Automatic diagnostic equipment and system for medical imaging

Abstract

The present invention discloses an automatic diagnostic device and system for medical imaging, which specifically relates to the field of image recognition technology. The device includes an image data acquisition module, a preliminary diagnosis module, a diagnosis result verification module, an image segmentation and quantification module, and a database module. The preliminary diagnosis module is used to predict the disease contained in the lung image obtained by the image data acquisition module. After obtaining the image data, the image features are extracted through preprocessing, convolution, and pooling operations. The extracted image features are compared with the first database for similarity. The type of disease is predicted based on the comparison result, and the predicted type of disease is transmitted to the diagnosis result verification module; the diagnosis result verification module is used to verify the accuracy of the diagnosis result, obtain the type of disease in the image data, and predict the severity of the disease; the image segmentation and quantification is used to perform quantitative analysis on the disease, segment the lesion in the diagnosis result, and obtain the location and size of the lesion.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and more particularly to an automatic diagnostic device and system for medical imaging. Background Art

[0002] Among current medical testing methods, medical imaging data accounts for 90% of medical information. It is the main source of information for disease screening and diagnosis and treatment, and is also an advanced means to assist clinical disease diagnosis and treatment.

[0003] Lung CT is a noninvasive examination primarily used to detect respiratory diseases such as pneumonia, bronchitis, bronchiectasis, lung tumors, and mediastinal tumors. It includes both plain and enhanced CT scans. For common inflammatory diseases, a single plain scan can generally confirm the diagnosis. When a plain scan reveals a lung mass or nodularity, an enhanced scan is typically performed. Enhanced CT scans, performed after intravenous injection of contrast agent, can assess changes in CT values ​​and provide a rough estimate of benign or malignant lesions. Lung CT is currently one of the most common clinical examinations.

[0004] Using window width technology, lung CT is divided into lung windows, mediastinal windows, and bone windows. The lung window primarily examines lesions within the lungs and trachea, including masses, ground-glass nodules, infections, and tracheal masses or stenosis. The mediastinal window primarily examines the size of the solid portion of the tumor or infection, whether it is enhanced, its relationship to blood vessels, whether the mediastinal lymph nodes are enlarged, whether there is pleural effusion, the size of the heart, and the presence of pericardial effusion. The bone window primarily examines the scanned chest, including the thoracic vertebrae and ribs, for fractures, bone destruction, or developmental abnormalities. Observing the entire lung CT scan requires comparing the lung window with the mediastinal window. The location and characterization of the lesion are determined by differences in image density and location, and a comprehensive analysis based on the clinical history is required for diagnosis. If the diagnosis is unclear on a plain scan, an enhanced scan can be performed to reveal the vascular involvement of the lesion and to determine the presence of lymph node enlargement.

[0005] At present, the diagnosis of lung CT images mainly relies on manual reading, which requires radiologists to have rich clinical experience. Sometimes repeated observations are needed to confirm the diagnosis, which is highly subjective. A large amount of repetitive reading work will increase the doctor's fatigue. In actual investigations, for the same CT image, different doctors or the same doctor at different times will have observer bias, leading to misdiagnosis and missed diagnosis of the disease.

[0006] Therefore, there is an urgent need for an automatic medical imaging diagnosis system that can automatically diagnose medical images, save patients' time, provide reference for doctors, and speed up the diagnosis process. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an automatic diagnostic device and system for medical imaging, which completes automatic diagnosis of lung CT through preliminary diagnosis, diagnosis verification and quantitative analysis of lesions of imaging data, so as to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: an automatic diagnostic system for medical imaging, comprising an image data acquisition module, a preliminary diagnosis module, a diagnosis result verification module, an image segmentation and quantification module, and a database module, wherein the image data acquisition module is used to acquire lung image data, determine whether the format of the image data meets the requirements, and transmit the acquired lung image data to the preliminary diagnosis module;

[0009] The preliminary diagnosis module is used to predict diseases contained in lung images. After obtaining image data, the image features are extracted through preprocessing, convolution, and pooling operations. The extracted image features are compared with the first database for similarity. The type of disease is predicted based on the comparison results, and the predicted disease type is transmitted to the diagnosis result verification module.

