Intelligent disease diagnosis system based on medical image data processing

Abstract

The invention discloses an intelligent disease diagnosis system based on medical image data processing, and relates to the technical field of medical data processing. The system comprises an image acquisition module, an image processing module, a diagnosis model training module and a diagnosis module which are connected in sequence, the image acquisition module is used for receiving a medical image uploaded by a patient through a mobile terminal and converting the medical image into a data image; the image processing module is used for preprocessing the data image and then importing the data image into the trained model; the diagnosis model training module is used for selecting a deep convolutional neural network as a component classifier and designing a diagnosis model; and the diagnosis module is used for performing disease diagnosis and category reasoning based on the diagnosis model. The diagnosis model is trained through the deep convolutional neural network, the medical image uploaded by the mobile terminal of the patient is imported into the trained diagnosis model, and the output result is sent to the mobile terminal of the patient, so that the treatment time of the patient is shortened, and the disease diagnosis accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical data processing, and in particular relates to an intelligent disease diagnosis system based on medical image data processing. Background Art

[0002] Medical imaging refers to the technology and process of obtaining images of internal tissues of the human body or a part of the human body in a non-invasive manner for medical treatment or medical research. It includes the following two relatively independent research directions: medical imaging systems and medical image processing. The former refers to the process of image formation, including the study of imaging mechanisms, imaging equipment, imaging system analysis and other issues; the latter refers to further processing of the acquired images, the purpose of which is either to restore the original unclear image, or to highlight certain feature information in the image, or to classify the image pattern, etc. As a science, medical imaging belongs to biological imaging, and includes imaging diagnosis, radiology, endoscopy, medical thermal imaging technology, medical photography and microscopy.

[0003] Among them, medical images can be produced by X-rays, which are mainly based on the penetration, differential absorption, photosensitivity and fluorescence of X-rays. Since X-rays are absorbed to different degrees when passing through the human body, such as the amount of X-rays absorbed by bones is more than that absorbed by muscles, the amount of X-rays after passing through the human body is different, thus carrying information on the density distribution of various parts of the human body. The intensity of the fluorescence or photosensitivity caused on the fluorescent screen or photographic film is quite different, so shadows of different densities will be displayed on the fluorescent screen or photographic film (after development and fixing). Based on the contrast of the shadows, combined with clinical manifestations, test results and pathological diagnosis, it can be judged whether a part of the human body is normal.

[0004] At present, patients usually take the X-ray films to see doctors for diagnosis. When there are many patients, they need to queue up and wait for the doctor, which consumes most of the time. Some seriously ill patients will have delayed treatment, which has an adverse effect on their health.

[0005] Therefore, there is an urgent need for an intelligent disease diagnosis system based on medical image data processing that can automatically diagnose medical images, save patients' time, provide reference for doctors, and speed up treatment progress. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent disease diagnosis system based on medical image data processing, which trains the diagnosis model through a deep convolutional neural network, imports the medical images uploaded by the patient's mobile terminal into the trained diagnosis model, and sends the output results to the patient's mobile terminal, solving the problem that existing hospital images require doctors' subjective diagnosis, which affects diagnostic efficiency and easily delays patient treatment.

[0007] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0008] The present invention is an intelligent disease diagnosis system based on medical image data processing, comprising an image acquisition module, an image processing module, a diagnosis model training module, and a diagnosis module connected in sequence; the image acquisition module is used to receive medical images uploaded by patients through mobile terminals, and convert them into data images and output them to the image processing module; the image processing module is used to pre-process the data images and import the pre-processed data images into a trained model; the diagnosis model training module is used to select a deep convolutional neural network as a component classifier and design a diagnosis model; the diagnosis module is used to diagnose and classify diseases based on the diagnosis model, and send the output results of the diagnosis model to the patient's mobile terminal through a wireless communication module.

[0009] As a preferred technical solution, the medical images include CT images, MRI or X-ray images.

