Gastric cancer image processing method based on AI prediction

Abstract

The invention discloses a gastric cancer image processing method based on AI prediction. The method comprises the following steps: S1, collecting and processing gastric tumor image data; s2, extracting lesion features in the image; s3, constructing a prediction model based on an AI technology; s4, training and verifying the prediction model; s5, applying the prediction model to actual diagnosis; s6, deeply interpreting and determining the prediction result, the prediction model is trained and verified, so that the prediction model can learn the ability to distinguish normal stomach tissues and lesion gastric cancer tissues, and the model can effectively identify and judge the gastric cancer lesion features in the lesion features; and the trained prediction model is verified, so that the accuracy, the stability and the generalization of the prediction model can be further ensured in a result verification mode, and the accuracy and the efficiency of subsequent gastric cancer diagnosis and treatment are further ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and particularly to a gastric cancer image processing method based on AI prediction. Background Art

[0002] The image processing of gastric cancer usually involves medical imaging technologies such as CT scans, MRI, ultrasound, and endoscopic imaging, etc., which are used to diagnose and monitor the condition of patients. These image processing technologies play an important role in the diagnosis, treatment planning, pathological analysis, and evaluation of treatment effects of gastric cancer. A gastric cancer pathological section image segmentation prediction method based on EfficientNet is disclosed in a Chinese patent, with the application number: 202210382588.6. This patent proposes to adopt a repeated learning strategy to process the results, generate new labels to supplement the pseudo regions in the original dataset, and further improve the generalization ability and accuracy of the model;

[0003] However, the current processing method cannot accurately identify the key features in the processed images based on the acquired image data information, which affects the accuracy of the actual prediction results and leads to the inability to provide accurate recognition results in the subsequent identification and diagnosis of gastric cancer lesion images. Summary of the Invention

[0004] The present invention provides a gastric cancer image processing method based on AI prediction, which can effectively solve the problem that the current processing method cannot accurately identify the key features in the processed images based on the acquired image data information, affecting the accuracy of the actual prediction results and leading to the inability to provide accurate recognition results in the subsequent identification and diagnosis of gastric cancer lesion images as proposed in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solution: A gastric cancer image processing method based on AI prediction, including the following steps:

[0006] S1. Collect and process gastric tumor image data;

[0007] S2. Extract the lesion features in the images;

[0008] S3. Construct a prediction model based on AI technology;

[0009] S4. Train and validate the prediction model;

[0010] S5. Apply the prediction model to actual diagnosis;

[0011] S6. Deeply interpret and determine the prediction results.

[0012] According to the above technical solution, in S1, it specifically includes collecting and obtaining gastric tumor image data, and processing the collected gastric tumor image data;

[0013] When collecting gastric tumor image data, the collected gastric tumor image data includes both diseased and non-diseased parts. The specific sources can come from medical institutions, clinical trials, and public information. Collect gastric cancer imaging data of different imaging modalities, specifically including data from pathological sections, ultrasound detection images, endoscopic examination images, and CT scan images;

[0014] When collecting gastric tumor image data, it is necessary to ensure that the collected images cover the gastric cancer conditions at different stages, different types, and different individuals.

[0015] According to the above technical solution, in S1, when processing the collected gastric tumor image data, mainly perform image quality inspection, image enhancement processing, image data cleaning, image cropping processing, and normalization processing on the collected original gastric tumor image data, so that the processed image data can be used for the training of the subsequent prediction model, improving the training efficiency and prediction accuracy of the prediction model;

[0016] Therefore, a segmentation algorithm with dynamically assigned weights is proposed. The whole image containing gastric cancer images is divided into several small grids according to pixels, and a rectangular coordinate system is established on the whole image, and each small grid is represented by coordinates. Combining gradient and texture features, the segmentation result formula of gastric cancer images is as follows:

[0017]

[0018] Where:

[0019] f seg (x, y) is the final segmentation result, which represents the segmentation result of the image at the position (x, y), usually 0 or 1, where 0 is the non-tumor area and 1 is the tumor area, and the position where 0 and 1 alternate is the edge position of the tumor area;

[0020] is the gradient amplitude of the pixel, that is, the edge intensity;

[0021] T(I(x, y)) is the texture response of the pixel, including GLCM contrast and entropy;

[0022] w g and w t are adaptive weights, dynamically adjusted to optimize the segmentation effect, adaptively weighing gradient and texture features, and can accurately segment the boundary of gastric tumors;

[0023]

[0024] w t =1 - w g

[0025] Wherein:

[0026] k is a parameter that controls the speed of weight adjustment;

[0027] u is a threshold used to determine the level of gradient change. Usually, u is the global mean of the gradient;

[0028] The weight w g and w t are dynamically calculated based on the local gradient information and texture information of the image. This can adaptively adjust the importance of gradients and textures in the segmentation process. When the edges of the image area are very obvious, the gradient weight w g will increase, making the segmentation more dependent on edge information. Conversely, in areas with blurred edges, texture information will receive more weight;

