Medical image management system based on SVM

Abstract

The invention relates to the technical field of machine learning, in particular to a medical image management system based on an SVM (Support Vector Machine). The method comprises the following steps: a feature construction expression unit performs multi-scale topological filtering on a medical image by using a persistent coherence analysis technology, extracts spatial structure features and spatio-temporal topological features of the medical image, and constructs a medical image comprehensive feature vector; the classification model training unit is used for training a support vector machine as a medical image classifier based on the comprehensive feature vector of the medical image, analyzing the spatial connectivity and the space-time consistency of the medical image in combination with a seepage theory, and assisting in optimizing the training of the support vector machine; the medical image classification unit performs classification operation on the medical images based on a medical image classifier; the storage data management unit stores and manages medical images. According to the medical image management system based on the SVM, medical image classification management is achieved through multi-scale topological feature extraction of a persistent coherence analysis technology and SVM regularization parameter optimization assisted by a seepage theory.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and more specifically, to a medical image management system based on SVM. Background Art

[0002] The medical image management system based on SVM aims to improve the classification accuracy of medical images and the data retrieval efficiency. Through persistent homology analysis technology, multi-scale topological filtering is performed on medical images, morphological, texture, and topological features are extracted and fused, and the regularization parameter optimization process of the support vector machine model is controlled to achieve efficient and accurate classification and management of medical images.

[0003] Existing medical image management systems usually have difficulty effectively capturing and integrating complex topological structures in medical images. Moreover, due to the spatial disconnection and spatio-temporal non-uniformity that easily occur during the classification management process of medical image data, problems such as low accuracy and high misclassification rate will occur when classifying multi-category images. Therefore, a medical image management system based on SVM is designed. Summary of the Invention

[0004] The purpose of the present invention is to provide a medical image management system based on SVM to solve the problems of low accuracy and high misclassification rate that occur when classifying multi-category images due to the spatial disconnection and spatio-temporal non-uniformity that easily occur during the classification management process of medical image data as mentioned in the above background art.

[0005] To achieve the above purpose, the present invention aims to provide a medical image management system based on SVM, including: A feature construction and representation unit that uses persistent homology analysis technology to perform multi-scale topological filtering on medical images, extracts the spatial structure features and spatio-temporal topological features of medical images, and constructs a comprehensive feature vector of medical images including morphological, texture, and topological features; It also includes a classification model training unit that trains a support vector machine as a medical image classifier based on the comprehensive feature vector of medical images, and analyzes the spatial connectivity and spatio-temporal consistency of medical images in combination with percolation theory to assist in optimizing the training process of the support vector machine; It also includes a medical image classification unit that performs classification operations on medical images based on the medical image classifier; It also includes a storage data management unit that is used to store and manage medical images.

[0006] As a further improvement of this technical solution, the feature construction and representation unit includes a multi-scale topological filtering module and a feature extraction and fusion module; Among them, the multi-scale topological filtering module performs multi-scale topological filtering on medical images using persistent homology analysis to construct a comprehensive topological feature set that includes the spatial structure features and spatio-temporal topological features of medical images; The feature extraction and fusion module is used to extract the morphological features and texture features of medical images, and fuse the morphological features, texture features with the comprehensive topological feature set constructed by the multi-scale topological filtering module to construct a comprehensive feature vector of medical images.

