Medical image feature extraction and disease prediction platform based on deep learning

Abstract

The invention relates to the technical field of medical images, and discloses a medical image feature extraction and disease prediction platform based on deep learning, which comprises a medical image scanning module, a feature extraction module, a disease prediction module and a system management module, wherein the medical image scanning module is used for scanning a medical image of a patient in a three-dimensional rendering mode and transmitting the medical image to the feature extraction module; and the feature extraction module is used for performing feature extraction on the medical image after three-dimensional body rendering by using a neural network construction method.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a medical imaging feature extraction and disease prediction platform based on deep learning. Background Art

[0002] With the rapid development of medical technology, medical imaging has become one of the important tools to assist in disease diagnosis and treatment. Through in-depth analysis of medical images, especially the structural features in three-dimensional medical images, doctors can find the potential pathological characteristics of neurological diseases (such as Alzheimer's disease, Parkinson's disease, etc.) earlier and more accurately. In recent years, the application of artificial intelligence and deep learning technology in the field of medical image analysis has gradually become popular, especially in the study of brain images, using neural networks to build predictive models has gradually become a mainstream trend. This field not only involves image processing, but also covers multidisciplinary cross-disciplinary technologies such as data mining, pattern recognition and pathology.

[0003] Traditional medical image analysis mainly relies on manual reading and rule-based image processing algorithms, such as edge detection, region segmentation, and texture analysis. These methods identify diseases by extracting grayscale values, textures, or shape features, but are often limited to the analysis of two-dimensional images and cannot fully reflect the characteristics of three-dimensional structures. In addition, experience-based manual diagnosis is not only time-consuming and subjective, but is also easily limited by the reader's knowledge and experience, resulting in inconsistent diagnostic results. Although some traditional methods attempt to improve diagnostic efficiency through model optimization, their ability to express complex data relationships is limited, and it is particularly difficult to capture the multi-level correlation information hidden in three-dimensional medical images.

[0004] Although the application of deep learning technology in medical image analysis has achieved initial results, the existing technology still has obvious deficiencies in the deep analysis of multidimensional data and the construction of efficient prediction models. For example, some methods fail to effectively utilize the global and local features of three-dimensional images, resulting in low feature extraction accuracy; other methods have a single way of processing data and cannot fully describe the changing relationships of complex biological structures. In addition, the design of some neural network models still has room for improvement in computing efficiency and resource utilization. To address these problems, this program proposes a medical image feature extraction and disease prediction platform based on three-dimensional volume rendering and neural networks. By optimizing the feature extraction and classification prediction process, it effectively makes up for the shortcomings of the existing technology and significantly improves the analysis accuracy, automation level and prediction reliability. Summary of the invention

[0005] The present invention mainly solves the technical problems existing in the above-mentioned prior art and provides a medical image feature extraction and disease prediction platform based on deep learning.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a medical image feature extraction and disease prediction platform based on deep learning, comprising: a medical image scanning module, a feature extraction module, a disease prediction module and a system management module; wherein the medical image scanning module is used to scan the patient's medical image by three-dimensional volume rendering and transmit it to the feature extraction module; the feature extraction module is used to extract features from the medical image after three-dimensional volume rendering by using a neural network construction method;

[0007] The specific method is:

[0008] S101. Under the premise of high-throughput measurement of different nuclei in the gray matter area of ​​the brain at different levels of brain imaging, quantitative analysis is performed using the measured gray matter density values ​​to represent the number density of different nuclei;

[0009] S102. On the basis of quantitative analysis of gray matter density, at different levels of brain images, the gray matter density value of each nucleus on the same level is calculated with the number density value of each nucleus in the same level to obtain the ratio of the gray matter density value of each nucleus on the same level to the number density of each nucleus in the same level;

[0010] S103, averaging the ratios of the gray matter density values ​​of different nuclei on the same level to the number densities of different nuclei in the level to obtain a gray matter density ratio of the brain imaging level;

[0011] S104, summarizing the gray matter density ratios of the brain image layers obtained in the above S103 to obtain the gray matter density ratios of different layers of the brain image, thereby deriving the value of the gray matter density ratio of the brain; the disease prediction module is used to construct a classification prediction model according to the gray matter density ratios of the brain image to predict nervous system diseases; the system management module is used to manage the parameters of the system to access and operate all modules in the system.

