A medical graphic image artificial intelligence modeling and recognition system

Abstract

The present invention discloses a medical graphic image artificial intelligence modeling and recognition system, specifically related to the field of intelligent modeling and recognition. The present invention includes a data acquisition module, an image data preprocessing module, an image segmentation module, a feature extraction module, an active learning model construction module, and a recognition and diagnosis module. The present invention preprocesses the CT images of patients by collecting them, including denoising and enhancement, and then segments the images to distinguish the lesions to be determined from the normal areas. Key features of the lesion areas are extracted using deep learning algorithms, and lesion values are calculated to determine the lesion areas. Then, the lesion features are input into the active learning model for training, and the most valuable samples are iteratively selected for annotation to optimize the model performance. Finally, the image to be recognized is input into the trained model, and the lesion type, location, size, and preliminary diagnosis report are output. The system also has a visualization display function, highlighting the lesion areas and providing a feedback and optimization loop.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent modeling and recognition, and more specifically, to an artificial intelligence modeling and recognition system for medical graphic images. Background Art

[0002] With the rapid development of technology, artificial intelligence (AI) technology has gradually penetrated into various industries, and the medical industry is one of the most significantly benefited fields; the application of AI in the medical field, especially in medical image diagnosis, has shown great potential and value.

[0003] Medical image diagnosis is an indispensable part of clinical medicine, which is of great significance for the early detection of diseases, accurate diagnosis, and selection of treatment plans. Although many studies and commercial applications have achieved certain results in the field of medical images, there are still some limitations, such as problems of data diversity, complexity, and model interpretability; this requires the construction of a medical image recognition system integrating advanced AI technology to achieve more efficient and accurate lesion recognition and diagnosis.

[0004] However, when the traditional manual analysis is actually used, there are still some disadvantages. For example, the traditional analysis method only segments the image once to determine the lesion area. This single segmentation method will lead to an increase in errors and the probability of misdiagnosis. Especially in the case of processing complex images, the lesion area may be partially blocked or there are similar tissue structures. Due to the lack of a secondary verification mechanism, the traditional method cannot adapt to the tiny changes in the image, further reducing the accuracy of diagnosis. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an artificial intelligence modeling and recognition system for medical graphic images to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Data acquisition module: used to obtain the CT scan medical image data of patients, and then through a high-speed data transmission interface, the CT image data is quickly transmitted from the hospital information system to the image data preprocessing module;

[0008] Image data preprocessing module: perform noise removal and image enhancement preprocessing on the input medical images;

[0009] Image segmentation module: used to segment the processed medical images, divide the segmented medical images into areas to be determined for lesions and normal areas, and transmit the areas to be determined for lesions in the medical images to the feature extraction module;

[0010] Feature extraction module: Using deep learning algorithms, extract key features from medical images within the lesion area to be determined, calculate the lesion value based on the calculated key features, and divide the lesion area to be determined into normal areas and lesion areas;

[0011] Active learning model construction module: Used to input the features and images of the lesion area into the constructed active learning model for training;

[0012] Recognition and diagnosis module: Input the medical image to be recognized into the trained active learning model, and the model outputs the recognition result.

[0013] Preferably, in the data acquisition module, the data acquisition module is docked with the hospital's information system to automatically collect and update the latest imaging data of patients; among them, the input format of image data supports DICOM, JPEG, and PNG medical imaging standard formats.

[0014] Preferably, in the image data preprocessing module, Gaussian filtering is used to remove random noise in the image data for noise removal; the image data after noise removal is subjected to histogram equalization to enhance image details.

[0015] Preferably, in the image segmentation module, the method of image segmentation is specifically as follows:

[0016] Step 1: Standardize the medical image and adjust the size of the medical image to a unified size;

[0017] Step 2: Select the number of clusters to be segmented and randomly initialize the fuzzy membership matrix. Each column of the membership matrix represents a cluster, and the sum of each row is 1, that is: , where, represents the membership degree of the a-th pixel in the medical image to the b-th cluster, and c represents the number of clusters;

[0018] Step 3: Calculate the cluster center, calculate the center of each cluster according to the membership matrix, and the calculation method of the center value of each cluster is specifically as follows:

[0019] , where, represents the center value of each cluster, D represents the total number of pixels in the image, represents the feature vector of the a-th pixel; represents the fuzzy membership degree of the a-th pixel relative to the b-th cluster, and this membership degree is adjusted by a fuzzy factor, d represents the fuzzy factor, and the value ranges from 1 to 2;

[0020] Step 4: Update the membership matrix according to the current cluster center, and the update calculation method is specifically as follows:

[0021] , where represents the updated membership matrix, c represents the number of clusters, and f represents the f-th cluster center; h represents the fuzzy factor, with a value between 1 and 2. represents the distance between the a-th pixel and the b-th cluster center. represents the distance between the a-th pixel and the f-th cluster center.

