A medical image analysis system based on artificial intelligence

Through an artificial intelligence-based medical image analysis system, combined with texture and shape feature extraction, the problem of insufficient adaptability and intelligence in the existing technology is solved, and efficient and accurate image analysis and early lesion recognition are achieved.

CN118570204BActive Publication Date: 2024-11-19LONGGANG DISTRICT CENT HOSPITAL OF SHENZHEN
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411046396.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-19
Estimated Expiration
2044-08-01

AI Technical Summary

Technical Problem

The existing medical imaging analysis methods have problems such as poor adaptability, low intelligence, low efficiency, easy to cause misjudgment, high labor intensity, poor human-computer interaction assistance capabilities, and weak early lesion recognition capabilities.

Method used

A medical image analysis system based on artificial intelligence was designed, including data preprocessing module, feature extraction module, diagnostic support module and learning and optimization module. Through texture feature extraction and shape extraction, image evaluation and use artificial intelligence models to perform Model adjustment, providing diagnostic support and early warning prompts.

Benefits of technology

It improves the accuracy and efficiency of medical imaging analysis, enhances the system's adaptability and intelligence, reduces the labor intensity of medical staff, and improves the early lesion recognition ability and human-computer interaction ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118570204B_ABST
    Figure CN118570204B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of medical equipment technology. The present invention provides a medical image analysis system based on artificial intelligence, including a data preprocessing module, a feature extraction module, a diagnosis support module, and a learning and optimization module. The data preprocessing module preprocesses the medical image data to segment the lesion area, the feature extraction module extracts features from the lesion area to obtain the state information corresponding to the lesion area, the diagnosis support module evaluates the state information based on the trained artificial intelligence model to form an evaluation result, and warns the medical staff of the evaluation result, and the learning and optimization module triggers the adjustment of the artificial intelligence model according to the medical staff's score on the evaluation result. The present invention ensures that the entire system has the advantages of strong medical image self-adaptation ability, high intelligence, reliable recognition efficiency, low labor intensity of medical staff, strong human-computer interaction ability and strong early lesion recognition ability through the cooperation of the feature extraction module and the diagnosis support module.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of medical equipment, and in particular to a medical image analysis system based on artificial intelligence. Background Art

[0002] Medical imaging technology is an indispensable part of modern medical diagnosis and treatment. Common medical images include X-rays, computed tomography (CT), magnetic resonance imaging (MRI), and ultrasonic imaging. These imaging technologies are widely used in the diagnosis and treatment planning of various diseases in clinical practice. Existing medical image quantitative analysis methods are less targeted, lack the connection with the information systems of different medical service institutions, lack the function of comparing with patient historical information and similar patient diagnosis and treatment information, and have weak intelligence. At the same time, the workload is large, the time cost is high, and it cannot meet the requirements of medical technicians and clinicians in terms of method optimization, result evaluation and analysis, and historical information management, and it is not suitable for promotion.

[0003] For example, the medical image analysis method and device disclosed in Chinese patent CN109830284A improves recognition efficiency by using an image recognition model, but the recognition accuracy and the self-adjustment ability of the model are poor, resulting in low recognition or auxiliary diagnosis capabilities.

[0004] However, traditional medical image analysis methods mainly rely on the experience and expertise of radiologists to manually observe and interpret images. This method has the following major problems:

[0005] Heavy workload and low efficiency: Hospitals generate a large amount of medical imaging data every day. Radiologists need to spend a lot of time viewing and analyzing these images, resulting in heavy workload and low efficiency.

[0006] Highly subjective and prone to error: Different doctors may interpret the same image differently, which is greatly affected by personal experience and fatigue. This subjective interpretation may lead to misdiagnosis or missed diagnosis, affecting the patient's treatment effect.

[0007] Limitations of diagnostic accuracy: Although experienced radiologists can provide high-quality diagnoses, manual analysis may not achieve sufficient accuracy and reliability in some complex cases, especially in the detection and identification of early lesions.

[0008] The present invention is made to solve the common problems in this field, such as poor adaptability, low intelligence, low efficiency, easy misjudgment, high labor intensity, poor human-computer interaction assistance capability and weak early lesion recognition capability. Summary of the invention

[0009] The purpose of the present invention is to propose a medical image analysis system based on artificial intelligence to address the current deficiencies.

[0010] In order to overcome the shortcomings of the prior art, the present invention adopts the following technical solutions:

[0011] A medical image analysis system based on artificial intelligence, the medical image analysis system comprising a server, the medical image analysis system further comprising a data preprocessing module, a feature extraction module, a diagnosis support module, and a learning and optimization module, the server being connected to the data preprocessing module, the feature extraction module, the diagnosis support module, and the learning and optimization module respectively;

[0012] The data preprocessing module preprocesses the medical image data to segment the lesion area, the feature extraction module extracts features from the lesion area to obtain status information corresponding to the lesion area, the diagnosis support module evaluates the status information based on the trained artificial intelligence model to form an evaluation result, and warns the medical staff of the evaluation result, and the learning and optimization module triggers the adjustment of the artificial intelligence model according to the medical staff's score on the evaluation result;

[0013] The feature extraction module includes a texture extraction unit and a shape extraction unit, wherein the texture extraction unit extracts texture features in the lesion area to obtain tissue structure state information of the lesion area, and the shape extraction unit extracts geometric shape features in the lesion area to obtain morphological state information of the lesion area;

[0014] The state information includes tissue structure state information and morphology state information.

[0015] Optionally, the texture feature extraction unit includes a grayscale co-occurrence matrix calculator, a local binary pattern extractor, a wavelet transform analyzer, a Gabor filter extractor, and a vector combiner. The grayscale co-occurrence matrix calculator is used to calculate the co-occurrence relationship of different grayscale levels in the image to analyze the distribution of pixel pairs at specific distances and angles and extract a first texture feature of the image. The local binary pattern extractor generates a local binary pattern by comparing each pixel in the image with its neighboring pixels, thereby capturing a second texture feature of local texture information. The wavelet transform analyzer extracts third texture features of different scales and directions by performing wavelet decomposition on the image. The Gabor filter extractor filters the image using a set of Gabor filters with different directions and frequencies to extract a fourth texture feature of a specific direction and scale.

[0016] Among them, the vector combiner combines the first texture feature extracted by the gray level co-occurrence matrix calculator, the second texture feature extracted by the local binary pattern extractor, the third texture feature extracted by the wavelet transform analyzer, and the fourth texture feature extracted by the Gabor filter to form a feature vector FeatureVector1 of the tissue structure state information.

[0017] Optionally, the shape extraction unit includes an edge detector, a contour extractor and a geometric feature calculator, the edge detector is used to detect the boundary of the lesion area, the contour extractor is used to extract the contour of the lesion area, and the geometric feature calculator is used to calculate the geometric shape features of the lesion area;

[0018] The geometric shape features extracted by the geometric feature calculator are combined to form a feature vector FeatureVector2 of the morphological state information.

[0019] Optionally, the geometric feature calculator calculates the centroid according to the following formula:

[0020] ;

[0021] Among them, (C x , C y ) is the coordinate of the center of mass, (x i ,y i ) is the coordinate of the i-th point on the contour, (x i+1 ,y i+1 ) is the coordinate of the i-th point on the contour, n is the total number of contour points, where (x n ,y n ) = (x0,y0) means that the polygon outlines are connected at the ends.