[0010] As a preferred technical solution, the preprocessing includes filtering the data image, denoising the filtered data image, and segmenting the denoised data image;

[0011] The filtering process adopts Gaussian filtering, uses a discretized two-dimensional Gaussian filtering function as a weight coefficient in a Gaussian kernel, discretizes the two-dimensional continuous normal distribution function, and obtains a weight matrix;

[0012] Among them, the two-dimensional Gaussian filter function formula is:

[0013]

[0014] In the formula, σ represents the standard deviation, (x, y) represents the pixel coordinates, and x, y are integers;

[0015] After the filtering process, the specific process of denoising the data image is as follows:

[0016] Step L1: Select two points on the data image as two diagonal points of a rectangle to construct a corresponding matrix;

[0017] Step L2: Change the value of the noise point in the matrix from 1 to 0;

[0018] Step L3: Create eight zero matrices in eight directions of the matrix and fuse them into a matrix of the same order as the data image matrix after filtering;

[0019] Step L4: Add the fused matrix to the filtered matrix, change all 1 values ​​in the added matrix to 0, and then change all 2 values ​​to 1;

[0020] After the denoising process, the specific steps of segmenting the data image are as follows:

[0021] Step Q1: Get the total average grayscale of the image:

[0022] u=w0*u0+w1*u1;

[0023] Step Q2: The variance of foreground and background images is:

[0024]

[0025] In the formula, the proportion of foreground points in the image is w0, and the average grayscale is u0; the proportion of background points in the image is w1, and the average grayscale is u1.

[0026] As a preferred technical solution, the method for obtaining the matrix in step L1 is as follows:

[0027] Step J1: Use the ginput function to determine the range to be selected, and select two diagonal corners within the range, with the coordinates of the two diagonal corners being F1(x1,y1) and F2(x2,y2) respectively;

[0028] Step J2: Calculate the length and width of the rectangle enclosed by the two corner points based on the values ​​of F1 and F2;

[0029] Step J3: read out the length and width of the original framed image as L and H respectively;

[0030] Step J4: Use F1(x1, y1), F2(x2, y2), L, and H to calculate the size of the matrix in four directions centered on the rectangle formed by the two corner points.

[0031] As a preferred technical solution, the training method of the diagnostic model training module includes: the diagnostic model training module uses ResNet or EfficientNet convolutional neural network as the backbone network for feature extraction; uses a feature pyramid network to expand ResNet or EfficientNet, uses a global average pooling operator to perform spatial scale dimension reduction on the feature vector of each layer of the feature pyramid, and obtains a feature vector with a width and height of 1, sets the output dimensions of the expanded features to be the same, and after the set scale features pass through the global average pooling operator, uses a fully connected layer to perform feature transformation on them, and uses an addition operation to fuse the expanded features, and continues to use the next fully connected layer to continue to perform feature transformation on the fused features, and finally outputs the extracted image features.

[0032] As an optimal technical solution, there are two model training methods:

[0033] Method 1:

[0034] Step S11: The preprocessed image is trained on the ResNet component, and the label is the same as the original image;

[0035] Step S12: obtain the feature layers of each component ResNet and splice them together to correspond to the labels of the original image;

[0036] Step S13: adding several fully connected layers to train the artificial neural network;

[0037] Step S14: Alternately adjust the feature layer and the artificial neural network according to the classification performance of the overall model;

[0038] Step S15: fix the parameters of each part and use the whole as a prediction model;

[0039] Method 2:

[0040] Step S21: remove the output layer of the component ResNet and concatenate the feature layers together;

[0041] Step S22: Add several more fully connected layers;

[0042] Step S23: adding an output layer;

[0043] Step S24: Train the model as a whole and directly implement end-to-end prediction.

[0044] As a preferred technical solution, the diagnosis module includes a first diagnosis module and a second diagnosis module; the first diagnosis module performs disease diagnosis reasoning based on a Bayesian network; the second diagnosis module performs disease diagnosis and attribute category reasoning based on a convolutional network;

[0045] The diagnosis module combines the outputs of the first diagnosis module and the second diagnosis module to obtain a final diagnosis result.

[0046] As a preferred technical solution, the first diagnosis module establishes a probabilistic dependency relationship model between disease categories and lesion attributes based on a Bayesian network;

[0047] The Bayesian network is represented as: <V B ,E B ,θ>, where <V B ,E B > represents a directed acyclic graph, V B represents a node set, E Brepresents the set of edges, θ represents the Bayesian network parameters composed of the conditional probability table of each node; the edge connecting two nodes is used to represent the probabilistic dependency relationship between the two nodes; first, the input image features are converted into probability scores of attributes, and sent to the Bayesian network as evidence for forward reasoning, and the input evidence of each node is combined with the conditional probability table to obtain the marginal posterior probability of each node. The marginal posterior probability P of node v0 is B (v0) is:

[0048] P B (v0)=∫...∫ V P(v0,v1,v2,...,v n )dv1...dv n

[0049]

[0050] Where n is the number of nodes, Parents(v i ) represents node v i The parent node of v i When there is no parent node, Parents(v i ) is empty.

