Intelligent case analysis method and system based on deep learning
Through intelligent case analysis methods based on deep learning, the historical data of thyroid or breast cases are integrated and analyzed, the data integration and analysis problems are solved, the utilization rate and diagnosis and treatment effect of case data are improved, and efficient and accurate case analysis is achieved.
Patent Information
- Application Number
- CN202410259966.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-07
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-03-07
AI Technical Summary
Existing medical data processing technologies are difficult to effectively integrate and analyze historical data of thyroid and breast cases, especially the data that combines pure images, plain text and images with text have differences in format, quality and integrity, which affects subsequent data analysis. and handle.
Using intelligent case analysis method based on deep learning, the historical data of thyroid or breast cases is obtained and preprocessed, and the data set of the same case types are divided into data sets, and a model is established in combination with deep learning networks to extract image features, text information and Feature fusion, generate auxiliary suggestions text, and conduct intelligent case analysis.
It improves the utilization rate and diagnosis and treatment effect of case data, enhances the generalization ability and robustness of the model, and provides an efficient, accurate and reliable intelligent case analysis method and system.
Smart Images

Figure CN118116576B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent case analysis, and in particular to an intelligent case analysis method and system based on deep learning. Background Art
[0002] With the continuous advancement of medical technology and the rapid development of digital medicine, a large amount of case data has been accumulated and stored in medical information systems. These data not only contain the patient's diagnostic information, but also reflect the doctor's treatment process and effect. They are valuable resources in medical research and clinical practice. Especially for thyroid and breast cases, due to their high incidence and complexity of diagnosis and treatment, the processing and analysis of relevant data is particularly important.
[0003] In the existing medical data processing technology, there are relatively mature methods for the acquisition and preprocessing of case history data. Case history data usually includes pure image case history data (such as B-ultrasound, CT, MRI and other medical images), pure text case history data (such as medical records written by doctors or electronically entered), and case history data combined with images and text (such as medical images with annotations or instructions). However, these data have differences in format, quality, integrity, etc., which affects the subsequent data analysis and processing. Therefore, there is an urgent need for an intelligent case analysis method that can effectively integrate and analyze these data to improve the utilization rate of case data and the diagnosis and treatment effect. Summary of the invention
[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.
[0005] In view of the above existing problems, the present invention is proposed.
[0006] Therefore, the present invention provides an intelligent case analysis method and system based on deep learning, which can solve the problems mentioned in the background technology.
[0007] In order to solve the above technical problems, the present invention provides the following technical solutions, an intelligent case analysis method based on deep learning, comprising:
[0008] Acquire thyroid or breast case history data and preprocess the case history data, wherein the case history data includes pure image case history data, pure text case history data, and case history data combining image and text;
[0009] Performing a classification operation on the preprocessed mapping relationship data set to divide the preprocessed mapping relationship data set into a first data set and a second data set;
[0010] The first data set is data in which there are at least two dates corresponding to the same person name and identity ID in the same case type, and the date interval is greater than two days, or a data set in which there are disease end marker words in the auxiliary suggestion text extracted from the information, and the second data set is the part of the mapping relationship data set excluding the first data set;
[0011] According to the first data set, a first model is established in combination with a deep learning network, wherein the input of the first model is picture case features, text case descriptions, and case types, and the output of the first model is auxiliary suggestion text;
[0012] The first model is scored, and if the score is not lower than a preset threshold, intelligent case analysis is performed based on the first model.
[0013] As a preferred solution of the intelligent case analysis method based on deep learning described in the present invention, it also includes:
[0014] Scoring the first model, if the score is lower than a preset threshold, expanding the second data set, combining the expanded second data set with the original first data set to form a new data set, re-establishing the second model based on the new data set combined with the deep learning network, and performing intelligent case analysis based on the second model;
[0015] The new data set is divided into a training set and a test set, wherein the training set contains at least data in which the same person name and identity ID correspond to at least three dates, and the date interval is greater than two days in all cases of the same type, and a data set in which the auxiliary suggestion text extracted from the information contains a disease end marker word, and the test set contains at least a data set in which the auxiliary suggestion text extracted from the information in the same case type in the second data set contains a disease end marker word;
[0016] The second model scoring score is not lower than the first model scoring score.
[0017] As a preferred solution of the intelligent case analysis method based on deep learning described in the present invention, wherein: the preprocessing includes image enhancement and case feature extraction of the pure image case history data, information extraction of the pure text case history data, and structural processing according to the results after information extraction, and information extraction of the image and text combined case history data, and structural processing according to the results after information extraction, and then image enhancement and case feature extraction;
[0018] The image enhancement includes using an improved edge detection algorithm to highlight the edge of the thyroid tissue, and using an improved wavelet transform to highlight the internal features of the breast lesion area;
[0019] The case feature extraction includes improving the neural network through specific features of thyroid or breast case images, and using the improved neural network model to extract case features;
[0020] The information extraction includes extracting seven types of structured data information, namely, date, person name, case type, age, identity ID, text case description and auxiliary suggestion text, from the case text data;
[0021] The preprocessing also includes establishing a mapping relationship data set based on the results of the structured processing, with case type as the first priority, age as the second priority, person name and identity ID as the third priority, and date as the fourth priority.
[0022] As a preferred solution of the intelligent case analysis method based on deep learning described in the present invention, the preprocessing further includes:
[0023] Using an improved direct mapping method to establish a mapping function, and establishing a mapping relationship between images and text data in the same case according to the mapping model;
[0024] Obtain a fixed-dimensional feature vector of the case image, convert the words in the case text into vector representations, and capture the contextual information of the text through a recurrent neural network structure to output the feature representation of the text;
[0025] The image feature vector and the text feature vector are fused, and the attention mechanism method is used to realize feature fusion to obtain the joint representation between the image and the text;
[0026] A fully connected neural network layer is designed as a mapping function to map the fused features to a target space, where the target space is a vector space of a fixed dimension and is used to represent the mapping relationship between images and texts.