[0022] Optionally, the diagnosis support module includes a diagnosis evaluation unit and a decision support unit, wherein the diagnosis evaluation unit evaluates the status information based on the artificial intelligence model to form an evaluation result, and the decision support unit issues an early warning prompt of the evaluation result to the medical staff;

[0023] The diagnostic evaluation unit integrates feature vectors from different sources into a comprehensive feature vector, standardizes the comprehensive feature vector, inputs the standardized comprehensive feature vector into an artificial intelligence model for evaluation, and outputs an evaluation result.

[0024] Optionally, the decision support unit generates a diagnosis report and early warning prompts based on the evaluation results output by the artificial intelligence model, and provides early warning prompts to the medical staff.

[0025] Optionally, the learning and optimization module includes a scoring collection unit and a learning optimization unit, the scoring collection unit collects the scores of the medical staff on the evaluation results, and the learning optimization unit triggers the adjustment of the artificial intelligence model according to the scoring data of the scoring collection unit;

[0026] The scoring collection unit includes a scoring interface, a scoring database and a scoring submitter. The scoring interface is used for the medical staff to score. The scoring database stores the scoring data of the medical staff. The scoring submitter is used to process the scoring data and save the scoring data in the scoring database.

[0027] The scoring data includes the scoring value, scoring time, medical staff ID, and case ID.

[0028] Optionally, the learning optimization unit includes a scoring analyzer, a model adjuster and a feedback device. The scoring analyzer is used to analyze the scoring data. The model adjuster calculates a trigger index ACT based on the scoring analysis results, and compares the trigger index ACT with a set trigger threshold Trigger. If the trigger index ACT exceeds the set trigger threshold Trigger, the artificial intelligence model is triggered to be retrained. After triggering the artificial intelligence model adjustment, the feedback device uses the feature vector corresponding to the medical image corresponding to the new scoring data and the original training data set to retrain the model.

[0029] Optionally, the retraining process includes: data preparation, model training and model validation.

[0030] The beneficial effects achieved by the present invention are:

[0031] 1. Through the cooperation between the texture feature extraction unit and the shape extraction unit, the features of the medical image can be extracted, ensuring that the entire system has the advantages of strong lesion recognition ability, good adaptability, high intelligence and high efficiency;

[0032] 2. The data preprocessing module preprocesses medical images, which improves the accuracy and reliability of texture feature extraction by the feature extraction module, ensuring that the entire system has the advantages of fast response speed and high feature extraction efficiency;

[0033] 3. Through the cooperation between the evaluation unit and the decision support unit, medical staff can obtain auxiliary evaluation results and cooperate with medical staff to analyze images, ensuring that the entire system has the advantages of strong medical image adaptability, high intelligence, reliable recognition efficiency, low labor intensity for medical staff, strong human-computer interaction and strong early lesion recognition capabilities;

[0034] 4. Through the mutual cooperation between the scoring acquisition unit and the learning optimization unit, the artificial intelligence model's ability to analyze medical images becomes more accurate, ensuring that the entire system has the advantages of strong adaptability, good judgment accuracy, reliable auxiliary judgment efficiency, strong human-computer interaction capability and high intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention can be further understood from the following description in conjunction with the accompanying drawings. The components in the figures are not necessarily drawn to scale, but the emphasis is placed on illustrating the principles of the embodiments. In different views, the same reference numerals designate the same parts.

[0036] Figure 1 It is an overall block diagram of the present invention.

[0037] Figure 2 It is a block diagram of the feature extraction module, diagnosis support module and medical staff / patient of the present invention.

[0038] Figure 3 It is a block diagram of the feature extraction module, diagnosis support module, learning optimization module and medical staff of the present invention.

[0039] Figure 4 It is a schematic diagram of the evaluation process of the score collection unit and the learning optimization unit of the present invention.

[0040] Figure 5 An exemplary schematic diagram of a medical image analysis report of the present invention. DETAILED DESCRIPTION

[0041] The following is an explanation of the embodiments of the present invention through specific embodiments. Those skilled in the art can understand the advantages and effects of the present invention from the contents disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and the details in this specification can also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. In addition, the drawings of the present invention are only simple schematic illustrations and are not depicted according to actual sizes. It is stated in advance. The following embodiments will further explain the relevant technical contents of the present invention in detail, but the disclosed contents are not intended to limit the scope of protection of the present invention.

[0042] Embodiment 1: According to Figure 1 , Figure 2 , Figure 3 , Figure 4 and Figure 5As shown, this embodiment provides a medical image analysis system based on artificial intelligence, the medical image analysis system includes a server, the medical image analysis system also includes a data preprocessing module, a feature extraction module, a diagnosis support module, and a learning and optimization module, the server is connected to the data preprocessing module, the feature extraction module, the diagnosis support module, and the learning and optimization module respectively, and the intermediate data and process data of the data preprocessing module, the feature extraction module, the diagnosis support module, and the learning and optimization module are stored in the database of the server for query;

[0043] The data preprocessing module preprocesses the medical image data to segment the lesion area, the feature extraction module extracts features from the lesion area to obtain status information corresponding to the lesion area, the diagnosis support module evaluates the status information based on the trained artificial intelligence model to form an evaluation result, and warns the medical staff of the evaluation result, and the learning and optimization module triggers the adjustment of the artificial intelligence model according to the medical staff's score on the evaluation result;

[0044] The medical image analysis system includes a central processing unit, which is respectively connected to a data preprocessing module, a feature extraction module, a diagnosis support module, and a learning and optimization module, and centrally controls the data preprocessing module, the feature extraction module, the diagnosis support module, and the learning and optimization module based on the central processing unit, and stores the control data of the central processing unit in a database of the server, so as to improve the analysis accuracy and analysis efficiency of the entire system for the medical image;

[0045] The data preprocessing module includes an image enhancement unit, an image normalization unit and an image segmentation unit. The image enhancement unit is used to improve image quality, remove noise and enhance contrast. The image normalization unit normalizes the image data processed by the image enhancement unit for subsequent processing. The image segmentation unit is used to process the normalized image data to segment out the lesion area.

[0046] The image enhancement unit includes a multi-format receiver, a denoising processor, a contrast enhancer and a sharpening processor. The multi-format receiver is used to receive a variety of medical image formats (such as DICOM, JPEG, PNG, etc.), and automatically identify and convert them into a unified format that can be analyzed by the system to reduce manual intervention. The denoising processor is used to remove noise from the medical image to improve the clarity of the image. The contrast enhancer is used to enhance the contrast of the image to make the lesion area more prominent. The sharpening processor is used to enhance the details of the image to make the edges clearer.

