] Medical image data is standardized to a specific range (such as 0 to 1). The formula is:

[0053]

[0054] Where μ is the mean value of the image, σ is the standard deviation, and the processed image is:

[0055] Acquire the patient's multimodal image data such as CT, MRI, and X-ray from the hospital's PACS, perform denoising and standardization on the acquired image data, eliminate unnecessary noise, and align image data of different modalities on the same dimension.

[0056] The multimodal feature extraction module sends the preprocessed images to the multi-branch convolutional neural network in the multimodal feature extraction module for feature extraction. The specific formula is as follows:

[0057] F modality =CNN modality (I modality )

[0058] Among them I modality Represents a medical image of a specific modality (such as CT, MRI, or X-ray), CNN modality represents the convolutional neural network designed for a specific modality, F modality represents the extracted feature vector.

[0059] CNN modality The convolutional neural network representing the corresponding modality usually includes a convolutional layer, an activation function, and a pooling layer. The convolutional layer extracts features through convolution operations. The formula is:

[0060] Z (l) =W (l) *A (l-1) +b (l)

[0061] Where Z (l) , W (l) , A (l-1) , b (l) They are respectively represented as the linear combination output of the lth layer, the convolution kernel weight, the activation value of the previous layer and the bias term, * represents the convolution operation,

[0062] The activation function usually uses the ReLU activation function, the formula is:

[0063] A (l) =ReLU(Z (l) )=max(0,Z (l) )

[0064] Pooling layer: used to reduce feature dimensions, the formula is:

[0065]

[0066] Where (i, j) is the area of ​​the pooling window. After extraction, the feature vector of each modality is:

[0067] F CT =CNN CT (I CT ),F MRI =CNN MRI (I MRI ),F X-ray =CNN X-ray (I X-ray )

[0068] The feature fusion module fuses the extracted features of different modalities to form a comprehensive feature vector. The fusion method uses weighted average or splicing. The specific formula is as follows:

[0069] F fused =α1F CT +α2F MRI +α3F X-ray

[0070] Among them, α1, α2, and α3 represent the weights of each modal feature, and F CT 、F MRI 、F X-ray are the feature vectors extracted from CT, MRI and X-ray images respectively, and F fused Represents the fused feature vector.

[0071] Convolutional neural networks are used to extract features from CT, MRI and X-ray images respectively to generate three feature vectors, which are then fused using a weighted method. The weight parameters can be adjusted according to actual clinical needs to adapt to the importance of images of different modalities.

[0072] The context information integration module combines the patient's clinical history, laboratory test results and other context information with the feature fusion module to extract different modal feature data to form a comprehensive feature vector. The formula is as follows:

[0073]

[0074] Where C represents the patient's contextual information, including medical history and laboratory test results. Indicates that the fusion feature is concatenated with the context information to form a more comprehensive feature vector. total The final feature vector containing all relevant features.

[0075] After feature fusion, contextual information such as the patient's medical history and laboratory test results are spliced ​​with the fused feature vector to form a more complete feature vector. This step can help the system judge abnormalities more accurately and comprehensively when analyzing cases.

[0076] The anomaly detection model module inputs the context information integration module into the deep learning model based on the integrated feature vector for anomaly detection. The anomaly detection formula is:

[0077] in Represents the probability that the input case is abnormal. RNN is a recursive neural network or a long short-term memory network, which is used to process sequence data or time series features. The total feature vector is input into the recursive neural network. RNN detects whether there is an abnormality based on the dependency of the historical data sequence, and then predicts the health status or abnormality type of the case. The state update formula of RNN is:

[0078] h t =f(W h h t-1 +W x x t +b)

[0079] where h t Indicates the current state, W h , W x are state weight and input weight respectively, b is bias term, and f is activation function;

[0080] The model training and optimization module is based on the detection data of the anomaly detection model module and historical case data. The cross entropy loss function is used to train the model, and L2 regularization is introduced to prevent the model from overfitting. The specific formula of the loss function is as follows:

[0081]

[0082] Where L represents the loss function, y i , Represent the true label and the predicted value of the model, λ represents the regularization coefficient, ||θ j || 2 represents the L2 norm of the model parameters, N is the number of samples, θ j It is a model parameter.