[0027] As a preferred solution of the intelligent case analysis method based on deep learning described in the present invention, wherein: the expansion of the second data set includes:
[0028] fusing feature maps of image data of the same dimension in the second data set by improved linear interpolation, and reconstructing them back to image space through a decoder to generate a new medical image;
[0029] The improved linear interpolation includes:
[0030]
[0031] Among them, F mixed Indicates the result after difference, F A 、F B They represent the feature maps of image data of the same dimension, α represents the difference parameter, and β represents the nonlinear degree control parameter;
[0032] When β = 0, the improved linear interpolation degenerates into the standard linear interpolation;
[0033] When β>0, the interpolation result will be biased towards F in a nonlinear manner according to the value of α. A or F B ;
[0034] According to the new medical image processed by improved linear interpolation and combined with the pre-trained NLP model, the text data in the second data set is expanded.
[0035] As a preferred solution of the intelligent case analysis method based on deep learning described in the present invention, the deep learning network includes:
[0036] The input layer of the deep learning network includes an image case feature input layer, a text case description input layer, and a case type input layer;
[0037] The input of the image case feature input layer is the preprocessed case image, and the output is a 2048-dimensional image feature vector;
[0038] The input of the text case description input layer is the preprocessed text case description, and the output is the hidden state of the last time step, which is a 256-dimensional text feature vector;
[0039] The input of the case type input layer is the case type represented by the embedding vector, and the embedding dimension is a 32-dimensional case type embedding vector;
[0040] The feature fusion layer of the deep learning network includes concatenating the image feature vector, the text feature vector and the case type embedding vector, outputting a vector of size 2336, and adding a fully connected layer after the feature fusion layer for dimensionality reduction, outputting a vector of size 512;
[0041] The hidden layer of the deep learning network uses two fully connected layers for nonlinear transformation. The input size of the first fully connected layer is 512 dimensions, the output size is 256 dimensions, and the ReLU activation function is used. The input size of the second fully connected layer is 256 dimensions, the output size is 128 dimensions, and the ReLU activation function is also used. A Dropout layer is added after each fully connected layer to prevent overfitting.
[0042] The output layer of the deep learning network uses a Transformer decoder structure to gradually generate text;
[0043] The connection method of the deep learning network includes parallel processing of the image case feature input layer, the text case description input layer and the case type input layer, and their outputs are spliced in the feature fusion layer. The output of the feature fusion layer is connected to the first hidden layer, and is connected to the subsequent hidden layers in turn. The output of the last hidden layer is connected to the output layer to generate the final auxiliary suggestion text.
[0044] As a preferred solution of the intelligent case analysis method based on deep learning described in the present invention, it also includes:
[0045] If there are still the same case data with the same person's name and ID after the date corresponding to the end of illness marker, then determine whether these cases are good-condition data. If so, when training the model, these data must appear in both the training set and the test set.
[0046] If not, the case data of the date corresponding to the disease end marker word is deleted from the first type of data set.
[0047] The intelligent case analysis method based on deep learning is characterized by including:
[0048] A data acquisition module, used to acquire thyroid or breast case history data and preprocess the case history data, wherein the case history data includes pure image case history data, pure text case history data, and case history data combining image and text;
[0049] A data set classification module is used to perform a classification operation on the preprocessed mapping relationship data set, and divide the preprocessed mapping relationship data set into a first data set and a second data set;
[0050] The first data set is data in which there are at least two dates corresponding to the same person name and identity ID in the same case type, and the date interval is greater than two days, or a data set in which there are disease end marker words in the auxiliary suggestion text extracted from the information, and the second data set is the part of the mapping relationship data set excluding the first data set;
[0051] A model building module, used to build a first model based on the first data set in combination with a deep learning network, wherein the input of the first model is picture case features, text case descriptions, and case types, and the output of the first model is auxiliary suggestion text;
[0052] The scoring and analysis module is used to score the first model. If the score is not lower than a preset threshold, intelligent case analysis is performed based on the first model.
[0053] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the above method when executing the computer program.
[0054] A computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the method described above when executed by a processor.
[0055] Beneficial effects of the present invention: The present invention proposes an intelligent case analysis method and system based on deep learning, obtains thyroid or breast case history data, and preprocesses the case history data, the case history data includes pure image case history data, pure text case history data and image and text combined case history data; the preprocessed mapping relationship data set is classified, and the preprocessed mapping relationship data set is divided into a first data set and a second data set; the first data set is data in which there are at least two dates corresponding to the same person name and identity ID in the same case type, and the date interval is greater than two days, or a data set in which there are disease end markers in the auxiliary suggestion text extracted from the information, and the second data set is the part of the mapping relationship data set excluding the first data set; according to the first data set, a first model is established in combination with a deep learning network, the input of the first model is picture case features, text case description and case type, and the output of the first model is auxiliary suggestion text; the first model is scored, and if the score is not lower than a preset threshold, intelligent case analysis is performed according to the first model. The present invention proposes an intelligent case analysis method and system based on deep learning, which can effectively analyze thyroid or breast cases. In addition, the present invention also improves the generalization ability and robustness of the model through data set classification operations. Therefore, the present invention provides an efficient, accurate and reliable method and system for intelligent case analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:
[0057] Figure 1 A first method flow chart of an intelligent case analysis method and system based on deep learning provided by one embodiment of the present invention;
[0058] Figure 2A second method flow chart of an intelligent case analysis method and system based on deep learning provided by one embodiment of the present invention;
[0059] Figure 3 An internal structure diagram of a computer device for an intelligent case analysis method and system based on deep learning provided in one embodiment of the present invention. DETAILED DESCRIPTION
[0060] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0061] Example 1
[0062] Reference Figure 1-3 , which is the first embodiment of the present invention, provides an intelligent case analysis method and system based on deep learning, including:
[0063] S101: Acquire thyroid or breast case history data and preprocess the case history data, where the case history data includes pure image case history data, pure text case history data, and case history data combining image and text;
[0064] Among them, the preprocessing includes image enhancement and case feature extraction for pure image case history data, information extraction for pure text case history data, and structural processing based on the results of information extraction, and information extraction for image and text combined case history data, and structural processing based on the results of information extraction, and then image enhancement and case feature extraction;
[0065] Specifically, image enhancement includes using an improved edge detection algorithm to highlight the edge of thyroid tissue and using an improved wavelet transform to highlight the internal features of breast lesion areas;
[0066] In the embodiment of the present application, using the improved edge detection algorithm to highlight the edge of the thyroid tissue includes:
[0067] Step 1: Preprocess the image case history data in the thyroid case to improve image quality and reduce noise and unnecessary details to facilitate subsequent edge detection.