[0022] Step 5: Check the convergence of the algorithm. Determine whether the algorithm converges by calculating the change amount of the membership matrix. The calculation method of the change amount of the membership matrix is specifically as follows:

[0023] , where represents the change amount of the membership matrix. represents the fuzzy membership matrix calculated after the current iteration. represents the fuzzy membership matrix calculated in the previous iteration step. That is, the membership values in the previous round.

[0024] Step 6: Generate a segmentation result according to the fuzzy membership matrix and perform segmentation using the binary method. The calculation method is specifically as follows:

[0025] , where represents the result of binary segmentation. represents the membership of the a-th pixel in the medical image to the b-th cluster.

[0026] Preferably, in the feature extraction module, the deep learning algorithm selects a convolutional neural network, and the gray distribution value, element density value, and texture value within the lesion area to be determined are extracted through the convolutional neural network.

[0027] Preferably, in the model training module, the method for constructing an active learning model is specifically as follows:

[0028] For the selected sample set T, first define a sampling strategy , which is used to calculate the probability of each sample in the sample set T being selected; subsequently, use the labeled sample set U to train the model to obtain a classifier w; this classifier w can predict the category of unlabeled samples, and according to the previously defined sampling strategy, select the most valuable samples from the prediction results for labeling; during this process, the number of labeled samples will be dynamically adjusted according to the sampling strategy; in each iteration, select the most valuable z samples for labeling, where the specific value of z will be flexibly adjusted according to the sampling strategy and the overall scale of the data set; such an iterative process will continue until the preset stop condition is met.

[0029] Preferably, in the recognition and diagnosis module, the image of the lesion area is input into the active learning model again for recognition. The model outputs the disease type, location and size within the lesion area, and generates a preliminary diagnosis report and suggestions.

[0030] Among the output results of the active learning model, first, the recognized lesion area is visually displayed. By highlighting the location of the lesion, medical staff can intuitively understand the distribution of the lesion. After the preliminary diagnosis report and suggestions are generated, the active learning model does not stop working immediately. Instead, it enters a feedback and optimization loop. The system compares this preliminary report with the preset high-precision medical diagnosis criteria, and evaluates the accuracy and integrity of the diagnosis through algorithms.

[0031] The technical effects and advantages of the present invention:

[0032] The present invention obtains CT image data from the hospital information system through a high-speed data transmission interface, and automatically collects and updates the latest image data of the patient. Then, preprocessing such as noise removal and image enhancement is performed on the input medical images. Gaussian filtering is used to eliminate random noise, and histogram equalization is performed to enhance image details. Next, the image is segmented into the area of the lesion to be determined and the normal area, and the area of the lesion to be determined in the medical image is transmitted to the feature extraction module. In the feature extraction module, a convolutional neural network is used to extract key features, and the lesion value is obtained according to the calculated key features, and the area of the lesion to be determined is divided into the normal area and the lesion area. Finally, the features and images of the lesion area are input into the constructed active learning model for training. The most valuable samples are selected for annotation through an iterative process until the preset stop condition is met. Finally, the medical image to be recognized is input into the trained active learning model, and the model outputs the recognition result and generates a preliminary diagnosis report and suggestions. The present invention calculates the lesion value according to the features extracted after image segmentation, and then distinguishes the area of the lesion to be determined again according to the lesion value. Through the mechanism of secondary verification, it adapts to the tiny changes in the image, reduces the error and misdiagnosis probability, and improves the accuracy of the diagnosis. Description of the Drawings

[0033] Figure 1 It is a schematic diagram of the module connection of the present invention. Detailed Embodiments

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0035] Please refer to Figure 1 As shown, the present invention provides a medical graphic image artificial intelligence modeling and recognition system, which includes a data acquisition module, an image data preprocessing module, an image segmentation module, a feature extraction module, an active learning model construction module, and a recognition and diagnosis module.

