Artificial intelligence-based gastric cancer risk quantitative scoring method, system and equipment

Abstract

The invention relates to the technical field of artificial intelligence, and provides a gastric cancer risk quantitative scoring method, system and equipment based on artificial intelligence, and the method comprises the steps: recognizing the image type of each image frame in alimentary canal endoscope image data; for the electronic dyeing image frame, identifying a part contour region and an intestinal contour region of the feature part in the image frame, determining an intestinal epithelial metaplasia grading category of the feature part according to an area proportion of the intestinal contour region in the part contour region, and performing electronic dyeing intestinal scoring on the image data; for the white light image frame, gastroscope part types included in the image frame and focus area types of all gastroscope parts are recognized; performing atrophy scoring on the image data according to the position distribution of the focus area of the atrophy type; performing table state scoring on the image data according to whether the plica enlargement, nodular gastritis and diffuse redness focus areas exist or not; and realizing gastric cancer risk quantitative scoring according to the electronic staining intestinal scoring, the atrophy scoring and the epistatic state scoring.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method, system, and device for quantifying and scoring the risk of gastric cancer. Background Technology

[0002] Gastric cancer is a disease that develops on the basis of chronic gastritis. With continued inflammation, gene mutations accumulate to a certain level, leading to cancer. Improving the diagnosis and management of this disease is crucial for early screening and prevention of early-stage cancer. The Kyoto Gastritis Classification and Scoring System identifies risk factors for gastric cancer; a high score reflects a higher risk of Helicobacter pylori infection and gastric cancer, helping to identify high-risk individuals for gastric cancer and early-stage gastric cancer under white light endoscopy. However, the Kyoto Gastritis Classification and Scoring System does not consider the characteristics of stained images of the digestive tract. While the existing Endoscopic Grading of Gastric Intestinal Metaplasia (EGGIM) system can improve the accuracy of endoscopic intestinal metaplasia, the EGGIM score only focuses on assessing the degree of intestinal metaplasia under the mucosa in electronic chromoendoscopic images; a higher grade indicates a higher risk of gastric cancer, without considering other risk factors for gastric cancer. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide an artificial intelligence-based method, system and device for quantifying gastric cancer risk and overcoming the above problems.

[0004] One aspect of the present invention provides an artificial intelligence-based method for quantitative scoring of gastric cancer risk, the method comprising: Acquire gastrointestinal endoscopic imaging data and identify the image category of each image frame in the imaging data, including white light category and electron staining category; For the electronically stained image frames to be identified in the image data, the region contour area and intestinal metaplasia contour area of ​​the feature parts in each electronically stained image frame to be identified are identified, and the intestinal metaplasia grade of the feature parts is determined according to the area ratio of the intestinal metaplasia contour area in the region contour area; the feature parts include the lesser curvature of the antrum, the greater curvature of the antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body. Electronic staining of intestinal metaplasia data was performed to score the intestinal metaplasia based on the grading of intestinal metaplasia in each characteristic location. For the white light image frames to be identified in the image data, identify the category of gastroscopy site and the type of lesion area in each gastroscopy site in each white light image frame to be identified. The lesion area types include atrophic type, fold swelling type, nodular gastritis type and diffuse redness type. Atrophy scores were assigned to imaging data based on the location and distribution of atrophic lesions at different endoscopic sites. The representational status of imaging data was scored based on the presence of lesions with folded swelling, nodular gastritis, and diffuse redness. Gastric cancer risk is quantified based on electronic staining scores for intestinal metaplasia, atrophy, and phenotypic status from imaging data.

[0005] Another aspect of the present invention provides an artificial intelligence-based gastric cancer risk quantification scoring system, the system comprising functional modules for implementing the above-described artificial intelligence-based gastric cancer risk quantification scoring method, specifically, the system comprising: The image category recognition module is used to acquire gastrointestinal endoscopy image data and identify the image category of each image frame in the image data. The image categories include white light category and electron staining category. The intestinal metaplasia grading and recognition module for stained images is used to identify the feature regions and intestinal metaplasia contour regions of the feature regions in each stained image frame to be identified in the image data. The intestinal metaplasia grading category of the feature regions is determined according to the area ratio of the intestinal metaplasia contour region to the feature region contour region. The feature regions include the lesser curvature of the antrum, the greater curvature of the antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body. The electronic staining intestinal metaplasia scoring module is used to score the intestinal metaplasia of image data based on the grading of intestinal metaplasia in various characteristic sites. The white light image recognition module is used to identify the gastroscopy site category and the lesion area type in each gastroscopy site in the white light image frame to be identified in the image data. The lesion area types include atrophic type, fold swelling type, nodular gastritis type and diffuse redness type. The atrophy scoring module is used to score the atrophy of imaging data based on the location and distribution of lesion areas of atrophy type at different gastroscopic sites. The phenotypic status scoring module is used to score the phenotypic status of imaging data based on the presence of lesion areas with fold swelling, nodular gastritis, and diffuse redness. The risk quantification module is used to quantify the risk of gastric cancer based on the electronic staining intestinal morphology score, atrophy score, and phenotypic status score of the imaging data.