[0068] The preprocessed image is grayscaled and the color image is converted into a grayscale image to reduce the computational complexity.
[0069] The grayscale formula can be used as follows:
[0070] I gray =0.2989×I R +0.5870×I G +0.1140×I B
[0071] Among them, I R ,I G ,I B are the pixel values of the red, green, and blue channels respectively.
[0072] Then perform Gaussian filtering on the grayscale image, where the filtering uses the following formula:
[0073]
[0074] Step 2: Perform gradient calculation on the image case history data in the processed thyroid case to calculate the gradient magnitude and direction of each pixel in the image;
[0075] The gradient operator uses the Sobel gradient operator to calculate the gradient.
[0076]
[0077]
[0078] Among them, G x and G y are the gradients in the x and y directions respectively.
[0079] Gradient magnitude and direction:
[0080]
[0081]
[0082] Step 3: Using improved edge detection, the edge detection algorithm is improved to enhance edge highlighting according to the characteristics of thyroid tissue.
[0083] Because the brightness and contrast of thyroid tissue can vary from image to image, an adaptive threshold was used to determine which gradient magnitudes corresponded to true edges;
[0084] T(x,y)=μ(x,y)+k×σ(x,y)
[0085] Here, μ(x, y) and σ(x, y) are the mean and standard deviation of the local neighborhood, respectively, and k is a constant determined experimentally.
[0086] Edge enhancement involves performing a nonlinear transformation on the gradient magnitude to enhance the contrast of edges.
[0087]
[0088] Among them, T min and T max are the minimum and maximum values of the edge amplitude determined by the adaptive threshold, and α and γ are parameters used to control the degree of enhancement.
[0089] Step 4: Post-process the results of the improved edge detection to refine the edges and remove unnecessary details and noise.
[0090] In an optional embodiment, edge refinement can be performed by applying non-maximum suppression (NMS) to refine the edge. Two thresholds T can also be used. low and T high To determine which edges are real edges. You can also use Hough Transform or other methods to connect broken edges.
[0091] It should be noted that the use of an improved edge detection algorithm is beneficial to improving the detection accuracy of thyroid tissue edges and is beneficial for subsequent case analysis.
[0092] In the embodiment of the present application, using improved wavelet transform to highlight the internal features of the breast lesion area includes:
[0093] Step 1: Preprocess the image case history data in breast cases to enhance the contrast of the lesion area and reduce noise and unnecessary details.
[0094] The grayscale formula can be used as follows:
[0095] I gray =0.2989×I R +0.5870×I G +0.1140×I B
[0096] Among them, I R ,I G ,I B are the pixel values of the red, green, and blue channels respectively.
[0097] And contrast enhancement is performed, using histogram equalization or contrast stretching technology to enhance the contrast of the lesion area.
[0098] Step 2: Use improved wavelet transform to better capture the internal features of the breast lesion area by adjusting the wavelet basis function and the number of decomposition layers.
[0099] Select wavelet basis function: According to the characteristics of breast lesions, select or design a wavelet basis function so that it can better match the characteristics of the lesion area. For example, you can choose a wavelet basis function with good localization characteristics, such as Daubechies, Symlets or Coiflets.
[0100] Furthermore, multi-scale decomposition: multi-level wavelet decomposition is performed on the preprocessed image to obtain approximate coefficients and detail coefficients of different scales. The appropriate number of decomposition levels is selected to capture the structural and texture characteristics inside the lesion area.
[0101] Approximation Coefficients:cA j
[0102] Detail Coefficients:cD j,h ,cD j,v ,cD j,d
[0103] Among them, j represents the number of decomposition levels, and h, v, and d represent the detail coefficients in the horizontal, vertical, and diagonal directions, respectively.
[0104] Furthermore, the improved detail coefficient processing performs nonlinear transformation on the detail coefficient to enhance the contrast of the internal features of the lesion area. The detail coefficient can be thresholded using an adaptive threshold method based on statistical characteristics to highlight the features of the lesion area.
[0105]
[0106] Where k represents the horizontal, vertical or diagonal direction, α and γ are parameters for adjusting contrast, and T j It is an adaptive threshold determined according to the statistical characteristics of the detail coefficients of the jth layer.
[0107] Step 3: Reconstruct the image after improved wavelet transform, and use the improved coefficients to reconstruct the image to highlight the internal features of the lesion area.
[0108]
[0109] Among them, J represents the maximum number of decomposition levels.
[0110] Step 4: Post-process the reconstructed image to further refine the features of the lesion area and remove noise and artifacts.
