A method and system for magnetic resonance-based analysis of brain disease categories

Abstract

The application relates to the field of disease category analysis, and discloses a brain disease category analysis method based on magnetic resonance, which comprises the following steps: performing image preprocessing on a magnetic resonance brain image of a brain patient to obtain a preprocessed image; calculating the brain tissue surface area of the brain patient and determining the brain tissue shape of the brain patient; identifying the brain tissue features of the brain patient, performing wavelet coefficient decomposition, and obtaining different frequency subband coefficients of the brain tissue features; calculating the statistical features of the different frequency subband coefficients, constructing a feature vector of the statistical features based on the statistical features, performing brain disease analysis on the brain patient, and obtaining a preliminary disease analysis category; extracting the texture features of the preprocessed image, performing feature fusion on the texture features and the feature vector, obtaining fused features, and performing accuracy analysis on the preliminary disease analysis category; and when the accuracy analysis result is optimal, constructing a disease analysis report. The application can improve the accuracy of brain disease category analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of disease category analysis, and in particular to a magnetic resonance-based brain disease category analysis method and system. BACKGROUND

[0002] In today's medical and biomedical field, the research and diagnosis of brain diseases has always been an important topic. With the continuous development of magnetic resonance imaging technology, it provides an important means for the diagnosis and analysis of brain diseases. In particular, the magnetic resonance-based brain disease category analysis method has attracted widespread attention because it can provide high-resolution brain structure and function information.

[0003] Currently, traditional brain disease category analysis methods mainly rely on clinical symptoms and neuropsychological tests. Although these methods can analyze brain disease categories to some extent, they have obvious limitations. Among them, the subjective judgment of clinical symptoms is easily affected by the experience of doctors and individual differences of patients, leading to uncertainty of the category analysis results; neuropsychological tests may be disturbed by the cooperation degree of patients and the test environment during the test, affecting the accuracy of the test results. SUMMARY

[0004] The present application provides a magnetic resonance-based brain disease category analysis method and system, which aims to improve the accuracy of brain disease category analysis.

[0005] To achieve the above-mentioned purpose, the present application provides a magnetic resonance-based brain disease category analysis method, comprising:

[0006] Collecting a magnetic resonance brain image of a brain patient, performing image preprocessing on the magnetic resonance brain image to obtain a preprocessed image;

[0007] Performing background separation on the preprocessed image to obtain a separated image, calculating the brain tissue volume of the brain patient based on the separated image, calculating the brain tissue surface area of the brain patient based on the separated image, and determining the brain tissue shape of the brain patient;

[0008] Identifying the brain tissue features of the brain patient based on the brain tissue volume, the brain tissue surface area, and the brain tissue shape, performing dimension reduction processing on the brain tissue features to obtain reduced dimension features, and performing wavelet coefficient decomposition on the reduced dimension features to obtain different frequency sub-band coefficients of the brain tissue features;

[0009] Identifying the statistical features of the different frequency sub-band coefficients, constructing a feature vector of the statistical features based on the statistical features, performing brain disease analysis on the brain patient based on the feature vector to obtain a preliminary disease analysis category;

[0010] Extracting a texture feature of the preprocessed image, performing feature fusion on the texture feature and the feature vector to obtain a fusion feature, performing accuracy analysis on the preliminary disease analysis category by using the fusion feature, and constructing a disease category analysis report of the brain patient when the accuracy analysis result is optimal.

[0011] Optionally, the image preprocessing on the magnetic resonance brain image to obtain a preprocessed image comprises:

[0012] Converting the magnetic resonance brain image into a converted image;

[0013] Identifying an original pixel value of the converted image;

[0014] Performing mean value filtering on the original pixel value to obtain a mean value pixel;

[0015] Determining a mean value image of the converted image based on the mean value pixel;

[0016] Performing image sharpening on the mean value image to obtain a preprocessed image.

[0017] Optionally, the background separation on the preprocessed image to obtain a separated image comprises:

[0018] Identifying an image size of the preprocessed image;

[0019] Constructing a full zero matrix of the preprocessed image based on the image size;

[0020] Constructing a background model of the preprocessed image by using the full zero matrix;

[0021] Performing background separation on the preprocessed image by using the background model to obtain a separated image.

[0022] Optionally, the background separation on the preprocessed image to obtain a separated image by using the background model comprises:

[0023] Performing image dilation on the preprocessed image to obtain a dilated image, and performing image erosion on the preprocessed image to obtain an eroded image;

[0024] Performing center processing on the dilated image by using the background model to obtain a center processed image;

[0025] Performing edge processing on the eroded image to obtain an edge processed image;

[0026] Performing image splicing on the center processed image and the edge processed image to obtain a separated image.