[0047] The denoising processor processes the medical image using a Gaussian filter, the contrast enhancer processes the denoised medical image to obtain medical image data with higher definition, and the sharpening processor detects edges in the image using a Laplace operator, and adds the Laplace operator result to the image corresponding to the original medical image to enhance the edge;

[0048] The image normalization unit normalizes the enhanced image data according to the following formula, specifically:

[0049] S11. Calculate the minimum and maximum values ​​of the image:

[0050] Traverse all pixels of the image and find the minimum value I min and the maximum value I max ;

[0051] S12. Apply the normalization formula:

[0052] For each pixel value I(x,y) in the image, apply the normalization formula:

[0053] ;

[0054] Where I(x,y) represents the pixel value of the image at position (x,y);

[0055] The image segmentation unit segments the lesion area in the medical image according to the following steps, specifically:

[0056] S21, threshold segmentation module:

[0057] The normalized image data (referring to the image corresponding to the normalized medical image)

[0058] Among them, a global threshold T is selected, and the area with pixel values ​​higher than the threshold is marked as the foreground (lesion area), and the area with pixel values ​​lower than the threshold is marked as the background;

[0059] S22, edge detection module: segmenting the area by detecting the edge of the image;

[0060] Perform edge detection on the segmented foreground (lesion area) by using Canny edge detection, wherein Canny edge detection is a technical means well known to those skilled in the art, and those skilled in the art can refer to relevant technical manuals to learn about the technology, so it will not be described in detail in this embodiment;

[0061] The data preprocessing module preprocesses the medical images, thereby improving the accuracy and reliability of the texture feature extraction by the feature extraction module, thereby ensuring that the entire system has the advantages of fast response speed and high feature extraction efficiency;

[0062] The feature extraction module includes a texture extraction unit and a shape extraction unit, wherein the texture extraction unit extracts texture features in the lesion area to obtain tissue structure state information of the lesion area, and the shape extraction unit extracts geometric shape features in the lesion area to obtain morphological state information of the lesion area;

[0063] Wherein, the state information includes tissue structure state information and morphology state information;

[0064] Optionally, the texture feature extraction unit includes a grayscale co-occurrence matrix calculator, a local binary pattern extractor, a wavelet transform analyzer, a Gabor filter extractor, and a vector combiner. The grayscale co-occurrence matrix calculator is used to calculate the co-occurrence relationship of different grayscale levels in the image to analyze the distribution of pixel pairs at specific distances and angles and extract a first texture feature of the image. The local binary pattern extractor generates a local binary pattern by comparing each pixel in the image with its neighboring pixels, thereby capturing a second texture feature of local texture information. The wavelet transform analyzer extracts third texture features of different scales and directions by performing wavelet decomposition on the image. The Gabor filter extractor filters the image using a set of Gabor filters with different directions and frequencies to extract a fourth texture feature of a specific direction and scale.

[0065] In this embodiment, the gray level co-occurrence matrix calculator calculates the gray level co-occurrence matrix of the lesion area, and selects different distances and directions (such as 0 degrees, 45 degrees, 90 degrees and 135 degrees), and the specific values ​​are selected by the system according to the actual situation;

[0066] Extract statistical features from the gray-level co-occurrence matrix, such as energy, entropy, contrast, and correlation;

[0067] The gray level co-occurrence matrix calculator calculates the first texture feature according to the following steps:

[0068] S31. Define the gray-level co-occurrence matrix:

[0069] Let P(i,j,d,θ) represent the number of co-occurrences of the first pixel with grayscale value i and the second pixel with grayscale value j in the pixel pair with distance d and direction θ; define the grayscale co-occurrence matrix G(i,j) as:

[0070] ;

[0071] Where i, j are grayscale values, d is the distance between pixel pairs, θ is the direction between pixel pairs, usually 0, 45, 90, and 135 degrees, and (x, y) represents the pixel position in the image;

[0072] Meanwhile, the distance d refers to the interval between two pixels when calculating the gray-level co-occurrence matrix; usually, the selected distance d should be determined according to the specific application scenario and the resolution of the image; the following are some common methods and considerations: 1) Application scenario: For images with higher resolution, a larger distance d can capture a wider range of texture information; for images with lower resolution, a smaller distance d is more suitable for capturing local details;

[0073] 2) Experience and experimentation: Common choices include d=1, 2, 3, 4, 5, etc. The distance that best suits a specific application can be determined through experimentation;

[0074] 3) Multi-scale analysis: In order to obtain the multi-scale texture features of the image, the gray-level co-occurrence matrix can be calculated using multiple distances d respectively, and then these features are combined for analysis;

[0075] The gray level co-occurrence matrix is ​​a conventional method in image processing technology. Therefore, in this embodiment, those skilled in the art can refer to relevant technical manuals to learn about the technology. Therefore, it will not be described in detail in this embodiment.

[0076] S32. Calculate entropy, contrast and correlation according to the gray-level co-occurrence matrix:

[0077] ;

[0078] Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image.

[0079] ;

[0080] Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image.

[0081] ;

[0082] Where i, j are grayscale values, N is the number of grayscale levels, G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image, μ x and μ yare the means of i and j respectively, σ x and σ y are the standard deviations of i and j respectively;

[0083] For the mean μ of gray level i x , calculated according to the following formula:

[0084] ;

[0085] Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image.

[0086] For the mean μ of gray level j y , calculated according to the following formula:

[0087] ;

[0088] Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image.

[0089] For the standard deviation σ of gray level i x , calculated according to the following formula:

[0090] ;

[0091] Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image.

[0092] For the standard deviation σ of gray level j y , calculated according to the following formula:

[0093] ;

[0094] Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image.

[0095] The local binary pattern extractor compares the grayscale value of each pixel in the lesion area with that of the neighboring pixels to generate a binary pattern;

[0096] Calculating a histogram of the local binary pattern as a second texture feature vector, wherein the feature vector is used to capture the local texture pattern in the image and is suitable for describing a tissue structure with rich details;

[0097] Specifically: the local binary pattern extractor calculates the histogram of the binary pattern according to the following steps:

[0098] S51, traverse the image:

[0099] For each pixel p, determine its R and P. In this embodiment, the position of the central pixel p is in practical applications, and each pixel in the image can be used as the central pixel p (in this embodiment, the LBP algorithm traverses each pixel of the image and calculates its corresponding LBP value);

[0100] The selection basis for the radius R of the neighborhood: The selected R value should be based on the resolution and detail characteristics of the image; for example, for high-resolution images, a larger R can capture more local texture information; for low-resolution images, a smaller R can capture details more effectively; Common choices: R = 1, 2, 3, etc.;

[0101] For the number of neighborhood pixels P, the selection criteria are: the value of P should balance computational complexity and feature representation capability; larger P values ​​can provide more details, but the computational complexity increases; common choices: P = 8, 16, etc.; generally used together with R, such as (8, 1) or (16, 2);

[0102] Get the gray value g of the center pixel p , in this embodiment, the gray value of position p in the image is directly used;

[0103] S52, calculate the gray value g of the neighborhood pixel n :

[0104] For each n (n=0,1,…,P−1), calculate the gray value g of the nth neighboring pixel n ;

[0105] Among them, the position of the neighborhood pixel is calculated according to R and P; the coordinates of the neighborhood pixel can be determined by the following formula:

[0106] ;

[0107] If the calculated position is not an integer, use an interpolation method (such as bilinear interpolation) to calculate the grayscale value;

[0108] S53, compare the gray value g of the neighboring pixels n and the gray value g of the center pixel p :

[0109] ;

[0110] S54, generate LBP binary pattern:

[0111] Convert the binary pattern to a decimal LBP value: ;

[0112] S55, calculating LBP histogram: counting the frequencies of all LBP values ​​in the image, and generating LBP histogram (in this embodiment, LBP histogram is used as an alternative for illustration);

[0113] The wavelet transform analyzer performs multi-scale wavelet decomposition on the lesion area to extract coefficients at different scales and directions;

[0114] Calculate the statistical characteristics of wavelet coefficients, such as mean and variance;

[0115] Among them, wavelet transform can decompose high-frequency and low-frequency information in images and effectively capture the third texture features at different scales;