[0007] As a further improvement of this technical solution, the multi-scale topological filtering module performs multi-scale topological filtering on medical images using persistent homology analysis to construct a comprehensive topological feature set that includes the spatial structure features and spatio-temporal topological features of medical images. The specific method steps are as follows: S1.1.1. Standardize the medical image : ; ; ; Among them, is the gray value of the medical image at the pixel position ; is the abscissa of the pixel of the medical image ; is the ordinate of the pixel of the medical image ; is the mean value of the medical image ; is the standard deviation of the medical image ; is the height of the medical image; is the width of the medical image; is the gray value of the standard medical image at the pixel position ; S1.1.2. Use the Canny edge detection algorithm to extract the edge points of the standard medical image : ; Among them, is the set of edge points of the standard medical image ; is the Canny edge detection algorithm operation; is the coordinate of the th edge point; is the total number of edge points; S1.1.3. Based on the set of edge points of the standard medical image ​ , construct a topological complex using the Vietoris-Rips complex , control the scale parameter : ; ; Among them, is the topological complex; is the simplex in the topological complex; is the maximum Euclidean distance between all pairs of points in the simplex ; is the coordinate of the th boundary point; S1.1.4. Calculate the persistent homology of the topological complex : ; Among them, is the persistent homology of the topological complex , including connected components and loops ; is the persistent homology operation; S1.1.5. Set the persistence threshold , and extract topological features above the persistence threshold from the persistent homology: ; Among them, is the th topological feature; is the persistence of the th topological feature; is the scale parameter under the topological feature set; S1.1.6. For multiple scale parameters , repeat S1.1.2 to S1.1.5, extract the topological feature set under the scale parameter , and construct a comprehensive topological feature set including the spatial structure features and spatio-temporal topological features of the medical image: ; Among them, is the comprehensive topological feature set including the spatial structure features and spatio-temporal topological features of the medical image; is the scale parameter label; is the number of scale parameters.

[0008] As a further improvement of the present technical solution, the feature extraction and fusion module is used to extract the morphological features and texture features of medical images, and fuse the morphological features, texture features with the comprehensive topological feature set constructed by the multi-scale topological filtering module to construct a comprehensive feature vector of medical images. The specific method steps are as follows: S1.2.1. Extract the morphological features of the standard medical image using Hu moments: ; wherein, is the morphological feature; is the eigenvalue of the first Hu moment, and so on, is the eigenvalue of the seventh Hu moment; is the operation of extracting the morphological features of the standard medical image using Hu moments; S1.2.2. Extract the texture features of the standard medical image using the gray-level co-occurrence matrix features. The texture features include contrast, correlation, energy, and homogeneity: ; ; ; ; ; wherein, is the texture feature; is the operation of extracting the texture features of the standard medical image using the gray-level co-occurrence matrix features; is the contrast in the texture features; is the correlation in the texture features; is the energy in the texture features; is the homogeneity in the texture features; wherein, is the number of gray levels; is the element of the gray-level co-occurrence matrix; is the gray level ; gray level ; is the gray level standard deviation; is the gray level standard deviation; is the gray level mean; is the gray level mean; S1.2.3. Use the Persistence Image coding method to convert the comprehensive topological feature set into dimensional topological features: ; Among them, is the topological feature of dimension ; is the operation of converting the comprehensive topological feature set by the Persistence Image encoding method; S1.2.4. Fuse the morphological features, texture features and topological features to construct a comprehensive feature vector of medical images: ; Among them, is the comprehensive feature vector of medical images, .

[0009] As a further improvement of this technical solution, in S1.2.4, the morphological features, texture features and topological features are fused to construct a comprehensive feature vector of medical images. The specific method is as follows: ; Among them, is the comprehensive feature vector of medical images, .

[0010] As a further improvement of this technical solution, the classification model training unit includes a classifier training module and a percolation theory optimization module; Among them, the classifier training module trains a support vector machine based on the comprehensive feature vector of medical images as a medical image classifier; The percolation theory optimization module uses percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images and assist in optimizing the training process of the support vector machine.

[0011] As a further improvement of this technical solution, the classifier training module trains a support vector machine based on the comprehensive feature vector of medical images as a medical image classifier. The specific method steps are as follows: S2.1.1. Use the comprehensive feature vector of medical images and the class label as the training data set: ; Among them, is the training data set; is the subscript of the training data; is the th comprehensive feature vector of medical images of the training data; is the th class label of the training data; is the total number of training data; S2.1.2. Define the SVM optimization objective function as follows: ; where is the weight vector of the SVM, ; is the bias term of the SVM; is the slack variable of the th training data; is the regularization parameter, ; S2.1.3. Transform the SVM optimization objective function into its dual form function: Dual form function: ; where is the Lagrange multiplier of the th training data; is the Lagrange multiplier of the th training data; is the class label of the th training data; is the transpose of the comprehensive medical image feature vector of the th training data; is the comprehensive medical image feature vector of the th training data; S2.1.4. By solving the dual form function, obtain the optimal Lagrange multiplier , and calculate the weight vector and the bias term : ; ; S2.1.5. Repeat S2.1.1 - S2.1.4 to calculate the weight vector and the bias term for the th class label; S2.1.6. Obtain the medical image classifier: ; where is the medical image classifier; is the th class label; is the total number of class labels; is the weight vector of the th class label; is the Category label offset term; As a further improvement of this technical solution, the percolation theory optimization module uses the percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images, and assists in optimizing the training process of the support vector machine. The specific method steps are as follows: S2.2.1. Set the distance threshold , and use the connectivity concept in the percolation theory to analyze the connectivity index of medical images in the training dataset: ; ; Among them, is the distance threshold; is the connectivity index, indicating the total number of sample pairs connected under the distance threshold ; is the adjacency matrix indicating whether the comprehensive feature vector of the medical image of the th training data and the comprehensive feature vector of the medical image of the th training data are within the distance threshold ; S2.2.2. Gradually adjust the distance threshold , detect the change of the connectivity index , identify the critical point , and use the critical point to adjust the regularization parameter of the medical image classifier to obtain the optimized regularization parameter : ; Among them, and are both constants for adjusting parameters according to the percolation critical point; S2.2.3. Use the optimized regularization parameter to optimize the medical image classifier again.