[0027] Optionally, the calculating the brain tissue volume of the brain patient based on the separated image comprises:

[0028] identifying an image voxel of the separated image;

[0029] calculating a voxel volume of the image voxel;

[0030] calculating the brain tissue volume of the brain patient based on the voxel volume.

[0031] Optionally, the identifying the brain tissue feature of the brain patient based on the brain tissue volume, the brain tissue surface area and the brain tissue shape comprises:

[0032] performing data standardization processing on the brain tissue volume, the brain tissue surface area and the brain tissue shape to obtain standardized data;

[0033] extracting normal brain tissue data corresponding to the brain patient;

[0034] comparing the standardized data with the normal brain tissue data to obtain a comparison result;

[0035] determining the brain tissue feature of the brain patient based on the comparison result.

[0036] Optionally, the wavelet coefficient decomposition of the reduced dimension feature to obtain different frequency sub-band coefficients of the brain tissue feature comprises:

[0037] obtaining an analysis requirement of the brain tissue feature;

[0038] configuring a wavelet function based on the analysis requirement, and setting a decomposition layer number of the reduced dimension feature;

[0039] performing wavelet transform on the reduced dimension feature based on the wavelet function and the decomposition layer number to obtain different frequency sub-band coefficients of the brain tissue feature.

[0040] Optionally, the extracting the texture feature of the pre-processed image comprises: converting the pre-processed image into a gray-scale image;

[0041] calculating a gray-scale number of the gray-scale image, and calculating a pixel distance of the gray-scale image;

[0042] calculating a texture feature value of the pre-processed image based on the gray-scale number and the pixel distance;

[0043] determining the texture feature of the pre-processed image based on the texture feature value.

[0044] Optionally, the feature fusion of the texture feature and the feature vector to obtain a fusion feature comprises:

[0045] convert the texture feature into a texture vector;

[0046] dimensionally convert the texture vector and the feature vector to obtain a dimensionally converted vector;

[0047] encode the dimensionally converted vector to obtain an encoded vector;

[0048] perform encoding fusion on the encoded vector by using a pre-constructed fusion matrix to obtain a fusion encoding;

[0049] perform encoding translation on the fusion encoding to obtain a fusion feature.

[0050] To solve the above problems, the application further provides a brain disease category analysis system based on magnetic resonance, which comprises:

[0051] an image preprocessing module, configured to collect a magnetic resonance brain image of a brain patient, perform image preprocessing on the magnetic resonance brain image, and obtain a preprocessed image;

[0052] a brain tissue shape analysis module, configured to perform background separation on the preprocessed image to obtain a separated image, calculate a brain tissue volume of the brain patient based on the separated image, calculate a brain tissue surface area of the brain patient based on the separated image, and determine a brain tissue shape recognition of the brain patient;

[0053] a feature decomposition module, configured to recognize brain tissue features of the brain patient based on the brain tissue volume, the brain tissue surface area and the brain tissue shape recognition, perform dimension reduction processing on the brain tissue features to obtain reduced dimension features, and perform wavelet coefficient decomposition on the reduced dimension features to obtain different frequency subband coefficients about the brain tissue features;

[0054] a disease category preliminary analysis module, configured to recognize statistical features of the different frequency subband coefficients, construct a feature vector of the statistical features based on the statistical features, perform brain disease analysis on the brain patient based on the feature vector, and obtain a preliminary disease analysis category;

[0055] a disease analysis report module, configured to extract texture features of the preprocessed image, perform feature fusion on the texture features and the feature vector to obtain a fusion feature, perform accuracy analysis on the preliminary disease analysis category by using the fusion feature, and construct a disease category analysis report of the brain patient when the accuracy analysis result is optimal.

[0056] The application can help users or doctors to analyze brain diseases more accurately by collecting the magnetic resonance brain image of the brain patient; optionally, the application can separate the background part in the image to enhance the analysis accuracy of the image target area by separating the background of the preprocessed image to obtain a separated image; the application can reflect the growth, atrophy or abnormal increase of brain tissue by calculating the brain tissue volume of the brain patient based on the separated image; the application can obtain the morphological characteristics of the brain tissue of the brain patient, and preliminarily understand whether the brain patient has a brain disease or which brain disease by identifying the brain tissue characteristics of the brain patient based on the brain tissue volume, the brain tissue surface area and the brain tissue shape; the application can describe the distribution of the sub-band coefficient by identifying the statistical characteristics of the different frequency sub-band coefficients, and provide valuable information for subsequent feature construction and classification; finally, the application can research and quantify the gray value distribution mode and spatial relationship of pixels or voxels in the brain image by extracting the texture features of the preprocessed image, so as to obtain feature information capable of describing the internal structure and organization characteristics of the image. Therefore, the application can improve the accuracy of brain disease category analysis. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 A flowchart of a brain disease category analysis method based on magnetic resonance provided by an embodiment of the application is shown in the figure.