[0006] In another aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the artificial intelligence-based gastric cancer risk quantification scoring method described above.

[0007] The artificial intelligence-based gastric cancer risk quantification scoring method, system, and device provided in this invention integrate the features of white light images and electronic staining images, while considering the degree of inflammation, atrophy, intestinal metaplasia, and intestinal epithelial metaplasia. Through AI recognition, sub-scores such as atrophy score from endoscopic white light images, intestinal epithelial metaplasia score from endoscopic electronic staining images, fold swelling, nodular gastritis, and diffuse redness are obtained, constructing a quantitative scoring system for the risk of gastric cancer in patients. This system more comprehensively assists in the quantitative assessment of gastric cancer risk, effectively improving the accuracy of auxiliary diagnosis and quantitative assessment of gastritis and related lesions, and has significant clinical application value.

[0008] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0009] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a flowchart of an artificial intelligence-based gastric cancer risk quantification scoring method according to an embodiment of the present invention; Figure 2 A schematic diagram of image data showing the outline of the gastric angle and the intestinal metaplasia region; Figure 3 This is a schematic diagram of the structure of an artificial intelligence-based gastric cancer risk quantification scoring system according to an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined.

[0012] Example 1 This invention provides an artificial intelligence-based method for quantifying and scoring the risk of gastric cancer, such as... Figure 1 As shown, the gastric cancer risk quantification scoring method based on artificial intelligence proposed in this invention includes the following steps: S11. Acquire gastrointestinal endoscopic image data and identify the image category of each image frame in the image data. The image categories include white light category and electron staining category.

[0013] S12. For the electronically stained image frames to be identified in the image data, identify the feature region and intestinal metaplasia contour region in each electronically stained image frame, and determine the intestinal metaplasia grading category of the feature region based on the area ratio of the intestinal metaplasia contour region to the feature region contour region. The feature regions include the lesser curvature of the gastric antrum, the greater curvature of the gastric antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body.

[0014] S13. Perform electronic staining of intestinal metaplasia scores based on the grading of intestinal metaplasia in each characteristic location.

[0015] S14. For the white light image frames to be identified in the image data, identify the category of the gastroscopy site and the type of lesion area present in each gastroscopy site in each white light image frame to be identified. Among them, the lesion area types include atrophic type, fold swelling type, nodular gastritis type, and diffuse redness type.

[0016] S15. Atrophy scores are calculated based on the location and distribution of atrophic lesions at different endoscopic sites.

[0017] S16. The phenotypic status of imaging data is scored based on the presence of lesions with fold swelling, nodular gastritis, and diffuse redness.

[0018] S17. Quantify the risk of gastric cancer based on the electronic staining scores for intestinal metaplasia, atrophy, and phenotypic status of the imaging data.

[0019] The artificial intelligence-based gastric cancer risk quantification scoring method provided in this invention integrates the features of white light images and electronic staining images, while considering the degree of inflammation, atrophy, intestinal metaplasia, and intestinal epithelial metaplasia. Through AI intelligent recognition, it obtains sub-scores such as atrophy score from endoscopic white light images, intestinal epithelial metaplasia score from endoscopic electronic staining images, fold swelling, nodular gastritis, and diffuse redness, thus constructing a quantitative scoring system for gastric cancer risk. This system provides a more comprehensive assessment of gastric cancer risk and effectively improves the accuracy of auxiliary diagnosis and quantitative assessment of gastritis and related lesions, demonstrating significant clinical application value.

[0020] In this embodiment of the invention, step S11, which involves acquiring gastrointestinal endoscopic image data and identifying the image category of each image frame in the image data, includes: using a pre-trained gastrointestinal endoscopic staining image recognition model to identify the image category of each image frame in the image data.

[0021] Specifically, the training process for the gastrointestinal endoscopy staining image recognition model is as follows: Upper gastrointestinal endoscopic image data is collected to construct a training dataset, including images categorized as white light, electronic staining, canonical staining, and indigo carmine staining. The electronic staining data includes various staining modes from different endoscopic equipment manufacturers, such as NBI, BLI, and OE electronic staining. Using this training dataset, a classification neural network model for image categories is trained. This model takes a gastrointestinal endoscopy image as input and outputs the staining category of the image frame. Based on the staining category of the image frame, it determines whether the image frame to be identified in the real-time acquired gastrointestinal endoscopy image data is a white light image frame or an electronic staining image frame.

[0022] In this embodiment of the invention, step S12, identifying the feature contour regions and intestinal contour regions of each electronically stained image frame to be identified, includes: using a pre-trained part segmentation model to identify the feature contour regions and intestinal contour regions of each image frame.

