Digestive endoscopy image-assisted identification and grading method for inflammatory bowel disease

Abstract

The invention discloses an inflammatory bowel disease digestive endoscopy image auxiliary identification and grading method, and relates to the technical field of medical treatment, and the method comprises the following steps: collecting an intestinal endoscopy video of a patient, selecting an ulcer image frame for processing, and intercepting an ulcer core area; the ulcer area image is converted into numerical features, the disorder and non-uniformity degree of the texture is calculated, and the edema condition is quantified; a patient fills in clinical information, and the system automatically converts the data into a standard format; combining the texture features with clinical data, and endowing different features with different importance weights; inputting the fused comprehensive data into a trained model, automatically outputting severity levels of diseases by the model, and giving suggestions; conclusions and treatment suggestions are displayed to doctors in real time through a webpage, and a report can be generated. According to the method, the characteristics in the endoscope image are extracted and quantified, and the characteristics and the clinical parameters are subjected to fusion analysis, so that grading of the severity of the inflammatory bowel disease and treatment decision support are realized.

Description

Technical Field

[0001] This invention belongs to the field of medical technology, and in particular relates to a method for the identification and grading of inflammatory bowel disease using digestive endoscopy images. Background Technology

[0002] Inflammatory bowel disease (IBD) is a group of chronic, nonspecific, immune-mediated inflammatory diseases of the gastrointestinal tract with unknown etiologies. It mainly includes Crohn's disease (CD) and ulcerative colitis (UC). It is characterized by a prolonged and recurrent course, with patients often presenting with abdominal pain, persistent diarrhea, bloody stools, weight loss, and fatigue. This disease is not simply a bowel lesion; the inflammatory process can penetrate the intestinal wall and can be accompanied by various extraintestinal complications, severely impacting patients' quality of life and requiring long-term, even lifelong, medical intervention and management.

[0003] Endoscopic imaging refers to dynamic video or static images of the mucosa within the digestive tract obtained directly during examination using endoscopic equipment such as electronic colonoscopes and gastroscopes. It is an indispensable "gold standard" tool for diagnosing and assessing inflammatory bowel disease. Endoscopic images can directly display macroscopic morphological changes in the mucosa, such as the characteristics, extent, distribution, and severity of lesions like congestion, edema, erosion, ulceration, stricture, polyps, and pseudopolyps. This provides physicians with the most direct visual evidence for disease classification, activity scoring (such as CDEIS and Mayo scores), and treatment planning.

[0004] Currently, the clinical assessment of the severity of inflammatory bowel disease (such as Crohn's disease), especially during digestive endoscopy, relies heavily on the physician's visual observation and subjective experience. Physicians need to make a comprehensive interpretation based on the morphological manifestations of the mucosa, such as ulcers, edema, and erosions, in endoscopic images. This method is subject to significant subjective variability. Physicians with different levels of experience may have inconsistent judgments on the severity of the same lesion, and even the same physician may make inaccurate judgments at different time points. This subjective assessment method, which relies on qualitative or semi-quantitative descriptions, lacks objective and quantifiable unified standards, limiting the repeatability and accuracy of endoscopic grading results, and potentially affecting the scientific rigor and optimality of subsequent treatment strategies. Therefore, the following solutions are proposed to address these issues. Summary of the Invention

[0005] The purpose of this invention is to provide a method for the identification and grading of inflammatory bowel disease using digestive endoscopy images. By extracting and quantifying local texture heterogeneity features (HTF) in endoscopic images and performing multimodal fusion analysis with key clinical parameters, it is possible to grade the severity of inflammatory bowel disease and support treatment decisions. This solves the problems of poor consistency and low repeatability of existing endoscopic assessments due to reliance on physicians' subjective experience and lack of unified quantitative standards.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention relates to a method for the auxiliary identification and grading of inflammatory bowel disease using digestive endoscopy images. The method specifically includes the following steps: Step S1, Image Acquisition and Preprocessing: Acquire the patient's intestinal endoscopy video, select the clearest ulcer image frame for noise reduction and contrast enhancement, and crop the core area of ​​the ulcer. Step S2, Texture Feature Extraction: Convert the ulcer area image into numerical features, calculate the disorder and unevenness of its texture, and quantify the edema that is difficult to judge with the naked eye; Step S3, Clinical Parameter Entry: After scanning the QR code, the patient fills in clinical information, including erythrocyte sedimentation rate and frequency of bowel movements, on the webpage. The system automatically converts this data into a standard format. Step S4, Feature Fusion and Weighting: Combine the texture features obtained from image analysis with the clinical data filled in by the patient, and assign different importance weights to different features through an algorithm; Step S5, Intelligent Hierarchical Decision-Making: Input the fused comprehensive data into the trained model. The model automatically outputs the severity level of the disease and provides suggestions on whether biological agents should be used. Step S6, Results Feedback Display: The conclusions and treatment suggestions are displayed to the doctor in real time via a webpage, and a detailed report containing the diagnostic basis can be generated.