[0111] It should be noted that the advantages of using improved wavelet transform to highlight the internal features of breast lesion areas include high-precision edge detection: the improved edge detection algorithm can better highlight the edge of thyroid tissue and improve the accuracy of edge detection. It also includes the ability to effectively extract lesion features: the use of improved wavelet transform can better capture the internal features of breast lesion areas, such as structure and texture. This helps to improve the accuracy and reliability of lesion detection. It also includes the ability to enhance the contrast of lesion areas: by preprocessing and contrast enhancement of lesion areas, the internal features of breast lesions can be better highlighted, which helps doctors to quickly discover and diagnose lesions. It also includes the ability to reduce noise and artifacts: by post-processing the reconstructed image, the features of the lesion area can be further refined, noise and artifacts can be removed, and the quality of the image can be improved. It also includes the ability to improve diagnostic efficiency: using improved wavelet transform to detect breast lesions can obtain more accurate diagnostic results in a shorter time and improve the work efficiency of doctors. It also includes personalized processing: according to the characteristics of different lesions, appropriate wavelet basis functions and decomposition layers can be selected to achieve personalized lesion detection.
[0112] In an optional embodiment, a morphological operation may be selected to apply a morphological filter (such as corrosion, dilation, opening operation, closing operation, etc.) to remove noise and artifacts while retaining the features inside the lesion area.
[0113] Furthermore, case feature extraction includes improving a neural network through specific characteristics of thyroid or breast case images, and using the improved neural network model to extract case features;
[0114] In an optional embodiment, to realize the pathological feature extraction of thyroid case images by improving the neural network, the following steps can be followed:
[0115] Step 1: Data collection and preprocessing: Collect a large number of thyroid case images, including normal and abnormal samples.
[0116] The images are preprocessed, such as denoising, contrast enhancement, and normalization, to improve image quality and reduce the complexity of model training.
[0117] Step 2: Annotation and data augmentation: The collected images are annotated, usually by experienced pathologists, to mark the regions of interest (ROI) and pathological features.
[0118] Using data augmentation techniques (such as rotation, scaling, flipping, etc.) to increase the diversity and quantity of the dataset helps the model generalize better.
[0119] Step 3: Neural Network Model Selection: Choose a neural network architecture suitable for image processing, such as a convolutional neural network (CNN).
[0120] You can consider using a pre-trained model (such as VGG16, ResNet, DenseNet, etc.) as a feature extractor and fine-tune it based on it.
[0121] Step 4: Network improvement: The neural network structure is improved according to the specific characteristics of the thyroid case images. For example, a deeper or wider network can be designed to capture more complex features.
[0122] An attention mechanism can also be introduced to enable the model to focus on key areas in the image, such as suspected lesions.
[0123] Multi-scale or multi-input networks can also be used to process image data of different resolutions and sources.
[0124] Specifically, when processing thyroid case images, the neural network structure can be improved according to the specific characteristics of such images to improve the accuracy of feature extraction and classification. The following are some possible specific characteristics and their suggestions for improving the neural network structure:
[0125] It should be noted that in thyroid case images, small-scale structures such as tiny calcifications and nodule boundaries may be the key to diagnosis.
[0126] An improvement to this part can be to design a multi-scale feature extraction module so that the network can focus on both large-scale and small-scale features. For example, using convolution kernels of different sizes or introducing multi-scale input / output branches in the network.
[0127] It should be noted that thyroid ultrasound images are often affected by acoustic shadowing, speckle noise, etc., resulting in low image contrast and blurred boundaries.
[0128] Improvements to this part can be to add a contrast enhancement module to the network, such as conditional generative adversarial network (cGAN) to enhance image quality; or design more robust convolutional layers, such as depthwise separable convolution, to reduce noise interference.
[0129] It should be noted that thyroid lesions have different morphologies, including solid, cystic, mixed, etc., requiring the network to have sufficient representation capabilities to capture these differences.
[0130] The improvement method for this part can be to introduce more complex network structures, such as Inception modules, residual connections (ResNet), dense connections (DenseNet), etc., to enhance the network's ability to learn diverse forms.
[0131] It should be noted that in real datasets, the distribution of normal and abnormal thyroid images may be unbalanced.
[0132] The improvement method for this part can be to balance the data set by oversampling the minority class or undersampling the majority class; use a weighted loss function in network training to give the minority class a higher weight.
[0133] Step 5: Feature extraction: Train the neural network model to learn and extract pathological features from the images.
[0134] During training, various regularization techniques (such as Dropout, L1 / L2 regularization) can be used to prevent overfitting.
[0135] The output of the intermediate layer is used as the feature representation of the image, which can be used for subsequent classification, regression or clustering tasks.
[0136] Step 6: Evaluation and Optimization: Use independent validation and test sets to evaluate the performance of the model.
[0137] Optimize the model based on the evaluation results, such as adjusting hyperparameters, changing the network structure, or introducing new training techniques.
[0138] Repeat the above steps until satisfactory performance is achieved.
[0139] Step 7: Deployment and Monitoring: Deploy the trained model to the production environment for actual thyroid case image analysis.
[0140] Monitor the performance of the model, and regularly update and optimize the model to adapt to new data and requirements.
[0141] In an optional embodiment, to realize the pathological feature extraction of breast case images by improving the neural network, the following steps can be followed:
[0142] Step 1: Data collection and preprocessing: Collect a large number of breast case images, including normal and abnormal samples, such as mammography (breast molybdenum target), ultrasound images, etc.
[0143] Perform image preprocessing such as denoising, contrast enhancement, normalization, image segmentation, etc. to improve image quality, reduce noise interference, and highlight the area of interest.
[0144] Step 2: Annotation and data enhancement: Invite experienced radiologists or pathologists to annotate the images and mark the lesion areas, such as masses, calcifications, etc.
[0145] Using data augmentation techniques to increase the diversity and quantity of the dataset, such as rotation, translation, scaling, flipping, etc., helps the model generalize better.
[0146] Step 3: Neural Network Model Selection and Design: Choose a neural network architecture suitable for image processing, such as a convolutional neural network (CNN) or its variants.