[0116] The method for extracting texture features of the lesion area by the wavelet transform analyzer comprises the following steps:

[0117] S61. Select wavelet function and decomposition series:

[0118] Choose an appropriate wavelet function (e.g., Daubechies wavelet) and decomposition series;

[0119] S62, perform wavelet decomposition: perform wavelet decomposition on the lesion area to obtain subband coefficient matrices W at different scales and directions s,d (x,y), where W s,d (x,y) is the wavelet coefficient of the image at position (x,y); here s and d represent the scale and direction of wavelet decomposition respectively;

[0120] Among them, s is the scale factor, which indicates the level of wavelet decomposition, and d is the direction factor, which indicates the direction of wavelet decomposition (for example, horizontal, vertical and diagonal directions). At the same time, the scale factor s and the direction factor d are set by the system or the operator according to the actual situation and input from the human interaction interface;

[0121] S63, extracting statistical features:

[0122] ;

[0123] Where Mean1 is the mean of the wavelet decomposition, Variance1 is the variance of the wavelet decomposition, N is the total number of pixels in the image, usually the width of the image multiplied by the height, (x, y) is the pixel coordinate in the image traversing the entire image range from 0 to N-1 (in this embodiment, it is assumed that the image is square and the size is N×N);

[0124] The Gabor filter extractor uses a set of Gabor filters of different frequencies and directions to convolve the lesion area; extracts the response features after filtering, such as amplitude and phase; wherein the Gabor filter can simulate the texture perception mechanism in the human visual system and is suitable for extracting the fourth texture feature specific to direction and scale, and thus is used in the analysis process of the medical image;

[0125] The method for extracting texture features of the lesion area using a Gabor filter extractor includes the following steps:

[0126] S71. Use a set of Gabor filters of different directions and frequencies to convolve the image and obtain the filter response R(x,y;λ,θ,ψ,σ,γ):

[0127] ;

[0128] Among them, G(x,y;λ,θ,ψ,σ,γ) is the Gabor filter convolution kernel function:

[0129] ;

[0130] in: , ;

[0131] Where x ′, y ′ is the position of the filter in the x and y directions, λ is the wavelength, which is used to control the period of the sine function, θ is the direction, which is used as the directional parameter of the Gabor filter, ψ is the phase offset, which is used to control the phase of the sine function, σ is the standard deviation of the Gaussian envelope, which is used to control the width of the Gaussian envelope, and γ is the spatial aspect ratio, which is used to control the elliptical shape of the Gaussian envelope;

[0132] In this embodiment, the parameters of the Gabor filter (wavelength λ, direction θ, phase shift ψ, standard deviation σ of the Gaussian envelope, spatial aspect ratio γ) are very important in image processing and texture analysis; the selection and determination of these parameters are usually based on application requirements and experience, can be set by the operator, and input from the human-computer interaction interface, which is well known to those skilled in the art, and therefore, in this embodiment, they are not described one by one;

[0133] The above parameters are set by the operator according to specific needs and experience. Different parameter combinations can extract texture features in different directions and scales. The operator can select the most suitable parameters through experiments. In addition, in this embodiment, examples commonly used by technicians in this field are provided, specifically:

[0134] Wavelength (λ): controls the period of the sine function; the longer the wavelength, the larger the corresponding filter response, which is suitable for extracting low-frequency features; common choices: range from a few pixels to tens of pixels, examples: λ=4, 8, 16;

[0135] Direction (θ): controls the direction of the filter; commonly used directions are 0 degrees, 45 degrees, 90 degrees and 135 degrees, and common choices are θ=0, π / 4, π / 2, 3π / 4 (corresponding to 0 degrees, 45 degrees, 90 degrees, 135 degrees);

[0136] Phase shift (ψ): controls the phase of the sine function; usually set to 0 or π / 2, corresponding to cosine and sine filters, respectively. Common choices: ψ=0, π / 2;

[0137] Standard deviation of the Gaussian envelope (σ): controls the width of the Gaussian envelope; a larger σ can capture more texture details. Common choices: σ usually ranges from 1 / 2 to 1 / 8 of λ, example: σ=λ / 2;

[0138] Spatial aspect ratio (γ): controls the elliptical shape of the Gaussian envelope; usually set between 0.5 and 1, common choices: γ=0.5, 1;

[0139] S72. Calculate the response characteristics (including the amplitude of the filter response and the phase of the filter response):

[0140] Amplitude of the filter response: ;

[0141] In the formula, R real (x,y) and R imag (x,y) represents the real and imaginary parts of the Gabor filter response, respectively. x and y are the pixel coordinates of the image, ranging from 0 to N−1 across the entire image.

[0142] Phase of the filter response: ;

[0143] In the formula, R real (x,y) and R imag (x,y) represent the real and imaginary parts of the Gabor filter response, respectively;

[0144] S73, calculating the mean and variance of the amplitude and phase as the fourth texture feature, specifically:

[0145] 1) Amplitude correlation: ;

[0146] Where Mean2 is the amplitude mean extracted by the Gabor filter, Variance2 is the amplitude variance extracted by the Gabor filter, N is the total number of pixels in the image, and M(x,y) is the amplitude of the filter response at position (x,y);

[0147] 2) Phase correlation: ;

[0148] Where Mean3 is the phase mean extracted by the Gabor filter, Variance3 is the phase variance extracted by the Gabor filter, N is the total number of pixels in the image, and P(x,y) is the phase of the filter response at position (x,y);

[0149] The vector combiner combines the first texture feature extracted by the gray-level co-occurrence matrix calculator, the second texture feature extracted by the local binary pattern extractor, the third texture feature extracted by the wavelet transform analyzer, and the fourth texture feature extracted by the Gabor filter to form a feature vector FeatureVector1=[Entropy, Contrast, Correlation, LBP Histogram, Mean1, Variance1, Mean2, Variance2, Mean3, Variance3] of the tissue structure state information;

[0150] Optionally, the shape extraction unit includes an edge detector, a contour extractor and a geometric feature calculator, the edge detector is used to detect the boundary of the lesion area, the contour extractor is used to extract the contour of the lesion area, and the geometric feature calculator is used to calculate the geometric shape features of the lesion area;

[0151] In this embodiment, the geometric shape features include: area, perimeter, centroid, circularity, aspect ratio and convex hull;

[0152] The area is calculated according to the following formula:

[0153] ;

[0154] In the formula, (x i ,y i ) is the coordinate of the i-th point on the contour, (x i+1 ,y i+1 ) is the coordinate of the i-th point on the contour, n is the total number of contour points, where (x n ,y n ) = (x0, y0) means that the polygon outlines are connected at the ends;

[0155] The perimeter is calculated according to the following formula:

[0156] ;

[0157] Among them, (x n ,y n ) = (x0, y0) means that the polygon outlines are connected at the ends;

[0158] Optionally, the geometric feature calculator calculates the centroid according to the following formula:

[0159] ;

[0160] In the formula, (C x , C y ) is the coordinate of the center of mass, (x i ,y i ) is the coordinate of the i-th point on the contour, (x i+1 ,y i+1 ) is the coordinate of the i-th point on the contour, n is the total number of contour points, where (x n ,y n ) = (x0, y0) means that the polygon outlines are connected at the ends;

[0161] The circularity is calculated according to the following formula:

[0162] ;

[0163] Where Area is the area of ​​the contour and Perimeter is the perimeter of the contour;