[0058] Figure 2 A functional module diagram of a brain disease category analysis system based on magnetic resonance provided by an embodiment of the application is shown in the figure.

[0059] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0060] It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application.

[0061] The embodiment of the present application provides a brain disease category analysis method based on magnetic resonance. The execution subject of the brain disease category analysis method based on magnetic resonance includes but is not limited to at least one of electronic devices such as a server, a terminal and the like which can be configured to execute the method provided by the embodiment of the present application. In other words, the brain disease category analysis method based on magnetic resonance can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster and the like. The server can be a stand-alone server, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0062] Referring to Figure 1 FIG. 1 is a flowchart of a brain disease category analysis method based on magnetic resonance provided by an embodiment of the present application. In the embodiment, the brain disease category analysis method based on magnetic resonance includes the following steps.

[0063] S1, acquiring a magnetic resonance brain image of a brain patient, performing image preprocessing on the magnetic resonance brain image to obtain a preprocessed image.

[0064] The embodiment of the present application can help users or doctors to analyze brain diseases more accurately by acquiring a magnetic resonance brain image of a brain patient. The magnetic resonance brain image refers to a brain image obtained by magnetic resonance imaging (MRI) technology, which can be acquired by a magnetic resonance imaging device.

[0065] Further, the embodiment of the present application can enhance image details or perform some image operations before analyzing the image, thereby reducing the operation difficulty of a certain step, by performing image preprocessing on the magnetic resonance brain image to obtain a preprocessed image. For example, in the present scheme, an image background segmentation operation is required, and therefore a mean filtering operation is performed during preprocessing to facilitate image background separation.

[0066] As an embodiment of the present application, the image preprocessing on the magnetic resonance brain image to obtain a preprocessed image includes: performing format conversion on the magnetic resonance brain image to obtain a converted image, identifying original pixel values of the converted image, performing mean filtering on the original pixel values by using the following formula to obtain mean pixel values:

[0067]

[0068] Wherein, L(x, y) represents the mean pixel, (x, y) represents a pixel point of the conversion image, m represents the number of rows of the pixel point (x, y) field, n represents the number of columns of the pixel point (x, y) field, f((x, y) represents the pixel value of the pixel point (x, y) in the conversion image,

[0069] Based on the mean pixel, a mean image of the conversion image is determined, and the mean image is image sharpened to obtain a pretreatment image.

[0070] Optionally, the format conversion of the magnetic resonance brain image to obtain the conversion image refers to converting the format of one image file into another format to meet different image processing requirements, which can be converted through the Adobe Photoshop tool, and the original pixel value of the conversion image can be obtained by reading the image and accessing the numerical value of each pixel through the opencv-python library in Python, and the image sharpening of the mean image to obtain the pretreatment image refers to enhancing the edge details of the image, which can be operated through the GIMP tool.

[0071] S2, background separation is performed on the pretreatment image to obtain a separation image, based on the separation image, the brain tissue volume of the brain patient is calculated, based on the separation image, the brain tissue surface area of the brain patient is calculated, and the brain tissue shape of the brain patient is determined.

[0072] The embodiment of the application can separate the background part in the image through the background separation of the pretreatment image to obtain a separation image, thereby enhancing the analysis accuracy of the image target differentiation.

[0073] As an embodiment of the application, the background separation of the pretreatment image to obtain a separation image comprises: identifying the image size of the pretreatment image, based on the image size, constructing a full zero matrix of the pretreatment image, and constructing a background model of the pretreatment image by using the full zero matrix and the following formula:

[0074]

[0075] Wherein, BM represents the background model, ω(A, B) represents the full zero matrix, A represents the length of the pretreatment image, B represents the width of the pretreatment image, F(i, j) represents the coordinates of the pixel point F in the pretreatment image, i represents the horizontal coordinate of the pixel point F, and j represents the vertical coordinate of the pixel point F.

[0076] The background model is used to separate the background of the pretreatment image to obtain a separation image.

[0077] The all-zero matrix is a matrix for initializing a data structure, which can be constructed by combining the image size of the preprocessed image with the python language set, and the background model is a method or representation for distinguishing the background and foreground in the image, which is constructed based on certain conditions and algorithms, such as a Gaussian mixture model.

[0078] Further, in another optional embodiment of the present application, the background separation of the preprocessed image by using the background model to obtain a separated image comprises: performing image dilation on the preprocessed image to obtain a dilated image, performing image erosion on the preprocessed image to obtain an eroded image, performing center processing on the dilated image by using the background model to obtain a center processed image, and performing edge processing on the eroded image to obtain an edge processed image, and performing image splicing on the center processed image and the edge processed image to obtain the separated image.