[0010] The diagnosis result verification module is used to verify the accuracy of the diagnosis result, obtain the disease type in the image data, and predict the severity of the disease;

[0011] The image segmentation and quantification is used to perform quantitative analysis on the disease, segment the lesions of the diagnosis results, obtain quantitative data of the lesions, and obtain the location and size of the lesions;

[0012] The database module is used to store a historical database of diseases, including a first database and a second database. The first database is used to store CT image data and corresponding diagnosis results, and the second database is used to store diagnosis indicators of diseases.

[0013] In a preferred embodiment, the preliminary diagnosis module extracts image features and compares the obtained features with the second database to predict the type of disease, including the following steps:

[0014] Step S01, image preprocessing: highlighting the image data features through image denoising and image enhancement technology, and dividing the lung image data into n regions, denoted as 1, 2, ...;

[0015] Step S02, feature extraction: extract features that affect each area through machine vision, extract information features from the texture, color, shape and gradient of the image data, and perform convolution and pooling operations on the input image to output a feature set, which includes a feature subset;

[0016] Step S03, feature localization: The region recommendation network generates candidate region blocks. Each region block is pooled to a uniform size and input into the fully connected layer for classification. The bounding box is used to accurately predict the position of the target and obtain the position corresponding to the feature subset. The position of the feature subset is the location of the lesion.

[0017] Step S04, similarity comparison: Calculate the similarity between the feature set in each region and the second database, and obtain diseases with similarity exceeding a preset value, which are recorded as the predicted disease set.

[0018] In a preferred embodiment, the diagnosis result verification module is used to verify whether the image contains disease A, including the following steps:

[0019] Step S11: Acquire characteristic parameters of each region of the image data to obtain the type of disease, divide the lung into n regions, and count the characteristic parameters of each disease in each region. The characteristic parameters of disease A in n regions are recorded as C1, where C1 = [c11, c12, ..., c1n];

[0020] Step S12: Count historical data to obtain the image data of disease A. The probability of lesions appearing in each area is regarded as the weight of disease A in the area to obtain the weight coefficient of each disease. The set of weight coefficients of disease A is recorded as Q1, Q1 = [q11, q12, ..., q1n]. The weight coefficient is calculated according to the area s1 and the number s2 of lesions in the area. The probability calculation formula satisfies Where k1 and k2 are the impact factor constants, is the average value of disease A in the region;

[0021] Step S13: multiply the obtained characteristic parameters by the weight coefficient corresponding to the disease. The preset disease discrimination standard is P, and the evaluation formula of disease A is F1(A)=C1×Q1≥P1, which indicates that the CT image diagnosis result has disease A, and the severity level of disease A satisfies the formula The P satisfies Where η1 is the influence coefficient constant.

[0022] In a preferred embodiment, the second database is used to express the feature set of disease A in each region, and obtaining the second database includes the following steps:

[0023] Step S21, sample data acquisition: obtain sample data from the first database, mark the lung CT image data as sample input X, and record the diagnosis result as sample output Y;

[0024] Step S22, classification of sample data: Classify samples according to disease type. Pulmonary diseases include infectious diseases, non-infectious diseases, and tumor diseases. Diseases are numbered sequentially to obtain diseases 1, 2, 3, ..., and a correspondence between lung diseases and image data is established. Using an image feature extraction algorithm, the image feature y1 corresponding to the disease x1 is extracted and denoted as F3(x1, y1). F3(x1, y1) represents a correspondence function between the disease and the image feature.

[0025] Step S23, feature extraction: A machine vision algorithm is used to extract features that affect the image, and a feature set B corresponding to disease A is obtained. Information features are mainly extracted from several aspects such as texture, color, shape, and gradient. The input image undergoes basic convolution and pooling operations to extract a feature map. Candidate region blocks are generated through a region recommendation network. Each region block is pooled to a uniform size and input into a fully connected layer for classification. The bounding box is used to accurately predict the location of the target.

[0026] Step S24: Based on medical knowledge, theoretical impact features of disease A are obtained. Based on theoretical imaging features and historical features, irrelevant features are eliminated. The filtered features are classified according to the regions they belong to, and recorded as regions 1, 2, ...;

[0027] Step S25: Analyze the weight of each image feature for disease A, obtain the weight coefficient Q of each regional feature for disease A, store the weight coefficient of each region of disease A in the first database, and so on to obtain the weight coefficients of all diseases in each region.