[0023] Specifically, the training process for the site segmentation model is as follows: Image data of intestinal metaplasia and non-intestinal metaplasia observed by upper gastrointestinal chromoendoscopy are collected to construct a training dataset. This dataset includes images of five sites: the lesser curvature of the gastric antrum, the greater curvature of the gastric antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body. Labelme and other mapping tools are used to annotate the dataset, marking the contour regions of these feature sites in the images. For images containing intestinal metaplasia, the intestinal metaplasia contour region also needs to be marked. Figure 2 As shown, Figure 2 Image data was used to annotate the contour regions of the gastric angle and intestinal metaplasia. An instance segmentation model was trained using the aforementioned training dataset. This model takes an electronic chromoendoscopy image as input and outputs the contour regions of the featured areas. If intestinal metaplasia is present, it also outputs the intestinal metaplasia contour regions.

[0024] In this embodiment of the invention, step S12, which determines the intestinal metaplasia grading category of the characteristic site based on the area ratio of the intestinal metaplasia contour region to the site contour region, includes the following steps not shown in the accompanying drawings: S121. A list of contour points for the part contour region and a list of contour points for the intestinal contour region obtained by the contour detection algorithm.

[0025] S122. Calculate the area of ​​the feature part based on the list of outline points of the part outline region.

[0026] S123. Create a blank image. Based on the list of contour points of the part contour region and the list of contour points of the intestinal contour region, draw the part contour region and the intestinal contour region onto the blank image respectively, and fill them with different masks.

[0027] S124. Calculate the ratio of the area of ​​the mask overlap region to the area of ​​the feature region to obtain the area ratio of the intestinal metaplasia contour region in the feature region contour region. Based on the area ratio of the intestinal metaplasia contour region in the feature region contour region, determine the intestinal metaplasia classification of the feature region.

[0028] In one specific embodiment, the intestinal metaplasia grading category of the feature site is determined based on the area ratio of the intestinal metaplasia contour region to the site contour region. Specifically, if the area ratio is >30%, the intestinal metaplasia grading category of the feature site is determined to be 2; if the area ratio is <=30%, the intestinal metaplasia grading category of the feature site is determined to be 1; if no intestinal metaplasia contour region is identified, the intestinal metaplasia grading category of the feature site is determined to be 0.

[0029] Specifically, for images containing intestinal metaplasia, after obtaining the contour regions of the affected area and the intestinal metaplasia contour region, the area ratio of the intestinal metaplasia region within the corresponding contour region is determined. Taking the gastric angle region as an example, the area ratio of the intestinal metaplasia lesion within the gastric angle region is calculated. This involves first calculating the area of ​​the gastric angle in the image using the contour region of the gastric angle, and then calculating the area of ​​the overlapping region between the two. This can be implemented using the OpenCV and NumPy libraries. Further, the gastric angle contour region contour point list stomach_contours can be obtained through contour detection algorithms (such as cv2.findContours()), where each contour consists of multiple points. Then, cv2.contourArea(cnt) is used to calculate the geometric area of ​​a single contour, and sum(...) is used to sum the areas of all detected gastric angle contours to obtain the total area stomach_area of ​​the gastric angle in the image, as follows: stomach_area = sum(cv2.contourArea(cnt) for cnt in stomach_contours); Finally, a blank image is created to draw the overlapping region. OpenCV is used to draw the outline of the gastric angle region, and then the outline of the intestinal metaplasia lesion is drawn. Different masks are used for filling; for example, the gastric angle region outline is filled with 1, and the intestinal metaplasia lesion is filled with 2. The masks are added together to obtain the overlapping region. NumPy is then used to count the number of pixels in the overlapping region, thus calculating the area of ​​the overlapping region. The details are as follows: overlap_area = np.sum((overlap_mask == 3).astype(np.uint8)); Finally, the percentage of intestinal metaplasia lesions within the gastric angle region was calculated: percentage = (overlap_area / stomach_area) * 100.

[0030] If intestinal metaplasia lesions are present and the area of ​​the intestinal metaplasia outline region in the gastric angle region is greater than 30%, then the intestinal metaplasia classification at the gastric angle region is determined to be 2. If intestinal metaplasia lesions are found and the area of ​​the intestinal metaplasia outline region in the gastric angle region is less than or equal to 30%, then the intestinal metaplasia classification at the gastric angle region is determined to be 1. If there are no intestinal metaplasia lesions, then the intestinal metaplasia classification at the gastric angle region is determined to be 0.

[0031] The above method enables the input of an endoscopic electronic staining image to obtain the characteristic regions of the image and the corresponding intestinal metaplasia grading category of the characteristic regions.

[0032] To accurately identify characteristic sites and their intestinal metaplasia grading categories from real-time acquired electron chromatogram (ECC) images during endoscopic examinations, it is essential to ensure that the ECC images to be identified are compliant and normal. However, during endoscopic examinations, issues such as severe image distortion due to endoscopic angle, excessively far or close lens distance, and incomplete site imaging are unavoidable. Without proper handling, these issues can lead to inaccurate intestinal metaplasia scoring. To address these problems, this invention, after determining the intestinal metaplasia grading category of the characteristic site based on the area ratio of the intestinal metaplasia contour region to the site contour region, further includes: extracting image features from the current ECC image frame to be identified, and determining whether the current ECC image is an abnormal image based on the extracted image features. If the current ECC image is determined to be an abnormal image, it is marked as invalid, meaning the calculation result of this frame is invalid and does not participate in the final ECC intestinal metaplasia scoring calculation. Otherwise, the current ECC image is determined to be a normal image, and the calculation result of this frame participates in the final ECC intestinal metaplasia scoring calculation.