[0147] According to the specific characteristics of breast case images, the neural network structure is designed. For example, an attention mechanism can be introduced to focus on the key areas in the image; multi-scale feature fusion can be used to capture lesions of different sizes; a deeper network structure can be designed to extract higher-level features, etc.
[0148] It should be noted that breast case images have some unique characteristics that can guide the improvement of the neural network structure to improve the accuracy of feature extraction and diagnosis. The following are some specific characteristics of breast case images and their suggestions for improving the neural network structure:
[0149] It should be noted that in breast case images, the lesion area may present a variety of different textures and forms, such as masses, calcifications, structural distortions, etc.
[0150] Improvements to this part could be to design convolutional layers or modules that can capture diverse texture and morphological features, such as using convolution kernels of different sizes and shapes, or introducing attention mechanisms to focus on important areas in the image.
[0151] It should be noted that microcalcifications or early tumors in breast images may be very small and blurry, making it difficult to distinguish from normal tissue.
[0152] Improvements to this part could be to enhance the detail capture capability of the network, for example by introducing high-resolution feature-preserving paths, multi-scale feature fusion, or using super-resolution techniques to improve sensitivity to small lesions.
[0153] It should be noted that the density of breast tissue may vary greatly between individuals and regions, which can affect the visibility of lesions and the difficulty of diagnosis.
[0154] An improvement method for this part may be to consider adding a preprocessing step of breast density standardization or correction in the neural network to reduce the interference of density differences on feature extraction.
[0155] It should be noted that breast diagnosis usually involves images of multiple modalities, such as mammography (mammography), ultrasound, MRI, etc., each of which provides different information.
[0156] An improvement to this part could be to design a neural network structure that can fuse multimodal image information, such as using a multi-input network, feature-level fusion, or decision-level fusion strategy.
[0157] Step 4: Feature extraction: Train the neural network model to learn and extract pathological features from breast case images.
[0158] The extracted features can include morphological features (such as edges, textures), spatial features (such as position, size), and high-level semantic features (such as lesion type).
[0159] Use the output of the intermediate layer or a specially designed feature extraction layer as the feature representation of the image for subsequent classification, regression, or clustering tasks.
[0160] Step 5: Evaluation and optimization: Use independent validation sets and test sets to evaluate the performance of the model. Common evaluation indicators include accuracy, sensitivity, specificity, etc.
[0161] Optimize the model based on the evaluation results, such as adjusting the network structure, improving the training strategy, introducing a new loss function, etc.
[0162] Consider using transfer learning techniques to leverage models pre-trained on other large image datasets to speed up the training process and improve performance.
[0163] Step 6: Deployment and monitoring: Deploy the trained model to the actual application environment to assist doctors in analyzing and diagnosing breast case images.
[0164] Monitor the performance of the model, and regularly update and optimize the model to adapt to new data and requirements to ensure the accuracy and stability of the model.
[0165] It should be noted that the use of case feature extraction includes improving the neural network through the specific characteristics of thyroid or breast case images, and using the improved neural network model for case feature extraction improves the model's ability to extract pathological features of thyroid and breast case images, enabling the model to more accurately identify and distinguish normal and abnormal cases. By improving the neural network structure in a targeted manner, the model's ability to learn different morphological and texture features is enhanced, and the generalization performance of the model in practical applications is improved.
[0166] It should be noted that information extraction includes extracting seven types of structured data information from case text data: date, person name, case type, age, identity ID, text case description, and auxiliary suggestion text;
[0167] It should be noted that the auxiliary suggestion text is the entire text of the doctor's suggestion part in the case, which may include relevant content such as surgical method, postoperative care, drug treatment, etc.
[0168] It should be noted that identity ID includes unique digital numbers such as ID card, medical record number, and medical insurance account number;
[0169] Furthermore, the preprocessing also includes establishing a mapping relationship data set based on the results of the structured processing, with case type as the first priority, age as the second priority, person name and identity ID as the third priority, and date as the fourth priority.
[0170] Table 1: Example data without structured processing after information extraction.
[0171]
[0172]
[0173] The third digit after the name represents the last Chinese character of the name, and Liu 2 has the same name but different diseases, and Wang 1 has the same name but different diseases;
[0174] Table 2: Breast disease mapping relationship dataset after priority processing.
[0175]
[0176]
[0177] Table 3: Thyroid disease mapping relationship dataset after priority processing.
[0178]
[0179] Furthermore, preprocessing also includes:
[0180] The mapping function is established using the improved direct mapping method, and the mapping relationship between the image and text data in the same case is established based on the mapping model;
[0181] Obtain a fixed-dimensional feature vector of the case image, convert the words in the case text into vector representations, and capture the contextual information of the text through a recurrent neural network structure to output the feature representation of the text;
[0182] The image feature vector and the text feature vector are fused, and the attention mechanism method is used to realize feature fusion to obtain the joint representation between the image and the text;
[0183] A fully connected neural network layer is designed as a mapping function to map the fused features to the target space. The target space is a vector space of fixed dimension, which is used to represent the mapping relationship between image and text.
[0184] In an optional embodiment, the mapping function may include an enhanced fully connected layer design, assuming that the fused feature vector is Where D z is the dimension of fusion features.
[0185] Going further, adaptive feature recalibration is performed:
[0186] First, an adaptive feature rescaling step is introduced to allow the network to learn the importance of each feature. This can be achieved by using a simple gating mechanism, such as the Sigmoid activation function:
[0187] g=σ(W g z+b g )
[0188] in: and are learnable weights and bias parameters. σ is the Sigmoid activation function, which is used to map feature importance to the (0,1) interval.
[0189] Then, the original feature vector is multiplied by the gated output to obtain the rescaled features:
[0190] Z'=z⊙g
[0191] where ⊙ represents element-wise multiplication.
[0192] Going one step further, an extended fully connected layer is used, which has multiple parallel linear transformations and concatenates their outputs. This allows the network to learn multiple different feature representations and increase the representation power of the model.