[0083] The model parameters are updated through the back propagation algorithm to minimize the loss function. The formula is:

[0084]

[0085] Where η is the learning rate, θ j It is the model parameter. It represents the gradient of the loss function L with respect to the parameter θ, that is, the rate of change. The gradient indicates the direction and magnitude of change of the loss function at a certain parameter point. It helps the model know in which direction the parameters need to be adjusted to minimize the loss function. The gradient descent algorithm is used to calculate the derivative of the loss function with respect to the model parameters, determine the fastest direction of error reduction, and update each parameter through this formula so that the loss function is gradually reduced, thereby improving the prediction accuracy of the model.

[0086] The model iterative training and verification module is based on the preliminary training results of the model training and optimization module and the system continuously performs iterative training and verification through new case data. It adopts a cyclic training and verification step, optimizes model parameters through the training set, and uses the verification set for testing. The model is continuously optimized in actual application.

[0087] The system trains the model based on massive historical case data, uses the cross-entropy loss function to measure the difference between the model prediction and the true label, and uses L2 regularization to prevent the model from overfitting. As new case data is input, the model is continuously iterated and updated to improve its adaptability and accuracy in clinical environments.

[0088] The real-time monitoring and feedback module is based on the model optimized by the model iteration training and verification module. It receives new case data in real time and detects anomalies based on the new model of the model iteration training and verification module. It also feeds back the detection results to doctors, prompting continuous optimization of the model to adapt to new data features and clinical environment.

[0089] The system receives new case data in real time and performs anomaly detection, and through the feedback mechanism, the system learns new patterns and information. The feedback update formula is:

[0090] θ←θ+Δθ

[0091] Here, Δθ is the parameter update generated based on the new data.

[0092] The system will feed back the detected abnormal results to the doctor through the user interface, providing intuitive images and annotations to help the doctor quickly understand the test results and make decisions. If the doctor finds that the test results do not match the actual situation, he or she can provide further information to the system through the feedback mechanism, and the system will optimize the model accordingly.

[0093] A case anomaly detection method based on deep learning, used to implement a case anomaly detection system based on deep learning according to any one of claims 1 to 9, comprising the following steps:

[0094] Step 1: Collect medical imaging data from different modalities and improve image quality through denoising and standardization steps;

[0095] Step 2: Use a deep convolutional neural network to extract features from images of different modalities and fuse them to form a comprehensive feature vector;

[0096] Step 3: Combine the patient’s clinical information with the fusion features to generate a comprehensive feature vector that considers more information for anomaly detection;

[0097] Step 4: Analyze the fused feature vector through the trained deep learning model and output the anomaly detection results;

[0098] Step 5: The system can monitor new cases in real time, automatically prompt abnormalities and continuously update the model based on doctor feedback to improve accuracy.

[0099] Working principle: First, the system obtains multimodal medical imaging data from the hospital's PACS system, including CT, MRI, and X-ray. Before these raw data are input into the deep learning model, they need to be standardized and denoised to ensure the consistency of data in different modalities. This process is implemented through the following function. The denoising formula is:

[0100] I denoised (x,y)=I noisey (x,y)-λ·▽ 2 I noisey (x,y)

[0101] Among them I denoised (x,y) represents the pixel value of the denoised image at the coordinate (x,y), I noisey (x,y) is the pixel value of the noisy input image at coordinate (x,y), λ is the denoising parameter, is the second-order gradient of the image, and x and y are used to represent the pixel coordinates in the image;

[0102] Medical image data is standardized to a specific range (such as 0 to 1). The formula is:

[0103]

[0104] Where μ is the mean value of the image, σ is the standard deviation, and the processed image is: The processed image data will be extracted by multimodal convolutional neural network (CNN) respectively. For each image type (such as CT, MRI, X-ray), the system uses independent CNN to extract features. The specific formula for the extracted features is as follows:

[0105] F modality =CNN modality (I modality )