[0033] Furthermore, in this embodiment, image features are extracted from the current electron staining image frame to be identified, and it is determined whether the current electron staining image to be identified is an abnormal image based on the extracted image features. Specifically, this includes the following steps (not shown in the accompanying drawings): S21. Convert the current electronic staining image frame to be identified into a grayscale image and scale it to a preset image size to obtain the target image. Calculate the average grayscale value of all pixels in the target image. Compare the grayscale value of each pixel in the target image with the average value. If the grayscale value of a pixel is greater than or equal to the average value, use the first preset value as the pixel feature of the current pixel. Otherwise, use the second preset value as the pixel feature of the current pixel. Combine the pixel features of all pixels in the target image into a target feature string. Use the target feature string obtained by the above-mentioned full-image mean hash algorithm as the target image feature of the current electronic staining image frame to be identified.

[0034] S22. Search for the standard image feature most similar to the target image feature from the preset standard part image feature database. The standard part image feature database includes standard image features corresponding to standard images of various feature parts belonging to different intestinal metaplasia grading categories. S23. Determine whether the feature region of the standard image corresponding to the most similar standard image feature found is consistent with the feature region of the electronic staining image frame to be identified. If they are inconsistent, the electronic staining image to be identified is determined to be an abnormal image.

[0035] The standard site image feature database needs to be pre-constructed. Constructing this database specifically involves acquiring historical electronically stained image frames from past gastrointestinal endoscopic examinations, including standard images of various feature sites belonging to different intestinal metaplasia grades. These historical electronically stained image frames contain standard images of five key feature sites: the lesser curvature of the antrum, the greater curvature of the antrum, the gastric angle, the lesser curvature of the body, and the greater curvature of the body, including images of intestinal metaplasia grades 0, 1, and 2. Each historical electron staining image frame is converted into a grayscale image and scaled to a preset image size to obtain a standard target image. The standard average value of grayscale values ​​of all pixels in the standard target image is calculated. The grayscale value of each pixel in the standard target image is compared with the standard average value. If the grayscale value of a pixel is greater than or equal to the standard average value, the first preset value is used as the pixel feature of the current pixel; otherwise, the second preset value is used as the pixel feature of the current pixel. The pixel features of all pixels in the standard target image are combined into a standard feature string. The standard feature string obtained by the above-mentioned full-image mean hashing algorithm is used as the standard image feature corresponding to the standard image of the feature part included in the current historical electron staining image frame.

[0036] In a specific example, the preset image size can be 8×8, with the first preset value being 1 and the second preset value being 0. Specifically, the acquired historical electronic staining image frames are converted into grayscale images, then scaled to an 8×8 image size. The average grayscale value of all 64 pixels is calculated as the standard average. The grayscale value of each pixel is compared with the average grayscale value. If it is greater than or equal to the average grayscale value, the pixel feature of the current pixel is recorded as 1; if it is less than the average grayscale value, the pixel feature of the current pixel is recorded as 0. The pixel features corresponding to the 64 pixel values ​​are combined to form a 64-bit integer, which serves as the standard feature string based on the full-image mean hash algorithm. This standard feature string is stored in the database.

[0037] In practical applications, the same method described above is used to calculate the feature string based on the full-image mean hash algorithm for the electronic staining image frame to be identified. This string is then compared with the standard image features of all corresponding categories in the database. Specifically, the standard image feature with the highest similarity can be extracted by calculating the Hamming distance. The feature region of the standard image corresponding to the most similar standard image feature is then determined to be consistent with the feature region of the electronic staining image frame to be identified. If they are inconsistent, the electronic staining image to be identified is determined to be an abnormal image. If they are consistent, the electronic staining image to be identified is a compliant and normal image.

[0038] In this embodiment of the invention, step S13, which involves scoring the intestinal metaplasia of the image data based on the intestinal metaplasia grading category of each feature region, includes: obtaining the highest score of the intestinal metaplasia grading category for each feature region as the final score for the corresponding feature region; summing the final scores for the five feature regions to obtain a total score; if the total score is greater than or equal to a preset intestinal metaplasia scoring threshold, the intestinal metaplasia score of the image data is determined to be 2 points; if the total score is less than the intestinal metaplasia scoring threshold but greater than 0 points, the intestinal metaplasia score of the image data is determined to be 1 point; and if the total score is equal to 0 points, the intestinal metaplasia score of the image data is determined to be 0 points. In a specific example, the intestinal metaplasia scoring threshold can be selected as 5 points. When the intestinal metaplasia grading category is 2, the intestinal metaplasia grading score is 2 points; if the area percentage is <= 30%, the intestinal metaplasia grading category of the feature region is determined to be 1, and the intestinal metaplasia grading score is 1 point; if no intestinal metaplasia contour area is identified, the intestinal metaplasia grading category of the feature region is determined to be 0. The final scores of the five feature sites are summed to obtain the total score. If the total score is greater than or equal to 5 points, the electronic staining score of the intestinal morphology of the image data is 2 points. If the total score is less than 5 points but greater than 0 points, the electronic staining score of the intestinal morphology of the image data is 1 point. If the total score is equal to 0 points, the electronic staining score of the intestinal morphology of the image data is 0 points.