[0012] Preferably, the medical imaging scanning module is used to perform a magnetic resonance scan on the patient's brain to obtain a magnetic resonance image of the patient's brain.

[0013] Preferably, the neural network construction method is as follows: for a medical image rendered by a three-dimensional volume, the feature extraction module processes the medical image rendered by the three-dimensional volume; for a three-dimensional matrix, three matrix values ​​located in the same plane in the three-dimensional matrix are sorted to obtain the minimum value of the three matrix values, and the minimum value of each plane in the three-dimensional matrix constitutes a new one-dimensional data matrix to obtain multiple one-dimensional data matrices; the multiple one-dimensional data matrices are input into the neural network as input values, and the relationship between the gradient of the elements in each one-dimensional data matrix and the adjacent matrices and the relationship between the gradients of the adjacent matrices are calculated by the neural network, and after the calculation is completed, a set of optimal values ​​is obtained, and the maximum connected area of ​​the medical image rendered by the three-dimensional volume is established with the optimal value set.

[0014] Preferably, the relationship between the gradient of each element in each one-dimensional data matrix and the adjacent matrix and the relationship between the gradients of adjacent matrices are calculated by a neural network, and the method for obtaining the set of optimal values ​​after the calculation is completed is: for each one-dimensional data matrix, a loss function L(w) of the data matrix is ​​established;

[0015] L(w)=α 1 L 1 (w)+α 2 L 2 (w)+…+α P-1 L P-1 (w);

[0016] where α p ∈[0,1] is a hyperparameter, ranging from 0 to 1; L p (w) is the loss function of the pth minimum value in the gradient descent process, p = 1, 2, ..., P-1, P is the total number of all minimum values ​​in the data matrix that require gradient descent; w is the parameter that needs to be descended in the data matrix; after the loss function is established, the expression of the gradient of the loss function to the parameter w is obtained through the loss function, and the loss function is processed by the gradient. When the loss function reaches the minimum value, it means that the processing is completed, and the optimal solution set Wp = {w1, w2, ..., wp} is output; the loss function of each one-dimensional data matrix input into the neural network is an expression of L(w), and according to the optimal solution set Wp of the input data matrix, the elements in each data matrix are optimized to obtain a new one-dimensional data matrix. When the loss function value of the new one-dimensional data matrix reaches the minimum or the number of iterations reaches the preset value, the optimization is terminated; the obtained multiple new one-dimensional data matrices are combined to obtain the optimal value set U = {u1, u2, ..., up}; wherein all element values ​​in the new one-dimensional data matrix are optimal values, p iis the total number of all optimal values ​​in the i-th new one-dimensional data matrix, 1≤i≤P, pi≥1, P is the total number of all parameters that need to be reduced in the i-th input data matrix; according to the position corresponding to the optimal value in the three-dimensional matrix, the set of optimal positions in the three-dimensional matrix is ​​obtained. Q = {q 1 ,q 2 ,...,q p}; where q j Represents the position in the p-th three-dimensional matrix, 1≤j≤pi, j is the total number of all optimal values ​​in the p-th new one-dimensional data matrix.

[0017] Preferably, the feature extraction module is also used to extract the maximum connected area of ​​the input 3D medical image through the above-mentioned neural network construction method. The specific method is: taking the pixel points in each two-dimensional slice in the 3D medical image as a one-dimensional data matrix, taking the pixel points as the minimum value, and performing gradient calculation after establishing the one-dimensional data matrix to obtain the 3D medical image with the maximum connected area.