[0029] After calculating the segmentation result f of each small grid, determine the gastric cancer area in the entire image. First, analyze the gastric cancer lesion situation in the entire image, denoted as h 1 ;

[0030] Then determine the area of the gastric cancer area in the entire image, and then select different segmentation positions based on the regional boundary and central area of the gastric cancer image. Analyze the gastric cancer lesion situation based on the data of the segmented gastric cancer image, denoted as h 2 ;

[0031] According to the above analysis results, evaluate the situation of gastric cancer image segmentation. The specific evaluation formula is as follows:

[0032]

[0033] Wherein:

[0034] Q represents the evaluation result. The smaller the value, the better the evaluation result;

[0035] S 整 represents the area of the entire image containing the gastric cancer image;

[0036] S 总 represents the total area of the gastric cancer area in S 整 ;

[0037] S (f+n1) represents that when S 总 occupies S 整 less than or equal to 50%, the edge of the gastric cancer area takes an additional n 1 small grids;

[0038] S (f+n2) represents that when S 总 occupies S 整 greater than 50%, the edge of the gastric cancer area takes n 2 small grids inward and outward respectively;

[0039] S 中心 represents the area of the central region of gastric cancer selected when it occupies more than 50% of S, and S 总 occupies S 整 when it is more than 50%, and S 中心 = 0.5S 总 ;

[0040] v represents the image processing speed per unit area;

[0041] h 1 represents the degree of gastric cancer carcinogenesis analyzed from the entire gastric cancer image;

[0042] h 2 represents the degree of gastric cancer carcinogenesis analyzed according to the segmented regions after image segmentation;

[0043] Select n 1 and n 2 values in sequence, calculate the value of h 2 and at the same time calculate the corresponding evaluation result Q. When h 1 = h 2 , that is, the segmented region can achieve the same analysis result as the entire image, then select a set of data with the smallest evaluation result Q from these data as a reference for subsequent image segmentation;

[0044] Use AI to select gastric cancer images for processing in the overall image, and judge the degree of gastric cancer lesions through big data. The judgment result is h 1 , and then predict the subsequent lesion rate under normal circumstances according to the average gastric cancer lesion data. The prediction result is P 平均 ;

[0045] The lesions of gastric cancer are affected by individual factors, specifically including genetic susceptibility, molecular characteristics, tumor stage, lifestyle, immune status, age and psychological state. According to the differences in individual factors, further calculate the lesion rate of individual data. The calculation formula is as follows:

[0046]

[0047] Among them:

[0048] P 个体 represents the individual gastric cancer lesion rate;

[0049] A i represents the development degree of an influencing factor of individual gastric cancer lesions, with a value range of (0,1). The value of A i is obtained by experts scoring based on individual data. The larger the value, the deeper the development degree;

[0050] αi Represents A i The weight of, with a value in (0, 1), α i Is obtained through statistical analysis training and determines the contribution degree of each factor to the prediction result.

[0051] According to the above technical solution, in S1, when performing image quality inspection, it aims to remove artifacts in the image, adjust brightness and contrast;

[0052] When performing image enhancement processing, it aims to perform flipping and rotation operations on the image data to adapt to the image processing ability of the subsequent prediction model, so as to enhance the image data;

[0053] When performing image data cleaning, it aims to remove noise and unnecessary information in the image to ensure image quality;

[0054] When performing image cropping processing, it aims to crop the obtained image data to adjust the size of the image data;

[0055] When performing normalization processing on the image, it aims to uniformly adjust the format of the image to reduce the differences between different image data.

[0056] According to the above technical solution, in S2, when extracting lesion features in the image, using deep learning technology indicates that the prediction model has high recognition accuracy, and useful lesion features are extracted from the gastric tumor image data, and the key information for tumor recognition is obtained by extracting lesion features;

[0057] Specifically, the processed gastric cancer image data is input into a deep learning network to extract deep features, specifically including morphological features, color features, and texture features of the tumor. Among them, the morphological features specifically include position information, shape information, and edge information.

[0058] According to the above technical solution, in S3, by constructing a prediction model based on AI technology, the prediction model can distinguish normal gastric tissue from diseased gastric cancer tissue, realize the prediction and diagnosis of early gastric cancer, and at the same time also include distinguishing the type of gastric cancer and segmenting and estimating the scope of canceration.

[0059] According to the above technical solution, in S4, it specifically includes training and validating the constructed prediction model;

[0060] When training the prediction model, the extracted lesion features are used to train the prediction model. Through training, the prediction model can learn the ability to distinguish normal gastric tissue from diseased gastric cancer tissue, so as to ensure that the model can effectively identify and judge the gastric cancer lesion features in the lesion features;

[0061] When validating the prediction model, it is mainly through setting up a control validation group to validate the constructed and trained prediction model, and evaluating the accuracy, stability and generalization of the prediction model through the validation;

[0062] The control validation group specifically includes a validation set and a test set. Both the validation set and the test set are input into the prediction model, and the prediction model is used to identify the validation set and the test set. There are gastric cancer lesion characteristics in the validation set, and there are some gastric cancer lesion characteristics in the test set. Through the validation, it is ensured that the prediction model can be widely identified and applied.