[0028] In a preferred embodiment, the first database is a hospital's historical image database, in which lung CT images and corresponding diagnosis results are obtained. After privacy protection, lung CT images and corresponding diagnosis results of a large number of cases are obtained. The lung CT image data is marked as sample input X, and the diagnosis results are recorded as sample output Y.

[0029] In a preferred embodiment, the image segmentation algorithm is based on 2D images and 3D images, wherein the 2D image is plain scan lung CT image data and the 3D image is lung enhanced CT image data, and the image segmentation algorithm comprises the following steps:

[0030] Step S31, perform 2D segmentation prediction: use the optimized 2D lesion detection model to detect the pixel points of the lesion, use the 2D segmentation result as a seed, and put the predicted lesion pixel points into the seed pixel set Z1;

[0031] Step S32: Perform 3D segmentation and optimize the network parameters using a similarity metric. The similarity metric is used to calculate the loss between the initial and final segmentation results. The 3D result is used as a neighboring element to check whether the eight neighboring pixels of the seed point should be included in the fusion result. The pixel predicted by the 3D segmentation is placed in the neighboring pixel set Z2.

[0032] Step S33, fusion segmentation: The segmentation result set is represented by Z3, and the 2D segmentation result is used as the initial value of the set Z3. Seed pixels are popped out from the seed pixel set Z1 in sequence to obtain the adjacent pixels a of each seed pixel. If the adjacent pixels are not inferior to the adjacent pixel set Z2, the adjacent pixels a are added to the segmentation result set Z3 to obtain the segmentation result set Z3, and the labeling of the lesion is completed.

[0033] In a preferred embodiment, the optimization of the 2D lesion detection model comprises the following steps:

[0034] Step S41, establish the direction field of the lesion: The lesion point is established by finding the two-dimensional direction vector of the nearest boundary point, with b as the internal center point and the direction field of the boundary point p as Use the direction field to complete the initial segmentation;

[0035] Step S42, optimized segmentation: optimize the initial segmentation by using feature correction and fusion prediction direction field, based on the original features and direction field learned by the network, and the enhanced features obtained after correction, F 0 is the original feature, F N is the N-step enhanced feature, px and py represent the x and y coordinates of the pixel point, and the feature of each point is obtained by bilinear interpolation. The enhanced feature of the lesion pixel satisfies the formula F k (p) = F k-1 (px+DF(p)·x,py+DF(p)·y), where N is a hyperparameter and 1≤k≤N;

[0036] Step S43: The direction field points to the center of the target. After N steps, the features of the boundary points are replaced by the features close to the center point, and an optimized segmentation result is obtained, that is, the precise location and size of the lesion.

[0037] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a device with automatic diagnosis function for medical imaging, comprising a user terminal, a server terminal, and a management terminal, wherein the user terminal is used to collect medical imaging data, convert it into image data, and transmit qualified medical imaging data to the server terminal, the user terminal includes a user login unit, and the user login unit obtains basic information of the user; the server terminal includes an automatic diagnosis model and a cloud database, the automatic diagnosis model collects medical imaging data, recognizes image features, obtains index parameters, compares the obtained index parameters with disease diagnosis standard data, and performs qualitative and quantitative analysis of the disease; the cloud database is used to store a disease diagnosis index database corresponding to lung diseases; the management terminal is used to manage the disease diagnosis index database and issue warnings to users based on the results of the automatic diagnosis model.

[0038] The device with automatic diagnosis function for medical imaging comprises the following steps:

[0039] Step 1: Data collection: After logging into the device client, the user uploads lung CT image data, converts qualified image data into image data, and transmits user basic data and image data to the automatic diagnosis model on the server;

[0040] Step 2: Preliminary diagnosis: The automatic diagnosis unit on the server automatically characterizes the disease and obtains the type of disease. The extracted features are compared with the disease diagnosis database to obtain the preset disease type.

[0041] Step 3: Verify the diagnosis results: verify the accuracy of the diagnosed disease and predict the progression of the disease;

[0042] Step 4: Quantitative analysis of lesions: Use image segmentation algorithms to obtain the location and size of lesions, and fusion algorithms to perform segmentation to obtain accurate positioning of lesions;

[0043] Step 5: Transmit the diagnosis results to users and managers and store them in the database.