[0036] Data acquisition module: used to obtain the CT scan medical image data of patients, and then through a high-speed data transmission interface, the CT image data is quickly transmitted from the hospital information system to the image data preprocessing module;

[0037] In the data acquisition module, the data acquisition module is docked with the hospital information system to automatically collect and update the latest image data of patients; among them, the input format of image data supports DICOM, JPEG, and PNG medical image standard formats.

[0038] Image data preprocessing module: preprocess the input medical images by removing noise and enhancing the images;

[0039] In the image data preprocessing module, Gaussian filtering is used to remove random noise in the image data; the image data after removing noise is subjected to histogram equalization to enhance image details;

[0040] The method of using Gaussian filtering to remove random noise in the image data is specifically as follows:

[0041] Step 1: Create a Gaussian filter according to the determined parameters; calculate a Gaussian weight matrix, and each element of the matrix represents the pixel weight at the corresponding position.

[0042] Step 2: Scan each pixel in the image with the created Gaussian filter. For each pixel, replace the value of the pixel with the weighted average gray value of the pixels in the neighborhood determined by the filter.

[0043] Step 3: Repeat the above steps until all pixels in the image have been processed, so as to obtain the denoised image.

[0044] Image segmentation module: used to segment the processed medical images, divide the segmented medical images into areas with lesions to be determined and normal areas, and transmit the areas with lesions to be determined in the medical images to the feature extraction module;

[0045] In the image segmentation module, the method of image segmentation is specifically as follows:

[0046] Step 1: Standardize the medical images and adjust the size of the medical images to a unified size;

[0047] Step 2: Select the number of segmented clusters and randomly initialize the fuzzy membership matrix. Each column of the membership matrix represents a cluster, and the sum of each row is 1, that is: , where represents the membership degree of the a-th pixel in the medical image to the b-th cluster, and c represents the number of clusters;

[0048] Step 3: Calculate the cluster centers. Calculate the center of each cluster according to the membership matrix. The specific calculation method of the center value of each cluster is:

[0049] , where represents the center value of each cluster, D represents the total number of pixels in the image, represents the feature vector of the a-th pixel; represents the fuzzy membership degree of the a-th pixel relative to the b-th cluster, and this membership degree is adjusted by the fuzzy factor. d represents the fuzzy factor, and its value ranges from 1 to 2;

[0050] Step 4: Update the membership matrix according to the current cluster centers. The specific update calculation method is:

[0051] , where represents the updated membership matrix, c represents the number of clusters, f represents the f-th cluster center; h represents the fuzzy factor, and its value ranges from 1 to 2, represents the distance between the a-th pixel and the b-th cluster center;

[0052] Step 5: Check the convergence of the algorithm. Determine whether the algorithm converges by calculating the change amount of the membership matrix. The specific calculation method of the change amount of the membership matrix is:

[0053] , where represents the change amount of the membership matrix, represents the fuzzy membership matrix calculated after the current iteration; represents the fuzzy membership matrix calculated in the previous iteration step. That is, the membership degree value in the previous round;

[0054] Compare the calculated change amount of the membership matrix with the preset change amount threshold. If the calculated change amount of the membership matrix is less than the preset change amount threshold, output a convergence instruction and stop the iteration; if the calculated change amount of the membership matrix is greater than the preset change amount threshold, output a non-convergence instruction and return to Step 3;

[0055] Step 6: Generate a segmentation result according to the fuzzy membership matrix and use the binary method for segmentation. The calculation method is:

[0056] , where represents the result of binary segmentation, and represents the membership degree of the ath pixel to the bth cluster in the medical image;

[0057] When the calculated membership degree is greater than or equal to a preset threshold, this area is the area of the lesion to be determined; if the calculated membership degree is less than the preset threshold, this area is a normal area;

[0058] According to steps 1 - 6, the medical image is divided into the area of the lesion to be determined and the normal area.