[0051] As a preferred technical solution, the system also includes a database and an entry module; the database is used to store disease name information and descriptive information corresponding to the disease name information, part information keywords corresponding to the part information, part extension words corresponding to the part information keywords, descriptive information corresponding to the disease, descriptive keywords of the descriptive information, and descriptive extension words of the descriptive keywords; the entry module is used to enter the stored disease information into the database.

[0052] The present invention has the following beneficial effects:

[0053] The present invention trains the diagnostic model through a deep convolutional neural network, imports the medical images uploaded by the patient's mobile terminal into the trained diagnostic model, and sends the output results to the patient's mobile terminal, thereby reducing the patient's consultation time and improving the accuracy of disease diagnosis.

[0054] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 A schematic diagram of the structure of an intelligent disease diagnosis system based on medical image data processing according to the present invention;

[0057] Figure 2 A specific flow chart for denoising a data image;

[0058] Figure 3 This is a schematic diagram of the structure of the neural network classifier ResNet of the present invention;

[0059] Figure 4 A step diagram of a model training method for the present invention;

[0060] Figure 5 This is another step diagram of implementing the model training method of the present invention. DETAILED DESCRIPTION

[0061] 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.

[0062] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0063] In order to make the purpose, technical solution and advantages of this application clearer, the following Figure 1 The implementation methods of the present application are described in further detail.

[0064] Before introducing the embodiments of the present application, the medical imaging artificial intelligence assisted diagnosis system is first described.

[0065] In recent computer vision research, artificial intelligence, especially deep neural networks, have achieved accuracy that exceeds that of classical machine learning models in a series of tasks such as image recognition, target detection, and semantic segmentation. At the same time, deep neural networks have been applied to the field of medical image analysis and have achieved good recognition results in the automatic diagnosis of some diseases, such as automatic detection of intracranial hemorrhage areas based on CT images, automatic classification and lesion localization of chest diseases based on X-ray images, automatic diagnosis of breast cancer based on mammography, and automatic screening of diabetes based on fundus images. The reason why deep learning can achieve good model performance in the field of medical image analysis is mainly due to its powerful image feature extraction and feature learning capabilities. It can learn representative and highly distinguishing image features of lesions from various medical images, thereby achieving high-precision disease diagnosis.

[0066] The existing medical imaging AI-assisted diagnosis system can transmit medical images to the AI ​​computing platform, call the corresponding deep learning model to automatically identify them, and then transmit the identification results to doctors for reference, query and review. However, the above system only uses a general deep learning model to identify and classify lesions. Since deep learning is essentially a data-driven model, its learning process is close to a black box. Therefore, the above model has poor verifiability and low diagnostic accuracy. Although doctors can also obtain the lesion recognition results given by the deep learning model, the recognition process of the model is invisible to doctors, and it is difficult for doctors to fully trust the results of automatic diagnosis. Therefore, in the actual work process, it takes time to carefully review the automatic diagnosis results, which is difficult to meet the generalization requirements of doctors for the reasoning process of artificial intelligence models. Specifically, verifiability means that the reasoning and decision-making process within the disease intelligent diagnosis model must not only be explainable, but also be consistent with the knowledge and rules in the medical field, so that its diagnosis results are verifiable. Taking the classification of benign and malignant lung nodules in medical imaging diagnosis as an example, the disease intelligent diagnosis model needs to first identify and grade the imaging properties of lung nodules like a doctor, and then learn the causal relationship between these lesion attributes (such as lobulation sign, burr sign, sphericity, etc.) and the benign and malignant nature of the lesions, so as to restore the doctor's diagnostic process and make the diagnostic results verifiable.