[0063] According to the above technical solution, in S4, after the prediction model is validated, when the prediction model accurately identifies the gastric cancer lesion characteristics in the validation set and can identify some gastric cancer lesion characteristics in the test set, it indicates that the prediction model has high recognition accuracy and can be applied in subsequent actual applications;

[0064] When the prediction model cannot identify the gastric cancer lesion characteristics in the validation set, it indicates that the prediction model cannot be applied in subsequent actual applications. At this time, the prediction model needs to be trained continuously until the prediction model can give an accurate validation result.

[0065] According to the above technical solution, in S5, specifically apply the trained and validated prediction model to the actual diagnosis environment of gastric cancer, use the prediction model to perform feature recognition and judgment on the newly obtained images, and obtain the prediction results to help medical staff better understand the condition;

[0066] When the prediction model is applied in actual practice, deploy the prediction model to the hospital's pathological diagnosis system, use the prediction model to process the image data in the hospital's pathological diagnosis system in real time, and assist doctors in making early gastric cancer diagnosis and treatment decisions by identifying the image data.

[0067] According to the above technical solution, in S6, it specifically refers to interpreting and determining the output result of the prediction model. Before determining the diagnosis result, it is necessary to combine the output result of the prediction model with the diagnosis opinions of clinical doctors with professional knowledge, and determine the final diagnosis result through combined comparative analysis.

[0068] Compared with the prior art, the beneficial effects of the present invention:

[0069] 1. By training and validating the prediction model, the prediction model can learn the ability to distinguish normal gastric tissue from diseased gastric cancer tissue, thereby ensuring that the model can effectively identify and judge the gastric cancer lesion characteristics in the lesion features, improving the prediction ability of the prediction model. And by validating the trained prediction model, it is convenient to further ensure the accuracy, stability and generalization of the prediction model in the way of result verification, and further guarantee the accuracy and efficiency of subsequent gastric cancer diagnosis and treatment.

[0070] 2. Utilize deep learning technology to automatically extract useful lesion features from gastric tumor image data. By extracting the lesion features as the key information for tumor recognition and combining with the prediction model to further distinguish normal gastric tissue from diseased gastric cancer tissue, it is convenient to effectively assist medical institutions in the early prediction and diagnosis of gastric cancer during the clinical diagnosis process, ensuring the accuracy of disease recognition.

[0071] 3. By deploying the prediction model into the hospital's pathological diagnosis system, the prediction model is used to process the image data in the hospital's pathological diagnosis system in real time. By identifying the image data to assist doctors in making early gastric cancer diagnosis and treatment decisions, and at the same time combining the output results of the prediction model with the diagnostic opinions of clinical doctors with professional knowledge, the final diagnostic result is determined through combined comparative analysis.

[0072] 4. By collecting gastric tumor image data covering gastric cancer cases at different stages, different types and different individuals, the comprehensiveness of image data collection is ensured, which is convenient for subsequent better substitution into the model for extensive training. And by processing the collected gastric tumor image data, the processed image data can be more accurately applied to the training of the subsequent prediction model, improving the training efficiency and prediction accuracy of the prediction model.

[0073] 5. First divide the image into several small grids according to pixels, then calculate and determine whether the small grid is a tumor area to determine the gastric cancer area and the edge position. Further, according to the area of the gastric cancer image, select the gastric cancer image area that needs to be calculated. Under the condition of ensuring the accuracy rate, select an appropriate data processing range to reduce unnecessary data processing, thereby improving the efficiency of gastric cancer image processing.

[0074] 6. By analyzing individual data, analyzing the weights of individual data, and combining with the development of individual data, predict the individual's gastric cancer lesion rate, which can more accurately achieve individual prediction. Compared with the average lesion rate of the usual prediction, it can better reflect individualization, can accurately serve individual prediction, reduce the impact of individual differences on the prediction result, and can show the difference between the individual and the average prediction. Description of the Drawings

[0075] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention.

[0076] In the accompanying drawings:

[0077] Figure 1 is a flowchart of the steps of the image processing method of the present invention. Detailed implementation manners

[0078] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to explain and illustrate the present invention, and are not used to limit the present invention.

[0079] Embodiment 1: As Figure 1 shown, the present invention provides a technical solution, an AI prediction-based gastric cancer image processing method, including the following steps:

[0080] S1. Collect and process gastric tumor image data;

[0081] S2. Extract lesion features in the image;

[0082] S3. Build a prediction model based on AI technology;

[0083] S4. Train and validate the prediction model;

[0084] S5. Apply the prediction model to actual diagnosis;

[0085] S6. Deeply interpret and determine the prediction results.