[0106] Among them I modality Represents a medical image of a specific modality (such as CT, MRI, or X-ray), CNN modality represents the convolutional neural network designed for a specific modality, F modality Represents the extracted feature vector. Then the system fuses the features of different modal images through a weighted method. Through feature fusion, the system can integrate a variety of medical imaging information and further enhance the accuracy of anomaly detection. The feature fusion formula is:

[0107] F fused =α1F CT +α2F MRI +α3F X-ray

[0108] Among them, α1, α2, and α3 represent the weights of each modal feature, and F CT 、F MRI 、F X-ray are the feature vectors extracted from CT, MRI and X-ray images respectively, and F fused Represents the fused feature vector. Based on feature fusion, the system further integrates the patient's contextual information, such as medical history, laboratory test data, etc., to build a more comprehensive case representation. This process is completed by splicing the image feature vector and contextual information: The specific formula is as follows:

[0109]

[0110] Where C represents the patient's contextual information, including medical history and laboratory test results. Indicates that the fusion feature is concatenated with the context information to form a more comprehensive feature vector. total The final feature vector containing all relevant features is fed into a recurrent neural network (RNN) for anomaly detection. RNN can effectively capture the temporal and sequential dependencies of data and predict whether there is an abnormality in the case. The output of RNN is the anomaly detection result: in Indicates the probability that the input case is abnormal. RNN is a recursive neural network or a long short-term memory network, which is used to process sequence data or time series features. The model is trained through historical case data, and the cross entropy loss function is used for error calculation. L2 regularization is used to prevent the model from overfitting. The specific formula of the loss function is as follows:

[0111]

[0112] Where L represents the loss function, y i , Represent the true label and the predicted value of the model, λ represents the regularization coefficient, ||θ j || 2 represents the L2 norm of the model parameters, N is the number of samples, θ j Represents model parameters. The system can receive new case data in real time and perform automatic detection, and supports a feedback mechanism. Doctors can make adjustments based on the test results and input the feedback into the system. The system uses these feedbacks for continuous learning and model optimization. Finally, the system feeds back the test results to the doctor, provides visual abnormality annotations and diagnostic suggestions, and helps doctors make clinical decisions quickly. The system's continuous learning ability enables it to adapt to the ever-changing case data and provide long-term and stable diagnostic support. The present invention uses deep learning technology to integrate multimodal imaging features and patient contextual information to achieve a highly automated abnormality detection process. The system improves the accuracy of diagnosis, reduces the risk of missed diagnosis and misdiagnosis, and can play a long-term role in actual clinical environments through real-time feedback and continuous learning.

[0113] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A case anomaly detection system based on deep learning, characterized by: The system comprises a data collection and preprocessing module, a multimodal feature extraction module, a feature fusion module, a context information integration module, an anomaly detection model module, a model training and optimization module, a model iteration training and verification module, and a real-time monitoring and feedback module; the data collection and preprocessing module receives medical image data of different modalities, and performs denoising and standardization on these images through the data preprocessing module; the multimodal feature extraction module sends the preprocessed images to the multi-branch convolutional neural network in the multimodal feature extraction module for feature extraction; the feature fusion module fuses the feature extraction data of different modalities; the context information integration module combines the patient's clinical information with the fused feature information; The features are integrated, and information from different sources is combined by vector splicing to form a more comprehensive feature vector; the anomaly detection and classification module sends the integrated feature vector to the deep learning model for anomaly detection, and the model training and optimization model uses the training data set for multiple iterations, and continuously adjusts the model parameters through the back propagation algorithm to minimize the loss function; the model iterative training and verification module adopts a cyclic training and verification step, optimizes the model parameters through the training set, and uses the verification set to test and evaluate the model performance until the model converges. The real-time monitoring and feedback module system can receive new case data in real time for anomaly detection and provide timely feedback on the detection results.