[0007] Furthermore, step S1, image acquisition preprocessing, specifically includes the following steps: Step S11: Acquire image data of the patient's colonic mucosa using a digestive endoscopy device in 1080p RGB video stream format, and select keyframes (including clear frames of the deepest ulcer lesions) as input; Step S12: Preprocess the keyframes using Gaussian filtering (standard deviation). Noise reduction was performed, and adaptive histogram equalization (CLAHE) was used to enhance mucosal texture contrast. The central region of the ulcer was cropped (radius). (pixels) are used as the region of interest (ROI); The purpose of this step is to obtain standardized images from the raw endoscopic video that can be used for analysis. It prepares high-quality, interference-free regions of interest (ROIs) for subsequent feature extraction by selecting keyframes containing the most severe lesions (deep ulcers) and processing them with noise reduction and contrast enhancement, thus ensuring the accuracy and consistency of the analysis basis.

[0008] Further, step S2, texture feature extraction, specifically includes the following steps: Step S21: Convert the ROI region to grayscale and calculate its gray-level co-occurrence matrix (GLCM), with the direction being... ,distance Pixel; Step S22: Calculate the 16-dimensional basic texture features in four directions: contrast, correlation, energy, and homogeneity; Step S23: Define the Local Texture Heterogeneity Feature (HTF) as follows: In the formula, This is a feature of local texture heterogeneity. The total number of pixels contained within the region of interest. For summation index variables, For the first The two-dimensional coordinates of each pixel in the image matrix Let a function represent the image in coordinates The pixel grayscale value at that location, The symbol is for partial differentials. In pixels A local window centered on (e.g.) The variance of all pixel grayscale values ​​within the window; Step S24: Concatenate the HTF with the 16-dimensional GLCM features to form a 17-dimensional texture feature vector; The purpose of this step is to quantify the microscopic morphological information in endoscopic images that is difficult for the human eye to interpret precisely. It calculates traditional texture features such as the gray-level co-occurrence matrix and combines them with the HTF formula to accurately characterize the heterogeneous changes such as edema and roughness of the mucosa around the ulcer, transforming subjective visual impressions into objective and quantifiable numerical features.

[0009] Furthermore, step S3, the clinical parameter entry, specifically includes the following steps: Step S31: The patient scans the QR code to access the H5 page and enters the following clinical parameters: Erythrocyte sedimentation rate (ESR, mm / h); C-reactive protein (CRP, mg / L); daily bowel movement frequency; ulcer depth (superficial / deep) in colonoscopy report; presence of intestinal stenosis (yes / no). Step S32: Perform Z-score normalization on continuous parameters (ESR, CRP, frequency of bowel movements): In the formula, These are standardized clinical parameter values. The original clinical parameter values ​​entered for the current patient. This is the arithmetic mean of all values ​​of a specific clinical parameter in the dataset used during model training. This represents the standard deviation of all values ​​of a specific clinical parameter in the dataset used for model training. The purpose of this step is to collect and standardize patients' clinical data for model analysis. It guides users to input key laboratory and examination indicators through a convenient QR code interface, and uses predefined statistical parameters (mean and standard deviation) to standardize these continuous variables, eliminating the influence of dimensions, so that they can be fused with image features on a unified scale.

[0010] Furthermore, step S4, feature fusion weighting, specifically includes the following steps: Step S41: Concatenate the 17-dimensional texture feature vector with the standardized clinical parameter vector to obtain the fused feature vector. (17-dimensional texture + 3-dimensional clinical parameters); Step S42: Weight the fused features using an attention mechanism: In the formula, This is the fused feature vector after being weighted by the attention mechanism. This is the original fused feature vector after concatenation. Multiplying corresponding elements of two vectors (or matrices) with the same dimension. It is a learnable weight matrix (the initial values ​​are fixed by the pre-trained model). For bias terms, Use the Sigmoid activation function; The purpose of this step is to effectively integrate information from different sources (image textures and clinical parameters). By using vector splicing and attention mechanisms for weighting, the model automatically learns and emphasizes feature combinations that are more critical to the final judgment, thereby achieving a "1+1>2" effect and improving the overall model's discriminative ability.