[0044] principle:

[0045] The CT image of each case is used as the input for the neural network model. Abnormal nodules and shadowed areas in the image are framed and used as the output of the neural network model. The system automatically calculates the mapping relationship between the input and output using a computer. The system output is expected to serve as a reference and prompt for clinical examination and treatment. The output includes the probability of disease, abnormal areas, and confidence level.

[0046] The diagnosis and localization of lung CT include

[0047] Technical effects and advantages of the present invention:

[0048] The database module of the present invention uses the CT image of each examined case as the input of the neural network model, frames the abnormal nodules and shadow areas in the image as the output of the neural network model, obtains the corresponding function of the disease input and output, extracts the features of the image data through the preliminary diagnosis module, compares the features with the second database, and uses the similarity to predict the type of disease of the image data; verifies the accuracy of the prediction result through the diagnosis result verification module, first obtains the feature parameters and the area where the features are located of the image data, obtains the weight coefficient of the disease in each area from the database, multiplies the area weight coefficient and the weight parameter to obtain the numerical value of the prediction result, and if the preset value is met, it indicates that the image data includes the disease, and the qualitative diagnosis of the disease is completed; obtains the accurate position and size of the lesion through the image segmentation and quantification module, and completes the quantitative diagnosis of the disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a system structure diagram of the present invention.

[0050] Figure 2 Flowchart of the preliminary diagnosis module of the present invention.

[0051] Figure 3 This is a flow chart for verifying the diagnostic results of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0053] The terms "module" and "system" used in this application are intended to include computer-related entities such as, but not limited to, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a module can be, but is not limited to: a process running on a processor, a processor, an object, an executable program, a thread of execution, a program, and / or a computer. For example, both an application running on a computing device and the computing device can be modules. One or more modules can be located within a process and / or thread of execution, and a module can also be located on one computer and / or distributed between two or more computers.

[0054] Example 1

[0055] like Figure 1This embodiment provides an automatic diagnostic system for medical imaging, specifically comprising an image data acquisition module, a preliminary diagnosis module, a diagnosis result verification module, an image segmentation and quantification module, and a database module. The image data acquisition module is used to acquire lung image data, determine whether the format of the image data meets the requirements, and transmit the acquired lung image data to the preliminary diagnosis module.

[0056] The preliminary diagnosis module is used to predict diseases contained in lung images. After obtaining image data, the image features are extracted through preprocessing, convolution, and pooling operations. The extracted image features are compared with the first database for similarity. The type of disease is predicted based on the comparison results, and the predicted disease type is transmitted to the diagnosis result verification module.

[0057] The diagnosis result verification module is used to verify the accuracy of the diagnosis result, obtain the disease type in the image data, and predict the severity of the disease;

[0058] The image segmentation and quantification is used to perform quantitative analysis on the disease, segment the lesions of the diagnosis results, obtain quantitative data of the lesions, and obtain the location and size of the lesions;

[0059] The database module is used to store a historical database of diseases, including a first database and a second database. The first database is used to store CT image data and corresponding diagnosis results, and the second database is used to store diagnosis indicators of diseases.

[0060] Furthermore, the preliminary diagnosis module extracts image features and compares the obtained features with the second database to predict the type of disease, including the following steps:

[0061] Step S01, image preprocessing: highlighting the image data features through image denoising and image enhancement technology, and dividing the lung image data into n regions, denoted as 1, 2, ...;

[0062] Step S02, feature extraction: extract features that affect each area through machine vision, extract information features from the texture, color, shape and gradient of the image data, and perform convolution and pooling operations on the input image to output a feature set, which includes a feature subset;

[0063] Step S03, feature localization: The region recommendation network generates candidate region blocks. Each region block is pooled to a uniform size and input into the fully connected layer for classification. The bounding box is used to accurately predict the position of the target and obtain the position corresponding to the feature subset. The position of the feature subset is the location of the lesion.

[0064] Step S04, similarity comparison: Calculate the similarity between the feature set in each region and the second database, and obtain diseases with similarity exceeding a preset value, which are recorded as the predicted disease set.