[0193] h i =W i z′+b i ,i=1,…,N
[0194] in: and are the weight and bias parameters of the ith linear transformation, and N is the number of parallel linear transformations, which is a hyperparameter and can be adjusted as needed.
[0195] Then, the outputs of all linear transformations are concatenated to form the final output vector:
[0196] h=[h1;h2;…;h N ]
[0197] Among them, [;] represents a vertical connection operation.
[0198] Furthermore, the dimension of the final output vector h is N×D h , where each h i They can all be regarded as feature representations in different subspaces.
[0199] Furthermore, the initialization and optimization of weights and biases are similar to the standard fully connected layer. The weight matrix Wi and the bias vector b i , and the weight W of the gating mechanism g and bias b g , are the parameters that the network needs to learn. They can be initialized randomly and updated using the back-propagation algorithm and the gradient descent optimizer.
[0200] Furthermore, the choice of loss function depends on the specific task. You can choose an appropriate loss function according to the requirements of the task, such as mean square error loss, cross entropy loss, etc., to guide the optimization of network parameters.
[0201] S102: performing a classification operation on the preprocessed mapping relationship data set to divide the preprocessed mapping relationship data set into a first data set and a second data set;
[0202] The first data set is data with at least two dates corresponding to the same person's name and identity ID in the same case type, and the date interval is greater than two days, or a data set with disease end marker words in the auxiliary suggestion text extracted from the information. The second data set is the part of the mapping relationship data set excluding the first data set;
[0203] It should be noted that the signs for the end of the illness are the signs that different doctors usually use, which may be "discharge recommended", "can be discharged", "agree to be discharged", "discharge recommended as the condition is stable", etc.
[0204] S103: Establishing a first model based on the first data set and in combination with a deep learning network, wherein the input of the first model is picture case features, text case description, and case type, and the output of the first model is auxiliary suggestion text;
[0205] Among them, the deep learning network includes:
[0206] The input layer of the deep learning network includes an image case feature input layer, a text case description input layer, and a case type input layer;
[0207] Furthermore, the input of the image case feature input layer is the preprocessed case image, and the output is a 2048-dimensional image feature vector;
[0208] Furthermore, the input of the text case description input layer is the preprocessed text case description, and the output is the hidden state of the last time step, which is a 256-dimensional text feature vector;
[0209] Furthermore, the input of the case type input layer is the case type represented by the embedding vector, and the embedding dimension is a 32-dimensional case type embedding vector;
[0210] Furthermore, the feature fusion layer of the deep learning network includes concatenating the image feature vector, text feature vector and case type embedding vector, outputting a vector of size 2336, and adding a fully connected layer after the feature fusion layer for dimensionality reduction, outputting a vector of size 512;
[0211] Furthermore, the hidden layer of the deep learning network uses two fully connected layers for nonlinear transformation. The first fully connected layer has an input size of 512 dimensions and an output size of 256 dimensions, using the ReLU activation function. The second fully connected layer has an input size of 256 dimensions and an output size of 128 dimensions, also using the ReLU activation function. A Dropout layer is added after each fully connected layer to prevent overfitting.
[0212] Furthermore, the output layer of the deep learning network uses the Transformer decoder structure to gradually generate text;
[0213] Furthermore, the connection method of the deep learning network includes parallel processing of the image case feature input layer, the text case description input layer and the case type input layer, and their outputs are spliced in the feature fusion layer. The output of the feature fusion layer is connected to the first hidden layer, and then connected to the subsequent hidden layers in turn. The output of the last hidden layer is connected to the output layer to generate the final auxiliary suggestion text.
[0214] It should be noted that if there are still the same case data related to the same person's name and identity ID after the date corresponding to the end of illness marker, it is judged whether this part of the case is good condition data. If so, when training the model, this part of the data needs to appear in both the training set and the test set;
[0215] If not, the case data of the date corresponding to the disease end marker word is deleted from the first type of data set.
[0216] S104: Score the first model. If the score is not lower than a preset threshold, perform intelligent case analysis based on the first model.
[0217] In an optional embodiment, the first model is trained using the first data set, and the model convergence is ensured.
[0218] Prepare an independent validation dataset, which should contain data with the same distribution as the first dataset but has not been used for training. The validation dataset may not contain data with at least three dates corresponding to the same person's name and ID in all cases of the same type, with a date interval greater than two days, and a dataset with disease end markers in the auxiliary suggestion text extracted from the information;
[0219] For each sample in the validation dataset, the first model is used to make predictions and generate auxiliary suggestion text.
[0220] Compare the model's predictions with the actual results in the validation dataset and calculate evaluation metrics.
[0221] The evaluation index calculation uses the weighted average method to calculate the comprehensive score:
[0222]
[0223] The evaluation indicators include accuracy, recall and precision, F1 score, and time interval accuracy. n represents the number of evaluation indicators, and i represents the evaluation indicator.