[0011] Furthermore, step S5, intelligent hierarchical decision-making, specifically includes the following steps: Will Input a pre-trained classification model (Support Vector Machine, SVM, kernel function RBF), and output two results: Disease severity: divided into mild (CDEIS<5), moderate (5≤CDEIS≤12), and severe (CDEIS>12); Upgrade criteria for biological agents: binary classification output (yes / no), the judgment criteria are as follows: In the formula, To predict the probability that the current patient has "severe" inflammatory bowel disease using a class model. To determine whether the model classifies patients as having "moderate" severity, The logical operator "AND". The raw C-reactive protein value entered for the patient. The specific scalar value of the local texture heterogeneity feature calculated in step S2. The high-confidence threshold for the probability of severe severity. The threshold value for CRP clinical indicators, This is the critical value for HTF characteristics (a value higher than this indicates significant heterogeneity in mucosal edema); The role of this step is to use a pre-trained classification model based on the fused multimodal features to not only output the severity level of the disease (mild, moderate, severe), but also to make a precise clinical recommendation on whether to upgrade to biologic therapy based on a comprehensive rule that combines imaging features, clinical indicators, and probability thresholds.

[0012] Furthermore, step S6, the result feedback display, specifically includes the following steps: The system returns results via a QR code page, displaying the disease severity level (mild / moderate / severe); recommendations for upgrading biological agents ("upgrade recommended" or "not recommended at this time"); and generating a structured report (PDF format) containing evidence of feature analysis (e.g., "HTF=0.41 around deep ulcers, indicating significant heterogeneity of mucosal edema"). The purpose of this step is to deliver the system's analysis results to doctors clearly and intuitively. It transforms complex algorithm results into effective information that clinicians can directly use for diagnosis and treatment decisions by generating easy-to-understand diagnostic conclusions and treatment recommendations, and providing structured reports containing key decision-making criteria.

[0013] The present invention has the following beneficial effects: 1. This invention transforms the traditionally subjective assessment of mucosal morphology, which relies on physicians' subjective experience, into quantifiable digital features by calculating the local texture heterogeneity characteristics of the area surrounding the ulcer in endoscopic images. This process eliminates interpretation differences between different observers and transforms texture changes that are difficult to distinguish precisely with the naked eye into mathematical model parameters. By integrating this objective indicator with clinical parameters, the system reduces its reliance on the experience of a single physician, enabling the endoscopic grading of inflammatory bowel disease to be established on an objective, uniform, and repeatable data basis, thereby improving the scientific rigor and reliability of the entire assessment process.

[0014] 2. This invention integrates imaging features with clinicopathological parameters across modalities, rather than simply referencing them side-by-side. Through a designed attention-based weighted algorithm, the system can adaptively weigh the contribution of imaging features and clinical indicators in the final decision, capturing the complex nonlinear relationship between the two. This deep fusion mechanism overcomes the limitations of decision-making based on a single data source, enabling the assessment of disease severity to include not only intuitive morphological evidence but also systemic biochemical inflammatory indicators and patient symptoms, forming a more comprehensive and three-dimensional evaluation system. The resulting treatment upgrade recommendations are more holistic and precise.

[0015] 3. This invention utilizes a lightweight QR code application interface to bring complex algorithm models and expert knowledge systems to primary healthcare settings. Primary care physicians do not need extensive expert experience; they can obtain decision support based on large-scale patient data models simply by collecting images and inputting parameters through a standardized process. This alleviates the disparity in treatment levels caused by uneven distribution of medical resources, provides primary care units with assessment tools that are consistent with those of higher-level hospitals, simplifies referral processes, and ensures that patients can receive high-quality preliminary diagnostic advice in a timely manner regardless of their location, thereby improving the overall efficiency and consistency of regional healthcare services.