[0067] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-2 It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0068] Embodiment 1

[0069] See also Figure 1 As shown, the present invention is an intelligent disease diagnosis system based on medical image data processing, including an image acquisition module, an image processing module, a diagnosis model training module, and a diagnosis module connected in sequence; the image acquisition module is used to receive medical images uploaded by patients through mobile terminals, and convert them into data images and output them to the image processing module; the image processing module is used to pre-process the data images and import the pre-processed data images into the trained model; the diagnosis model training module is used to select a deep convolutional neural network as a component classifier and design a diagnosis model; the diagnosis module is used to diagnose and classify diseases based on the diagnosis model, and send the output results of the diagnosis model to the patient's mobile terminal through the wireless communication module.

[0070] Medical images include CT images, MRI or X-ray images.

[0071] The preprocessing includes: filtering the data image, denoising the filtered data image, and segmenting the denoised data image;

[0072] The filtering process adopts Gaussian filtering, and uses the discretized two-dimensional Gaussian filtering function as the weight coefficient in the Gaussian kernel to discretize the two-dimensional continuous normal distribution function to obtain a weight matrix;

[0073] Among them, the two-dimensional Gaussian filter function formula is:

[0074]

[0075] In the formula, σ represents the standard deviation, (x, y) represents the pixel coordinates, and x, y are integers;

[0076] See also Figure 2 As shown in the figure, after filtering, the specific process of denoising the data image is as follows:

[0077] Step L1: Select two points on the data image as two diagonal points of a rectangle to construct a corresponding matrix;

[0078] Step L2: Change the value of the noise point in the matrix from 1 to 0;

[0079] Step L3: Create eight zero matrices in eight directions of the matrix and fuse them into a matrix of the same order as the data image matrix after filtering;

[0080] Step L4: Add the fused matrix to the filtered matrix, change all 1 values ​​in the added matrix to 0, and then change all 2 values ​​to 1;

[0081] Image segmentation processing is to take an optimal threshold to divide the original image into foreground color and background color. The optimal threshold is set according to the medical image. The inter-class variance of the two parts is denoted as t, which is the segmentation threshold of foreground color and background color. The proportion of foreground points in the image is w0, and the average grayscale is u0; the proportion of background points in the image is w1, and the average grayscale is u1. The total average grayscale of the image is

[0082] After denoising, the specific steps for segmenting the data image are as follows:

[0083] Step Q1: Get the total average grayscale of the image:

[0084] u=w0*u0+w1*u1;

[0085] Step Q2: The variance of foreground and background images is:

[0086]

[0087] In the formula, the proportion of foreground points in the image is w0, and the average grayscale is u0; the proportion of background points in the image is w1, and the average grayscale is u1.

[0088] There are three common filtering methods for smoothing data images, namely: mean filtering, median filtering and Gaussian filtering. This embodiment repeatedly tests the three filtering methods and compares the test results; finally, Gaussian filtering is selected to process the data image;

[0089] Image Gaussian smoothing is a method of smoothing data images based on the idea of ​​neighborhood averaging, assigning different weights to pixels at different positions. Formula 1) is a two-dimensional Gaussian function:

[0090]

[0091] In the formula, σ represents the standard deviation, (x, y) represents the pixel coordinates, and x, y are integers;

[0092] Discretize the two-dimensional Gaussian filter function, and use the function value as the weight coefficient in the Gaussian kernel. The Gaussian template discretizes the two-dimensional continuous normal distribution function to obtain a (2k+1)×(2k+1)-order weight matrix.

[0093]

[0094] The value of each pixel in the neighborhood can be determined by formula 2), and the variance σ 2 The size of has a great influence on the weight of the Gaussian template. It is particularly important to select a suitable template. In this embodiment, a 3*3 template is selected to process the data image.

[0095] When using mean filtering, the edge information and feature information in the image will be "blurred", and many features will be lost. Gaussian filtering has the same degree of smoothness in all directions, does not change the edge direction of the original image, and ensures the characteristics of feature points and edges. The processing effect of median filtering is similar to that of Gaussian filtering, but the operation time in Matlab is too long compared to Gaussian filtering. Therefore, the embodiment selects Gaussian filtering to smooth the data image.