[0018] Preferably, the method further includes: dividing the sum of all differences in the grayscale value differences of each adjacent pixel point in the 3D medical image of the obtained maximum connected area by the number of pixels of the same pixel in the 3D medical image, and establishing a feature extraction expression according to the calculated grayscale gradient difference sum; the specific method of calculating the expression is as follows:

[0019]

[0020] Among them, a represents the number of pixels in the 3D medical image, E represents the feature extraction expression of the 3D medical image of the extracted maximum connected area, i and j represent the position coordinates of the pixel points, G i,j Represented by (x i ,y j ,z) is the three-dimensional gray value of the pixel point, N(i,j) represents the set of adjacent pixels of the pixel point (i,j), |G i,j -G i′,j′ | represents the absolute difference of the grayscale values ​​of adjacent pixels; the calculated 3D medical image feature extraction expression is used to construct the brain gray matter density ratio, and according to the relationship between the obtained brain gray matter density ratio and neurological diseases, a classification prediction model is constructed to obtain an indication of whether the 3D medical image suffers from a neurological disease.

[0021] Preferably, the neural network construction method specifically includes:

[0022] Step 1: Determine the loss function L(w) based on the value of all the parameters p to be optimized in the input data matrix. After determining the loss function, establish an expression for the gradient of the loss function with respect to the parameter w.

[0023] Step 2: Set the gradient of the loss function obtained in step 1 to parameter w equal to 0, assign the gradient value to the corresponding parameter w, assign the parameter w, and obtain multiple new one-dimensional data matrices;

[0024] Step 3: Combine the calculated multiple new one-dimensional data matrices to obtain the optimal value set U = {u 1 ,u 2 ,...,u p}; Judge the two-dimensional matrix of each pixel in the 3D medical image of the maximum connected area. If the pixel u 1 ,u 2 ,...,u p is 0, then the two-dimensional matrix of each pixel in the image is changed to 0;

[0025] Step 4: When it is determined that the gradient difference at adjacent pixel points x and y in the data matrix is ​​0 and the two-dimensional matrix value at x and y is 1, there are adjacent pixels in the 3D medical image of the maximum connected area in x and y.

[0026] Preferably, the data sets for constructing, training and validating the classification prediction model are all from the internationally used database ADNI.

[0027] Preferably, step 4 comprises:

[0028] Step 4(a): firstly obtain a 3D medical image of the target to be predicted, and process the 3D medical image of the target to be predicted by using a Gaussian filter;

[0029] Step 4(b), extracting the largest connected region of the input 3D medical image by using the above neural network construction method;

[0030] Step 4(c), establishing a three-dimensional gray value feature, comparing the brain gray matter density with the density of the cerebral cortical nuclei, and establishing a brain gray matter density ratio;

[0031] Step 4(d): Based on the established brain gray matter density ratio, a classification prediction model is constructed, and the model includes a test set, a training set, and a validation set.

[0032] Preferably, the parameter w in the neural network is updated and iterated by the gradient descent method. The specific update and iteration form is as follows:

[0033] w (n+1) =w (n) -α▽ w L(w (n) );

[0034] Among them, n represents the number of iterative updates. In each iterative update process, the gradient expression of the loss function with respect to the parameter w needs to be assigned to the corresponding parameter w to update the parameter w. In the gradient descent method, the update method of the parameter w is as follows:

[0035] w″=w-α▽ w L(w);

[0036] Among them, w″ is the updated parameter value, w is the parameter value before the update, α is a hyperparameter with a value range of [0,1], which indicates the multiple of the distance moved by parameter w in each iteration, ▽ w L(w) is the gradient of the loss function L(w) with respect to the parameter w in the nth iteration of gradient descent;

[0037] The gradient ▽ w The calculation method of L(w) is to derive the loss function with respect to the parameter w, that is:

[0038]

[0039] Using the gradient, the parameter w is iteratively updated through the above formula until the termination condition is met. When the values ​​of all elements in multiple new one-dimensional data matrices obtained by the gradient descent method are equal to 0, the gradient update process is terminated; wherein, the algorithm of step 1 is updated through the above update method.

[0040] Beneficial Effects

[0041] The present invention provides a medical image feature extraction and disease prediction platform based on deep learning. It has the following beneficial effects:

[0042] (1) This deep learning-based medical image feature extraction and disease prediction platform uses a 3D volume rendering combined with a neural network method to efficiently extract features from medical images. In particular, in the calculation of gray matter density ratios, it achieves quantitative and ratio analysis of gray matter areas in the brain and provides more detailed data expression. This method can deeply explore the key features related to neurological diseases in medical images and significantly improve the depth and quality of analysis.