[0012] As a further improvement of this technical solution, the medical image classification unit performs classification operations on medical images based on the medical image classifier. The specific method steps are as follows: S3.1. Receive the comprehensive feature vector of the medical image from the feature construction and representation unit, and verify the dimension and data integrity of the comprehensive feature vector of the medical image: ; Among them, is used to verify the dimension and data integrity of the input data; S3.2. Use the medical image classifier Calculate the comprehensive feature vector of medical images for the decision function value: ; wherein, is the comprehensive feature vector of medical images at the category label decision function value; S3.3. Determine the category label of the medical image according to the comprehensive feature vector of the medical image at the category label decision function value : ; wherein, is the category label.

[0013] As a further improvement of the present technical solution, the storage data management unit includes a medical image storage module and a data management module; wherein, the medical image storage module receives the medical images from the feature construction and representation unit , assigns a unique identifier to each medical image, and stores the medical images in a database; the data management module is used to receive the medical image retrieval request from the user interface and retrieve the classified medical images for the user; The medical image retrieval request includes query conditions and retrieval parameters.

[0014] Compared with the prior art, the beneficial effects of the present invention: 1. In the SVM-based medical image management system, multi-scale topological feature extraction is performed based on the persistent homology analysis technology, which can effectively capture the complex spatial structure and topological information in medical images and significantly improve the classification accuracy.

[0015] 2. In the SVM-based medical image management system, the spatial connectivity and spatio-temporal consistency of medical images are analyzed through percolation theory, and the regularization parameters of the support vector machine are optimized to improve the classification processing ability of the medical image classifier. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the overall flow block diagram of the present invention; The meanings of the various reference numerals in the figure are as follows: 1. Feature construction and representation unit; 2. Classification model training unit; 3. Medical image classification unit; 4. Storage data management unit; 11. Multi-scale topological filtering module; 12. Feature extraction and fusion module; 21. Classifier training module; 22. Percolation theory optimization module; 41. Medical image storage module; 42. Data management module. Detailed implementation

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Please refer to Figure 1 As shown, a medical image management system based on SVM is provided, including: a feature construction and representation unit 1. The feature construction and representation unit 1 uses persistent homology analysis technology to perform multi-scale topological filtering on medical images, extracts the spatial structure features and spatio-temporal topological features of medical images, and constructs a comprehensive feature vector of medical images including shape, texture, and topological features. The feature construction and representation unit 1 includes a multi-scale topological filtering module 11 and a feature extraction and fusion module 12; Among them, the multi-scale topological filtering module 11 uses persistent homology analysis to perform multi-scale topological filtering on medical images, and constructs a comprehensive topological feature set including the spatial structure features and spatio-temporal topological features of medical images; The feature extraction and fusion module 12 is used to extract the shape features and texture features of medical images, and fuse the shape features, texture features with the comprehensive topological feature set constructed by the multi-scale topological filtering module 11 to construct a comprehensive feature vector of medical images.