[0079] The image dilation on the preprocessed image to obtain a dilated image refers to increasing the size of the object in the image, which can be realized by a convolutional neural network operation, the image erosion on the preprocessed image to obtain an eroded image refers to reducing the size of the object in the image, which can be realized by the convolutional neural network operation, the center processing on the dilated image by using the background model to obtain a center processed image refers to a method for highlighting the center part of the image, which can be obtained by image sharpening of the dilated image by using the background model, and the edge processing on the eroded image to obtain an edge processed image refers to highlighting the edge part of the eroded image, which can be realized by an edge detection algorithm.

[0080] Further, the calculation of the brain tissue volume of the brain patient based on the separated image can reflect the growth, atrophy or abnormal enlargement of the brain tissue, for example, in Alzheimer's disease, the reduction of the hippocampal volume can be an important indicator of disease progression.

[0081] As an embodiment of the present application, the calculation of the brain tissue volume of the brain patient based on the separated image comprises: identifying the image voxel of the separated image, calculating the voxel volume of the image voxel, and calculating the brain tissue volume of the brain patient based on the voxel volume.

[0082] The voxel refers to a pixel in a three-dimensional space, similar to a pixel in a two-dimensional image, and the pixel is the smallest unit of a two-dimensional image, while the voxel is the smallest unit of a three-dimensional image, such as a magnetic resonance image or a computed tomography image.

[0083] Optionally, the identification of image voxels in the separated image is calculated by importing the separated image into the image processing tool in MATLAB software. The voxel volume of the identified image voxels can be obtained by reading the parameter settings of magnetic resonance imaging, including pixel pitch (in the x, y, and z directions) and slice thickness. The brain tissue volume of the brain patient is calculated based on the voxel volume by multiplying the number of voxels by the volume of a single voxel. For example, if the brain tissue region in the separated image contains 1000 voxels, the pixel pitch in the x and y directions is 1 mm, and the slice thickness in the z direction is 2 mm, then the volume of a single voxel is 1 mm × 1 mm × 2 mm = 2 mm. 3 The total volume of brain tissue is 1000 × 2 mm. 3 =2000mm 3 .

[0084] Furthermore, in this embodiment of the invention, calculating the surface area of ​​the brain tissue of the patient based on the separated images and determining the shape of the brain tissue can help assess the degree and complexity of wrinkles on the surface of the brain tissue. For example, certain brain developmental abnormalities may lead to changes in the surface area of ​​the cerebral cortex, and the shape can quantitatively describe the shape characteristics of the brain tissue. For example, tumors may cause distortion of the shape of the surrounding normal brain tissue, and the shape can more accurately describe this change.

[0085] Optionally, the calculation of the brain tissue surface area of ​​the brain patient based on the separated image can be achieved by triangulating the corresponding brain tissue surface in the separated image into a triangular mesh, calculating the area of ​​the triangles and summing them to obtain the surface area of ​​the brain tissue. The shape of the brain tissue of the brain patient can be obtained by extracting shape description parameters such as eccentricity, roundness, and rectangularity of the corresponding brain tissue in the separated image and calculating the values ​​of the shape description parameters.

[0086] S3. Based on the brain tissue volume, brain tissue surface area, and brain tissue shape, identify the brain tissue characteristics of the patient, perform dimensionality reduction processing on the brain tissue characteristics to obtain dimensionality-reduced features, and perform wavelet coefficient decomposition on the dimensionality-reduced features to obtain different frequency sub-band coefficients related to the brain tissue characteristics.

[0087] This invention, through identifying the brain tissue characteristics of a brain patient based on the brain tissue volume, brain tissue surface area, and brain tissue shape, can obtain morphological characteristics of the brain tissue of the brain patient, and make a preliminary understanding of whether the brain patient exists or what kind of brain disease may exist.

[0088] As an embodiment of the present application, the brain tissue features of the brain patient are identified based on the brain tissue volume, the brain tissue surface area and the brain tissue shape, comprising: performing data standardization processing on the brain tissue volume, the brain tissue surface area and the brain tissue shape to obtain standardized data, extracting normal brain tissue data corresponding to the brain patient, comparing the standardized data with the normal brain tissue data to obtain a comparison result, and determining the brain tissue features of the brain patient based on the comparison result.

[0089] The data standardization processing refers to converting data into the same format for facilitating analysis together, which can be processed by vector conversion. The brain tissue features of the brain patient are determined based on the comparison result by comparing the size changes of the standardized data and the corresponding data of the normal brain tissue data to obtain a conclusion. For example, if the corresponding brain tissue surface area data vector in the standardized data is smaller than the corresponding brain tissue surface area data vector in the normal brain tissue data, there is a possibility of brain atrophy. In the analysis of other brain diseases, such as the study of Parkinson's disease patients, it is found that the volume of the substantia nigra is significantly smaller than that of the normal population. For another example, in the analysis of brain tumors, by observing the changes of the surface area and shape index of the tumor and its surrounding tissue, the doctor can determine the invasion degree of the tumor and the influence on the surrounding tissue.