[0043] (2) This deep learning-based medical image feature extraction and disease prediction platform uses a classification prediction model to seamlessly combine feature extraction with disease prediction. Through automated gradient optimization and connected region extraction, it simplifies the disease detection process and reduces the complexity of manual intervention. The high-precision performance of the model in predicting neurological diseases provides doctors with reliable decision support.

[0044] (3) This deep learning-based medical image feature extraction and disease prediction platform uses the international authoritative database ADNI as the basis for model training and verification to ensure that the method has good generalization capabilities. Through multiple rounds of parameter optimization and gradient descent algorithms, the model stability is improved, the impact of data noise or outliers on the results is reduced, and the prediction results are more consistent and credible.

[0045] (4) This deep learning-based medical image feature extraction and disease prediction platform uses gradient descent method and multiple rounds of parameter iterative optimization to ensure the stability of model training results and reduce errors caused by data noise or outliers.

[0046] (5) This deep learning-based medical image feature extraction and disease prediction platform optimizes the feature expression of medical images by extracting the maximum connected area and analyzing the grayscale value of 3D medical images, providing a more intuitive reference for subsequent clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the implementation of the present invention or the technical solution in the prior art, the following briefly introduces the drawings required for the implementation or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0048] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0049] Figure 1 This is a functional schematic diagram of the system modules of the present invention;

[0050] Figure 2 A flowchart of gradient optimization of a one-dimensional data matrix by a neural network of the present invention;

[0051] Figure 3 A flow chart of extracting the maximum connected area by the feature extraction module of the present invention;

[0052] Figure 4 The neural network construction method and data processing flow chart of the present invention;

[0053] Figure 5 This is a flow chart for optimizing the gradient descent method of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Example: Medical image feature extraction and disease prediction platform based on deep learning, such as Figures 1 to 5 As shown, it includes: a medical image scanning module, a feature extraction module, a disease prediction module and a system management module; wherein the medical image scanning module is used to scan the patient's medical image by three-dimensional volume rendering and transmit it to the feature extraction module; the feature extraction module is used to extract features from the medical image after three-dimensional volume rendering by using a neural network construction method;

[0056] The specific method is:

[0057] S101. Under the premise of high-throughput measurement of different nuclei in the gray matter area of ​​the brain at different levels of brain imaging, quantitative analysis is performed using the measured gray matter density values ​​to represent the number density of different nuclei;

[0058] S102. On the basis of quantitative analysis of gray matter density, at different levels of brain images, the gray matter density value of each nucleus on the same level is calculated with the number density value of each nucleus in the same level to obtain the ratio of the gray matter density value of each nucleus on the same level to the number density of each nucleus in the same level;

[0059] S103, averaging the ratios of the gray matter density values ​​of different nuclei on the same level to the number densities of different nuclei in the level to obtain a gray matter density ratio of the brain imaging level;

[0060] S104, summarizing the gray matter density ratios of the brain image layers obtained in the above S103 to obtain the gray matter density ratios of different layers of the brain image, thereby deriving the value of the gray matter density ratio of the brain; the disease prediction module is used to construct a classification prediction model according to the gray matter density ratios of the brain image to predict nervous system diseases; the system management module is used to manage the parameters of the system to access and operate all modules in the system.

[0061] As a further limitation of this solution, the medical imaging scanning module is used to perform a magnetic resonance scan on the patient's brain to obtain a magnetic resonance image of the patient's brain.

[0062] As a further limitation of the present scheme, the neural network construction method is as follows: for a medical image rendered by a three-dimensional volume, the feature extraction module processes the medical image rendered by the three-dimensional volume; for a three-dimensional matrix, three matrix values ​​located in the same plane in the three-dimensional matrix are sorted to obtain the minimum value of the three matrix values, and the minimum value of each plane in the three-dimensional matrix constitutes a new one-dimensional data matrix to obtain multiple one-dimensional data matrices; multiple one-dimensional data matrices are input into the neural network as input values, and the relationship between the gradient of the elements in each one-dimensional data matrix and the adjacent matrices and the relationship between the gradients of the adjacent matrices are calculated through the neural network, and after the calculation is completed, a set of optimal values ​​is obtained, and the maximum connected area of ​​the medical image rendered by the three-dimensional volume is established with the optimal value set.