[0019] In this embodiment, the multi-scale topological filtering module 11 uses persistent homology analysis to perform multi-scale topological filtering on medical images, and constructs a comprehensive topological feature set including the spatial structure features and spatio-temporal topological features of medical images. The specific method steps are as follows: S1.1.1. Standardize the medical image : ; ; ; Among them, is the gray value of the medical image at the pixel position ; For medical images The abscissa of the pixel; For medical images The ordinate of the pixel; For medical images The mean value of; For medical images The standard deviation of; The height of the medical image; The width of the medical image; For the standard medical image At the pixel position The gray value of; S1.1.2. Extract the edge points of the standard medical image using the Canny edge detection algorithm: The edge points: ; Among them, For the standard medical image The set of edge points; The Canny edge detection algorithm operation; For the coordinates of the The total number of edge points; S1.1.3. Based on the set of edge points of the standard medical image , use the Vietoris-Rips complex to construct the topological complex , controlling the scale parameter : ; ; Among them, The topological complex; The simplex in the topological complex; For the simplex The maximum Euclidean distance between all pairs of points in; For the coordinates of the S1.1.4. Calculate the persistent homology of the topological complex : ; Among them, The persistent homology of the topological complex , including the connected components and loops ; The persistent homology operation; S1.1.5. Set the persistence threshold , extracting topological features above the persistence threshold from persistent homology: ; wherein, is the th topological feature; is the persistence of the th topological feature; is the set of topological features under the scale parameter ; S1.1.6. For multiple scale parameters , repeat S1.1.2 to S1.1.5 to extract the set of topological features under the scale parameter , and construct a comprehensive topological feature set containing the spatial structure features and spatio-temporal topological features of medical images: ; wherein, is the comprehensive topological feature set containing the spatial structure features and spatio-temporal topological features of medical images; is the scale parameter label; is the number of scale parameters.

[0020] In this embodiment, the feature extraction and fusion module 12 is used to extract the morphological features and texture features of medical images, and fuse the morphological features, texture features with the comprehensive topological feature set constructed by the multi-scale topological filtering module 11 to construct a comprehensive feature vector of medical images. The specific method steps are as follows: S1.2.1. Use Hu moments to extract the morphological features of standard medical images: ; wherein, is the morphological feature; is the eigenvalue of the first Hu moment, and so on, is the eigenvalue of the seventh Hu moment; is the operation of using Hu moments to extract the morphological features of standard medical images; S1.2.2. Use gray-level co-occurrence matrix features to extract the texture features of standard medical images. The texture features include contrast, correlation, energy, and homogeneity: ; ; ; ; ; wherein, is the texture feature; Operation for extracting texture features of standard medical images using gray-level co-occurrence matrix features; is the contrast in the texture features; is the correlation in the texture features; is the energy in the texture features; is the homogeneity in the texture features; where, is the number of gray levels; is the element of the gray-level co-occurrence matrix; is the gray level ; gray level ; is the gray level standard deviation; is the gray level standard deviation; is the gray level mean; is the gray level mean; S1.2.3. Use the Persistence Image encoding method to convert the comprehensive topological feature set into dimensional topological features: ; where, is dimensional topological features, ; is the operation of converting the comprehensive topological feature set by the Persistence Image encoding method; In this embodiment, the operation of converting the comprehensive topological feature set by the Persistence Image encoding method is as follows: For the th topological feature and the persistence of the th topological feature perform coordinate mapping, convert the mapped coordinates into image pixel values through the Gaussian kernel function, superimpose all feature points to form an image of a fixed size, and finally convert it into dimensional topological features ; S1.2.4. Fuse the morphological features, texture features, and topological features to construct a comprehensive feature vector of medical images.

[0021] In this embodiment S1.2.4, the method of fusing the morphological features, texture features, and topological features to construct a comprehensive feature vector of medical images is as follows: ; Among them, is the comprehensive feature vector of medical images, .

[0022] It further includes a classification model training unit 2. The classification model training unit 2 trains a support vector machine as a medical image classifier based on the comprehensive feature vector of medical images, and analyzes the spatial connectivity and spatio-temporal consistency of medical images in combination with percolation theory to assist in optimizing the training process of the support vector machine; The classification model training unit 2 includes a classifier training module 21 and a percolation theory optimization module 22; Among them, the classifier training module 21 trains a support vector machine as a medical image classifier based on the comprehensive feature vector of medical images; The percolation theory optimization module 22 uses percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images to assist in optimizing the training process of the support vector machine.