[0016] 4. This invention integrates scattered, unstructured clinical consultation, endoscopic observation, and pathology report data into a standardized data acquisition and analysis process. All input parameters undergo predefined standardized processing and are parsed using a fixed algorithm model, ensuring that regardless of the operator, the input data can be processed and interpreted according to a unified standard. This method forcibly standardizes the key data items required for diagnosis, reduces evaluation bias caused by data omissions or inconsistent formats, and helps to form a unified diagnosis and treatment path and data standard within medical institutions and even between different institutions.

[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating a method for endoscopic-assisted identification and grading of inflammatory bowel disease according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 As shown, this invention provides a method for the endoscopic-assisted identification and grading of inflammatory bowel disease, comprising the following steps: Step S1, Image Acquisition and Preprocessing: Acquire the patient's intestinal endoscopy video, select the clearest ulcer image frame for noise reduction and contrast enhancement, and crop the core area of ​​the ulcer. Step S1, image acquisition preprocessing specifically includes the following steps: Step S11: Acquire image data of the patient's colonic mucosa using a digestive endoscopy device in 1080p RGB video stream format, and select keyframes (including clear frames of the deepest ulcer lesions) as input; Step S12: Preprocess the keyframes using Gaussian filtering (standard deviation). Noise reduction was performed, and adaptive histogram equalization (CLAHE) was used to enhance mucosal texture contrast. The central region of the ulcer was cropped (radius). (pixels) are used as the region of interest (ROI).

[0022] Step S2, Texture Feature Extraction: Convert the ulcer area image into numerical features, calculate the disorder and unevenness of its texture, and quantify the edema that is difficult to judge with the naked eye; Step S2, texture feature extraction specifically includes the following steps: Step S21: Convert the ROI region to grayscale and calculate its gray-level co-occurrence matrix (GLCM), with the direction being... ,distance Pixel; Step S22: Calculate the 16-dimensional basic texture features in four directions: contrast, correlation, energy, and homogeneity; Step S23: Define the Local Texture Heterogeneity Feature (HTF) as follows: In the formula, This is a feature of local texture heterogeneity. The total number of pixels contained within the region of interest. For summation index variables, For the first The two-dimensional coordinates of each pixel in the image matrix Let a function represent the image in coordinates The pixel grayscale value at that location, The symbol is for partial differentials. In pixels A local window centered on (e.g.) The variance of all pixel grayscale values ​​within the window; Step S24: Concatenate the HTF with the 16-dimensional GLCM features to form a 17-dimensional texture feature vector.

[0023] Step S3, Clinical Parameter Entry: After scanning the QR code, the patient fills in clinical information, including erythrocyte sedimentation rate and frequency of bowel movements, on the webpage. The system automatically converts this data into a standard format. Step S3, clinical parameter entry, specifically includes the following steps: Step S31: The patient scans the QR code to access the H5 page and enters the following clinical parameters: Erythrocyte sedimentation rate (ESR, mm / h); C-reactive protein (CRP, mg / L); daily bowel movement frequency; ulcer depth (superficial / deep) in colonoscopy report; presence of intestinal stenosis (yes / no). Step S32: Perform Z-score normalization on continuous parameters (ESR, CRP, frequency of bowel movements): In the formula, These are standardized clinical parameter values. The original clinical parameter values ​​entered for the current patient. This is the arithmetic mean of all values ​​of a specific clinical parameter in the dataset used during model training. This represents the standard deviation of all values ​​for a specific clinical parameter in the dataset used for model training.

[0024] Step S4, Feature Fusion and Weighting: Combine the texture features obtained from image analysis with the clinical data filled in by the patient, and assign different importance weights to different features through an algorithm; Step S4, feature fusion weighting, specifically includes the following steps: Step S41: Concatenate the 17-dimensional texture feature vector with the standardized clinical parameter vector to obtain the fused feature vector. (17-dimensional texture + 3-dimensional clinical parameters); Step S42: Weight the fused features using an attention mechanism: In the formula, This is the fused feature vector after being weighted by the attention mechanism. This is the original fused feature vector after concatenation. Multiplying corresponding elements of two vectors (or matrices) with the same dimension. It is a learnable weight matrix (the initial values ​​are fixed by the pre-trained model). For bias terms, This is the Sigmoid activation function.