[0063] As a further limitation of the present scheme, the relationship between the gradient of each element in each one-dimensional data matrix and the adjacent matrix and the relationship between the gradients of the adjacent matrices are calculated by a neural network, and after the calculation, a set of optimal values ​​is obtained by: for each one-dimensional data matrix, a loss function L(w) of the data matrix is ​​established;

[0064] L(w)=α 1 L 1 (w)+α 2 L 2 (w)+…+α P-1 L P-1 (w);

[0065] where α p ∈[0,1] is a hyperparameter, ranging from 0 to 1; L p (w) is the loss function of the pth minimum value in the gradient descent process, p = 1, 2, ..., P-1, P is the total number of all minimum values ​​in the data matrix that require gradient descent; w is the parameter that needs to be descended in the data matrix; after the loss function is established, the expression of the gradient of the loss function to the parameter w is obtained through the loss function, and the loss function is processed by the gradient. When the loss function reaches the minimum value, it means that the processing is completed, and the optimal solution set Wp = {w1, w2, ..., wp} is output; the loss function of each one-dimensional data matrix input into the neural network is an expression of L(w), and according to the optimal solution set Wp of the input data matrix, the elements in each data matrix are optimized to obtain a new one-dimensional data matrix. When the loss function value of the new one-dimensional data matrix reaches the minimum or the number of iterations reaches the preset value, the optimization is terminated; the obtained multiple new one-dimensional data matrices are combined to obtain the optimal value set U = {u1, u2, ..., up}; wherein all element values ​​in the new one-dimensional data matrix are optimal values, p iis the total number of all optimal values ​​in the i-th new one-dimensional data matrix, 1≤i≤P, pi≥1, P is the total number of all parameters that need to be reduced in the i-th input data matrix; according to the position corresponding to the optimal value in the three-dimensional matrix, the set of optimal positions in the three-dimensional matrix is ​​obtained. Q = {q 1 ,q 2 ,...,q p}; where q j Represents the position in the p-th three-dimensional matrix, 1≤j≤pi, j is the total number of all optimal values ​​in the p-th new one-dimensional data matrix.

[0066] As a further limitation of the present scheme, the feature extraction module is also used to extract the maximum connected area of ​​the input 3D medical image through the above-mentioned neural network construction method. The specific method is: taking the pixel points in each two-dimensional slice of the 3D medical image as a one-dimensional data matrix, taking the pixel points as the minimum value, establishing the one-dimensional data matrix, and then performing gradient calculation to obtain the 3D medical image with the maximum connected area.

[0067] As a further limitation of the present solution, the present invention further includes: dividing the sum of all differences in the grayscale value differences of each adjacent pixel point in the 3D medical image of the obtained maximum connected area by the number of pixels of the same pixel in the 3D medical image, and establishing a feature extraction expression according to the calculated sum of grayscale gradient differences; the specific method of calculating the expression is as follows:

[0068]

[0069] Among them, a represents the number of pixels in the 3D medical image, E represents the feature extraction expression of the 3D medical image of the extracted maximum connected area, i and j represent the position coordinates of the pixel points, G i,j Represented by (x i ,y j ,z) is the three-dimensional gray value of the pixel point, N(i,j) represents the set of adjacent pixels of the pixel point (i,j), |G i,j -G i′,j′ | represents the absolute difference of the grayscale values ​​of adjacent pixels; the calculated 3D medical image feature extraction expression is used to construct the brain gray matter density ratio, and according to the relationship between the obtained brain gray matter density ratio and neurological diseases, a classification prediction model is constructed to obtain an indication of whether the 3D medical image suffers from a neurological disease.