[0059] Feature extraction module: Using a deep learning algorithm, extract key features from the medical image within the area of the lesion to be determined, obtain the lesion value based on the calculated key features, and divide the area of the lesion to be determined into a normal area and a lesion area;

[0060] In the said feature extraction module, the deep learning algorithm selects a convolutional neural network, and extracts the gray - level distribution value, element density value, and texture value within the area of the lesion to be determined through the convolutional neural network;

[0061] The calculation method of the gray - level distribution value is specifically:

[0062] , where H represents the gray - level distribution value within the image of the area of the lesion to be determined, represents the gray - level value at the position within the image of the area of the lesion to be determined; represents the position of the centroid on the x - axis, represents the position of the centroid on the y - axis, and qj and qk together form the centroid ; represents the sum of all gray - level values in the image of the area of the lesion to be determined;

[0063] The calculation method of the position of the centroid on the x - axis is specifically:

[0064] , where represents the position of the centroid on the x - axis, represents the gray - level value at the position within the area of the lesion to be determined;

[0065] The calculation method of the position of the centroid on the y - axis is specifically:

[0066] , where represents the position of the centroid on the y - axis, represents the gray - level value at the position within the area of the lesion to be determined;

[0067] The specific calculation method for the element density value in the image of the lesion area to be determined is as follows:

[0068] , where L represents the element density value in the image of the lesion area to be determined, represents the linear attenuation coefficient of the tissue to X-rays, represents the linear attenuation coefficient of water to X-rays;

[0069] The attenuation coefficient of water is set as the reference standard. The tissue with the highest density in the body has the greatest absorption and attenuation of X-rays, so its element density value is the highest and is set as +1000 Hu;

[0070] The density of air is the lowest, and its absorption and attenuation of X-rays are the smallest, so its CT value is the lowest and is set as -1000 Hu.

[0071] The density of other tissues in the human body is between that of air and dense bone, so their element density values are between -1000 Hu and +1000 Hu;

[0072] The specific calculation method for the texture value in the image of the lesion area to be determined is as follows:

[0073] , where M represents the texture value in the image of the lesion area to be determined, N represents the contrast, P represents the correlation, Q represents the entropy, 、 、 represent the weight coefficients;

[0074] The specific calculation method for the contrast is as follows:

[0075] , where N represents the contrast, represents the frequency of pixel pairs in the gray-level co-occurrence matrix, and m and n represent the gray values of two pixels;

[0076] The specific calculation method for the correlation is as follows:

[0077] , where P represents the correlation, represents the frequency of pixel pairs in the gray-level co-occurrence matrix, m and n represent the gray values of two pixels, represents the mean value of the row, represents the mean value of the column, represents the standard deviation of the row, represents the standard deviation of the column;

[0078] , where Q represents the entropy, It represents the frequency of pixel pairs in the gray-level co-occurrence matrix;

[0079] Lesion area recognition module: It is used to calculate the lesion value by calculating the extracted moment of inertia value, element density value, and texture value, and determine the lesion area according to the lesion value;

[0080] In the said lesion area recognition module, the calculation method of the lesion value is specifically as follows:

[0081] , where Z represents the lesion value, H represents the moment of inertia value, L represents the element density value in the image of the lesion area to be determined, and M represents the texture value in the image of the lesion area to be determined;

[0082] If the calculated lesion value is greater than the preset lesion value, a lesion instruction is output, and this lesion area to be determined is the lesion area; if the calculated lesion value is less than the preset lesion value, a non-lesion instruction is output, and this lesion area to be determined is the normal area;

[0083] Transmit the characteristics and images of the lesion area to the active learning model training module for active learning.

[0084] Active learning model construction module: It is used to input the characteristics and images of the lesion area into the constructed active learning model and train it;

[0085] In the said active learning model construction module, the method of constructing the active learning model is specifically as follows:

[0086] For the selected sample set T, first define a sampling strategy , which is used to calculate the probability of each sample in the sample set T being selected; subsequently, use the labeled sample set U to train the model to obtain a classifier w; this classifier w can predict the category of unlabeled samples, and select the most valuable samples from the prediction results for labeling according to the previously defined sampling strategy; during this process, the number of labeled samples will be dynamically adjusted according to the sampling strategy; in each iteration, select the most valuable z samples for labeling, where the specific value of z will be flexibly adjusted according to the sampling strategy and the overall scale of the data set; such an iterative process will continue until the preset stop condition is met;