[0023] In this embodiment, the classifier training module 21 trains a support vector machine as a medical image classifier based on the comprehensive feature vector of medical images. The specific method steps are as follows: S2.1.1. Take the comprehensive feature vector of medical images and the class label as the training data set: ; Among them, is the training data set; is the subscript of the training data; is for the th comprehensive feature vector of medical images of the training data; is for the th class label of the training data; is the total number of training data; S2.1.2. Define the SVM optimization objective function as follows: ; Among them, is the weight vector of the SVM, ; is the bias term of the SVM; is the th slack variable of the training data; is the regularization parameter, ; S2.1.3. Convert the SVM optimization objective function into a dual form function: Dual form function: ; wherein, is the Lagrange multiplier of the -th training data; is the Lagrange multiplier of the -th training data; is the class label of the -th training data; is the transpose of the comprehensive medical image feature vector of the -th training data; is the comprehensive medical image feature vector of the -th training data; S2.1.4. By solving the dual-form function, obtain the optimal Lagrange multiplier , calculate the weight vector and the bias term : ; ; S2.1.5. Repeat S2.1.1 - S2.1.4 to calculate the weight vector and the bias term of the -th class label; S2.1.6. Obtain the medical image classifier: ; wherein, is the medical image classifier; is the -th class label; is the total number of class labels; is the weight vector of the -th class label; is the bias term of the -th class label; In this embodiment, the percolation theory optimization module 22 uses the percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images, and assist in optimizing the training process of the support vector machine. The specific method steps are as follows: S2.2.1. Set the distance threshold , and use the connectivity concept in the percolation theory to analyze the connectivity index of medical images in the training dataset: ; ; wherein, is the distance threshold; is the connectivity index, indicating the total number of sample pairs connected under the distance threshold ; is the The adjacency matrix between the comprehensive feature vectors of the medical images of the -th training data and the comprehensive feature vectors of the medical images of the -th training data within the distance threshold; S2.2.2. Gradually adjust the distance threshold , detect the change of the connectivity index , identify the critical point , and use the critical point to adjust the regularization parameter of the medical image classifier to obtain the optimized regularization parameter : ; wherein, and are both constants for adjusting parameters according to the percolation critical point; S2.2.3. Use the optimized regularization parameter to optimize the medical image classifier again.

[0024] It further includes a medical image classification unit 3, and the medical image classification unit 3 classifies medical images based on the medical image classifier; In this embodiment, the medical image classification unit 3 classifies medical images based on the medical image classifier, and the specific method steps are as follows: S3.1. Receive the comprehensive feature vector of the medical image from the feature construction and representation unit 1 , and verify the dimension and data integrity of the comprehensive feature vector of the medical image : ; wherein, is used to verify the dimension and data integrity of the input data; S3.2. Use the medical image classifier to calculate the decision function value of the comprehensive feature vector of the medical image : ; wherein, is the decision function value of the comprehensive feature vector of the medical image for the -th category label; S3.3. Determine the category label of the medical image according to the decision function value of the comprehensive feature vector of the medical image for the -th category label : ; Among them, is the category label.

[0025] It also includes a storage data management unit 4, and the storage data management unit 4 is used for storing and managing medical images.

[0026] In this embodiment, the storage data management unit 4 includes a medical image storage module 41 and a data management module 42; Among them, the medical image storage module 41 receives medical images from the feature construction representation unit 1 , assigns a unique identifier to each medical image, and stores the medical images in the database; The data management module 42 is used to receive a medical image retrieval request from the user interface and retrieve the classified medical images for the user; The medical image retrieval request includes query conditions and retrieval parameters.

[0027] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A medical image management system based on SVM, characterized in that, Including: A feature construction and representation unit (1), which uses persistent homology analysis technology to perform multi-scale topological filtering on medical images, extracts the spatial structure features and spatio-temporal topological features of medical images, and constructs a comprehensive feature vector of medical images including morphological, texture, and topological features; A classification model training unit (2), which trains a support vector machine as a medical image classifier based on the comprehensive feature vector of medical images, and combines percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images to assist in optimizing the training process of the support vector machine; A medical image classification unit (3), which performs classification operations on medical images based on the medical image classifier; A storage data management unit (4), which is used to store and manage medical images.