[0070] As a further limitation of this solution, the neural network construction method specifically includes:

[0071] Step 1: Determine the loss function L(w) based on the value of all the parameters p to be optimized in the input data matrix. After determining the loss function, establish an expression for the gradient of the loss function with respect to the parameter w.

[0072] Step 2: Set the gradient of the loss function obtained in step 1 to parameter w equal to 0, assign the gradient value to the corresponding parameter w, assign the parameter w, and obtain multiple new one-dimensional data matrices;

[0073] Step 3: Combine the calculated multiple new one-dimensional data matrices to obtain the optimal value set U = {u 1 ,u 2 ,...,u p}; Judge the two-dimensional matrix of each pixel in the 3D medical image of the maximum connected area. If the pixel u 1 ,u 2 ,...,u p is 0, then the two-dimensional matrix of each pixel in the image is changed to 0;

[0074] Step 4: When it is determined that the gradient difference at adjacent pixel points x and y in the data matrix is ​​0 and the two-dimensional matrix value at x and y is 1, there are adjacent pixels in the 3D medical image of the maximum connected area in x and y.

[0075] As a further limitation of this scheme, the data sets for constructing, training and validating the classification prediction model are all from the internationally used database ADNI.

[0076] As a further limitation of this solution, step 4 includes:

[0077] Step 4(a): firstly obtain a 3D medical image of the target to be predicted, and process the 3D medical image of the target to be predicted by using a Gaussian filter;

[0078] Step 4(b), extracting the largest connected region of the input 3D medical image by using the above neural network construction method;

[0079] Step 4(c), establishing a three-dimensional gray value feature, comparing the brain gray matter density with the density of the cerebral cortical nuclei, and establishing a brain gray matter density ratio;

[0080] Step 4(d): Based on the established brain gray matter density ratio, a classification prediction model is constructed, and the model includes a test set, a training set, and a validation set.

[0081] As a further limitation of this scheme, the parameter w in the neural network is updated and iterated by the gradient descent method. The specific update iteration form is as follows:

[0082] w (n+1) =w(n) -α▽ w L(w (n) );

[0083] Among them, n represents the number of iterative updates. In each iterative update process, the gradient expression of the loss function with respect to the parameter w needs to be assigned to the corresponding parameter w to update the parameter w. In the gradient descent method, the update method of the parameter w is as follows:

[0084] w″=w-α▽ w L(w);

[0085] Among them, w″ is the updated parameter value, w is the parameter value before the update, α is a hyperparameter with a value range of [0,1], which indicates the multiple of the distance moved by parameter w in each iteration, ▽ w L(w) is the gradient of the loss function L(w) with respect to the parameter w in the nth iteration of gradient descent;

[0086] The gradient ▽ w The calculation method of L(w) is to derive the loss function with respect to the parameter w, that is:

[0087]

[0088] Using the gradient, the parameter w is iteratively updated through the above formula until the termination condition is met. When the values ​​of all elements in multiple new one-dimensional data matrices obtained by the gradient descent method are equal to 0, the gradient update process is terminated; wherein, the algorithm of step 1 is updated through the above update method.

[0089] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A medical image feature extraction and disease prediction platform based on deep learning, characterized by: include: A medical image scanning module, a feature extraction module, a disease prediction module and a system management module; wherein the medical image scanning module is used to scan the patient's medical image by three-dimensional volume rendering and transmit it to the feature extraction module; the feature extraction module is used to extract features from the medical image after three-dimensional volume rendering by using a neural network construction method; The specific method is: S101. Under the premise of high-throughput measurement of different nuclei in the gray matter area of ​​the brain at different levels of brain imaging, quantitative analysis is performed using the measured gray matter density values ​​to represent the number density of different nuclei; S102. On the basis of quantitative analysis of gray matter density, at different levels of brain images, the gray matter density value of each nucleus on the same level is calculated with the number density value of each nucleus in the same level to obtain the ratio of the gray matter density value of each nucleus on the same level to the number density of each nucleus in the same level; S103, averaging the ratios of the gray matter density values ​​of different nuclei on the same level to the number densities of different nuclei in the level to obtain a gray matter density ratio of the brain imaging level; S104, summarizing the gray matter density ratios of the brain image layers obtained in the above S103 to obtain the gray matter density ratios of different layers of the brain image, thereby deriving the value of the gray matter density ratio of the brain; the disease prediction module is used to construct a classification prediction model according to the gray matter density ratios of the brain image to predict nervous system diseases; the system management module is used to manage the parameters of the system to access and operate all modules in the system.