[0096] The matrix of step L1 is obtained as follows:

[0097] Step J1: Use the ginput function to determine the range to be selected, and select two diagonal corners within the range, with the coordinates of the two diagonal corners being F1(x1,y1) and F2(x2,y2) respectively;

[0098] Step J2: Calculate the length and width of the rectangle enclosed by the two corner points based on the values ​​of F1 and F2;

[0099] Step J3: read out the length and width of the original framed image as L and H respectively;

[0100] Step J4: Use F1(x1, y1), F2(x2, y2), L, and H to calculate the size of the matrix in four directions centered on the rectangle formed by the two corner points.

[0101] The training method of the diagnosis model training module includes: the diagnosis model training module uses the ResNet or EfficientNet convolutional neural network as the backbone network for feature extraction; uses a feature pyramid network to expand ResNet or EfficientNet, uses a global average pooling operator to perform spatial scale dimension reduction on the feature vector of each layer of the feature pyramid, and obtains a feature vector with a width and a height of 1, sets the output dimensions of the expanded features to be the same, uses a fully connected layer to transform the set scale features after the global average pooling operator, uses an addition operation to fuse the expanded features, and continues to use the next fully connected layer to continue to transform the fused features, and finally outputs the extracted image features.

[0102] Image features are extracted through component classifiers, and then spliced ​​or otherwise combined; the component classifier is a deep convolutional network ResNet.

[0103] See also Figure 3 As shown in the figure, the deep convolutional network ResNet is used as a component classifier. A ResNet is composed of multiple stacked modules. At this time, the features obtained by the component classifier are selected as the last fully connected layer of the convolutional network. For each ResNet, this feature is 2048-dimensional.

[0104] The features obtained by the ResNet component are concatenated together as the features of the skin disease image to obtain the classification probability; the concatenated feature layer can be directly connected to the output probability layer, or a fully connected layer can be added between the two layers, that is, an additional artificial neural network is designed.

[0105] There are two ways to train the model:

[0106] See also Figure 4 As shown, method one:

[0107] Step S11: The preprocessed image is trained on the ResNet component, and the label is the same as the original image;

[0108] Step S12: obtain the feature layers of each component ResNet and splice them together to correspond to the labels of the original image;

[0109] Step S13: adding several fully connected layers to train the artificial neural network;

[0110] Step S14: Alternately adjust the feature layer and the artificial neural network according to the classification performance of the overall model;

[0111] Step S15 fixes the parameters of each part and takes the whole as a prediction model;

[0112] See also Figure 5 As shown, method 2:

[0113] Step S21: remove the output layer of the component ResNet and concatenate the feature layers together;

[0114] Step S22: Add several more fully connected layers;

[0115] Step S23: adding an output layer;

[0116] Step S24: Train the model as a whole and directly implement end-to-end prediction.

[0117] The diagnosis module includes a first diagnosis module and a second diagnosis module; the first diagnosis module performs disease diagnosis reasoning based on a Bayesian network; the second diagnosis module performs disease diagnosis and attribute category reasoning based on a convolutional network;

[0118] The diagnosis module combines the outputs of the first diagnosis module and the second diagnosis module to obtain a final diagnosis result.

[0119] The first diagnosis module establishes a probabilistic dependency model between disease categories and lesion attributes based on a Bayesian network;

[0120] The Bayesian network is represented as: B = <V B ,E B ,θ>, where <V B ,E B > represents a directed acyclic graph, V B represents a node set, E B represents the set of edges, θ represents the Bayesian network parameters composed of the conditional probability table of each node; the edge connecting two nodes is used to represent the probabilistic dependency relationship between the two nodes; first, the input image features are converted into probability scores of attributes, and sent to the Bayesian network as evidence for forward reasoning, and the input evidence of each node is combined with the conditional probability table to obtain the marginal posterior probability of each node. The marginal posterior probability P of node v0 is B (v0) is:

[0121] P B (v0)=∫...∫ V P(v0,v1,v2,...,v n )dv1...dv n

[0122]

[0123] Where n is the number of nodes, Parents(v i ) represents node v i The parent node of v i When there is no parent node, Parents(v i ) is empty.

[0124] The system also includes a database and an input module; the database is used to store disease name information and description information corresponding to the disease name information, location information keywords corresponding to the location information, location extension words corresponding to the location information keywords, description information corresponding to the disease, description keywords of the description information, and description extension words of the description keywords; the input module is used to enter the stored disease information into the database.