[0025] Step S5, Intelligent Hierarchical Decision-Making: Input the fused comprehensive data into the trained model. The model automatically outputs the severity level of the disease and provides suggestions on whether biological agents should be used. Step S5, intelligent hierarchical decision-making specifically includes the following steps: Will Input a pre-trained classification model (Support Vector Machine, SVM, kernel function RBF), and output two results: Disease severity: divided into mild (CDEIS<5), moderate (5≤CDEIS≤12), and severe (CDEIS>12); Upgrade criteria for biological agents: binary classification output (yes / no), the judgment criteria are as follows: In the formula, To predict the probability that the current patient has "severe" inflammatory bowel disease using a class model. To determine whether the model classifies patients as having "moderate" severity, The logical operator "AND". The raw C-reactive protein value entered for the patient. The specific scalar value of the local texture heterogeneity feature calculated in step S2. The high-confidence threshold for the probability of severe severity. The threshold value for CRP clinical indicators, This is the critical value for HTF characteristics (a value higher than this indicates significant heterogeneity in mucosal edema).

[0026] Step S6, Results Feedback Display: The conclusions and treatment suggestions are displayed to the doctor in real time via a webpage, and a detailed report containing the diagnostic basis can be generated.

[0027] Step S6, the result feedback display specifically includes the following steps: The system returns results via a QR code page, displaying the disease severity level (mild / moderate / severe); recommendations for upgrading biological agents ("upgrade recommended" or "not recommended for now"); and generating a structured report (PDF format) containing the basis for feature analysis (e.g., "HTF=0.41 around deep ulcer, indicating significant heterogeneity of mucosal edema").

[0028] One specific application of this embodiment is: Background: A primary care hospital performed a colonoscopy on a patient suspected of having Crohn's disease. After discovering a deep ulcer lesion, the endoscopist used the system of this invention for auxiliary identification and classification.

[0029] Implementation steps: 1. Endoscopic image acquisition The physician acquires endoscopic images of the deep ulcer area in the descending colon of the patient (resolution: 1920×1080) and selects a clear image containing the ulcer and surrounding mucosa. 2. Image preprocessing and ROI extraction The system will execute automatically: Gaussian filtering ( Noise reduction; CLAHE enhances contrast (cropping and limiting) Grid size ); Center point of ulcer With the center as the cutting radius The circular ROI region of a pixel; 3. HTF Feature Calculation Calculations for the ROI region: Gray variance exist The average value within the window is ; Second-order mixed partial derivatives The average value is ; Substitute into the HTF formula: 4. Clinical parameter entry After scanning the QR code, the doctor enters the following data: ESR = 28 mm / h; CRP = 14 mg / L; Number of bowel movements per day = 6; Ulcer depth = deep; Intestinal stricture = no; 5. Data standardization System call training set parameters ( ; , ; , Standardize: 6. Multimodal feature fusion Construct a 20-dimensional fusion feature vector (17-dimensional texture + 3-dimensional clinical parameters), where: Texture features include HTF=0.041 and other 16-dimensional GLCM features (assuming energy=0.15, contrast=45.2, etc.). Clinical features ; Through pre-trained attention weight matrix (Example value: identity matrix) Weighted: (in practical applications) (This is a simplified demonstration as it is not an identity matrix.) 7. Hierarchy and Decision-Making SVM model output: Severe probability ; Calculate the decision conditions: ; CRP=14>10 and HTF=0.041>0.035, therefore the upgrade flag is triggered. 8. Results Feedback The system returned a page displaying: Severity: Moderate (CDEIS=9.2); Treatment recommendation: Upgrade to a biological agent is recommended; Analysis basis: Significant heterogeneity of mucosal texture around deep ulcers (HTF=0.041>threshold 0.035), and CRP>10mg / L.

[0030] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0031] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for the endoscopic-assisted identification and grading of inflammatory bowel disease, characterized in that, The auxiliary identification and classification method specifically includes the following steps: Step S1, Image Acquisition and Preprocessing: Acquire the patient's intestinal endoscopy video, select the clearest ulcer image frame for noise reduction and contrast enhancement, and crop the core area of ​​the ulcer. Step S2, Texture Feature Extraction: Convert the ulcer area image into numerical features, calculate the disorder and unevenness of its texture, and quantify the edema that is difficult to judge with the naked eye; Step S3, Clinical Parameter Entry: After scanning the QR code, the patient fills in clinical information, including erythrocyte sedimentation rate and frequency of bowel movements, on the webpage. The system automatically converts this data into a standard format. Step S4, Feature Fusion and Weighting: Combine the texture features obtained from image analysis with the clinical data filled in by the patient, and assign different importance weights to different features through an algorithm; Step S5, Intelligent Hierarchical Decision-Making: Input the fused comprehensive data into the trained model. The model automatically outputs the severity level of the disease and provides suggestions on whether biological agents should be used. Step S6, Results Feedback Display: The conclusions and treatment suggestions are displayed to the doctor in real time via a webpage, and a detailed report containing the diagnostic basis can be generated.