2. The medical image management system based on SVM according to claim 1, wherein: The feature construction and representation unit (1) includes a multi-scale topological filtering module (11) and a feature extraction and fusion module (12); Among them, the multi-scale topological filtering module (11) uses persistent homology analysis to perform multi-scale topological filtering on medical images, and constructs a comprehensive topological feature set including the spatial structure features and spatio-temporal topological features of medical images; The feature extraction and fusion module (12) is used to extract the morphological features and texture features of medical images, and fuse the morphological features, texture features with the comprehensive topological feature set constructed by the multi-scale topological filtering module (11) to construct a comprehensive feature vector of medical images.

3. The medical image management system based on SVM according to claim 2, characterized in that: The multi-scale topological filtering module (11) uses persistent homology analysis to perform multi-scale topological filtering on medical images, and constructs a comprehensive topological feature set including the spatial structure features and spatio-temporal topological features of medical images. The specific method steps are as follows: S1.1.

1. Standardize the medical image : ； ； ； Among them, is the medical image at the pixel position gray value; is the medical image abscissa of the pixel; is the medical image ordinate of the pixel; is the medical image mean value; is the medical image standard deviation; is the height of the medical image; is the width of the medical image; is the standard medical image at the pixel position gray value; S1.1.

2. Extract the edge points of the standard medical image using the Canny edge detection algorithm : ； Among them, is the set of edge points of the standard medical image ; is the operation of the Canny edge detection algorithm; is the -th coordinate of the edge point; is the total number of edge points; S1.1.

3. Based on the standard medical images of the edge point set , construct a topological complex using the Vietoris-Rips complex , and control the scale parameter : ； ； Among them, is a topological complex; is a simplex in the topological complex; is the simplex is the maximum Euclidean distance between all pairs of points in; is the coordinate of the th edge point; S1.1.

4. Calculate the topological complex of persistent homology: ； Among them, is the persistent homology of the topological complex, including connected components and loops ; is a persistent homology operation; S1.1.

5. Set the persistence threshold , and extract topological features higher than the persistence threshold from persistent homology: ； Among them, is the th topological feature; is the persistence of the th topological feature; is the set of topological features at the scale parameter ; S1.1.

6. For multi-scale parameters , repeat S1.1.2 to S1.1.5 to extract the topological feature sets under the scale parameters , and construct a comprehensive topological feature set that includes the spatial structure features and spatio-temporal topological features of medical images: ； Among them, is a comprehensive topological feature set containing the spatial structure features and spatio-temporal topological features of medical images; is the scale parameter label; is the number of scale parameters.

4. The medical image management system based on SVM according to claim 3, characterized in that: The feature extraction and fusion module (12) is used to extract the morphological features and texture features of medical images, and fuse the morphological features, texture features with the comprehensive topological feature set constructed by the multi-scale topological filtering module (11) to construct a comprehensive feature vector of medical images. The specific method steps are as follows: S1.2.

1. Use Hu moments to extract the morphological features of standard medical images: ； Among them, is the morphological feature; is the eigenvalue of the first Hu moment, and so on, is the eigenvalue of the seventh Hu moment; is the operation of extracting the morphological features of standard medical images by Hu moments; S1.2.

2. Use gray-level co-occurrence matrix features to extract the texture features of standard medical images. The texture features include contrast, correlation, energy, and homogeneity: ； ； ； ； ； Among them, is the texture feature; is the operation of extracting the texture feature of the standard medical image by using the gray-level co-occurrence matrix feature; is the contrast in the texture feature; is the correlation in the texture feature; is the energy in the texture feature; is the homogeneity in the texture feature; Among them, is the number of gray levels; is an element of the gray-level co-occurrence matrix; is the gray level ; gray level ; is the standard deviation of the gray level ; is the standard deviation of the gray level ; is the mean of the gray level ; is the mean of the gray level ; S1.2.

3. Convert the comprehensive topological feature set into topological features of dimensions by using the Persistence Image encoding method: ； Among them, is the topological feature of dimension ; is the operation of converting the comprehensive topological feature set by the Persistence Image encoding method; S1.2.

4. Fuse the morphological features, texture features, and topological features to construct a comprehensive feature vector of medical images.

