Intestinal stoma mucous membrane state evaluation system based on color comparison

Abstract

The invention relates to the technical field of medical image processing, in particular to an enterostomy mucous membrane state evaluation system based on color comparison, which comprises a multi-modal data acquisition module used for acquiring and analyzing a mucous membrane image sequence, ambient light characteristics and basic physiological data and generating acquisition data records with aligned timestamps; and the individualized benchmark modeling module is used for establishing a patient specific color reference benchmark through HSL color space analysis on the basis of the collected data records and the preoperative physiological indexes, and generating an initial adaptive color gamut threshold model. According to the method, the specific initial color reference is established for each patient, and the sliding window algorithm is combined to continuously monitor the color gamut drift so as to dynamically update the reference, so that accurate self-adaption to individual physiological differences of the patients and postoperative slow recovery changes is realized, and therefore, the objectivity and accuracy of evaluation are greatly improved; the problem of misjudgment caused by individual differences or normal evolution of physiological states in a traditional fixed threshold value method is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a color-matching-based system for assessing the condition of enterostomy mucosa. Background Technology

[0002] Enterostomy is a routine surgical procedure for treating gastrointestinal diseases such as colorectal cancer and inflammatory bowel disease. Close monitoring of the stoma mucosa postoperatively is crucial for ensuring smooth patient recovery and preventing serious complications. As a direct exposure of intestinal tissue to the body surface, the color, luster, and moisture of the stoma mucosa are core physiological indicators reflecting local blood circulation and tissue vitality. Especially in the early postoperative period, continuous and dynamic observation of changes in mucosal color from normal pink or red to abnormal hues is of vital clinical significance for the timely detection and treatment of emergencies such as ischemia and necrosis.

[0003] Existing technologies for assessing the condition of enterostomy mucosa mostly rely on fixed, universally applicable healthy color thresholds derived from population statistics. These technologies ignore the significant individual differences in the baseline color of healthy mucosa caused by factors such as skin color and hemoglobin levels among different patients. This makes it easy to generate a large number of false positive or false negative warnings because it is difficult to distinguish between normal individual differences or physiological changes and true pathological abnormalities, which seriously affects the accuracy of assessment and the reliability of clinical decision-making. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a color comparison-based enterostomy mucosal status assessment system, which aims to improve the problem that most existing technologies rely on fixed, population-based statistically derived universal healthy color thresholds, ignoring the significant individual differences in the baseline color of healthy mucosa caused by factors such as skin color and hemoglobin levels among different patients.

[0005] In a first aspect, the present invention provides the following technical solution: a color comparison-based enterostomy mucosal status assessment system, comprising:

[0006] The multimodal data acquisition module is used to acquire and analyze mucosal image sequences, ambient light features, and basic physiological data, and generate timestamp-aligned acquisition data records.

[0007] The individualized baseline modeling module, based on the collected data records and preoperative physiological indicators, establishes a patient-specific color reference baseline through HSL color space analysis and generates an initial adaptive color gamut threshold model.

[0008] The environmental noise decoupling module, based on the collected data records and the adaptive color gamut threshold model, analyzes the ambient light characteristics and performs physical calibration to generate a standard mucosal image after environmental noise decoupling;

[0009] The dynamic threshold update module, based on the standard mucosal image sequence and the current adaptive color gamut threshold model, monitors color gamut drift through a sliding window algorithm. When the drift exceeds the preset baseline ratio, it triggers model iteration optimization to generate an updated version of the adaptive color gamut threshold model.

[0010] The mucosal state decision module, based on the standard mucosal image and the adaptive color gamut threshold model, generates a risk level heatmap by comparing the distribution of abnormal color spots, and outputs clinical warning signals and quantitative assessment reports.

[0011] Preferably, the acquisition and analysis of mucosal image sequences, ambient light features, and basic physiological data specifically includes:

[0012] The acquired raw mucosal images were subjected to standardized preprocessing, including noise suppression based on Gaussian filtering, detail enhancement based on contrast-limited adaptive histogram equalization, and image distortion correction.

[0013] A semantic segmentation network was used to extract the stoma mucosa region from the image and exclude the interference areas of the surrounding skin and stoma instruments.

[0014] The ambient light sensor data is analyzed to obtain the color temperature, illuminance, and spectral characteristics of the current scene, and the color temperature, illuminance, and spectral characteristics are spatiotemporally aligned and bound with the corresponding image data frames.

[0015] Preferably, the generation of timestamp-aligned collected data records specifically includes:

[0016] The acquisition time points of the camera and the ambient light sensor are synchronized by hardware triggering signals;

[0017] The mucosal image sequence is decomposed into frames, and a timestamp is added to each frame.

[0018] The color temperature and illuminance data streams continuously collected by the ambient light sensor are segmented according to the same time base.

[0019] Establish a mapping relationship between image frames and ambient light data segments to form timestamp-aligned acquisition data records.

[0020] Preferably, the step of establishing a patient-specific color reference standard based on the collected data records and preoperative physiological indicators through HSL color space analysis specifically includes:

[0021] During the patient's postoperative stable period, multiple sets of standard mucosal images were collected under standard lighting conditions and processed by the environmental noise decoupling module to construct an individual initial sample set.