[0086] Based on the above technical solution, in S1, it specifically includes collecting and obtaining gastric tumor image data, and processing the collected gastric tumor image data;

[0087] When collecting gastric tumor image data, the collected gastric tumor image data includes both diseased and non-diseased ones. The specific sources can come from medical institutions, clinical trials, and public information. Collect gastric cancer imaging data of different imaging modalities, specifically including data from pathological sections, ultrasound detection images, endoscopic examination images, and CT scan images;

[0088] When collecting gastric tumor image data, it is necessary to ensure that the collected images cover gastric cancer cases in different stages, different types, and different individuals.

[0089] Based on the above technical solution, in S1, when processing the collected gastric tumor image data, mainly perform image quality inspection, image enhancement processing, image data cleaning, image cropping processing, and normalization processing on the collected original gastric tumor image data, so that the processed image data can be used for the training of the subsequent prediction model, improving the training efficiency and prediction accuracy of the prediction model.

[0090] Therefore, a segmentation algorithm with dynamically assigned weights is proposed. The entire image containing gastric cancer images is divided into several small grids according to pixels, and a rectangular coordinate system is established on the entire image. Each small grid is represented by coordinates (x, y). Combining gradient and texture features, the segmentation result formula of the gastric cancer image is as follows:

[0091]

[0092] Where:

[0093] f seg (x, y) is the final segmentation result, which represents the segmentation result of the image at the position (x, y), usually 0 or 1, where 0 is the non-tumor area and 1 is the tumor area, and the position where 0 and 1 alternate is the edge position of the tumor area;

[0094] is the gradient magnitude of the pixel, that is, the edge intensity;

[0095] T(I(x, y)) is the texture response of the pixel, including GLCM contrast and entropy;

[0096] w g and w t are adaptive weights, dynamically adjusted to optimize the segmentation effect, adaptively weighing the gradient and texture features, and can accurately segment the boundary of gastric tumors;

[0097]

[0098] w t = 1 - w g

[0099] Where:

[0100] k is a parameter that controls the speed of weight adjustment;

[0101] u is a threshold used to determine the change level of the gradient. Usually, u is the global mean of the gradient;

[0102] The weights w g and w t are dynamically calculated according to the local gradient information and texture information of the image, so that the importance of the gradient and texture in the segmentation process can be adjusted adaptively. When the edge of the image area is very obvious, the gradient weight wg It will become larger, making the segmentation more dependent on edge information. Conversely, in areas with blurred edges, texture information will be given more weight;

[0103] After calculating the segmentation result f of each small grid, determine the gastric cancer area in the entire image. First, analyze the gastric cancer lesion situation in the entire image, denoted as h 1 ;

[0104] Then determine the area of the gastric cancer area in the entire image. Then, based on the regional boundary and central area of the gastric cancer image, select different segmentation positions, and analyze the gastric cancer lesion situation according to the data of the segmented gastric cancer image, denoted as h 2 ;

[0105] According to the above analysis results, evaluate the situation of gastric cancer image segmentation. The specific evaluation formula is as follows:

[0106]

[0107] Among them:

[0108] Q represents the evaluation result. The smaller the value, the better the evaluation result;

[0109] S 整 represents the area of the entire image containing the gastric cancer image;

[0110] S 总 represents the total area of the gastric cancer area in S 整 ;

[0111] S (f+n1) represents that when S 总 occupies S 整 less than or equal to 50%, the edge of the gastric cancer area takes an additional n 1 small grids;

[0112] S (f+n2) represents that when S 总 occupies S 整 greater than 50%, the edge of the gastric cancer area takes n 2 small grids inward and outward respectively;

[0113] S 中心 represents that when S 总 occupies S 整 greater than 50%, the area of the selected gastric cancer central area, and S 中心 = 0.5S 总 ;

[0114] v represents the image processing speed per unit area;

[0115] h 1 represents the degree of gastric cancer carcinogenesis analyzed through the entire gastric cancer image;

[0116] h 2 represents the degree of gastric cancer carcinogenesis analyzed based on the segmented regions after image segmentation;

[0117] At S 总 ≤0.5S 整 when, first select the gastric cancer image region in the whole image, and then take n more grids outward from the edge position of the gastric cancer image region 1 = 1, expand the selection range of the gastric cancer image region, first calculate h 2 , and then calculate the Q value. Take n 1 = 2, n 1 = 3, n 1 = 4...... Calculate the corresponding h 2 and Q value. When h 1 = h 2 , it means that the segmented region can achieve the same analysis result as the whole image. Select the data with the smallest evaluation result Q and the corresponding n 1 value as the reference data for subsequent gastric cancer image segmentation, so as to help select a smaller gastric cancer image region, increase the efficiency of image processing, and achieve the same analysis effect at the same time;