2. The method for endoscopic-assisted identification and grading of inflammatory bowel disease according to claim 1, characterized in that, Step S1, image acquisition preprocessing, specifically includes the following steps: Step S11: Acquire image data of the patient's colonic mucosa using a digestive endoscopy device in 1080p RGB video stream format, and select keyframes as input; Step S12: Preprocess the keyframes by using Gaussian filtering for noise reduction, adaptive histogram equalization to enhance the contrast of the mucosal texture, and cropping the central region of the ulcer as the region of interest.

3. The method for endoscopic-assisted identification and grading of inflammatory bowel disease according to claim 1, characterized in that, Step S2, texture feature extraction, specifically includes the following steps: Step S21: Convert the ROI region to grayscale and calculate its gray-level co-occurrence matrix, with the direction being... ,distance Pixel; Step S22: Calculate the 16-dimensional basic texture features in four directions, including contrast, correlation, energy, and homogeneity; Step S23: Define the local texture heterogeneity feature as: In the formula, This is a feature of local texture heterogeneity. The total number of pixels contained within the region of interest. For summation index variables, For the first The two-dimensional coordinates of each pixel in the image matrix Let a function represent the image in coordinates The pixel grayscale value at that location, The symbol is for partial differentials. In pixels The variance of the grayscale values ​​of all pixels within a local window centered on the center; Step S24: Concatenate the HTF with the 16-dimensional GLCM features to form a 17-dimensional texture feature vector.

4. The method for endoscopic-assisted identification and grading of inflammatory bowel disease according to claim 1, characterized in that, Step S3, clinical parameter entry, specifically includes the following steps: Step S31: The patient scans the QR code to access the H5 page and enters the following clinical parameters: Erythrocyte sedimentation rate (ESR); C-reactive protein (CRP); daily bowel movement frequency; ulcer depth in colonoscopy report; presence of intestinal stenosis; Step S32: Perform Z-score normalization on the continuous parameters: In the formula, These are standardized clinical parameter values. The original clinical parameter values ​​entered for the current patient. This is the arithmetic mean of all values ​​of a specific clinical parameter in the dataset used during model training. This represents the standard deviation of all values ​​for a specific clinical parameter in the dataset used for model training.

5. The method for endoscopic-assisted identification and grading of inflammatory bowel disease according to claim 1, characterized in that, Step S4, feature fusion weighting, specifically includes the following steps: Step S41: Concatenate the 17-dimensional texture feature vector with the standardized clinical parameter vector to obtain the fused feature vector. ; Step S42: Weight the fused features using an attention mechanism: In the formula, This is the fused feature vector after being weighted by the attention mechanism. This is the original fused feature vector after concatenation. Multiply the corresponding elements of two vectors with the same dimension. For learnable weight matrix, For bias terms, This is the Sigmoid activation function.

6. The method for endoscopic-assisted identification and grading of inflammatory bowel disease according to claim 1, characterized in that, Step S5, intelligent hierarchical decision-making, specifically includes the following steps: Will Input a pre-trained classification model and output two results: Disease severity: classified as mild, moderate, and severe; Upgrade criteria for biological agents: binary classification output, the judgment criteria are as follows: In the formula, To predict the probability that the current patient has "severe" inflammatory bowel disease using a class model. To determine whether the model classifies patients as having "moderate" severity, The logical operator "AND". The raw C-reactive protein value entered for the patient. The specific scalar value of the local texture heterogeneity feature calculated in step S2. The high-confidence threshold for the probability of severe severity. The threshold value for CRP clinical indicators, This is the critical value for HTF characteristics.

7. The method for endoscopic-assisted identification and grading of inflammatory bowel disease according to claim 1, characterized in that, Step S6, the result feedback display, specifically includes the following steps: The system returns results via a QR code page, displaying the disease severity level; recommendations for upgrading biological agents; and generating a structured report containing the basis for feature analysis.

Images