[0125] When the user inputs the diseased part information as hand, and the patient's condition description information is hand joint pain in the present invention, the processor first compares the hand with the part word bag feature information in the first text feature vector, outputs the first text feature vector containing the hand as a third text feature vector, then compares the joint pain with the description word bag feature information in the third text feature vector, converts the third text feature vector containing the word "joint" and "pain" into a fourth text feature vector, and then converts the second data image corresponding to the fourth text feature vector into a third data image.

[0126] Through comparison, it is found that the diseases corresponding to the first text feature information [(hand, palm, finger), (joint, connection, link)] may be the diseases suffered by the patient. Then, the X-ray film of the disease corresponding to the first text feature information [(hand, palm, finger), (joint, connection, link)], that is, the third data image, is scaled to overlap with the X-ray film of the patient taken in the hospital, that is, the first data image, by the fourth method, and the intersection points of the two images are compared. If the intersection point of the third data image with the first data image is less than or equal to the first preset threshold, that is, less than 3, then it is proved that the disease corresponding to the third data image is the disease suffered by the patient. If the intersection point is greater than 3, it is proved that the patient's X-ray film is inconsistent with the X-ray film stored in the database, and a follow-up signal is output to allow the doctor to conduct a follow-up diagnosis.

[0127] Embodiment 2

[0128] The present invention is an intelligent disease diagnosis system based on medical image data processing, which can be used to execute the method content of Example 1 of the present invention.

[0129] In the present invention, patients can not only upload medical images to the system for diagnosis through mobile terminals; when there are no medical images, they can directly take pictures of the diseased parts of the body and upload them to the system, such as skin disease pictures, STD pictures, wound pictures, etc.; the diagnosis model trained in Example 1 can also be used to diagnose diseases, but because it is a human body picture, each person's skin color, hair density, and light will have an impact, and the system needs to perform preliminary processing after uploading the picture, such as: image color balancing, hair removal, and disease area detection, etc.;

[0130] Image color balance: Use simplestcolorbalance or grayworld algorithm to improve image quality;

[0131] Hair removal is done through a mask reconstruction algorithm, which is used to reduce the noise of the image;

[0132] Disease area detection is performed by overfeat or RCNN algorithm for skin lesion location and refinement. After the image processing is completed, the processing flow is consistent with that of the first embodiment, and no further explanation is given here.

[0133] It is worth noting that in the above system embodiment, the various units included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0134] In addition, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiments can be completed by instructing related hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

[0135] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent disease diagnosis system based on medical image data processing, comprising an image acquisition module, an image processing module, a diagnosis model training module, a diagnosis module, a database and an input module connected in sequence, characterized in that: The image acquisition module is used to receive the medical images uploaded by the patient through the mobile terminal, convert them into digital images and output them to the image processing module; The image processing module is used to preprocess the data image and import the preprocessed data image into the trained model; The diagnostic model training module is used to select a deep convolutional neural network as a component classifier and design a diagnostic model; The diagnosis module is used to diagnose and classify the disease based on the diagnosis model, and send the output result of the diagnosis model to the patient's mobile terminal through the wireless communication module; The database is used to store disease name information and description information corresponding to the disease name information, part information keywords corresponding to the part information, part extension words corresponding to the part information keywords, description information corresponding to the disease, description keywords of the description information, and description extension words of the description keywords; The input module is used to input the stored disease information into the database.

2. The disease intelligent diagnosis system based on medical image data processing according to claim 1 is characterized in that: The medical images include CT images, MRI or X-ray images.

3. The disease intelligent diagnosis system based on medical image data processing according to claim 1 is characterized in that: The preprocessing includes filtering the data image, denoising the filtered data image, and segmenting the denoised data image; The filtering process adopts Gaussian filtering, uses a discretized two-dimensional Gaussian filtering function as a weight coefficient in a Gaussian kernel, discretizes the two-dimensional continuous normal distribution function, and obtains a weight matrix; Among them, the two-dimensional Gaussian filter function formula is: In the formula, σ represents the standard deviation, (x, y) represents the pixel coordinates, and x, y are integers; After the filtering process, the specific process of denoising the data image is as follows: Step L1: Select two points on the data image as two diagonal points of a rectangle to construct a corresponding matrix; Step L2: Change the value of the noise point in the matrix from 1 to 0; Step L3: Create eight zero matrices in eight directions of the matrix and fuse them into a matrix of the same order as the data image matrix after filtering; Step L4: Add the fused matrix to the filtered matrix, change all 1 values ​​in the added matrix to 0, and then change all 2 values ​​to 1; After the denoising process, the specific steps of segmenting the data image are as follows: Step Q1: Get the total average grayscale of the image: u=w0*u0+w1*u1; Step Q2: The variance of foreground and background images is: g=w0*(u0-u)*(u0-u)+w1*(u1-u)*(u1-u) =w0*w1*(u0-u1)*(u0-u1); In the formula, the proportion of foreground points in the image is w0, and the average grayscale is u0; the proportion of background points in the image is w1, and the average grayscale is u1.