[0164] The aspect ratio is calculated according to the following formula:

[0165] ;

[0166] Where W is the width of the contour, that is, the width of the minimum bounding rectangle, and H is the height of the contour, that is, the height of the minimum bounding rectangle;

[0167] The convex hull area is calculated according to the following formula. For a polygonal convex hull consisting of m points, the area formula of the convex hull is the same as the calculation formula of the contour area, that is:

[0168] ;

[0169] Where m is the total number of convex hull points, (x i ,y i ) is the coordinate of the i-th point on the convex hull, (x m ,y m) = (x0, y0) means that the polygon outlines are connected at the ends;

[0170] The convexity is calculated according to the following formula:

[0171] ;

[0172] Where, Area is the contour area, Convex Hull Area is the convex hull area;

[0173] The geometric shape features extracted by the geometric feature calculator are combined to form a feature vector FeatureVector2=[Area, Perimeter, C x , C y ,Aspect Ratio,Convex HullArea,Convexity];

[0174] In this embodiment, the features of the medical image can be extracted through the cooperation between the texture feature extraction unit and the shape extraction unit, ensuring that the entire system has the advantages of strong lesion recognition ability, good adaptability, high intelligence and high efficiency;

[0175] Optionally, the diagnosis support module includes a diagnosis evaluation unit and a decision support unit, wherein the diagnosis evaluation unit evaluates the status information based on the artificial intelligence model to form an evaluation result, and the decision support unit issues an early warning prompt of the evaluation result to the medical staff;

[0176] The diagnostic evaluation unit integrates feature vectors from different sources into a comprehensive feature vector, standardizes the comprehensive feature vector, inputs the standardized comprehensive feature vector into an artificial intelligence model for evaluation, and outputs an evaluation result;

[0177] The diagnostic evaluation unit normalizes the comprehensive feature vector pair according to the following formula:

[0178] ;

[0179] In the formula, NormalizedFeatureVector is the standardized or normalized comprehensive feature vector, CompositeFeatureVector is the comprehensive feature vector, σ V is the standard deviation of the comprehensive eigenvector, μ V is the mean of the comprehensive eigenvector, satisfying:

[0180] ;

[0181] Where N is the dimension of the composite feature vector, CompositeFeatureVector i is the i-th eigenvalue of the comprehensive eigenvector;

[0182] The standard deviation of the comprehensive feature vector is calculated according to the following formula:

[0183] ;

[0184] Where N is the dimension of the composite feature vector, CompositeFeatureVector i is the i-th eigenvalue of the comprehensive eigenvector, μ V is the mean of the comprehensive eigenvector;

[0185] The diagnostic evaluation unit inputs the standardized or normalized comprehensive feature vector into the artificial intelligence model, and outputs an evaluation result Output through the artificial intelligence model;

[0186] Among them, the artificial intelligence model includes but is not limited to the following commonly used models: deep learning model, support vector machine and other models; at the same time, the artificial intelligence model is a technical means well known to technicians in this field, and technicians in this field can refer to relevant technical manuals to learn about the technology, so it is not repeated one by one in this embodiment; Optionally, the support decision unit generates a diagnosis report and early warning prompts according to the evaluation results output by the artificial intelligence model, and provides early warning prompts to the medical staff;

[0187] The decision support unit obtains the evaluation unit Output and the monitoring threshold Warning set by the system. If the evaluation result Output is greater than or equal to the monitoring threshold Warning set by the system, it indicates that there is an abnormality in the lesion area corresponding to the medical image;

[0188] If the evaluation result Output is lower than the monitoring threshold Warning set by the system, it means that the lesion area corresponding to the medical image is in a non-dangerous state;

[0189] The monitoring threshold Warning set by the system is set by the system or the administrator / operator according to the actual medical image category. This is a technical means well known to those skilled in the art. Those skilled in the art can refer to relevant technical manuals to learn about the technology, so it will not be described in detail in this embodiment.

[0190] For example: if the type of the output evaluation result is: "probability";

[0191] 1) For lung medical images used for lung nodule detection: if the output probability of a malignant nodule medical image exceeds 0.7, it is considered to be at high risk;

[0192] 2) Chronic obstructive pulmonary disease (COPD) detection: If the output probability of COPD medical imaging exceeds 0.6, it is considered to be high risk;

[0193] 3) Breast medical imaging is used to detect breast cancer: if the output probability of a malignant tumor medical imaging exceeds 0.8, it is considered to be a high risk;

[0194] 4) Cardiac medical images are used to detect heart disease: if the output probability of a heart disease image exceeds 0.75, it is considered to be at high risk;

[0195] Of course, in this embodiment, the type of the output result of the evaluation result is determined according to the artificial intelligence model used, that is, the output evaluation result is not limited to: "probability", which is also well known to those skilled in the art, and thus will not be described in detail in this embodiment;

[0196] After the decision support unit determines the danger category, it sends a warning label corresponding to the danger category to the medical staff for warning;

[0197] In this embodiment, different danger categories are provided with corresponding warning labels;

[0198] Through the mutual cooperation between the evaluation unit and the decision support unit, the medical staff can obtain auxiliary evaluation results and cooperate with the medical staff to analyze the images, ensuring that the entire system has the advantages of strong medical image adaptability, high intelligence, reliable recognition efficiency, low labor intensity for medical staff, strong human-computer interaction ability and strong early lesion recognition ability;

[0199] Optionally, the learning and optimization module includes a scoring collection unit and a learning optimization unit, the scoring collection unit collects the scores of the medical staff on the evaluation results, and the learning optimization unit triggers the adjustment of the artificial intelligence model according to the scoring data of the scoring collection unit;

[0200] The scoring collection unit includes a scoring interface, a scoring database and a scoring submitter. The scoring interface is used for the medical staff to score. The scoring database stores the scoring data of the medical staff. The scoring submitter is used to process the scoring data and save the scoring data in the scoring database.

[0201] The scoring data includes the scoring value, scoring time, medical staff ID, and case ID;

[0202] The scoring interface is provided on the service terminal, and is used by the medical staff to provide feedback on the scoring data of the evaluation result. The service terminal may be provided on a web page or in a program of a mobile terminal. This is a technical means well known to those skilled in the art, and therefore will not be described in detail in this embodiment.

[0203] Optionally, the learning optimization unit includes a scoring analyzer, a model adjuster and a feedback device, the scoring analyzer is used to analyze the scoring data, the model adjuster calculates a trigger index ACT based on the scoring analysis result, and compares the trigger index ACT with a set trigger threshold Trigger, if the trigger index ACT exceeds the set trigger threshold Trigger, then the artificial intelligence model is triggered to be retrained, and after the artificial intelligence model adjustment is triggered, the feedback device uses the feature vector corresponding to the medical image corresponding to the new scoring data and the original training data set to retrain the model;

[0204] Optionally, the retraining process includes: data preparation, model training and model verification;

[0205] The score analyzer obtains the score data and calculates the current average score mean μ according to the following formula: score :

[0206] ;

[0207] Where M is the total number of ratings, r i Score the i-th item;

[0208] The score analyzer calculates the standard deviation σ of the current score according to the following formula score :

[0209] ;

[0210] Where M is the total number of ratings, r i is the i-th score, μ score is the current average score;

[0211] The model adjuster obtains the current average score mean μ score , and the standard deviation of the current rating σ score And the trigger index ACT is calculated according to the following formula:

[0212] ;

[0213] In the formula, μ baseline is the mean of the benchmark scores, which is equal to the mean score at the beginning of model training or the mean score at the previous time point, σ scoreis the standard deviation of the benchmark score, which is equal to the standard deviation of the score at the beginning of model training or the standard deviation of the score at the previous time point;

[0214] If the trigger index ACT exceeds the monitoring threshold Trigger set by the system, retraining of the artificial intelligence model is triggered;

[0215] If the trigger index ACT is lower than the monitoring threshold Trigger set by the system, it means that the current artificial intelligence model meets expectations and there is no need to retrain the artificial intelligence model;

[0216] Meanwhile, in this embodiment, the manner and method for training the artificial intelligence model is a technical means well known to those skilled in the art, and those skilled in the art can refer to relevant technical manuals to learn about the technology, so in this embodiment, no further description is given;

[0217] In this embodiment, through the mutual cooperation between the scoring collection unit and the learning optimization unit, the artificial intelligence model's ability to analyze the medical images is made more accurate, ensuring that the entire system has the advantages of strong adaptability, good judgment accuracy, reliable auxiliary judgment efficiency, strong human-computer interaction capability and high intelligence.

[0218] Embodiment 2: This embodiment should be understood to include all the features of any of the above embodiments, and further improve upon them. Figure 1 , Figure 2 , Figure 3 , Figure 4 and Figure 5 As shown, the medical image analysis system further includes a report review module, which verifies the identity of the medical staff or the patient and allows them to query the medical image analysis report after the verification;

[0219] The report review module also includes a report review unit and a security protection unit. The security protection unit verifies the identity of the report reviewer or inquirer and grants an access verification code based on the identity verification data. The report review unit verifies the access verification code. If the access verification code is consistent with the access verification code in the server database, the report reviewer or inquirer is allowed to review or query the report.

[0220] The security protection unit includes a local identity verification terminal and an access authorization device. The local identity verification terminal is used by the medical staff or the patient to submit their identity ID. The access authorization device generates an access verification code based on the submitted identity ID and transmits the generated access verification code to the identity verification terminal and the database of the server for storage;

[0221] Wherein, the local identity verification terminal is set in the public space of the medical institution and is used by the medical staff or the patient to query or operate;

[0222] The access authorizer generates an access verification code according to the following formula:

[0223] ;

[0224] Where AccessCode(i) is the value corresponding to the i-th digit of the access verification code, D m D is the last four digits extracted from the medical staff's ID. p is the last four digits extracted from the patient's identity ID, i is the bit index of the access verification code (from 1 to k, i∈{1,2,…,k}), T is the last four digits of the current timestamp, and R is the number of visits to the local identity verification terminal on the same day;

[0225] The identity ID includes but is not limited to the following: ID card, hospitalization card, medical card or hospitalization number and other IDs with identity identification properties;

[0226] The newly generated access verification code must be inconsistent with the previously generated access verification code to be valid, otherwise a new one will be generated;

[0227] The report review unit obtains the access verification code submitted by the local verification terminal and compares it with the access verification code stored in the database of the cloud server. If the two are consistent, the inquirer is allowed to query or call the medical image analysis report;

[0228] If the two are inconsistent, the query of the medical imaging analysis report is not allowed;

[0229] In this embodiment, through the cooperation between the report review module and the diagnosis support module, the security of the evaluation results is made higher, so that the entire system has the advantages of good data security protection effect, high intelligence and good human-computer interaction assistance capability.

[0230] The contents disclosed above are only preferred feasible embodiments of the present invention, and do not limit the protection scope of the present invention. Therefore, all equivalent technical changes made using the contents of the present invention specification and drawings are included in the protection scope of the present invention. In addition, the elements therein can be updated as technology develops.