[0039] Specifically, the system identifies endoscopic image data in real time and processes images from electronic staining endoscopy models such as NBI, BLI, and OE. The process is as follows: Each frame of the video is analyzed, and a pre-trained digestive endoscopy staining image recognition model is invoked to verify whether the image frame is an electronic staining category image. If it is electronically stained, a pre-trained site segmentation model is invoked to identify the contour regions of characteristic sites and intestinal metaplasia contour regions in each image frame. Further processing yields the recognition result for each frame, namely, five evaluation sites and their corresponding intestinal metaplasia grading categories. Then, based on a standard site image feature database, it is determined whether the image frame is an abnormal image frame containing non-standard images of characteristic sites. If the site image in an image frame is not a standard image site, the calculation result for that frame is invalid. Finally, for each feature region, the highest category of the intestinal metaplasia grading in the image is taken as the score for that region. The scores of the five regions are then summed up. When the total score is ≥5, the EGGIM score for this examination is 2 points. When the total score is less than 5 but greater than 0, the EGGIM score for this examination is 1 point. When the total score is less than 1, the EGGIM score for this examination is 0 points.

[0040] In this embodiment of the invention, step S14, identifying the gastroscopy site category included in each white light image frame to be identified, specifically includes: using a pre-trained gastroscopy site recognition model to identify the gastroscopy site category included in each image frame.

[0041] Specifically, the training process of the gastroscopy site recognition model is as follows: A training dataset is constructed by collecting upper gastrointestinal endoscopic image data. The gastroscopy site is divided into 30 sites: pharynx, upper esophagus, middle esophagus, lower esophagus, cardia, lesser curvature of the gastric fundus, upper inverted lesser curvature of the gastric body, upper lesser curvature of the gastric body, lower lesser curvature of the gastric body, anterior wall of the gastric fundus, posterior wall of the gastric fundus, greater curvature of the gastric fundus, upper inverted anterior wall of the gastric body, upper inverted posterior wall of the gastric body, upper greater curvature of the gastric body, upper anterior wall of the gastric body, upper posterior wall of the gastric body, lower anterior wall of the gastric body, lower posterior wall of the gastric body, lower greater curvature of the gastric body, anterior angle of the gastric angle, middle angle of the gastric angle, posterior angle of the gastric angle, lesser curvature of the gastric antrum, greater curvature of the gastric antrum, anterior and posterior angle of the gastric antrum, pylorus, duodenal bulb, and descending duodenum. The images in the training dataset are labeled with gastroscopy site categories. Based on the above training dataset, a classification neural network model for site recognition is trained to output the gastroscopy site categories included in the image from a given digestive endoscopy image.

[0042] In this embodiment of the invention, step S14, identifying the types of lesion regions present in the gastroscopy site included in each white light image frame to be identified, includes: using a pre-trained lesion region segmentation model to identify the types of lesion regions present in the gastroscopy site included in the image frame.

[0043] Specifically, the training process for the lesion region segmentation model is as follows: A training dataset is constructed by collecting upper gastrointestinal endoscopic images, including case data of non-atrophic gastritis and atrophic gastritis. The boundaries of lesion regions in the images are delineated, including atrophic areas, enlarged folds, nodular gastritis areas, and diffuse reddened areas, thus labeling the contours and types of lesion regions in the training data. Based on the above training dataset, the lesion region segmentation model is trained, enabling it to derive the contours and types of different lesion regions from a single white light endoscopic image.

[0044] In this embodiment of the invention, the atrophy score of the image data is determined based on the location distribution of the atrophic lesion area at different gastroscopic sites. This includes: performing Kimura Takemoto classification on the image data according to the location of the atrophic lesion area at different gastroscopic sites; if no atrophic lesion area is identified or the Kimura Takemoto classification is C1, the atrophy score of the image data is determined to be 0 points; if the Kimura Takemoto classification is C2 or C3, the atrophy score of the image data is determined to be 1 point; if the Kimura Takemoto classification is O1, O2, or O3, the atrophy score of the image data is determined to be 2 points.