[0087] Among them, the calculation method in the active learning model is specifically as follows:

[0088] , where, represents selecting the sample with the highest score in the unlabeled sample set, J represents the unlabeled sample, and w represents the classifier; represents the classifier performance metric;

[0089] , where U represents the labeled sample set, represents the upper limit of the number of labeled samples;

[0090] Select the most valuable samples for labeling according to the sampling strategy. The calculation method is as follows:

[0091] , where T represents the samples after labeling, represents the sampling strategy;

[0092] The calculation method for the finally selected labeled sample set is as follows:

[0093] , where, represents the finally selected labeled sample set, represents the classifier performance metric;

[0094] Active learning is an iterative process until a stopping criterion is reached.

[0095] Recognition and Diagnosis Module: Input the medical image to be recognized into the trained active learning model, and the model outputs the recognition result;

[0096] In the recognition and diagnosis module, input the lesion area image into the active learning model again for recognition. The model outputs the disease type, location, and size within the lesion area, and generates a preliminary diagnosis report and suggestions;

[0097] In the output result of the active learning model, first visualize the recognized lesion area. By highlighting the location of the lesion, medical staff can intuitively understand the distribution of the lesions. After the preliminary diagnosis report and suggestions are generated, the active learning model does not stop working immediately; instead, it enters a feedback and optimization loop. The system compares this preliminary report with the preset high-precision medical diagnosis criteria, and evaluates the accuracy and completeness of the diagnosis through algorithms.

[0098] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A medical image artificial intelligence modeling and recognition system, characterized in that: include: Data acquisition module: used to obtain the patient's CT scan medical image data, and then quickly transmit the CT image data from the hospital information system to the image data preprocessing module through a high-speed data transmission interface; Image data preprocessing module: performs noise removal and image enhancement preprocessing on the input medical image; Image segmentation module: used for performing image segmentation on the medical image after image processing, dividing the segmented medical image into a lesion area to be determined and a normal area, and transmitting the lesion area to be determined in the medical image to the feature extraction module; Feature extraction module: It uses deep learning algorithms to extract key features from medical images in the lesion area to be determined, calculates the lesion value based on the key features, and divides the lesion area to be determined into a normal area and a lesion area; The specific method for calculating the element density value in the image of the lesion area to be determined is: , Expressed as the linear attenuation coefficient of tissue to X-rays, Expressed as the linear attenuation coefficient of water for X-rays; The specific method for calculating the texture value M in the image of the lesion area to be determined is: , N represents contrast, P represents correlation, Q represents entropy, , , Expressed as weight coefficient; , It is represented as the frequency of pixel pairs in the gray-level co-occurrence matrix, and m and n are represented as the gray values ​​of two pixels; , is represented as the mean of the rows, is represented as the mean of the column, Expressed as the standard deviation of the row, Expressed as the standard deviation of the column; ； Active learning model building module: used to input the features and images of the lesion area into the constructed active learning model and perform training; Recognition and diagnosis module: The medical image to be identified is input into the trained active learning model, and the model outputs the recognition result.

2. The medical image artificial intelligence modeling and recognition system according to claim 1, characterized in that: In the data acquisition module, the data acquisition module is connected to the hospital's information system to automatically collect and update the patient's latest imaging data; wherein, the image data input format supports DICOM, JPEG, and PNG medical imaging standard formats.

3. The medical image artificial intelligence modeling and recognition system according to claim 1, characterized in that: In the image data preprocessing module, noise removal uses Gaussian filtering to eliminate random noise in image data; and histogram equalization is performed on the image data after noise removal to enhance image details.