[0065] Further, such as Figure 3 As shown, the diagnosis result verification module is used to verify whether the image contains disease A, including the following steps:

[0066] Step S11: Acquire characteristic parameters of each region of the image data to obtain the type of disease, divide the lung into n regions, and count the characteristic parameters of each disease in each region. The characteristic parameters of disease A in n regions are recorded as C1, where C1 = [c11, c12, ..., c1n];

[0067] Step S12: Count historical data to obtain the image data of disease A. The probability of lesions appearing in each area is regarded as the weight of disease A in the area to obtain the weight coefficient of each disease. The set of weight coefficients of disease A is recorded as Q1, Q1 = [q11, q12, ..., q1n]. The weight coefficient is calculated according to the area s1 and the number s2 of lesions in the area. The probability calculation formula satisfies Where k1 and k2 are the impact factor constants, is the average value of disease A in the region;

[0068] Step S13: multiply the obtained characteristic parameters by the weight coefficient corresponding to the disease. The preset disease discrimination standard is P, and the evaluation formula of disease A is F1(A)=C1×Q1≥P1, which indicates that the CT image diagnosis result has disease A, and the severity level of disease A satisfies the formula The P satisfies Where η1 is the influence coefficient constant.

[0069] Furthermore, the second database is used to express the feature set of disease A in each region, and obtaining the second database includes the following steps:

[0070] Step S21, sample data acquisition: obtain sample data from the first database, mark the lung CT image data as sample input X, and record the diagnosis result as sample output Y;

[0071] Step S22, classification of sample data: Classify samples according to disease type. Pulmonary diseases include infectious diseases, non-infectious diseases, and tumor diseases. Diseases are numbered sequentially to obtain diseases 1, 2, 3, ..., and a correspondence between lung diseases and image data is established. Using an image feature extraction algorithm, the image feature y1 corresponding to the disease x1 is extracted and denoted as F3(x1, y1). F3(x1, y1) represents a correspondence function between the disease and the image feature.

[0072] Step S23, feature extraction: A machine vision algorithm is used to extract features that affect the image, and a feature set B corresponding to disease A is obtained. Information features are mainly extracted from several aspects such as texture, color, shape, and gradient. The input image undergoes basic convolution and pooling operations to extract a feature map. Candidate region blocks are generated through a region recommendation network. Each region block is pooled to a uniform size and input into a fully connected layer for classification. The bounding box is used to accurately predict the location of the target.

[0073] Step S24: Based on medical knowledge, theoretical impact features of disease A are obtained. Based on theoretical imaging features and historical features, irrelevant features are eliminated. The filtered features are classified according to the regions they belong to, and recorded as regions 1, 2, ...;

[0074] Step S25: Analyze the weight of each image feature for disease A, obtain the weight coefficient Q of each regional feature for disease A, store the weight coefficient of each region of disease A in the first database, and so on to obtain the weight coefficients of all diseases in each region.

[0075] Furthermore, the first database is a historical image database of a hospital, in which lung CT images and corresponding diagnosis results are obtained. After privacy protection, lung CT images and corresponding diagnosis results of a large number of cases are obtained. The lung CT image data is marked as sample input X, and the diagnosis results are recorded as sample output Y.

[0076] Furthermore, the image segmentation algorithm is based on 2D images and 3D images, wherein the 2D image is plain scan lung CT image data and the 3D image is lung enhanced CT image data. The image segmentation algorithm includes the following steps:

[0077] Step S31, perform 2D segmentation prediction: use the optimized 2D lesion detection model to detect the pixel points of the lesion, use the 2D segmentation result as a seed, and put the predicted lesion pixel points into the seed pixel set Z1;

[0078] Step S32: Perform 3D segmentation and optimize the network parameters using a similarity metric. The similarity metric is used to calculate the loss between the initial and final segmentation results. The 3D result is used as a neighboring element to check whether the eight neighboring pixels of the seed point should be included in the fusion result. The pixel predicted by the 3D segmentation is placed in the neighboring pixel set Z2.

[0079] Step S33, fusion segmentation: The segmentation result set is represented by Z3, and the 2D segmentation result is used as the initial value of the set Z3. Seed pixels are popped out from the seed pixel set Z1 in sequence to obtain the adjacent pixels a of each seed pixel. If the adjacent pixels are not inferior to the adjacent pixel set Z2, the adjacent pixels a are added to the segmentation result set Z3 to obtain the segmentation result set Z3, and the labeling of the lesion is completed.