2. The medical image feature extraction and disease prediction platform based on deep learning according to claim 1, characterized in that: The medical imaging scanning module is used to perform a magnetic resonance scan on the patient's brain to obtain a magnetic resonance image of the patient's brain.

3. The medical image feature extraction and disease prediction platform based on deep learning according to claim 1, characterized in that: The neural network construction method is as follows: for a medical image rendered by a three-dimensional volume, the feature extraction module processes the medical image rendered by the three-dimensional volume; for a three-dimensional matrix, three matrix values ​​located in the same plane in the three-dimensional matrix are sorted to obtain the minimum value of the three matrix values, and the minimum value of each plane in the three-dimensional matrix is ​​used to form a new one-dimensional data matrix to obtain multiple one-dimensional data matrices; Multiple one-dimensional data matrices are input into the neural network as input values. The relationship between the gradient of the elements in each one-dimensional data matrix and the adjacent matrices and the relationship between the gradients of the adjacent matrices are calculated by the neural network. After the calculation is completed, a set of optimal values ​​is obtained, and the maximum connected area of ​​the medical image after three-dimensional volume rendering is established with the optimal value set.

4. The medical image feature extraction and disease prediction platform based on deep learning according to claim 3, characterized in that: in, The relationship between the gradient of each element in the one-dimensional data matrix and the adjacent matrix and the relationship between the gradients of the adjacent matrices are calculated by the neural network. After the calculation is completed, the method for obtaining the set of optimal values ​​is as follows: for each one-dimensional data matrix, the loss function L(w) of the data matrix is ​​established; L(w)=α1L1(w)+α2L2(w)+…+α P-1 L P-1 (w); where α p ∈[0,1] is a hyperparameter, ranging from 0 to 1; L p (w) is the loss function of the pth minimum value in the gradient descent process, p = 1, 2, ..., P-1, P is the total number of all minimum values ​​in the data matrix that require gradient descent; w is the parameter that needs to be descended in the data matrix; after the loss function is established, the expression of the gradient of the loss function to the parameter w is obtained through the loss function, and the loss function is processed by the gradient. When the loss function reaches the minimum value, it means that the processing is completed, and the optimal solution set Wp = {w1, w2, ..., wp} is output; the loss function of each one-dimensional data matrix input into the neural network is an expression of L(w), and according to the optimal solution set Wp of the input data matrix, the elements in each data matrix are optimized to obtain a new one-dimensional data matrix. When the loss function value of the new one-dimensional data matrix reaches the minimum or the number of iterations reaches the preset value, the optimization is terminated; the obtained multiple new one-dimensional data matrices are combined to obtain the optimal value set U = {u1, u2, ..., up}; wherein all element values ​​in the new one-dimensional data matrix are optimal values, p i is the total number of all optimal values ​​in the i-th new one-dimensional data matrix, 1≤i≤P, pi≥1, P is the total number of all parameters that need to be reduced in the i-th input data matrix; according to the position corresponding to the optimal value in the three-dimensional matrix, the set of optimal positions in the three-dimensional matrix is ​​obtained Q={q1,q2,...,q p }; where q j Represents the position in the p-th three-dimensional matrix, 1≤j≤pi, j is the total number of all optimal values ​​in the p-th new one-dimensional data matrix.

5. The medical image feature extraction and disease prediction platform based on deep learning according to claim 4, characterized in that: The feature extraction module is also used to extract the maximum connected area of ​​the input 3D medical image through the above-mentioned neural network construction method. The specific method is: taking the pixel points in each two-dimensional slice in the 3D medical image as a one-dimensional data matrix, taking the pixel points as the minimum value, and performing gradient calculation after establishing the one-dimensional data matrix to obtain the 3D medical image with the maximum connected area.