5. The medical image management system based on SVM according to claim 4, wherein: In S1.2.4, the morphological features, texture features, and topological features are fused to construct a comprehensive feature vector of medical images. The specific method is as follows: ； Among them, is the comprehensive feature vector of medical images, .

6. The medical image management system based on SVM according to claim 5, wherein: The classification model training unit (2) includes a classifier training module (21) and a percolation theory optimization module (22); Among them, the classifier training module (21) trains a support vector machine as a medical image classifier based on the comprehensive feature vector of medical images; The percolation theory optimization module (22) uses percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images to assist in optimizing the training process of the support vector machine.

7. The SVM-based medical image management system according to claim 6, characterized in that: The classifier training module (21) trains a support vector machine based on the comprehensive feature vector of medical images as a medical image classifier. The specific method steps are as follows: S2.1.

1. Use the comprehensive medical image feature vector and the class label as the training data set: ； Among them, is the training dataset; is the subscript of the training data; is the comprehensive medical image feature vector of the th training data; is the class label of the th training data; is the total number of training data; S2.1.

2. Define the SVM optimization objective function as follows: ； Among them, is the weight vector of the SVM, ; is the bias term of the SVM; is the slack variable of the th training data; is the regularization parameter, ; S2.1.

3. Convert the SVM optimization objective function into a dual-form function: Dual-form function: ； wherein, is the Lagrange multiplier of the th training data; is the Lagrange multiplier of the th training data; is the class label of the th training data; is the transpose of the comprehensive medical image feature vector of the th training data; is the comprehensive medical image feature vector of the th training data; S2.1.

4. Obtain the optimal Lagrange multipliers by solving the dual-form function , calculate the weight vector and the bias term : ； ; S2.1.

5. Repeat S2.1.1 - S2.1.4 to calculate the weight vector of the category - specific label and the bias term ; S2.1.

6. Obtain the medical image classifier: ； Among them, is a medical image classifier; is the category label of the total number of category labels; is the weight vector of the category label of the is the bias term of the category label of the 8. The medical image management system based on SVM according to claim 7, wherein: The percolation theory optimization module (22) uses percolation theory to analyze the spatial connectivity and spatio-temporal consistency of medical images, and assists in optimizing the training process of the support vector machine. The specific method steps are as follows: S2.2.

1. Set the distance threshold , and using the concept of connectivity in percolation theory, analyze the connectivity index of medical images in the training dataset: ； ； wherein, is the distance threshold; is the connectivity index, representing the total number of sample pairs connected under the distance threshold ; is the adjacency matrix indicating whether the comprehensive medical image feature vectors of the th training data and the comprehensive medical image feature vectors of the th training data are within the distance threshold ; S2.2.

2. Gradually adjust the distance threshold , and detect the change of the connectivity index to identify the critical point , and use the critical point to adjust the regularization parameter of the medical image classifier to obtain the optimized regularization parameter : ； Among them, and are both constants for adjusting parameters according to the percolation critical point; S2.2.

3. Optimize the medical image classifier again using the optimized regularization parameter , and optimize the medical image classifier again 9. The medical image management system based on SVM according to claim 8, characterized in that: The medical image classification unit (3) classifies medical images based on the medical image classifier. The specific method steps are as follows: S3.

1. Receive the comprehensive medical image feature vector from the feature construction and representation unit (1) , and verify the dimension and data integrity of the comprehensive medical image feature vector : ； Among them, used to verify the dimension and data integrity of the input data; S3.

2. Use a medical image classifier Calculate the comprehensive feature vector of the medical image and the decision function value: ； Among them, is the comprehensive feature vector of medical images at the category label decision function value; S3.

3. Determine the class label of the medical image according to the comprehensive feature vector of the medical image At the class label decision function value of the category, determine the class label of the medical image: ； Among them, is the category label.

10. The medical image management system based on SVM according to claim 9, wherein: The stored data management unit (4) includes a medical image storage module (41) and a data management module (42); Among them, the medical image storage module (41) receives medical images from the feature construction representation unit (1) , assigns a unique identifier to each medical image, and stores the medical images in a database; The data management module (42) is used to receive a medical image retrieval request from the user interface and retrieve the classified medical images for the user; The medical image retrieval request includes query conditions and retrieval parameters.

Images