[0045] Specifically, the system uses real-time identification of endoscopic video data and AI to identify atrophic gastritis in images under a white light endoscopy model. Based on the Kimura Takemoto classification rules, an atrophy score is derived. The process involves analyzing each frame of the video, calling a digestive endoscopy staining image recognition model to verify if the image frame is a white light category image. If so, a gastroscopy site recognition model is used to identify the site type. Simultaneously, a lesion region segmentation model is used to identify whether each frame contains atrophic lesions. Each frame will then determine the location and whether atrophy is present. For example, it might identify the current location as the gastric angle with atrophic areas. The Kimura Takemoto classification divides atrophic gastritis into six severity levels based on the extent of spread at the gastroscopy site: C1, C2, C3, O1, O2, and O3. The Kimura Takemoto classification criteria are shown in Table 1. Table 1 Kimura Takemoto's Classification Standards Kimura Takemoto Type Part C1 Anterior and posterior gastric antrums, lesser curvature and greater curvature of the gastric antrum C2 Anterior gastric angle, middle gastric angle, posterior gastric angle C3 mid-upper inverted minor curvature O1 lesser curvature of the stomach O2 greater curvature of the gastric fundus, anterior wall of the gastric fundus, posterior wall of the gastric fundus O3 The upper and middle major curves of the body and the lower major curvature of the body If atrophic gastritis is identified in one of the locations corresponding to Kimura's C1 classification, the result is C1. For example, if three images are captured in the video frame of the pre-antral area, and chronic atrophic gastritis is identified in one of them, then the pre-antral area is classified as having chronic atrophic gastritis, and the Kimura classification is C1. If chronic atrophic gastritis is also identified in the location corresponding to C2 when C1 is met, then the Kimura classification is C2, and so on. The atrophy score is summarized from the identification results of the entire examination video, and the atrophy scoring criteria are shown in Table 2. Table 2 Atrophy Scoring Criteria atrophy score Fraction No atrophic gastritis, C1 0 points C2, C3 1 point O1, O2, O3 2 points In this embodiment of the invention, step S16 involves scoring the representational status of image data based on the presence or absence of lesion areas of the folded swelling type, nodular gastritis type, and diffuse redness type. Specifically, this includes: calling a lesion area segmentation model to identify whether each frame of the image has lesion areas such as folded swelling, nodular gastritis, and diffuse redness, and then assigning a risk score. No swelling of folds was found during examination: 0 points; Examination revealed enlarged folds: 1 point; Examination revealed no nodular gastritis: 0 points; The examination revealed nodular gastritis: 1 point; No diffuse redness detected: 0 points; The examination revealed diffuse redness: 1 point.

[0046] In this embodiment of the invention, step S17 involves a quantitative assessment of gastric cancer risk based on the electronic staining intestinal metaplasia score, atrophy score, and phenotypic status score from the imaging data. Specifically, this includes combining the electronic staining intestinal metaplasia score, the atrophy score based on the Kimura Takemoto classification, the fold swelling score, the nodular gastritis score, and the diffuse redness score to summarize five sub-scoring modules. The specific quantitative scoring standards are detailed in Table 3. The quantitative score for the risk of gastric cancer in patients with gastritis ranges from 0 to 7 points, with scores above 4 indicating high risk and scores below 4 indicating low risk.

[0047] Table 3 Quantitative Scoring Criteria Sub-scoring module score atrophy score 0~2 points Electronic staining EGGIM score 0~2 points Swelling of folds score 0~1 point Nodular gastritis score 0~1 point Diffuse redness score 0~1 point This invention presents an artificial intelligence-based method for quantifying gastric cancer risk. It combines the Kyoto Gastritis Classification Score and the EGGIM scoring system, integrating features from white light images and electronic staining images. It also considers the degree of inflammation, atrophy, intestinal metaplasia, and intestinal epithelial metaplasia to construct an artificial intelligence-based gastric cancer risk quantification scoring system. This system provides a more comprehensive assessment of the patient's condition and aims to improve the accuracy of diagnosis and assessment of gastritis and related lesions, thus having significant clinical application value.

[0048] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0049] Example 2 Another embodiment of the present invention provides an artificial intelligence-based gastric cancer risk quantification scoring system, the system including a functional module for implementing the artificial intelligence-based gastric cancer risk quantification scoring method as described in any of the preceding claims. Figure 3The schematic diagram illustrates the structure of an artificial intelligence-based gastric cancer risk quantification scoring system provided by an embodiment of the present invention. (Refer to...) Figure 3 An embodiment of the present invention provides an artificial intelligence-based gastric cancer risk quantification scoring system, specifically comprising an image category recognition module 301, a stained image intestinal metaplasia grading recognition module 302, an electronic staining intestinal metaplasia scoring module 303, a white light image recognition module 304, an atrophy scoring module 305, a phenotypic status scoring module 306, and a risk quantification module 307, wherein: The image category recognition module 301 is used to acquire gastrointestinal endoscopy image data and identify the image category of each image frame in the image data. The image categories include white light category and electron staining category. The intestinal metaplasia grading recognition module 302 for stained image is used to identify the feature region and intestinal metaplasia contour region of each electronically stained image frame to be identified in the image data, and to determine the intestinal metaplasia grading category of the feature region based on the area ratio of the intestinal metaplasia contour region to the feature region contour region; the feature regions include the lesser curvature of the gastric antrum, the greater curvature of the gastric antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body; The electronic staining intestinal metaplasia scoring module 303 is used to score the intestinal metaplasia of image data based on the grading of intestinal metaplasia in various characteristic sites. The white light image recognition module 304 is used to identify the gastroscopy site category and the lesion area type in each gastroscopy site in the white light image frame to be identified in the image data. The lesion area types include atrophic type, fold swelling type, nodular gastritis type and diffuse redness type. The atrophy scoring module 305 is used to score the atrophy of imaging data based on the location and distribution of lesion areas of atrophy type at different gastroscopic sites. The phenotypic status scoring module 306 is used to score the phenotypic status of imaging data based on the presence or absence of lesion areas with fold swelling, nodular gastritis, and diffuse redness. The risk quantification module 307 is used to quantify the risk of gastric cancer based on the electronic staining intestinal metaplasia score, atrophy score, and phenotypic status score of the imaging data.