[0090] Further, the embodiment of the present application can reduce the data dimension and improve the data analysis efficiency by performing dimension reduction processing on the brain tissue features to obtain reduced dimension features, which can be performed by principal component analysis.

[0091] Further, the embodiment of the present application can improve the analysis accuracy by analyzing the brain tissue features in different frequency domains by performing wavelet coefficient decomposition on the reduced dimension features to obtain different frequency sub-band coefficients of the brain tissue features. The sub-band coefficient refers to a representation method for decomposing original feature data into different frequency components. Different frequency sub-band coefficients reflect the feature information of data at different scales or resolutions. For example, low-frequency sub-band coefficients usually contain the overall trend and general outline of data, while high-frequency sub-band coefficients reflect the details, mutations and noise of data.

[0092] As an embodiment of the present application, the wavelet coefficient decomposition is performed on the reduced dimension features to obtain different frequency sub-band coefficients of the brain tissue features, comprising: obtaining an analysis requirement of the brain tissue features, configuring a wavelet function based on the analysis requirement, setting the decomposition layer number of the reduced dimension features, performing wavelet transform on the reduced dimension features based on the wavelet function and the decomposition layer number to obtain different frequency sub-band coefficients of the brain tissue features.

[0093] The analysis requirement refers to that a user selects a more suitable wavelet decomposition function in combination with image quality, for example, if the image resolution is low or the noise is large, the user tends to select a wavelet function with strong denoising ability and low sensitivity to details, and the wavelet function is selected according to the user's demand, and the wavelet function includes Daubechies wavelet, Symlet wavelet and the like, the decomposition layer number of the dimensionality reduction feature can be set according to the user's demand, the more the decomposition layer number is, the more complex the decomposition is, and the longer the decomposition time is, the dimensionality reduction feature is subjected to wavelet transform, and different frequency subband coefficients about the brain tissue feature are obtained by decomposition of the wavelet function, for example, the user selects Daubechies wavelet as a base function, and the decomposition layer number is 3. First, the dimensionality reduction feature is convolved with the first layer of wavelet base functions, and then down-sampling is performed to obtain the first layer of approximation coefficients and detail coefficients; then, the first layer of approximation coefficients is subjected to the above process repeatedly to obtain the second layer of coefficients; and in this way, the decomposition process is completed.

[0094] S4, identifying statistical features of the different frequency subband coefficients, constructing a feature vector of the statistical features based on the statistical features, performing brain disease analysis on the brain of the patient based on the feature vector, and obtaining a preliminary disease analysis category.

[0095] The statistical features of the different frequency subband coefficients can describe the distribution of the subband coefficients, and provide valuable information for subsequent feature construction and classification.

[0096] As an embodiment of the present application, the statistical features of the different frequency subband coefficients include:

[0097] The mean of the different frequency subband coefficients is calculated by the following formula:

[0098]

[0099] Wherein, α represents the mean, N represents the decomposition layer number of the subband coefficient, x i represents the i th subband coefficient;

[0100] The variance of the different frequency subband coefficients is calculated by the following formula:

[0101]

[0102] Wherein, β represents the variance, N represents the decomposition layer number of the subband coefficient, x i represents the i th subband coefficient, and α represents the mean of the subband coefficient;

[0103] The skewness of the different frequency subband coefficients is calculated by the following formula:

[0104]

[0105] wherein γ represents skewness, N represents the number of decomposition layers of the sub-band coefficients, x i represents the i-th sub-band coefficient, α represents the mean of the sub-band coefficients, and δ represents the standard deviation of the sub-band coefficients.

[0106] The statistical characteristics of the different frequency sub-band coefficients are determined based on the mean, the variance, and the skewness.

[0107] Further, the feature vector of the statistical characteristics is constructed based on the statistical characteristics, so that data can be format-converted, that is, statistically analyzed.

[0108] Optionally, the feature vector of the statistical characteristics is constructed by performing standardization processing on the mean, the variance, and the skewness through Z-score standardization, and then combining the standardized mean, variance, and skewness into a feature vector.

[0109] Further, the brain disease of the brain patient is analyzed based on the feature vector, so that a preliminary disease analysis category is obtained, which can provide an objective reference basis for doctors, help them more accurately analyze the disease category that the patient may have, and enable prediction based on subtle feature changes in the early stage of the disease, which is helpful for timely discovery of potential disease risks.