[0080] Furthermore, the optimization of the 2D lesion detection model includes the following steps:

[0081] Step S41, establish the direction field of the lesion: The lesion point is established by finding the two-dimensional direction vector of the nearest boundary point, with b as the internal center point and the direction field of the boundary point p as Use the direction field to complete the initial segmentation;

[0082] Step S42, optimized segmentation: optimize the initial segmentation by using feature correction and fusion prediction direction field, based on the original features and direction field learned by the network, and the enhanced features obtained after correction, F 0 is the original feature, F N is the N-step enhanced feature, px and py represent the x and y coordinates of the pixel point, and the feature of each point is obtained by bilinear interpolation. The enhanced feature of the lesion pixel satisfies the formula F k (p) = F k-1 (px+DF(p)·x,py+DF(p)·y), where N is a hyperparameter and 1≤k≤N;

[0083] Step S43: The direction field points to the center of the target. After N steps, the features of the boundary points are replaced by the features close to the center point, and an optimized segmentation result is obtained, that is, the precise location and size of the lesion.

[0084] This embodiment provides Figure 2 A device with automatic diagnosis function for medical imaging is shown, including a user terminal, a server terminal, and a management terminal. The user terminal is used to collect medical imaging data, convert it into image data, and transmit qualified medical imaging data to the server terminal. The user terminal includes a user login unit, which obtains basic user information; the server terminal includes an automatic diagnosis model and a cloud database. The automatic diagnosis model collects medical imaging data, recognizes image features, obtains indicator parameters, compares the obtained indicator parameters with disease diagnosis standard data, and performs qualitative and quantitative analysis of the disease; the cloud database is used to store a disease diagnosis indicator database corresponding to lung diseases; the management terminal is used to manage the disease diagnosis indicator database and issue warnings to users based on the results of the automatic diagnosis model.

[0085] The device with automatic diagnosis function for medical imaging comprises the following steps:

[0086] Step 1: Data collection: After logging into the device client, the user uploads lung CT image data, converts qualified image data into image data, and transmits user basic data and image data to the automatic diagnosis model on the server;

[0087] Step 2: Preliminary diagnosis: The automatic diagnosis unit on the server automatically characterizes the disease and obtains the type of disease. The extracted features are compared with the disease diagnosis database to obtain the preset disease type.

[0088] Step 3: Verify the diagnosis results: verify the accuracy of the diagnosed disease and predict the progression of the disease;

[0089] Step 4: Quantitative analysis of lesions: Use image segmentation algorithms to obtain the location and size of lesions, and fusion algorithms to perform segmentation to obtain accurate positioning of lesions;

[0090] Step 5: Transmit the diagnosis results to users and managers and store them in the database.

[0091] In summary: the present invention uses the database module to use the CT image of each examined case as the input of the neural network model, frames the abnormal nodules and shadow areas in the image as the output of the neural network model, obtains the corresponding function of the disease input and output, extracts the features of the image data through the preliminary diagnosis module, compares the features with the second database, and uses the similarity to predict the type of disease of the image data; verifies the accuracy of the prediction result through the diagnosis result verification module, first obtains the feature parameters and the area where the features are located of the image data, obtains the weight coefficient of the disease in each area from the database, multiplies the regional weight coefficient and the weight parameter to obtain the numerical value of the prediction result, if it meets the preset value, it indicates that the image data includes the disease, and completes the qualitative diagnosis of the disease; obtains the accurate position and size of the lesion through the image segmentation and quantification module, completes the quantitative diagnosis of the disease, and solves the problem raised by the present invention.

[0092] The embodiments of the present invention merely provide one implementation method, rather than specifically limiting the protection scope of the present invention.