[0118] At S 总 > 0.5S 整 when, first select the central region of the gastric cancer image region in the whole image, and then take 1 grid more outward and 1 grid more inward from the edge position of the gastric cancer image region, that is, n 2 = ±1, form an annular selection range at the edge of the gastric cancer image region. First calculate h 2 through the central region area and the annular selection range, and then calculate the Q value. Take n 2 = ±2, n 2 = ±3, n 2 = ±4...... Calculate the corresponding h 2 and Q value. When h 1 = h 2 , it means that the segmented region can achieve the same analysis result as the whole image. Select the data with the smallest evaluation result Q and the corresponding n 2 value as the reference data for subsequent gastric cancer image segmentation, so as to help select a smaller gastric cancer image region, increase the efficiency of image processing, and achieve the same analysis effect at the same time;

[0119] Through AI, select gastric cancer images for processing in the overall image, and judge the degree of gastric cancer lesions through big data. The judgment result is h 1 , and then predict the subsequent lesion rate under normal circumstances according to the average gastric cancer lesion data. The prediction result is P平均 ;

[0120] The lesions of gastric cancer are affected by individual factors, specifically including genetic susceptibility, molecular characteristics, tumor stage, lifestyle, immune status, age, and psychological state. According to the differences in individual factors, the lesion rate of individual data is further calculated, and the calculation formula is as follows:

[0121]

[0122] Where:

[0123] P 个体 represents the individual gastric cancer lesion rate;

[0124] A i represents the degree of development of an influencing factor of individual gastric cancer lesions, with a value range of (0, 1). The value of A i is obtained by experts scoring based on individual data. The larger the value, the deeper the degree of development;

[0125] α i represents the weight of A i with a value range of (0, 1). α i is obtained through statistical analysis training and determines the contribution degree of each factor to the prediction result;

[0126] By analyzing the individual data A i and adding weights, the rate of individual gastric cancer lesions is comprehensively calculated and predicted. Compared with the prediction using average data under normal circumstances, it is more in line with the individual development situation, can predict the individual situation more accurately, and through the comparison of the data of P 个体 and P 平均 , the lesion rate of the individual relative to the average value can be seen, and the individual lesion situation can be understood more comprehensively.

[0127] Based on the above technical solution, in S1, when performing image quality inspection, it aims to remove artifacts in the image, adjust brightness and contrast;

[0128] When performing image enhancement processing, it aims to perform flipping and rotation operations on the image data to adapt to the image processing ability of the subsequent prediction model and enhance the image data;

[0129] When performing image data cleaning, it aims to remove noise and unnecessary information in the image to ensure image quality;

[0130] When performing image cropping processing, it aims to crop the acquired image data to adjust the size of the image data;

[0131] When normalizing an image, it aims to uniformly adjust the image format to reduce the differences between different image data.

[0132] Based on the above technical solution, in S2, when extracting the lesion features in the image, using deep learning technology indicates that the prediction model has high recognition accuracy. Useful lesion features are extracted from the gastric tumor image data, and the extracted lesion features are used as the key information for tumor recognition.

[0133] Specifically, the processed gastric cancer image data is input into a deep learning network to extract deep features, specifically including the morphological features, color features, and texture features of the tumor. Among them, the morphological features specifically include position information, shape information, and edge information.

[0134] Based on the above technical solution, in S3, by constructing a prediction model based on AI technology, the prediction model can distinguish normal gastric tissue from diseased gastric cancer tissue, realize the prediction and diagnosis of early gastric cancer, and also include distinguishing the types of gastric cancer and segmenting and estimating the scope of canceration.

[0135] Based on the above technical solution, in S4, it specifically includes training and validating the constructed prediction model.

[0136] When training the prediction model, the extracted lesion features are used to train the prediction model. Through training, the prediction model can learn the ability to distinguish normal gastric tissue from diseased gastric cancer tissue, so as to ensure that the model can effectively identify and judge the gastric cancer lesion features in the lesion features.

[0137] When validating the prediction model, it is mainly through setting a control validation group to validate the constructed and trained prediction model, and through validation to evaluate the accuracy, stability, and generalization of the prediction model.

[0138] The control validation group specifically includes a validation set and a test set. Both the validation set and the test set are input into the prediction model, and the prediction model is used to identify the validation set and the test set. Among them, there are gastric cancer lesion features in the validation set, and there are some gastric cancer lesion features in the test set. Through validation, it is ensured that the prediction model can be widely recognized and applied.

[0139] Based on the above technical solution, in S4, after the prediction model is validated, when the prediction model accurately identifies that there are gastric cancer lesion features in the validation set and can identify that there are some gastric cancer lesion features in the test set, it indicates that the prediction model has high recognition accuracy and can be used for subsequent practical applications.