4. The disease intelligent diagnosis system based on medical image data processing according to claim 3 is characterized in that: The method for obtaining the matrix of step L1 is as follows: Step J1: Use the ginput function to determine the range to be selected, and select two diagonal corners within the range, with the coordinates of the two diagonal corners being F1(x1,y1) and F2(x2,y2) respectively; Step J2: Calculate the length and width of the rectangle enclosed by the two corner points based on the values ​​of F1 and F2; Step J3: read out the length and width of the original framed image as L and H respectively; Step J4: Use F1(x1, y1), F2(x2, y2), L, and H to calculate the size of the matrix in four directions centered on the rectangle formed by the two corner points.

5. The disease intelligent diagnosis system based on medical image data processing according to claim 1 is characterized in that: The training method of the diagnostic model training module includes: the diagnostic model training module uses a ResNet or EfficientNet convolutional neural network as a backbone network for feature extraction; uses a feature pyramid network to expand ResNet or EfficientNet, uses a global average pooling operator to perform spatial scale dimension reduction on the feature vector of each layer of the feature pyramid, and obtains a feature vector with a width and a height of 1, sets the output dimensions of the expanded features to be the same, uses a fully connected layer to perform feature transformation on the set scale features after the global average pooling operator, uses an addition operation to fuse the expanded features, and continues to use the next fully connected layer to continue to perform feature transformation on the fused features, and finally outputs the extracted image features.

6. The disease intelligent diagnosis system based on medical image data processing according to claim 5 is characterized in that: There are two ways to train the model: Method 1: Step S11: The preprocessed image is trained on the ResNet component, and the label is the same as the original image; Step S12: obtain the feature layers of each component ResNet and splice them together to correspond to the labels of the original image; Step S13: adding several fully connected layers to train the artificial neural network; Step S14: Alternately adjust the feature layer and the artificial neural network according to the classification performance of the overall model; Step S15: fix the parameters of each part and use the whole as a prediction model; Method 2: Step S21: remove the output layer of the component ResNet and concatenate the feature layers together; Step S22: Add several more fully connected layers; Step S23: adding an output layer; Step S24: Train the model as a whole and directly implement end-to-end prediction.

7. The disease intelligent diagnosis system based on medical image data processing according to claim 1 is characterized in that: The diagnosis module includes a first diagnosis module and a second diagnosis module; the first diagnosis module performs disease diagnosis reasoning based on a Bayesian network; the second diagnosis module performs disease diagnosis and attribute category reasoning based on a convolutional network; The diagnosis module combines the outputs of the first diagnosis module and the second diagnosis module to obtain a final diagnosis result.

8. The disease intelligent diagnosis system based on medical image data processing according to claim 7 is characterized in that: The first diagnosis module establishes a probabilistic dependency relationship model between disease categories and lesion attributes based on a Bayesian network; The Bayesian network is represented as: <V B ,E B ,θ>, where <V B ,E B > represents a directed acyclic graph, V B represents a node set, E B represents the set of edges, θ represents the Bayesian network parameters composed of the conditional probability table of each node; the edge connecting two nodes is used to represent the probabilistic dependency relationship between the two nodes; first, the input image features are converted into probability scores of attributes, and sent to the Bayesian network as evidence for forward reasoning, and the input evidence of each node is combined with the conditional probability table to obtain the marginal posterior probability of each node. The marginal posterior probability P of node v0 is B (v0) is: P B (v0)=∫...∫ V P(v0,v1,v2,...,v n )dv1...dv n Where n is the number of nodes, Parents(v i ) represents node v i The parent node of v i When there is no parent node, Parents(v i ) is empty.

Images