2. The case anomaly detection system based on deep learning according to claim 1, characterized in that: The data collection and preprocessing module includes a data collection unit and a data processing unit. The data collection unit includes a storage device. The storage device first receives medical image data of different modalities and represents the collected image data information coordinates as (x, y). The data processing unit includes a processor. The processor performs denoising and standardization preprocessing on the medical image data collected by the data collection unit. The denoising processing formula is: Among them I denoised (x,y) represents the pixel value of the denoised image at the coordinate (x,y), I noisey (x,y) is the pixel value of the noisy input image at coordinate (x,y), λ is the denoising parameter, is the second-order gradient of the image, and x and y are used to represent the pixel coordinates in the image; Medical image data is standardized to a specific range (such as 0 to 1). The formula is: Where μ is the mean value of the image, σ is the standard deviation, and the processed image is:

3. The case anomaly detection system based on deep learning according to claim 2, characterized in that: The multimodal feature extraction module sends the preprocessed images to the multi-branch convolutional neural network in the multimodal feature extraction module for feature extraction. The specific formula is as follows: F modality =CNN modality (I modality ) Among them I modality Represents a medical image of a specific modality (such as CT, MRI, or X-ray), CNN modality represents the convolutional neural network designed for a specific modality, F modality represents the extracted feature vector.

4. The case anomaly detection system based on deep learning according to claim 3, characterized in that: The feature fusion module fuses the extracted features of different modalities to form a comprehensive feature vector. The fusion method adopts weighted average or splicing. The specific formula is as follows: F fused =α1F CT +α2F MRI +α3F X-ray Among them, α1, α2, and α3 represent the weights of each modal feature, and F CT 、F MRI 、F X-ray are the feature vectors extracted from CT, MRI and X-ray images respectively, and F fused Represents the fused feature vector.

5. The case anomaly detection system based on deep learning according to claim 4, characterized in that: The context information integration module combines the patient's clinical history, laboratory test results and other context information with the feature fusion module to extract different modal feature data to form a comprehensive feature vector. The formula is as follows: Where C represents the patient's contextual information, including medical history and laboratory test results. Indicates that the fusion feature is concatenated with the context information to form a more comprehensive feature vector. total The final feature vector containing all relevant features.

6. The case anomaly detection system based on deep learning according to claim 5, characterized in that: The anomaly detection model module inputs the context information integration module into the deep learning model based on the integrated feature vector to perform anomaly detection. The anomaly detection formula is: in It represents the probability that the input case is abnormal. RNN is a recursive neural network or a long short-term memory network, which is used to process sequence data or time series features.

7. The case anomaly detection system based on deep learning according to claim 6, characterized in that: The model training and optimization module is based on the detection data of the anomaly detection model module and the historical case data. The cross entropy loss function is used to train the model, and L2 regularization is introduced to prevent the model from overfitting. The specific formula of the loss function is as follows: Where L represents the loss function, y i , Represent the true label and the predicted value of the model, λ represents the regularization coefficient, ||θ j || 2 represents the L2 norm of the model parameters, N is the number of samples, θ j It is a model parameter.

8. The case anomaly detection system based on deep learning according to claim 7, characterized in that: The model iterative training and verification module is based on the preliminary training results of the model training and optimization module and the system continuously performs iterative training and verification through new case data. It adopts a cyclic training and verification step, optimizes model parameters through the training set, and uses the verification set for testing. The model is continuously optimized in practical applications.

9. The case anomaly detection system based on deep learning according to claim 1, characterized in that: The real-time monitoring and feedback module is based on the model optimized by the model iteration training and verification module, receives new case data in real time, detects abnormalities based on the new model of the model iteration training and verification module, and feeds back the detection results to the doctor, so as to continuously optimize the model to adapt to new data features and clinical environment.

10. A method for case anomaly detection based on deep learning, used to implement a case anomaly detection system based on deep learning according to any one of claims 1 to 9, comprising the following steps: Step 1: Collect medical imaging data from different modalities and improve image quality through denoising preprocessing and standardization steps; Step 2: Use a deep convolutional neural network to extract features from images of different modalities and fuse them to form a comprehensive feature vector; Step 3: Combine the patient’s clinical information with the fusion features to generate a comprehensive feature vector that considers more information for anomaly detection; Step 4: Analyze the fused feature vector through the trained deep learning model and output the anomaly detection results; Step 5: The system can monitor new cases in real time, automatically prompt abnormalities and continuously update the model based on doctor feedback to improve accuracy.

Images