[0140] Based on the above technical solution, in S5, specifically apply the trained and verified prediction model to the actual diagnosis environment of gastric cancer, use the prediction model to perform feature recognition and judgment on the newly obtained images, and obtain the prediction results to help medical staff better understand the condition;

[0141] When the prediction model is specifically applied in practice, deploy the prediction model to the hospital's pathological diagnosis system, and use the prediction model to process the image data in the hospital's pathological diagnosis system in real time. Assist doctors in making early gastric cancer diagnosis and treatment decisions by identifying the image data.

[0142] Based on the above technical solution, in S6, it specifically refers to interpreting and determining the output result of the prediction model. Before determining the diagnosis result, it is necessary to combine the output result of the prediction model with the diagnosis opinions of clinical doctors with professional knowledge, and determine the final diagnosis result through combined comparative analysis.

[0143] Example 2: An AI-based gastric cancer image processing method includes the following steps:

[0144] S1. Collect and process gastric tumor image data;

[0145] S2. Extract the lesion features in the images;

[0146] S3. Build a prediction model based on AI technology;

[0147] S4. Train and verify the prediction model;

[0148] S5. Apply the prediction model to actual diagnosis;

[0149] S6. Deeply interpret and determine the prediction results.

[0150] Based on the above technical solution, in S1, it specifically includes collecting gastric tumor image data and processing the collected gastric tumor image data;

[0151] When collecting gastric tumor image data, the collected gastric tumor image data includes those with lesions and those without lesions. The specific sources can come from medical institutions, clinical trials, and public materials. Collect gastric cancer imaging data of different imaging modalities, specifically including data from pathological sections, ultrasound detection images, endoscopic examination images, and CT scan images;

[0152] When collecting gastric tumor image data, it is necessary to ensure that the collected images cover gastric cancer cases at different stages, of different types, and of different individuals.

[0153] Based on the above technical solution, in S1, when processing the collected gastric tumor image data, the original collected gastric tumor image data is mainly subjected to image quality inspection, image enhancement processing, image data cleaning, image cropping processing, and normalization processing, so that the processed image data can be used for the training of the subsequent prediction model, improving the training efficiency and prediction accuracy of the prediction model.

[0154] Based on the above technical solution, in S1, when performing image quality inspection, it aims to remove artifacts in the image, adjust brightness and contrast;

[0155] When performing image enhancement processing, it aims to perform flipping and rotation operations on the image data to adapt to the image processing capabilities of the subsequent prediction model, so as to enhance the image data;

[0156] When performing image data cleaning, it aims to remove noise and unnecessary information in the image to ensure image quality;

[0157] When performing image cropping processing, it aims to crop the obtained image data to adjust the size of the image data;

[0158] When performing normalization processing on the image, it aims to uniformly adjust the format of the image to reduce the differences between different image data.

[0159] Based on the above technical solution, in S2, when extracting lesion features in the image, using deep learning technology shows that the prediction model has high recognition accuracy. Useful lesion features are extracted from the gastric tumor image data, and the extracted lesion features are used as key information for tumor recognition;

[0160] Specifically, the processed gastric cancer image data is input into a deep learning network to extract deep features, specifically including morphological features, color features, and texture features of the tumor. Among them, the morphological features specifically include position information, shape information, and edge information.

[0161] Based on the above technical solution, in S3, by constructing a prediction model based on AI technology, the prediction model can distinguish normal gastric tissue from diseased gastric cancer tissue, realize the prediction and diagnosis of early gastric cancer, and also include differentiating the types of gastric cancer and segmenting and estimating the scope of canceration.

[0162] Based on the above technical solution, in S4, it specifically includes training and validating the constructed prediction model;

[0163] When training the prediction model, the extracted lesion features are used to train the prediction model. Through training, the prediction model can learn the ability to distinguish normal gastric tissue from diseased gastric cancer tissue, so as to ensure that the model can effectively identify and judge the gastric cancer lesion features in the lesion features.

[0164] When validating the prediction model, the constructed and trained prediction model is mainly validated by setting up a control validation group. Through validation, the accuracy, stability, and generalization of the prediction model are evaluated.

[0165] The control validation group specifically includes a validation set and a test set. Both the validation set and the test set are input into the prediction model, and the prediction model is used to identify the validation set and the test set. There are gastric cancer lesion features in the validation set, and there are some gastric cancer lesion features in the test set. Through validation, it is ensured that the prediction model can be widely recognized and applied.

[0166] Based on the above technical solution, in S4, after the prediction model is validated, when the prediction model fails to identify the gastric cancer lesion features in the validation set, it means that the prediction model cannot be used for subsequent actual applications. At this time, the prediction model needs to be trained continuously until the prediction model can give an accurate validation result.

[0167] Based on the above technical solution, in S5, the trained and validated prediction model is specifically applied to the actual diagnosis environment of gastric cancer. With the help of the prediction model, feature recognition and judgment are carried out on the newly obtained images, and prediction results are obtained to help medical staff better understand the condition.