[0050] In this embodiment of the invention, the system further includes an abnormal image judgment module (not shown in the accompanying drawings). The abnormal image judgment module is used to extract image features from the current electron staining image frame to be identified after the intestinal metaplasia grading recognition module determines the intestinal metaplasia grading category of the feature site based on the area ratio of the intestinal metaplasia contour region to the site contour region, and judges whether the current electron staining image to be identified is an abnormal image based on the extracted image features. If the current electron staining image to be identified is determined to be an abnormal image, the current electron staining image frame to be identified is marked as invalid.

[0051] In the specific implementation process of Embodiment 2, you can refer to Embodiment 1, and it has the corresponding technical effects.

[0052] Example 3 This invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the various embodiments of the artificial intelligence-based gastric cancer risk quantification scoring method. For example... Figure 1 Steps S11-S17 are shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above embodiments of the artificial intelligence-based gastric cancer risk quantification scoring system, for example... Figure 3 The modules shown are: image category recognition module 301, stained image intestinal metaplasia grading recognition module 302, electronic staining intestinal metaplasia scoring module 303, white light image recognition module 304, atrophy scoring module 305, phenotypic status scoring module 306, and risk quantification module 307.

[0053] In its specific implementation, Example 3 can be referred to Example 1 and has the corresponding technical effects. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, any of the claimed embodiments can be used in any combination.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based method for quantitative scoring of gastric cancer risk, characterized in that, The method includes: Acquire gastrointestinal endoscopic imaging data and identify the image category of each image frame in the imaging data, including white light category and electron staining category; For the electronically stained image frames to be identified in the image data, the region contour area and intestinal metaplasia contour area of ​​the feature parts in each electronically stained image frame to be identified are identified, and the intestinal metaplasia grade of the feature parts is determined according to the area ratio of the intestinal metaplasia contour area in the region contour area; the feature parts include the lesser curvature of the antrum, the greater curvature of the antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body. Electronic staining of intestinal metaplasia data was performed to score the intestinal metaplasia based on the grading of intestinal metaplasia in each characteristic location. For the white light image frames to be identified in the image data, identify the category of gastroscopy site and the type of lesion area in each gastroscopy site in each white light image frame to be identified. The lesion area types include atrophic type, fold swelling type, nodular gastritis type and diffuse redness type. Atrophy scores were assigned to imaging data based on the location and distribution of atrophic lesions at different endoscopic sites. The representational status of imaging data was scored based on the presence of lesions with folded swelling, nodular gastritis, and diffuse redness. Gastric cancer risk is quantified based on electronic staining scores for intestinal metaplasia, atrophy, and phenotypic status from imaging data.

2. The method according to claim 1, characterized in that, The determination of the intestinal metaplasia grading category of a characteristic site based on the area ratio of the intestinal metaplasia contour region to the site contour region includes: The list of contour points for the part contour region and the list of contour points for the intestinal contour region obtained by the contour detection algorithm. Calculate the area of ​​the feature region based on the list of contour points of the region. Create a blank image. Based on the list of contour points for the part contour region and the list of contour points for the intestinal transformation contour region, draw the part contour region and the intestinal transformation contour region onto the blank image respectively, and fill them with different masks. The ratio of the area of ​​the mask overlap region to the area of ​​the feature region is calculated to obtain the area ratio of the intestinal metaplasia contour region in the feature region contour region. Based on the area ratio of the intestinal metaplasia contour region in the feature region contour region, the intestinal metaplasia classification of the feature region is determined.

3. The method according to claim 1 or 2, characterized in that, After determining the intestinal metaplasia grading category of a characteristic site based on the area ratio of the intestinal metaplasia contour region to the site contour region, the method further includes: Image features are extracted from the current electron staining image frame to be identified, and the extracted image features are used to determine whether the current electron staining image to be identified is an abnormal image. If the current electron staining image to be identified is determined to be an abnormal image, the current electron staining image frame to be identified is marked as invalid.