Claims

1. A medical image analysis system based on artificial intelligence, the medical image analysis system comprising a server, characterized in that: The medical image analysis system further comprises a data preprocessing module, a feature extraction module, a diagnosis support module, and a learning and optimization module, and the server is connected to the data preprocessing module, the feature extraction module, the diagnosis support module, and the learning and optimization module respectively; The data preprocessing module preprocesses the medical image data to segment the lesion area, the feature extraction module extracts features from the lesion area to obtain status information corresponding to the lesion area, the diagnosis support module evaluates the status information based on the trained artificial intelligence model to form an evaluation result, and warns the medical staff of the evaluation result, and the learning and optimization module triggers the adjustment of the artificial intelligence model according to the medical staff's score on the evaluation result; The feature extraction module includes a texture extraction unit and a shape extraction unit, wherein the texture extraction unit extracts texture features in the lesion area to obtain tissue structure state information of the lesion area, and the shape extraction unit extracts geometric shape features in the lesion area to obtain morphological state information of the lesion area; Wherein, the state information includes tissue structure state information and morphology state information; The texture extraction unit includes a grayscale co-occurrence matrix calculator, a local binary pattern extractor, a wavelet transform analyzer, a Gabor filter extractor, and a vector combiner. The grayscale co-occurrence matrix calculator is used to calculate the co-occurrence relationship of different grayscale levels in the image to analyze the distribution of pixel pairs at specific distances and angles and extract the first texture feature of the image. The local binary pattern extractor generates a local binary pattern by comparing each pixel in the image with its neighboring pixels, thereby capturing the second texture feature of local texture information. The wavelet transform analyzer extracts the third texture feature in different scales and directions by performing wavelet decomposition on the image. The Gabor filter extractor filters the image using a group of Gabor filters with different directions and frequencies to extract the fourth texture feature in a specific direction and scale. The vector combiner combines the first texture feature extracted by the gray-level co-occurrence matrix calculator, the second texture feature extracted by the local binary pattern extractor, the third texture feature extracted by the wavelet transform analyzer, and the fourth texture feature extracted by the Gabor filter to form a feature vector FeatureVector1 of the tissue structure state information; The shape extraction unit extracts shape features of the lesion area according to the following steps: S1, processing the lesion area with an edge detection algorithm to detect the boundary of the lesion area; S2, using the result of edge detection to perform contour extraction, the contour extraction uses a contour extraction algorithm, and outputs a set of contour points; S3, calculating the geometric features according to the set of output contour points; The learning and optimization module includes a scoring collection unit and a learning optimization unit, wherein the scoring collection unit collects the scores of the medical staff on the evaluation results, and the learning optimization unit triggers the adjustment of the artificial intelligence model according to the scoring data of the scoring collection unit; The scoring collection unit includes a scoring interface, a scoring database and a scoring submitter. The scoring interface is used for the medical staff to score. The scoring database stores the scoring data of the medical staff. The scoring submitter is used to process the scoring data and save the scoring data in the scoring database. The scoring data includes the scoring value, scoring time, medical staff ID, and case ID; The learning optimization unit includes a scoring analyzer, a model adjuster and a feedback device. The scoring analyzer is used to analyze the scoring data. The model adjuster calculates a trigger index ACT based on the scoring analysis result, and compares the trigger index ACT with a set trigger threshold Trigger. If the trigger index ACT exceeds the set trigger threshold Trigger, the artificial intelligence model is triggered to be retrained. After the artificial intelligence model is adjusted, the feedback device uses the feature vector corresponding to the medical image corresponding to the new scoring data and the original training data set to retrain the model. The retraining process includes: data preparation, model training and model verification; The data preprocessing module includes an image enhancement unit, an image normalization unit and an image segmentation unit. The image enhancement unit is used to improve image quality, remove noise and enhance contrast. The image normalization unit normalizes the image data processed by the image enhancement unit for subsequent processing. The image segmentation unit is used to process the normalized image data to segment out the lesion area. The image enhancement unit includes a multi-format receiver, a denoising processor, a contrast enhancer and a sharpening processor. The multi-format receiver is used to receive multiple medical image formats, and automatically identify and convert them into a unified format that can be analyzed by the system. The denoising processor is used to remove noise from the medical image. The contrast enhancer is used to enhance the contrast of the image to make the lesion area more prominent. The sharpening processor is used to enhance the details of the image. The denoising processor processes the medical image using a Gaussian filter, the contrast enhancer processes the medical image after denoising, and the sharpening processor detects edges in the image using a Laplace operator, and adds the Laplace operator result to the image corresponding to the original medical image to enhance the edge; The image normalization unit normalizes the enhanced image data according to the following formula, specifically: S11. Calculate the minimum and maximum values ​​of the image: Traverse all pixels of the image and find the minimum value I min and the maximum value I max ; S12. Apply the normalization formula: For each pixel value I(x,y) in the image, apply the normalization formula: ; Where I(x,y) represents the pixel value of the image at position (x,y); The image segmentation unit segments the lesion area in the medical image according to the following steps, specifically: S21, threshold segmentation module: For the normalized image data, a global threshold T is selected, and the area with pixel values ​​higher than the threshold is marked as foreground, and the area with pixel values ​​lower than the threshold is marked as background; S22, edge detection module: segmenting the area by detecting the edge of the image; The gray level co-occurrence matrix calculator calculates the gray level co-occurrence matrix of the lesion area, selects different distances and directions; extracts statistical features from the gray level co-occurrence matrix; The gray level co-occurrence matrix calculator calculates the first texture feature according to the following steps: S31. Define the gray-level co-occurrence matrix: Let P(i,j,d,θ) represent the number of co-occurrences of the first pixel with grayscale value i and the second pixel with grayscale value j in the pixel pair with distance d and direction θ; define the grayscale co-occurrence matrix G(i,j) as: ; Where i, j are grayscale values, d is the distance between pixel pairs, θ is the direction between pixel pairs, which can be 0, 45, 90, and 135 degrees, and (x, y) represents the pixel position in the image; Meanwhile, the distance d refers to the interval between two pixels when calculating the gray-level co-occurrence matrix; S32. Calculate entropy, contrast and correlation according to the gray-level co-occurrence matrix: ; Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image. ; Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image. ; Where i, j are grayscale values, N is the number of grayscale levels, G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image, μ x and μ y are the means of i and j respectively, σ x and σ y are the standard deviations of i and j respectively; For the mean μ of gray level i x , calculated according to the following formula: ; Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image. For the mean μ of gray level j y , calculated according to the following formula: ; Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image. For the standard deviation σ of gray level i x , calculated according to the following formula: ; Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image. For gray level j, the standard deviation σ y , calculated according to the following formula: ; Where i, j are the grayscale values, N is the number of grayscale levels, and G(i, j) is the grayscale co-occurrence matrix, which represents the co-occurrence relationship between pixel pairs with grayscale levels i and j in the image. The local binary pattern extractor compares the grayscale value of each pixel in the lesion area with that of the neighboring pixels to generate a binary pattern; Calculating a histogram of the local binary pattern as a second texture feature vector, wherein the feature vector is used to capture the local texture pattern in the image and is suitable for describing a tissue structure with rich details; The local binary pattern extractor calculates a histogram of binary patterns according to the following steps: S51, traverse the image: For each pixel p, determine the radius R of its neighborhood and the number P of neighboring pixels; Directly use the gray value of position p in the image as the gray value g of the central pixel p ; S52, calculate the gray value g of the neighborhood pixel n : For each n (n=0,1,…,P−1), calculate the gray value g of the nth neighboring pixel n ; Among them, the position of the neighborhood pixel is calculated according to R and P; the coordinates of the neighborhood pixel are determined by the following formula: ; If the calculated position is not an integer, use interpolation to calculate the grayscale value; S53, compare the gray value g of the neighboring pixels n and the gray value g of the center pixel p : ; S54, generate LBP binary pattern: Convert the binary pattern to a decimal LBP value: ; S55, calculating LBP histogram: counting the frequencies of all LBP values ​​in the image to generate LBP histogram; The wavelet transform analyzer performs multi-scale wavelet decomposition on the lesion area to extract coefficients at different scales and directions; Calculate the statistical characteristics of wavelet coefficients; Among them, wavelet transform can decompose high-frequency and low-frequency information in images and effectively capture the third texture features at different scales; The method for extracting texture features of the lesion area by the wavelet transform analyzer comprises the following steps: S61. Select wavelet function and decomposition series: S62, perform wavelet decomposition: perform wavelet decomposition on the lesion area to obtain subband coefficient matrices W at different scales and directions s,d (x,y), where W s,d (x,y) is the wavelet coefficient of the image at position (x,y); here s and d represent the scale and direction of wavelet decomposition respectively; Among them, s is the scale factor, which indicates the level of wavelet decomposition, and d is the direction factor, which indicates the direction of wavelet decomposition. At the same time, the scale factor s and the direction factor d are set by the system or the operator according to the