[0168] When the prediction model is specifically applied, the prediction model is deployed to the hospital's pathological diagnosis system. The prediction model is used to process the image data in the hospital's pathological diagnosis system in real time, and by identifying the image data, it assists doctors in making early gastric cancer diagnosis and treatment decisions.

[0169] Based on the above technical solution, in S6, it specifically refers to interpreting and determining the output result of the prediction model. Before determining the diagnosis result, the output result of the prediction model needs to be combined with the diagnosis opinions of clinical doctors with professional knowledge, and the final diagnosis result is determined through combined comparative analysis.

[0170] Finally, it should be noted that the above are only preferred examples of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A gastric cancer image processing method based on AI prediction, characterized by: The steps include: S1. Collect and process gastric tumor image data; S2, extracting lesion features in the image; S3. Build a prediction model based on AI technology; S4, training and validating the prediction model; S5. Apply the prediction model to actual diagnosis; S6. Conduct in-depth interpretation and confirmation of the prediction results.

2. The method for gastric cancer image processing based on AI prediction according to claim 1, characterized in that: The S1 specifically includes collecting and acquiring gastric tumor image data, and processing the collected gastric tumor image data; When collecting gastric tumor image data, the collected gastric tumor image data includes lesions and non-lesions, and the specific sources can be from medical institutions, clinical trials and public information. Gastric cancer image data of different imaging methods are collected, including pathological sections, ultrasound detection images, endoscopic examination images and CT scan images; When collecting gastric tumor image data, it is necessary to ensure that the collected images cover gastric cancer conditions of different stages, types and individuals.

3. The gastric cancer image processing method based on AI prediction according to claim 2, characterized in that: In S1, when processing the collected gastric tumor image data, mainly image quality inspection, image enhancement processing, image data cleaning, image cropping processing and normalization processing are performed on the collected original gastric tumor image data, so that the processed image data can be used for subsequent prediction model training, thereby improving the training efficiency and prediction accuracy of the prediction model; Therefore, a segmentation algorithm with dynamic weight allocation is proposed. The entire image containing gastric cancer images is divided into several small grids according to pixels, and a rectangular coordinate system is established on the entire image. The coordinates are used to represent each small grid. Combining gradient and texture features, the segmentation result formula of gastric cancer images is as follows: in: f seg (x, y) is the final segmentation result, which represents the segmentation result of the image at the position (x, y), usually 0 or 1, where 0 is the tumor-free area and 1 is the tumor area. The position where 0 and 1 alternate is the edge position of the tumor area; is the gradient amplitude of the pixel, which is the edge strength; T(I(x, y)) is the texture response of the pixel, including GLCM contrast and entropy; w g and w t It is an adaptive weight that is dynamically adjusted to optimize the segmentation effect, adaptively weighing gradient and texture features, and can accurately segment the boundaries of gastric tumors; In t =1-in g in: k is a parameter that controls the speed of weight adjustment; u is a threshold used to determine the level of gradient change. Usually, u is the global mean of the gradient; Weight w g and w t The dynamic calculation is based on the local gradient information and texture information of the image, so that the importance of gradient and texture in the segmentation process can be adaptively adjusted. When the edge of the image area is very obvious, the gradient weight w g will become larger, making the segmentation more dependent on edge information. Conversely, in areas with blurred edges, texture information will receive more weight; After calculating the segmentation result f of each small grid, the gastric cancer area in the whole image is determined. The gastric cancer lesion is first analyzed through the whole image, which is recorded as h1; Then determine the area of ​​the gastric cancer region in the entire image, and then select different segmentation positions according to the regional boundary and central area of ​​the gastric cancer image, and analyze the gastric cancer lesion according to the segmented gastric cancer image data, which is recorded as h2; According to the above analysis results, the gastric cancer image segmentation is evaluated. The specific evaluation formula is as follows: in: Q represents the evaluation result. The smaller the value, the better the evaluation result. S 整 Represents the area of ​​the entire image containing the gastric cancer image; S 总 Indicates that in S 整 Total area of ​​the mid-gastric cancer region; S (f+n1) Indicates S 总 Occupies S 整 When it is less than or equal to 50%, the edge of the gastric cancer area has n1 more small grids outward; S (f+n2) Indicates S 总 Occupies S 整 When it is greater than 50%, the edge of the gastric cancer area is divided into n2 small grids inward and outward respectively; S 中心 Indicates that in S 总 Occupies S 整 When it is greater than 50%, the area of ​​the central region of gastric cancer is selected, and S 中心 =0.5S 总 ; v represents the image processing speed per unit area; h1 represents the degree of gastric cancer malignancy analyzed through the entire gastric cancer image; h2 represents the degree of gastric cancer malignancy analyzed based on the segmented areas after image segmentation; Select n1 and n2 values ​​in turn, calculate the value of h2, and calculate the corresponding evaluation result Q at the same time. When h1=h2, the segmented area can achieve the same analysis result as the whole image. Then select a group of data with the smallest evaluation result Q from these data as a reference for subsequent image segmentation. Through AI, gastric cancer images are selected for processing in the overall image, and the degree of gastric cancer lesions is judged through big data. The judgment result is h1. Then, the subsequent lesion rate under normal circumstances is predicted based on the average gastric cancer lesion data. The prediction result is P 平均 ; Gastric cancer lesions are affected by individual factors, including genetic susceptibility, molecular characteristics, tumor stage, lifestyle, immune status, age and psychological state. Based on the differences in individual factors, the lesion rate of individual data is further calculated. The calculation formula is as follows: in: P 个体 represents the individual gastric cancer lesion rate; A i It represents the development degree of an influencing factor of individual gastric cancer lesions, with a value of (0,1). i The value is obtained by experts through scoring individual data. The larger the value, the deeper the degree of development. α i Indicates A i The weight of is (0,1), α i It is obtained through training through statistical analysis and determines the contribution of each factor to the prediction results.