[0224] In an optional embodiment, the following steps may also be included:
[0225] S101: Acquire thyroid or breast case history data and preprocess the case history data, where the case history data includes pure image case history data, pure text case history data, and case history data combining image and text;
[0226] S102: performing a classification operation on the preprocessed mapping relationship data set to divide the preprocessed mapping relationship data set into a first data set and a second data set;
[0227] The first data set is data with at least two dates corresponding to the same person's name and identity ID in the same case type, and the date interval is greater than two days, or a data set with disease end marker words in the auxiliary suggestion text extracted from the information. The second data set is the part of the mapping relationship data set excluding the first data set;
[0228] S103: Establishing a first model based on the first data set and in combination with a deep learning network, wherein the input of the first model is picture case features, text case description, and case type, and the output of the first model is auxiliary suggestion text;
[0229] S104: scoring the first model, and if the score is not lower than a preset threshold, performing intelligent case analysis according to the first model;
[0230] S105: scoring the first model. If the score is lower than a preset threshold, expanding the second data set, combining the expanded second data set with the original first data set to form a new data set, re-establishing the second model based on the new data set combined with the deep learning network, and performing intelligent case analysis based on the second model;
[0231] The expanding the second data set includes:
[0232] The image data of the same dimension in the second data set is fused with feature maps by improving linear interpolation, and reconstructed back to image space through a decoder to generate a new medical image;
[0233] Specifically, the improved linear interpolation includes:
[0234]
[0235] Among them, F mixed Indicates the result after difference, F A 、F B They represent the feature maps of image data of the same dimension, α represents the difference parameter, and β represents the nonlinear degree control parameter;
[0236] When β = 0, the improved linear interpolation degenerates into the standard linear interpolation;
[0237] When β>0, the interpolation result will be biased towards F in a nonlinear manner according to the value of α. A or F B ;
[0238] It should be noted that the text data in the second data set is expanded based on the new medical image processed by improved linear interpolation and combined with the pre-trained NLP model.
[0239] It should be noted that the new data set is divided into a training set and a test set. The training set contains at least data in which there are at least three dates corresponding to the same person name and identity ID in all cases of the same type, and the date interval is greater than two days, and the data set contains the end-of-disease marker words in the auxiliary suggestion text extracted from the information. The test set includes at least the data set in which the end-of-disease marker words are present in the auxiliary suggestion text extracted from the information in the same case type in the second data set.
[0240] It should be noted that the second model scoring score is not lower than the first model scoring score.
[0241] In summary, the present invention proposes an intelligent case analysis method based on deep learning, obtains thyroid or breast case history data, and preprocesses the case history data, the case history data includes pure image case history data, pure text case history data and image and text combined case history data; the preprocessed mapping relationship data set is classified, and the preprocessed mapping relationship data set is divided into a first data set and a second data set; the first data set is data in which there are at least two dates corresponding to the same person name and identity ID in the same case type, and the date interval is greater than two days, or a data set in which the disease end marker word exists in the auxiliary suggestion text extracted from the information, and the second data set is the part of the mapping relationship data set except the first data set; according to the first data set, a first model is established in combination with a deep learning network, the input of the first model is the picture case feature, the text case description and the case type, and the output of the first model is the auxiliary suggestion text; the first model is scored, and if the score is not lower than the preset threshold, the intelligent case analysis is performed according to the first model. The present invention proposes an intelligent case analysis method and system based on deep learning, which can effectively analyze thyroid or breast cases. In addition, the present invention also improves the generalization ability and robustness of the model through data set classification operations. Therefore, the present invention provides an efficient, accurate and reliable method and system for intelligent case analysis.
[0242] In a preferred embodiment, the intelligent case analysis method based on deep learning includes:
[0243] A data acquisition module is used to acquire thyroid or breast case history data and preprocess the case history data, wherein the case history data includes pure image case history data, pure text case history data, and case history data combining image and text;
[0244] A data set classification module is used to perform a classification operation on the preprocessed mapping relationship data set, and divide the preprocessed mapping relationship data set into a first data set and a second data set;
[0245] The first data set is data with at least two dates corresponding to the same person's name and identity ID in the same case type, and the date interval is greater than two days, or a data set with disease end marker words in the auxiliary suggestion text extracted from the information. The second data set is the part of the mapping relationship data set excluding the first data set;
[0246] A model building module, used to build a first model based on the first data set in combination with a deep learning network, wherein the input of the first model is picture case features, text case descriptions, and case types, and the output of the first model is auxiliary suggestion text;
[0247] The scoring and analysis module is used to score the first model. If the score is not lower than a preset threshold, intelligent case analysis is performed based on the first model.
[0248] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.
[0249] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, an intelligent case analysis method based on deep learning is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse.
[0250] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
[0251] Acquire thyroid or breast case history data and preprocess the case history data, the case history data including pure image case history data, pure text case history data and case history data combining image and text;
[0252] Performing a classification operation on the preprocessed mapping relationship data set to divide the preprocessed mapping relationship data set into a first data set and a second data set;
[0253] The first data set is data with at least two dates corresponding to the same person's name and identity ID in the same case type, and the date interval is greater than two days, or a data set with disease end marker words in the auxiliary suggestion text extracted from the information. The second data set is the part of the mapping relationship data set excluding the first data set;
[0254] According to the first data set, a first model is established in combination with a deep learning network, the input of the first model is picture case features, text case descriptions and case types, and the output of the first model is auxiliary suggestion text;
[0255] The first model is scored, and if the score is not lower than the preset threshold, intelligent case analysis is performed based on the first model.
[0256] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
[0257] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.
[0258] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0259] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0260] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0261] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.