[0110] Optionally, the brain disease of the brain patient is analyzed based on the feature vector, so that a preliminary disease analysis category is obtained by inputting the feature vector into a pre-configured deep learning model, inputting an analysis label corresponding to the feature vector into the pre-configured deep learning model, and performing preliminary disease analysis on the brain patient based on the analysis label.

[0111] S5, texture features of the preprocessed image are extracted, the texture features are fused with the feature vector to obtain fused features, the fused features are used to analyze the accuracy of the preliminary disease analysis category, and when the accuracy analysis result is optimal, a disease category analysis report of the brain patient is constructed.

[0112] The texture features of the preprocessed image are extracted, so that the distribution mode of the gray value of pixels or voxels in the brain image, the spatial relationship, and the like can be researched and quantified to obtain feature information capable of describing the internal structure and organization characteristics of the image.

[0113] As an embodiment of the present application, the extracting the texture feature of the preprocessed image comprises: converting the preprocessed image into a gray image, calculating the number of gray levels of the gray image, and calculating the pixel distance of the gray image, calculating the texture feature value of the preprocessed image based on the number of gray levels and the pixel distance, and determining the texture feature of the preprocessed image based on the texture feature value.

[0114] Optionally, the converting the preprocessed image into a gray image refers to converting a color image or an image with multiple color channels into an image with only gray levels, which can be converted by a script built by Java, the number of gray levels of the gray image refers to the number of different gray values in the gray image, which can be calculated by a gray level co-occurrence matrix, the calculating the pixel distance of the gray image refers to calculating the distance between pixels, which can be calculated by a gray level co-occurrence matrix, and the calculating the texture feature value of the preprocessed image based on the number of gray levels and the pixel distance refers to calculating the contrast, correlation and entropy of the preprocessed image based on the number of gray levels and the pixel distance, wherein the contrast represents the difference degree of the gray levels in the image, the correlation is used to measure the linear relationship between the gray levels in the image, and the entropy reflects the complexity and randomness of the image. By calculating these features, the texture, definition and complexity of the image can be described and analyzed.

[0115] Further, the embodiment of the present application can combine or combine multiple features from different feature extraction methods or algorithms to obtain more comprehensive and more accurate image representation by the feature fusion of the texture feature and the feature vector.

[0116] As an embodiment of the present application, the feature fusion of the texture feature and the feature vector to obtain a fusion feature comprises: converting the texture feature into a texture vector, performing dimension conversion on the texture vector and the feature vector to obtain a dimension conversion vector, encoding the dimension conversion vector to obtain an encoded vector, using a pre-constructed fusion matrix to perform encoding fusion on the encoded vector to obtain a fusion encoding, and performing encoding translation on the fusion encoding to obtain a fusion feature.

[0117] Optionally, the dimension conversion on the texture vector and the feature vector to obtain a dimension conversion vector can be realized by a full connection layer operation in a Transformer model, the encoding on the dimension conversion vector to obtain an encoded vector can be realized by an image feature encoder, the encoding on the dimension conversion vector to obtain an encoded vector refers to replacing a vector by a special character or symbol, which can be realized by binary encoding, and the encoding fusion on the encoded vector by using a pre-constructed fusion matrix to obtain a fusion encoding can be realized by a plurality of multi-head attention layers in the Transformer model.

[0118] Further, the accuracy analysis on the preliminary disease analysis category by using the fusion feature can further improve the accuracy of disease analysis.

[0119] Optionally, the accuracy analysis on the preliminary disease analysis category by using the fusion feature is obtained by secondary classification by a support vector machine.

[0120] It should be noted that when the accuracy analysis result is optimal, the disease category analysis report of the brain patient is constructed, which means that the accuracy reaches a preset threshold of 95%, and if less than 95%, expert knowledge and clinical data are introduced for comprehensive analysis to determine the final brain disease category, and then the disease analysis report of the brain patient is constructed according to the final analysis structure.

[0121] The present application can help users or doctors to analyze brain diseases more accurately by collecting magnetic resonance brain images of brain patients, and optionally, the present application can separate the background part of the image to enhance the analysis accuracy of the image target by separating the background of the preprocessed image to obtain a separated image, the present application can reflect the growth, atrophy or abnormal enlargement of brain tissue by calculating the brain tissue volume of the brain patient based on the separated image, the present application can obtain the morphological characteristics of the brain tissue of the brain patient by identifying the brain tissue features of the brain patient based on the brain tissue volume, the brain tissue surface area and the brain tissue shape, and preliminarily understand whether the brain patient has a brain disease or not, the present application can describe the distribution of sub-band coefficients by identifying the statistical characteristics of the different frequency sub-band coefficients, and provide valuable information for subsequent feature construction and classification, and finally, the present application can research and quantify the gray value distribution pattern, spatial relationship and the like of pixels or voxels in the brain image by extracting the texture features of the preprocessed image, so as to obtain feature information capable of describing the internal structure and organization characteristics of the image. Therefore, the present application can improve the accuracy of brain disease category analysis.