4. The method according to claim 3, characterized in that, Image features are extracted from the current electron staining image frame to be identified, and the extracted image features are used to determine whether the current electron staining image to be identified is an anomalous image, including: The current electronic staining image frame to be identified is converted into a grayscale image and scaled to a preset image size to obtain the target image. The average grayscale value of all pixels in the target image is calculated. The grayscale value of each pixel in the target image is compared with the average value. If the grayscale value of a pixel is greater than or equal to the average value, the first preset value is used as the pixel feature of the current pixel. Otherwise, the second preset value is used as the pixel feature of the current pixel. The pixel features of all pixels in the target image are combined into a target feature string. The target feature string is used as the target image feature of the current electronic staining image frame to be identified. The standard image feature that is most similar to the target image feature is searched from the preset standard site image feature database. The standard site image feature database includes standard image features corresponding to standard images of various feature sites belonging to different intestinal metaplasia grading categories. Determine whether the feature region of the standard image corresponding to the most similar standard image feature found is consistent with the feature region of the electronic staining image frame to be identified. If they are inconsistent, the electronic staining image to be identified is determined to be an abnormal image.

5. The method according to claim 4, characterized in that, The method further includes: pre-constructing a standard body part image feature database; Construct a database of standard body part image features, including: Acquire historical electronically stained image frames, including standard images of various characteristic sites belonging to different grades of intestinal metaplasia, collected during historical gastrointestinal endoscopy examinations. Each historical electron staining image frame is converted into a grayscale image and scaled to a preset image size to obtain a standard target image. The standard average value of grayscale values ​​of all pixels in the standard target image is calculated. The grayscale value of each pixel in the standard target image is compared with the standard average value. If the grayscale value of a pixel is greater than or equal to the standard average value, the first preset value is used as the pixel feature of the current pixel; otherwise, the second preset value is used as the pixel feature of the current pixel. The pixel features of all pixels in the standard target image are combined into a standard feature string. The standard feature string is used as the standard image feature corresponding to the standard image of the feature region included in the current historical electron staining image frame.

6. The method according to claim 2, characterized in that, The electronic staining scoring of intestinal metaplasia based on the grading of intestinal metaplasia in various characteristic sites includes: The highest score of the intestinal metaplasia grading category for each feature site is obtained as the final score for the corresponding feature site. The final scores of the five feature sites are summed to obtain the total score. If the total score is greater than or equal to the preset intestinal metaplasia scoring threshold, the electronic staining intestinal metaplasia score of the image data is determined to be 2 points. If the total score is less than the intestinal metaplasia scoring threshold but greater than 0 points, the electronic staining intestinal metaplasia score of the image data is determined to be 1 point. If the total score is equal to 0 points, the electronic staining intestinal metaplasia score of the image data is determined to be 0 points.

7. The method according to claim 1, characterized in that, Atrophy scoring was performed on imaging data based on the location and distribution of atrophic lesions at different endoscopic sites, including: Based on the location of the atrophic lesion area at different gastroscopic sites, the imaging data are classified using the Kimura Takemoto classification. If no atrophic lesion area is identified or the Kimura Takemoto classification is C1, the atrophy score of the imaging data is determined to be 0 points. If the Kimura Takemoto classification is C2 or C3, the atrophy score of the imaging data is determined to be 1 point. If the Kimura Takemoto classification is O1, O2, or O3, the atrophy score of the imaging data is determined to be 2 points.

8. An artificial intelligence-based gastric cancer risk quantification scoring system, characterized in that, The system includes: The image category recognition module is used to acquire gastrointestinal endoscopy image data and identify the image category of each image frame in the image data. The image categories include white light category and electron staining category. The intestinal metaplasia grading and recognition module for stained images is used to identify the feature regions and intestinal metaplasia contour regions of the feature regions in each stained image frame to be identified in the image data. The intestinal metaplasia grading category of the feature regions is determined according to the area ratio of the intestinal metaplasia contour region to the feature region contour region. The feature regions include the lesser curvature of the antrum, the greater curvature of the antrum, the gastric angle, the lesser curvature of the gastric body, and the greater curvature of the gastric body. The electronic staining intestinal metaplasia scoring module is used to score the intestinal metaplasia of image data based on the grading of intestinal metaplasia in various characteristic sites. The white light image recognition module is used to identify the gastroscopy site category and the lesion area type in each gastroscopy site in the white light image frame to be identified in the image data. The lesion area types include atrophic type, fold swelling type, nodular gastritis type and diffuse redness type. The atrophy scoring module is used to score the atrophy of imaging data based on the location and distribution of lesion areas of atrophy type at different gastroscopic sites. The phenotypic status scoring module is used to score the phenotypic status of imaging data based on the presence of lesion areas with fold swelling, nodular gastritis, and diffuse redness. The risk quantification module is used to quantify the risk of gastric cancer based on the electronic staining intestinal morphology score, atrophy score, and phenotypic status score of the imaging data.

9. The system according to claim 8, characterized in that, The system also includes: The abnormal image judgment module is used to extract image features from the current electron staining image frame to be identified after the intestinal metaplasia grading recognition module determines the intestinal metaplasia grading category of the feature site based on the area ratio of the intestinal metaplasia contour region in the site contour region. Based on the extracted image features, it judges whether the current electron staining image to be identified is an abnormal image. If the current electron staining image to be identified is determined to be an abnormal image, the current electron staining image frame to be identified is marked as invalid.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any one of claims 1-7.

Images