6. The medical image feature extraction and disease prediction platform based on deep learning according to claim 5, characterized in that: Also includes: The sum of all differences in the grayscale value differences of each adjacent pixel point in the 3D medical image of the obtained maximum connected area is divided by the number of pixels of the same pixel in the 3D medical image, and a feature extraction expression is established according to the calculated sum of grayscale gradient differences; The specific method of calculating the expression is as follows: Among them, a represents the number of pixels in the 3D medical image, E represents the feature extraction expression of the 3D medical image of the extracted maximum connected area, i and j represent the position coordinates of the pixel points, and G i,j Represented by (x i ,y j ,z) is the three-dimensional gray value of the pixel point, N(i,j) represents the set of adjacent pixels of the pixel point (i,j), |G i,j -G i′,j′ | represents the absolute difference of the grayscale values ​​of adjacent pixels; the calculated 3D medical image feature extraction expression is used to construct the brain gray matter density ratio, and according to the relationship between the obtained brain gray matter density ratio and neurological diseases, a classification prediction model is constructed to obtain an indication of whether the 3D medical image suffers from a neurological disease.

7. The medical image feature extraction and disease prediction platform based on deep learning according to claim 6, characterized in that: The neural network construction method specifically includes: Step 1: Determine the loss function L(w) based on the value of all the parameters p to be optimized in the input data matrix. After determining the loss function, establish an expression for the gradient of the loss function with respect to the parameter w. Step 2: Set the gradient of the loss function obtained in step 1 to parameter w equal to 0, assign the gradient value to the corresponding parameter w, assign the parameter w, and obtain multiple new one-dimensional data matrices; Step 3: Combine the calculated multiple new one-dimensional data matrices to obtain the optimal value set U = {u1,u2,…,u p }; Judge the two-dimensional matrix of each pixel in the 3D medical image of the maximum connected area. If the pixel u1,u2,…,u p is 0, then the two-dimensional matrix of each pixel in the image is changed to 0; Step 4: When it is determined that the gradient difference at adjacent pixel points x and y in the data matrix is ​​0 and the two-dimensional matrix value at x and y is 1, there are adjacent pixels in the 3D medical image of the maximum connected area in x and y.

8. The deep learning-based medical image feature extraction and disease prediction platform according to claim 7, characterized in that: The data sets for constructing, training and validating the classification prediction model are all from the internationally used database ADNI.

9. The medical image feature extraction and disease prediction platform based on deep learning according to claim 7, characterized in that: The step 4 comprises: Step 4(a): firstly obtain a 3D medical image of the target to be predicted, and process the 3D medical image of the target to be predicted by using a Gaussian filter; Step 4(b), extracting the largest connected region of the input 3D medical image by using the above neural network construction method; Step 4(c), establishing a three-dimensional gray value feature, comparing the brain gray matter density with the density of the cerebral cortical nuclei, and establishing a brain gray matter density ratio; Step 4(d): Based on the established brain gray matter density ratio, a classification prediction model is constructed, and the model includes a test set, a training set, and a validation set.

10. The medical image feature extraction and disease prediction platform based on deep learning according to claim 5, characterized in that: The parameter w in the neural network is updated and iterated by the gradient descent method. The specific update iteration form is as follows: Among them, n represents the number of iterative updates. In each iterative update process, the gradient expression of the loss function with respect to the parameter w needs to be assigned to the corresponding parameter w to update the parameter w. In the gradient descent method, the update method of the parameter w is as follows: Where w″ is the updated parameter value, w is the parameter value before the update, and α is a hyperparameter with a value range of [0,1], which indicates the distance multiple of the parameter w moves in each iteration. is the gradient of the loss function L(w) with respect to the parameter w in the nth iteration of gradient descent; The gradient The calculation method is to obtain the derivative of the loss function with respect to the parameter w, that is: Using the gradient, the parameter w is iteratively updated through the above formula until the termination condition is met. When the values ​​of all elements in multiple new one-dimensional data matrices obtained by the gradient descent method are equal to 0, the gradient update process is terminated; wherein, the algorithm of step 1 is updated through the above update method.

Images