[0122] AsFigure 2 Fig. 1 is a functional module diagram of a brain disease category analysis system based on magnetic resonance according to an embodiment of the present application.

[0123] The brain disease category analysis system 200 based on magnetic resonance can be installed in an electronic device. According to the functions implemented, the brain disease category analysis system 200 based on magnetic resonance can include an image preprocessing module 201, a brain tissue shape analysis module 202, a feature decomposition module 203, a disease category preliminary analysis module 204, and a disease analysis report module 205. The modules of the present application can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0124] In the present embodiment, the functions of each module / unit are as follows:

[0125] The image preprocessing module 201 is configured to acquire a magnetic resonance brain image of a brain patient, perform image preprocessing on the magnetic resonance brain image, and obtain a preprocessed image.

[0126] The brain tissue shape analysis module 202 is configured to perform background separation on the preprocessed image to obtain a separated image, calculate a brain tissue volume of the brain patient based on the separated image, calculate a brain tissue surface area of the brain patient based on the separated image, and determine a brain tissue shape identification of the brain patient.

[0127] The feature decomposition module 203 is configured to identify brain tissue features of the brain patient based on the brain tissue volume, the brain tissue surface area, and the brain tissue shape identification, perform dimension reduction processing on the brain tissue features to obtain reduced dimension features, perform wavelet coefficient decomposition on the reduced dimension features to obtain different frequency sub-band coefficients of the brain tissue features.

[0128] The disease category preliminary analysis module 204 is configured to identify statistical features of the different frequency sub-band coefficients, construct a feature vector of the statistical features based on the statistical features, perform brain disease analysis on the brain patient based on the feature vector, and obtain a preliminary disease analysis category.

[0129] The disease analysis report module 205 is configured to extract texture features of the preprocessed image, perform feature fusion on the texture features and the feature vector to obtain fused features, perform accuracy analysis on the preliminary disease analysis category using the fused features, and construct a disease category analysis report of the brain patient when the accuracy analysis result is optimal.

[0130] In detail, each module in the brain disease category analysis system 200 based on magnetic resonance in the embodiments of the present application adopts the same technical means as the brain disease category analysis method based on magnetic resonance in the drawings when in use, and can produce the same technical effects, which will not be described here again.

[0131] In several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other manners. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. There can be another division manner in actual implementation.

[0132] In addition, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. Multiple units or systems stated in the system claims can also be implemented by one unit or system through software or hardware. The first, second, etc. words are used to indicate names and not to indicate any specific order.

[0133] Finally, it should be noted that the above embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the present application. Although the present application 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 application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for classifying brain diseases based on magnetic resonance imaging, characterized in that, The method includes: Magnetic resonance imaging (MRI) images of brain patients are acquired, and the MRI images are preprocessed to obtain preprocessed images. Background separation is performed on the preprocessed image to obtain a separated image. Based on the separated image, the brain tissue volume of the brain patient is calculated, the brain tissue surface area of ​​the brain patient is calculated based on the separated image, and the brain tissue shape of the brain patient is determined. Based on the brain tissue volume, brain tissue surface area, and brain tissue shape, the brain tissue characteristics of the brain patient are identified. The brain tissue characteristics are then subjected to dimensionality reduction processing to obtain dimensionality-reduced features. Wavelet coefficient decomposition is then performed on the dimensionality-reduced features to obtain different frequency sub-band coefficients related to the brain tissue characteristics. The statistical characteristics of the coefficients of the different frequency subbands are identified. Based on the statistical characteristics, a feature vector of the statistical characteristics is constructed. Based on the feature vector, the brain disease analysis of the brain patient is performed to obtain a preliminary disease analysis category. The statistical characteristics of the coefficients of the different frequency sub-bands are determined based on the mean, the variance, and the skewness. Extract the texture features of the preprocessed image, fuse the texture features with the feature vector to obtain fused features, use the fused features to perform accuracy analysis on the preliminary disease analysis category, and when the accuracy analysis result is excellent, construct the disease category analysis report of the brain patient; The step of fusing the texture features with the feature vector to obtain fused features includes: Convert the texture features into texture vectors; The texture vector and the feature vector are subjected to dimensionality transformation to obtain a dimensionality transformation vector; The dimension transformation vector is encoded to obtain the encoded vector; The encoded vector is fused using a pre-constructed fusion matrix to obtain the fused code; The fusion code is encoded and translated to obtain the fusion feature; The step of performing background separation on the preprocessed image to obtain a separated image includes: identifying the image size of the preprocessed image, constructing an all-zero matrix of the preprocessed image based on the image size, and constructing a background model of the preprocessed image using the all-zero matrix in combination with the following formula: ； in, Represents the background model. A matrix of all zeros, where A represents the length of the preprocessed image and B represents the width of the preprocessed image. The coordinates of pixel F in the preprocessed image. Let j represent the x-coordinate of pixel F, and j represent the y-coordinate of pixel F. The background model is used to separate the background from the preprocessed image to obtain a separated image.