4. The medical image artificial intelligence modeling and recognition system according to claim 1, characterized in that: In the image segmentation module, the image segmentation method is specifically as follows: Step 1: Standardize the medical image and adjust the size of the medical image to a uniform size; Step 2: Select the number of clusters to be segmented and randomly initialize the fuzzy membership matrix. Each column of the membership matrix represents a cluster and the sum of each row is 1, that is: ,in, It is represented as the membership of the a-th pixel in the medical image to the b-th cluster, and c is the number of clusters; Step 3: Calculate the cluster center. Calculate the center of each cluster based on the membership matrix. The calculation method of the center value of each cluster is as follows: ,in, is represented by the center value of each cluster, D is represented by the total number of pixels in the image, Represented as the feature vector of the a-th pixel; It is represented as the fuzzy membership of the a-th pixel relative to the b-th cluster, and this membership is adjusted by the fuzzy factor. d is represented as the fuzzy factor, which takes a value between 1 and 2. Step 4: Update the membership matrix according to the current cluster center. The updated calculation method is as follows: ,in, It is represented as the updated membership matrix, c is the number of clusters, f is the center of the fth cluster, and h is the fuzzy factor, which takes a value between 1 and 2. It is expressed as the distance between the ath pixel and the bth cluster center, It is expressed as the distance between the ath pixel and the fth cluster center; Step 5: Check the convergence of the algorithm. Determine whether the algorithm converges by calculating the change in the membership matrix. The specific method for calculating the change in the membership matrix is: ,in, Expressed as the change in the membership matrix, Represented as the fuzzy membership matrix calculated after the current iteration; Represents the fuzzy membership matrix calculated in the previous iteration step; that is, the membership value of the previous round; Step 6: Generate segmentation results based on the fuzzy membership matrix and use the binary method for segmentation. The specific calculation method is: ,in, Represented as the result of binary segmentation, It is expressed as the membership of the a-th pixel to the b-th cluster in the medical image; When the membership degree is calculated If the calculated membership degree is greater than or equal to the preset threshold, then this area is the lesion area to be determined; If it is smaller than the preset threshold, the area is a normal area.

5. The medical image artificial intelligence modeling and recognition system according to claim 1, characterized in that: In the feature extraction module, the deep learning algorithm selects a convolutional neural network, and the moment of inertia value, element density value, and texture value in the lesion area to be determined are extracted through the convolutional neural network; The calculation method of the grayscale distribution value is as follows: , where H represents the grayscale distribution value in the image of the lesion area to be determined, Represented as the image of the lesion area to be determined Gray value of the position; It is represented as the position of the center of mass on the x-axis, It is expressed as the position of the center of mass on the y-axis. qj and qk are combined to form the center of mass ; It is represented as the sum of all gray values ​​in the image of the lesion area to be determined.

6. The medical image artificial intelligence modeling and recognition system according to claim 5, characterized in that: The method for calculating the position of the center of mass on the x-axis is as follows: ,in, It is represented as the position of the center of mass on the x-axis, Indicated as the area of ​​lesion to be determined Gray value of the position; The calculation method of the center of mass position on the y-axis is as follows: ,in, It is represented as the position of the center of mass on the y-axis, Indicated as the area of ​​lesion to be determined Grayscale value of the position.

7. The medical image artificial intelligence modeling and recognition system according to claim 1, characterized in that: In the active learning model construction module, the method for constructing the active learning model is specifically as follows: The selected sample set T first defines a sampling strategy , this strategy is used to calculate the probability of each sample in the sample set T being selected; then, the labeled sample set U is used to train the model to obtain a classifier w; this classifier w can predict the category of unlabeled samples, and according to the previously defined sampling strategy, the most valuable samples are selected from the prediction results for labeling; in this process, the number of labeled samples will be dynamically adjusted according to the sampling strategy; in each iteration, the most valuable z samples are selected for labeling, where the specific value of z will be flexibly adjusted according to the sampling strategy and the overall size of the data set; this iterative process will continue until the preset stop condition is met; Among them, the calculation method in the active learning model is specifically as follows: ,in, It means selecting the sample with the highest score in the unlabeled sample set, J represents the unlabeled sample, and w represents the classifier; Expressed as a classifier performance metric; , where U represents the set of labeled samples, It is expressed as the upper limit of the number of labeled samples; According to the sampling strategy, the most valuable samples are selected for annotation. The specific calculation method is as follows: , where T represents the labeled sample. It is represented as a sampling strategy; The calculation method of the final selected labeled sample set is as follows: ,in, Represents the final selected set of labeled samples, It is expressed as a classifier performance metric.

8. The medical image artificial intelligence modeling and recognition system according to claim 1, characterized in that: In the recognition and diagnosis module, the image of the lesion area is input into the active learning model again for recognition. The model outputs the disease type, location and size in the lesion area, and generates a preliminary diagnosis report and suggestions.

Images