actual situation and input from the human-computer interaction interface; S63, extracting statistical features: ; Where Mean1 is the mean of the wavelet decomposition, Variance1 is the variance of the wavelet decomposition, N is the total number of pixels in the image, is the width multiplied by the height of the image, (x, y) is the pixel coordinate in the image, traversing the entire image range from 0 to N−1; The Gabor filter extractor uses a set of Gabor filters of different frequencies and directions to convolve the lesion area; extracts the response features after filtering; wherein the Gabor filter simulates the texture perception mechanism in the human visual system and is suitable for extracting the fourth texture features specific to directionality and scale; The method for extracting texture features of the lesion area using a Gabor filter extractor includes the following steps: S71. Use a set of Gabor filters of different directions and frequencies to convolve the image and obtain the filter response R(x,y;λ,θ,ψ,σ,γ): ; Among them, G(x,y;λ,θ,ψ,σ,γ) is the Gabor filter convolution kernel function: ; Where x ′, y ′ is the position of the filter in the x and y directions, λ is the wavelength, which is used to control the period of the sine function, θ is the direction, which is used as the directional parameter of the Gabor filter, ψ is the phase offset, which is used to control the phase of the sine function, σ is the standard deviation of the Gaussian envelope, which is used to control the width of the Gaussian envelope, and γ is the spatial aspect ratio, which is used to control the elliptical shape of the Gaussian envelope; S72. Calculate response characteristics: Amplitude of the filter response: ; In the formula, R real (x,y) and R imag (x,y) represents the real and imaginary parts of the Gabor filter response, respectively. x and y are the pixel coordinates of the image, ranging from 0 to N−1 across the entire image. Phase of the filter response: ; In the formula, R real (x,y) and R imag (x,y) represent the real and imaginary parts of the Gabor filter response respectively; S73, calculating the mean and variance of the amplitude and phase as the fourth texture feature, specifically: 1) Amplitude correlation: ; Where Mean2 is the amplitude mean extracted by the Gabor filter, Variance2 is the amplitude variance extracted by the Gabor filter, N is the total number of pixels in the image, and M(x,y) is the amplitude of the filter response at position (x,y); 2) Phase correlation: ; Where Mean3 is the phase mean extracted by the Gabor filter, Variance3 is the phase variance extracted by the Gabor filter, N is the total number of pixels in the image, and P(x,y) is the phase of the filter response at position (x,y); The vector combiner combines the first texture feature extracted by the gray-level co-occurrence matrix calculator, the second texture feature extracted by the local binary pattern extractor, the third texture feature extracted by the wavelet transform analyzer, and the fourth texture feature extracted by the Gabor filter to form a feature vector FeatureVector1=[Entropy, Contrast, Correlation, LBP Histogram, Mean1, Variance1, Mean2, Variance2, Mean3, Variance3] of the tissue structure state information; The shape extraction unit includes an edge detector, a contour extractor and a geometric feature calculator, wherein the edge detector is used to detect the boundary of the lesion area, the contour extractor is used to extract the contour of the lesion area, and the geometric feature calculator is used to calculate the geometric shape features of the lesion area; The geometric shape features include: area, perimeter, centroid, circularity, aspect ratio and convexity of convex hull; The area is calculated according to the following formula: ; In the formula, (x i ,y i ) is the coordinate of the i-th point on the contour, (x i+1 ,y i+1 ) is the coordinate of the i-th point on the contour, n is the total number of contour points, where (x n ,y n ) = (x0, y0) means that the polygon outlines are connected at the ends; The perimeter is calculated according to the following formula: ; Among them, (x n ,y n ) = (x0, y0) means that the polygon outlines are connected at the ends; Optionally, the geometric feature calculator calculates the centroid according to the following formula: ; In the formula, (C x , C y ) is the coordinate of the center of mass, (x i ,y i ) is the coordinate of the i-th point on the contour, (x i+1 ,y i+1 ) is the coordinate of the i-th point on the contour, n is the total number of contour points, where (x n ,y n ) = (x0, y0) means that the polygon outlines are connected at the ends; The circularity is calculated according to the following formula: ; Where Area is the area of ​​the contour and Perimeter is the perimeter of the contour; The aspect ratio is calculated according to the following formula: ; Where W is the width of the contour, that is, the width of the minimum bounding rectangle, and H is the height of the contour, that is, the height of the minimum bounding rectangle; The convex hull area is calculated according to the following formula. For a polygonal convex hull consisting of m points, the area formula of the convex hull is the same as the calculation formula of the contour area, that is: ; Where m is the total number of convex hull points, (x i ,y i ) is the coordinate of the i-th point on the convex hull, (x m ,y m ) = (x0, y0) means that the polygon outlines are connected at the ends; The convexity is calculated according to the following formula: ; Where, Area is the contour area, Convex Hull Area is the convex hull area; The geometric shape features extracted by the geometric feature calculator are combined to form a feature vector FeatureVector2=[Area, Perimeter, C x , C y , Aspect Ratio, Convex Hull Area, Convexity]; The diagnosis support module includes a diagnosis evaluation unit and a decision support unit, wherein the diagnosis evaluation unit evaluates the status information based on the artificial intelligence model to form an evaluation result, and the decision support unit issues an early warning prompt of the evaluation result to the medical staff; The diagnostic evaluation unit integrates feature vectors from different sources into a comprehensive feature vector, standardizes the comprehensive feature vector, inputs the standardized comprehensive feature vector into an artificial intelligence model for evaluation, and outputs an evaluation result; The diagnostic evaluation unit normalizes the comprehensive feature vector pair according to the following formula: ; In the formula, NormalizedFeatureVector is the standardized or normalized comprehensive feature vector, CompositeFeatureVector is the comprehensive feature vector, σ V is the standard deviation of the comprehensive eigenvector, μ V is the mean of the comprehensive eigenvector, satisfying: ; Where N is the dimension of the composite feature vector, CompositeFeatureVector i is the i-th eigenvalue of the comprehensive eigenvector; The standard deviation of the comprehensive feature vector is calculated according to the following formula: ; Where N is the dimension of the composite feature vector, CompositeFeatureVector i is the i-th eigenvalue of the comprehensive eigenvector, μ V is the mean of the comprehensive eigenvector; The diagnostic evaluation unit inputs the standardized or normalized comprehensive feature vector into the artificial intelligence model, and outputs an evaluation result Output through the artificial intelligence model; The decision support unit generates a diagnosis report and a warning prompt according to the evaluation result output by the artificial intelligence model, and provides a warning prompt to the medical staff; The decision support unit obtains the evaluation unit Output and the monitoring threshold Warning set by the system. If the evaluation result Output is greater than or equal to the monitoring threshold Warning set by the system, it indicates that there is an abnormality in the lesion area corresponding to the medical image; If the evaluation result Output is lower than the monitoring threshold Warning set by the system, it means that the lesion area corresponding to the medical image is in a non-dangerous state; The score analyzer obtains the score data and calculates the current average score mean μ according to the following formula: score : ; Where M is the total number of ratings, r i Score the i-th item; The score analyzer calculates the standard deviation σ of the current score according to the following formula score : ; Where M is the total number of ratings, r i is the i-th score, μ score is the current average score; The model adjuster obtains the current average score mean μ score , and the standard deviation of the current rating σ score And the trigger index ACT is calculated according to the following formula: ; In the formula, μ baseline is the mean of the benchmark scores, which is equal to the mean score at the beginning of model training or the mean score at the previous time point, σ score is the standard deviation of the benchmark score, which is equal to the standard deviation of the score at the beginning of model training or the standard deviation of the score at the previous time point; If the trigger index ACT exceeds the monitoring threshold Trigger set by the system, retraining of the artificial intelligence model is triggered; If the trigger index ACT is lower than the monitoring threshold Trigger set by the system, it means that the current artificial intelligence model meets expectations and there is no need to retrain the artificial intelligence model; The medical image analysis system further comprises a report review module, which verifies the identity of the medical staff or the patient and allows them to query the medical image analysis report after the verification; The report review module also includes a report review unit and a security protection unit. The security protection unit verifies the identity of the report reviewer or inquirer and grants an access verification code according to the identity verification data. The report review unit verifies the access verification code. If the access verification code is consistent with the access verification code in the server database, the report reviewer or inquirer is allowed to review or query the report. The security protection unit includes a local identity verification terminal and an access authorization device. The local identity verification terminal is used by the medical staff or the patient to submit their identity ID. The access authorization device generates an access verification code based on the submitted identity ID and transmits the generated access verification code to the identity verification terminal and the database of the server for storage; Wherein, the local identity verification terminal is set in the public space of the medical institution and is used by the medical staff or the patient to query or operate; The access authorizer generates an access verification code according to the following formula: ; Where AccessCode(i) is the value corresponding to the i-th digit of the access verification code, D m D is the last four digits extracted from the medical staff's ID. p is the last four digits extracted from the patient's identity ID, i is the bit index of the access verification code (from 1 to k, i∈{1,2,…,k}), T is the last four digits of the current timestamp, and R is the number of visits to the local identity verification terminal on the same day; The newly generated access verification code must be inconsistent with the previously generated access verification code to be valid, otherwise a new one will be generated; The report review unit obtains the access verification code submitted by the local verification terminal and compares it with the access verification code stored in the database of the cloud server. If the two are consistent, the inquirer is allowed to query or call the medical image analysis report; If the two are inconsistent, querying the medical imaging analysis report is not allowed.