2. The method for classifying brain diseases based on magnetic resonance imaging as described in claim 1, characterized in that, The step of preprocessing the magnetic resonance brain image to obtain a preprocessed image includes: The magnetic resonance brain images are converted to a new format to obtain converted images; Identify the original pixel values ​​of the transformed image; The original pixel values ​​are subjected to mean filtering to obtain mean pixels; Based on the mean pixels, determine the mean image of the transformed image; The mean image is sharpened to obtain a preprocessed image.

3. The method for classifying brain diseases based on magnetic resonance imaging as described in claim 2, characterized in that, The step of using the background model to separate the background from the preprocessed image to obtain a separated image includes: The preprocessed image is dilated to obtain a dilated image, and the preprocessed image is eroded to obtain an eroded image; The background model is used to center the dilated image to obtain a centered image; The eroded image is subjected to edge processing to obtain an edge-processed image; The center-processed image and the edge-processed image are stitched together to obtain a separate image.

4. The method for classifying brain diseases based on magnetic resonance imaging as described in claim 1, characterized in that, The calculation of the brain tissue volume of the brain patient based on the separated images includes: Identify the image voxels of the separated image; Calculate the voxel volume of the image voxel; The brain tissue volume of the patient was calculated based on the voxel volume.

5. The method for classifying brain diseases based on magnetic resonance imaging as described in claim 1, characterized in that, The method of identifying brain tissue characteristics of a patient based on brain tissue volume, brain tissue surface area, and brain tissue shape includes: The brain tissue volume, brain tissue surface area, and brain tissue shape are standardized to obtain standardized data. Extract normal brain tissue data corresponding to the brain patients; The standardized data is compared with the normal brain tissue data to obtain the comparison results; The brain tissue characteristics of the patient were determined based on the comparison results.

6. The method for classifying brain diseases based on magnetic resonance imaging as described in claim 1, characterized in that, The wavelet coefficient decomposition of the dimensionality-reduced features yields different frequency sub-band coefficients for the brain tissue features, including: Analysis requirements for obtaining the characteristics of the brain tissue; Based on the aforementioned analysis requirements, the wavelet function is configured, and the number of decomposition layers for the dimensionality reduction features is set. Based on the wavelet function and the number of decomposition layers, wavelet transform is performed on the dimensionality-reduced features to obtain different frequency sub-band coefficients for the brain tissue features.

7. The method for classifying brain diseases based on magnetic resonance imaging as described in claim 1, characterized in that, The extraction of texture features from the preprocessed image includes: Calculate the number of gray levels in the grayscale image and calculate the pixel distance of the grayscale image; The texture feature value of the preprocessed image is calculated based on the number of gray levels and the pixel distance; The texture features of the preprocessed image are determined based on the texture feature values.

8. A magnetic resonance imaging-based brain disease category analysis system, characterized in that, The system is used to perform a magnetic resonance-based brain disease category analysis method as described in any one of claims 1-7, the system comprising: The image preprocessing module is used to acquire magnetic resonance brain images of brain patients, and to perform image preprocessing on the magnetic resonance brain images to obtain preprocessed images; The brain tissue shape analysis module is used to perform background separation on the preprocessed image to obtain a separated image, calculate the brain tissue volume of the brain patient based on the separated image, calculate the brain tissue surface area of ​​the brain patient based on the separated image, and determine the brain tissue shape of the brain patient. The feature decomposition module is used to identify the brain tissue features of the brain patient based on the brain tissue volume, the brain tissue surface area and the brain tissue shape, perform dimensionality reduction processing on the brain tissue features to obtain dimensionality-reduced features, and perform wavelet coefficient decomposition on the dimensionality-reduced features to obtain different frequency sub-band coefficients related to the brain tissue features. The preliminary disease category analysis module is used to identify the statistical characteristics of the coefficients of the different frequency subbands, construct a feature vector of the statistical characteristics based on the statistical characteristics, and perform brain disease analysis on the brain patients based on the feature vector to obtain a preliminary disease analysis category. The disease analysis report module is used to extract the texture features of the preprocessed image, fuse the texture features with the feature vector to obtain fused features, use the fused features to perform accuracy analysis on the preliminary disease analysis category, and construct the disease category analysis report for the brain patient when the accuracy analysis result is excellent.

Images