4. The method for gastric cancer image processing based on AI prediction according to claim 3, characterized in that: In S1, when performing image quality inspection, it aims to remove artifacts and adjust brightness and contrast in the image; When performing image enhancement processing, the image data is flipped and rotated to enhance the image data in order to adapt to the image processing capabilities of the subsequent prediction model; When performing image data cleaning, the aim is to remove noise and unnecessary information in the image to ensure image quality; When performing image cropping processing, the purpose is to crop the acquired image data to adjust the size of the image data; When normalizing an image, the goal is to uniformly adjust the format of the image to reduce the differences between different image data.

5. The gastric cancer image processing method based on AI prediction according to claim 1, characterized in that: In S2, when extracting the lesion features in the image, the deep learning technology is used to extract useful lesion features from the gastric tumor image data after the prediction model has a high recognition accuracy, and the extracted lesion features are used as key information for tumor recognition; Specifically, the processed gastric cancer image data is input into a deep learning network to extract deep features, including morphological features, color features, and texture features of the tumor, wherein the morphological features specifically include position information, shape information, and edge information.

6. The method for gastric cancer image processing based on AI prediction according to claim 1, characterized in that: In S3, a prediction model based on AI technology is constructed so that the prediction model can distinguish between normal gastric tissue and diseased gastric cancer tissue, thereby achieving early prediction and diagnosis of gastric cancer. It also includes distinguishing the types of gastric cancer and segmenting and estimating the range of canceration.

7. The gastric cancer image processing method based on AI prediction according to claim 1, characterized in that: The S4 specifically includes training and verifying the constructed prediction model; When training the prediction model, the extracted lesion features are used to train the prediction model. Through training, the prediction model can learn the ability to distinguish normal gastric tissue from lesion gastric cancer tissue, thereby ensuring that the model can effectively identify and judge the gastric cancer lesion features in the lesion features; When verifying the prediction model, the constructed and trained prediction model is mainly verified by setting up a control verification group, and the accuracy, stability and generalization of the prediction model are evaluated through verification; The control validation group specifically includes a validation set and a test set. Both the validation set and the test set are input into the prediction model, and the prediction model is used to identify the validation set and the test set. The validation set contains gastric cancer lesion features, and the test set contains some gastric cancer lesion features. Verification is used to ensure that the prediction model can be widely recognized and applied.

8. The method for gastric cancer image processing based on AI prediction according to claim 7, characterized in that: In S4, after the prediction model is verified, when the prediction model accurately identifies the gastric cancer lesion features in the verification set and can identify some gastric cancer lesion features in the test set, it means that the prediction model has a high recognition accuracy and can be used for subsequent practical applications; When the prediction model cannot identify the presence of gastric cancer lesion features in the validation set, it means that the prediction model cannot be used in subsequent practical applications. At this time, the prediction model needs to be further trained until the prediction model can provide accurate verification results.

9. The method for gastric cancer image processing based on AI prediction according to claim 1, characterized in that: In S5, the trained and verified prediction model is specifically applied to the actual diagnosis environment of gastric cancer, and the prediction model is used to perform feature recognition and judgment on the newly acquired image to obtain the prediction result, so as to help medical personnel better understand the condition; When the prediction model is put into practical application, it is deployed in the hospital's pathology diagnosis system. The prediction model is used to process the image data in the hospital's pathology diagnosis system in real time, and the image data is identified to assist doctors in making diagnosis and treatment decisions for early gastric cancer.

10. The gastric cancer image processing method based on AI prediction according to claim 1, characterized in that: In the above S6, it specifically refers to interpreting and determining the output results of the prediction model. Before determining the diagnosis result, it is necessary to combine the output results of the prediction model with the diagnosis opinions of clinicians with professional knowledge, and determine the final diagnosis result through combined comparative analysis.

Images