[0093] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An automatic diagnostic system for medical imaging, characterized in that: It includes image data acquisition module, preliminary diagnosis module, diagnosis result verification module, image segmentation and quantification module, and database module. The image data acquisition module is used to acquire lung image data, determine whether the format of the image data meets the requirements, and transmit the acquired lung image data to the preliminary diagnosis module; The preliminary diagnosis module is used to predict diseases contained in lung images. After obtaining image data, the image features are extracted through preprocessing, convolution, and pooling operations. The extracted image features are compared with the first database for similarity. The type of disease is predicted based on the comparison results, and the predicted disease type is transmitted to the diagnosis result verification module. The diagnosis result verification module is used to verify the accuracy of the diagnosis result, obtain the disease type in the image data, and predict the severity of the disease; The diagnostic result verification module is used to verify whether the image contains disease A, including the following steps: Step S11: Acquire characteristic parameters of each region of the image data to obtain the type of disease, divide the lung into n regions, and count the characteristic parameters of each disease in each region. The characteristic parameters of disease A in n regions are recorded as C1, where C1 = [c11, c12, ..., c1n]; Step S12: Count historical data to obtain the image data of disease A. The probability of lesions appearing in each area is regarded as the weight of disease A in the area to obtain the weight coefficient of each disease. The set of weight coefficients of disease A is recorded as Q1, Q1 = [q11, q12, ..., q1n]. The weight coefficient is calculated according to the area s1 and the number s2 of lesions in the area. The probability calculation formula satisfies Where k1 and k2 are the impact factor constants, is the average value of disease A in the region; Step S13: multiply the obtained characteristic parameters by the weight coefficient corresponding to the disease. The preset disease discrimination standard is P, and the evaluation formula of disease A is F1(A)=C1×Q1≥P1, which indicates that the CT image diagnosis result has disease A, and the severity level of disease A satisfies the formula The P satisfies Where η1 is the influence coefficient constant; The image segmentation and quantification is used to perform quantitative analysis on the disease, segment the lesions of the diagnosis results, obtain quantitative data of the lesions, and obtain the location and size of the lesions; The database module is used to store a historical database of diseases, including a first database and a second database. The first database is used to store CT image data and corresponding diagnosis results, and the second database is used to store diagnosis indicators of diseases.

2. The automatic medical imaging diagnostic system according to claim 1, characterized in that: The preliminary diagnosis module extracts image features and compares the obtained features with the second database to predict the type of disease, including the following steps: Step S01, image preprocessing: highlighting the image data features through image denoising and image enhancement technology, and dividing the lung image data into n regions, denoted as 1, 2, ...; Step S02, feature extraction: extract features that affect each area through machine vision, extract information features from the texture, color, shape and gradient of the image data, and perform convolution and pooling operations on the input image to output a feature set, which includes a feature subset; Step S03, feature localization: The region recommendation network generates candidate region blocks. Each region block is pooled to a uniform size and input into the fully connected layer for classification. The bounding box is used to accurately predict the position of the target and obtain the position corresponding to the feature subset. The position of the feature subset is the location of the lesion. Step S04, similarity comparison: Calculate the similarity between the feature set in each region and the second database, and obtain diseases with similarity exceeding a preset value, which are recorded as the predicted disease set.

3. The automatic medical imaging diagnostic system according to claim 1, characterized in that: The second database is used to describe the feature set of disease A in each region, and obtaining the second database includes the following steps: Step S21, sample data acquisition: obtain sample data from the first database, mark the lung CT image data as sample input X, and record the diagnosis result as sample output Y; Step S22, classification of sample data: Classify samples according to disease type. Pulmonary diseases include infectious diseases, non-infectious diseases, and tumor diseases. Diseases are numbered sequentially to obtain diseases 1, 2, 3, ..., and a correspondence between lung diseases and image data is established. Using an image feature extraction algorithm, the image feature y1 corresponding to the disease x1 is extracted and denoted as F3(x1, y1). F3(x1, y1) represents a correspondence function between the disease and the image feature. Step S23, feature extraction: A machine vision algorithm is used to extract features that affect the image, and a feature set B corresponding to disease A is obtained. Information features are mainly extracted from several aspects such as texture, color, shape, and gradient. The input image undergoes basic convolution and pooling operations to extract a feature map. Candidate region blocks are generated through a region recommendation network. Each region block is pooled to a uniform size and input into a fully connected layer for classification. The bounding box is used to accurately predict the location of the target. Step S24: Based on medical knowledge, theoretical impact features of disease A are obtained. Based on theoretical imaging features and historical features, irrelevant features are eliminated. The filtered features are classified according to the regions they belong to, and recorded as regions 1, 2, ...; Step S25: Analyze the weight of each image feature for disease A, obtain the weight coefficient Q of each regional feature for disease A, store the weight coefficient of each region of disease A in the first database, and so on to obtain the weight coefficients of all diseases in each region.