[0262] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
Claims
1. An intelligent case analysis method based on deep learning, characterized in that: include: Acquire thyroid or breast case history data and preprocess the case history data, wherein the case history data includes pure image case history data, pure text case history data, and case history data combining image and text; Performing a classification operation on the preprocessed mapping relationship data set to divide the preprocessed mapping relationship data set into a first data set and a second data set; The first data set is data in which there are at least two dates corresponding to the same person name and identity ID in the same case type, and the date interval is greater than two days, or a data set in which there are disease end marker words in the auxiliary suggestion text extracted from the information, and the second data set is the part of the mapping relationship data set excluding the first data set; According to the first data set, a first model is established in combination with a deep learning network, wherein the input of the first model is picture case features, text case descriptions, and case types, and the output of the first model is auxiliary suggestion text; Scoring the first model, and if the score is not lower than a preset threshold, performing intelligent case analysis based on the first model; Scoring the first model, if the score is lower than a preset threshold, expanding the second data set, combining the expanded second data set with the original first data set to form a new data set, re-establishing the second model based on the new data set combined with the deep learning network, and performing intelligent case analysis based on the second model; The new data set is divided into a training set and a test set, wherein the training set contains at least data in which the same person name and identity ID correspond to at least three dates, and the date interval is greater than two days in all cases of the same type, and a data set in which the auxiliary suggestion text extracted from the information contains a disease end marker word, and the test set contains at least a data set in which the auxiliary suggestion text extracted from the information in the same case type in the second data set contains a disease end marker word; The second model scoring score is not lower than the first model scoring score; The preprocessing includes image enhancement and case feature extraction for the pure image case history data, information extraction for the pure text case history data, and structural processing according to the results of information extraction, and information extraction for the image and text combined case history data, and structural processing according to the results of information extraction, and then image enhancement and case feature extraction; The image enhancement includes using an improved edge detection algorithm to highlight the edge of the thyroid tissue, and using an improved wavelet transform to highlight the internal features of the breast lesion area; The case feature extraction includes improving the neural network through specific features of thyroid or breast case images, and using the improved neural network model to extract case features; The improved neural network model designs a multi-scale feature extraction module, which enables the network to focus on both large-scale and small-scale features at the same time; Add a contrast enhancement module to the network to reduce noise interference; Balance the dataset by oversampling the minority class or undersampling the majority class. Use a weighted loss function in network training to give higher weights to the minority class. The information extraction includes extracting seven types of structured data information, namely, date, person name, case type, age, identity ID, text case description and auxiliary suggestion text, from the case text data; The preprocessing also includes establishing a mapping relationship data set with case type as the first priority, age as the second priority, person name and identity ID as the third priority, and date as the fourth priority according to the result of the structured processing; The pre-processing further comprises: Using an improved direct mapping method to establish a mapping function, and establishing a mapping relationship between images and text data in the same case according to the mapping function; Obtain a fixed-dimensional feature vector of the case image, convert the words in the case text into vector representations, and capture the contextual information of the text through a recurrent neural network structure to output the feature representation of the text; The image feature vector and the text feature vector are fused, and the attention mechanism method is used to realize feature fusion to obtain the joint representation between the image and the text; A fully connected neural network layer is designed as a mapping function to map the fused features to a target space, where the target space is a vector space of a fixed dimension and is used to represent the mapping relationship between images and texts.
2. The intelligent case analysis method based on deep learning according to claim 1, characterized in that: The expanding the second data set comprises: fusing feature maps of image data of the same dimension in the second data set by improved linear interpolation, and reconstructing them back to image space through a decoder to generate a new medical image; The improved linear interpolation includes: Among them, F mixed Indicates the result after difference, F A 、F B They represent the feature maps of image data of the same dimension, α represents the difference parameter, and β represents the nonlinear degree control parameter; When β = 0, the improved linear interpolation degenerates into the standard linear interpolation; When β>0, the interpolation result will be biased towards F in a nonlinear manner according to the value of α. A or F B ; According to the new medical image processed by improved linear interpolation and combined with the pre-trained NLP model, the text data in the second data set is expanded.
3. The intelligent case analysis method based on deep learning according to claim 2, characterized in that: The deep learning network includes: The input layer of the deep learning network includes an image case feature input layer, a text case description input layer, and a case type input layer; The input of the image case feature input layer is the preprocessed case image, and the output is a 2048-dimensional image feature vector; The input of the text case description input layer is the preprocessed text case description, and the output is the hidden state of the last time step, which is a 256-dimensional text feature vector; The input of the case type input layer is the case type represented by the embedding vector, and the embedding dimension is a 32-dimensional case type embedding vector; The feature fusion layer of the deep learning network includes concatenating the image feature vector, the text feature vector and the case type embedding vector, outputting a vector of size 2336, and adding a fully connected layer after the feature fusion layer for dimensionality reduction, outputting a vector of size 512; The hidden layer of the deep learning network uses two fully connected layers for nonlinear transformation. The first fully connected layer has an input size of 512 dimensions and an output size of 256 dimensions, and uses a ReLU activation function. The second fully connected layer has an input size of 256 dimensions and an output size of 128 dimensions, and also uses a ReLU activation function. A Dropout layer is added after each fully connected layer to prevent overfitting. The output layer of the deep learning network uses a Transformer decoder structure to gradually generate text; The connection method of the deep learning network includes parallel processing of the image case feature input layer, the text case description input layer and the case type input layer, and their outputs are spliced in the feature fusion layer. The output of the feature fusion layer is connected to the first hidden layer, and is connected to the subsequent hidden layers in turn. The output of the last hidden layer is connected to the output layer to generate the final auxiliary suggestion text.
4. The intelligent case analysis method based on deep learning according to claim 3, characterized in that: Also includes: If there are still the same case data with the same person's name and ID after the date corresponding to the end of illness marker, then determine whether these cases are good-condition data. If so, when training the model, these data must appear in both the training set and the test set. If not, the case data of the date corresponding to the disease end marker word is deleted from the first data set.
5. A system using the intelligent case analysis method based on deep learning as claimed in claim 1, characterized in that: include: A data acquisition module, used to acquire thyroid or breast case history data and preprocess the case history data, wherein the case history data includes pure image case history data, pure text case history data, and case history data combining image and text; A data set classification module is used to perform a classification operation on the preprocessed mapping relationship data set, and divide the preprocessed mapping relationship data set into a first data set and a second data set; The first data set is data in which there are at least two dates corresponding to the same person name and identity ID in the same case type, and the date interval is greater than two days, or a data set in which there are disease end marker words in the auxiliary suggestion text extracted from the information, and the second data set is the part of the mapping relationship data set excluding the first data set; A model building module, used to build a first model based on the first data set in combination with a deep learning network, wherein the input of the first model is picture case features, text case descriptions, and case types, and the output of the first model is auxiliary suggestion text; The scoring and analysis module is used to score the first model. If the score is not lower than a preset threshold, intelligent case analysis is performed based on the first model.
6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.