4. The automatic medical imaging diagnostic system according to claim 1, characterized in that: The first database is a historical image database of the hospital, which contains lung CT images and corresponding diagnosis results. After privacy protection, a large number of lung CT images and corresponding diagnosis results of cases are obtained. The lung CT image data is marked as sample input X, and the diagnosis results are recorded as sample output Y.

5. The automatic medical imaging diagnostic system according to claim 1, characterized in that: The image segmentation algorithm is based on 2D images and 3D images, wherein the 2D image is plain scan lung CT image data and the 3D image is lung enhanced CT image data. The image segmentation algorithm includes the following steps: Step S31, perform 2D segmentation prediction: use the optimized 2D lesion detection model to detect the pixel points of the lesion, use the 2D segmentation result as a seed, and put the predicted lesion pixel points into the seed pixel set Z1; Step S32: Perform 3D segmentation and optimize the network parameters using a similarity metric. The similarity metric is used to calculate the loss between the initial and final segmentation results. The 3D result is used as a neighboring element to check whether the eight neighboring pixels of the seed point should be included in the fusion result. The pixel predicted by the 3D segmentation is placed in the neighboring pixel set Z2. Step S33, fusion segmentation: The segmentation result set is represented by Z3, and the 2D segmentation result is used as the initial value of the set Z3. Seed pixels are popped out from the seed pixel set Z1 in sequence to obtain the adjacent pixels a of each seed pixel. If the adjacent pixels are not inferior to the adjacent pixel set Z2, the adjacent pixels a are added to the segmentation result set Z3 to obtain the segmentation result set Z3, and the labeling of the lesion is completed.

6. The automatic medical imaging diagnostic system according to claim 5, characterized in that: The optimization of the 2D lesion detection model includes the following steps: Step S41, establishing the direction field of the lesion: the lesion point is established by finding the two-dimensional direction vector of the nearest boundary point, with b as the internal center point and the direction field of the boundary point p as , and the initial segmentation is completed using the direction field; Step S42, optimized segmentation: Optimize the initial segmentation using feature correction and fusion prediction direction field. Based on the original features and direction field learned by the network, enhanced features are obtained after correction. F0 is the original feature, FN is the N-step enhanced feature, px and py represent the x and y coordinates of the pixel point, and the feature of each point is obtained by bilinear interpolation. The enhanced feature of the lesion pixel satisfies the formula, where N is a hyperparameter and 1≤k≤N; Step S43: The direction field points to the center of the target. After N steps, the features of the boundary points are replaced by the features close to the center point, and an optimized segmentation result is obtained, that is, the precise location and size of the lesion.

7. A device with automatic diagnostic function for medical imaging, characterized in that: An automatic diagnostic system for medical imaging as described in any one of claims 1-6: comprising a user terminal, a server terminal, and a management terminal, wherein the user terminal is used to collect medical imaging data, convert it into image data, and transmit qualified medical imaging data to the server terminal, and the user terminal includes a user login unit, and the user login unit obtains basic information of the user; the server terminal includes an automatic diagnosis model and a cloud database, the automatic diagnosis model collects medical imaging data, recognizes image features, obtains indicator parameters, compares the obtained indicator parameters with disease diagnosis standard data, and performs qualitative and quantitative analysis of the disease; the cloud database is used to store a disease diagnosis indicator database corresponding to lung diseases; the management terminal is used to manage the disease diagnosis indicator database and issue warnings to users based on the results of the automatic diagnosis model.

8. The device with automatic medical imaging diagnostic function according to claim 7, characterized in that: The following steps are involved: Step 1: Data collection: After logging into the device client, the user uploads lung CT image data, converts qualified image data into image data, and transmits user basic data and image data to the automatic diagnosis model on the server; Step 2: Preliminary diagnosis: The automatic diagnosis unit on the server automatically characterizes the disease and obtains the type of disease. The extracted features are compared with the disease diagnosis database to obtain the preset disease type. Step 3: Verify the diagnosis results: verify the accuracy of the diagnosed disease and predict the progression of the disease; Step 4: Quantitative analysis of lesions: Use image segmentation algorithms to obtain the location and size of lesions, and fusion algorithms to perform segmentation to obtain accurate positioning of lesions; Step 5: Transmit the diagnosis results to users and managers and store them in the database.

Images