[0022] The standard mucosal image is converted from the RGB color space to the HSL color space, and the hue and saturation component values ​​of all pixels in the stoma mucosal area are extracted.

[0023] Based on the patient's skin color characteristics in the preoperative physiological indicators, the extracted hue and saturation components were initially calibrated for cluster centers, and a clustering algorithm was applied to construct a color gamut cluster representing the patient's healthy mucosal state.

[0024] Based on the distribution of the color gamut clusters in the HSL color space, the statistical boundaries of their hue and saturation are calculated to establish a patient-specific initial adaptive color gamut threshold model.

[0025] Preferably, the step of analyzing ambient light features and performing physical calibration based on the acquired data records and the adaptive color gamut threshold model to generate a standard mucosal image after decoupling from ambient noise specifically includes:

[0026] The color temperature and illuminance data captured synchronously with the environmental image frames are parsed from the collected data records;

[0027] Based on the reference values ​​of the standard color check card, a mapping model is constructed from the device-dependent color space under the current ambient light conditions to the device-independent color space under standard lighting conditions;

[0028] The mapping model is applied to perform pixel-level color correction on the original mucosal image;

[0029] The corrected image is converted to a color representation under standard reference lighting conditions, and the output is a standard mucosal image after the environmental noise is decoupled.

[0030] Preferably, the step of monitoring color gamut drift using a sliding window algorithm based on the standard mucosal image sequence and the current adaptive color gamut threshold model specifically includes:

[0031] During system operation, recently generated standard mucosal images are cached in chronological order to form a fixed-capacity sliding window data set;

[0032] For each standard mucosal image in the sliding window data set, extract the hue and saturation values ​​of all pixels in the stoma mucosal area in the HSL color space;

[0033] Calculate the hue and saturation distribution characteristics of all mucosal pixels within the sliding window data set;

[0034] The distribution features are compared with the baseline distribution features defined by the current adaptive color gamut threshold model, and the feature drift is calculated.

[0035] Preferably, the step of triggering model iteration optimization and generating an updated adaptive color gamut threshold model when the drift exceeds the baseline ratio specifically includes:

[0036] The calculated feature drift is compared with a preset baseline ratio threshold calculated based on historical system operation data.

[0037] When the feature drift exceeds a preset baseline ratio threshold, the model optimization process is triggered.

[0038] The statistical distribution parameters of the color gamut cluster established by the individualized benchmark modeling module are updated using the sliding window dataset.

[0039] Based on the updated color gamut cluster distribution, the statistical boundaries of hue and saturation are recalculated.

[0040] Generate the updated adaptive color gamut threshold model and replace the original model.

[0041] Preferably, the step of generating a risk level heatmap by comparing the distribution of abnormal color spots based on the standard mucosal image and the adaptive color gamut threshold model specifically includes:

[0042] Pixel-level analysis is performed on the standard mucosal image, and the hue and saturation values ​​in the HSL color space of each pixel are compared with the normal color gamut range defined by the adaptive color gamut threshold model.

[0043] Identify and mark abnormal pixels that exceed the normal color gamut range, and classify them into different risk levels according to the degree of deviation;

[0044] Abnormal pixels of different risk levels are aggregated according to their spatial location to generate a risk level heatmap superimposed on the original image.

[0045] Based on the area proportion and spatial distribution characteristics of each level region in the risk level heatmap, a quantitative assessment report containing risk scores and abnormal area location information is generated.

[0046] When the risk score exceeds a preset threshold, a corresponding clinical warning signal is generated.

[0047] Secondly, the present invention provides the following technical solution: a method for assessing the state of enterostomy mucosa based on color comparison, the method comprising the following steps:

[0048] S1. Collect and analyze mucosal image sequences, ambient light characteristics, and basic physiological data to generate timestamp-aligned data records;

[0049] S2. Based on the collected data records and preoperative physiological indicators, establish a patient-specific color reference benchmark through HSL color space analysis and generate an initial adaptive color gamut threshold model.

[0050] S3. Based on the collected data records and the adaptive color gamut threshold model, analyze the ambient light characteristics and perform physical calibration to generate a standard mucosal image after environmental noise decoupling.

[0051] S4. Based on the standard mucosal image sequence and the current adaptive color gamut threshold model, monitor the color gamut drift through the sliding window algorithm. When the drift exceeds the preset baseline ratio, trigger the model iterative optimization to generate an updated adaptive color gamut threshold model.

[0052] S5. Based on the standard mucosal image and the adaptive color gamut threshold model, a risk level heatmap is generated by comparing the distribution of abnormal color spots, and a clinical early warning signal and quantitative assessment report are output.

[0053] The present invention has the following beneficial effects:

[0054] 1. In this invention, by establishing a specific initial color benchmark for each patient and continuously monitoring color gamut drift using a sliding window algorithm to dynamically update the benchmark, precise adaptation to individual physiological differences and slow postoperative recovery changes is achieved, thereby greatly improving the objectivity and accuracy of the assessment and effectively avoiding the misjudgment problem caused by individual differences or normal evolution of physiological state in traditional fixed threshold methods.

[0055] 2. In this invention, a color mapping model is constructed using a standard color check card to perform physical color calibration on the original mucosal image. This successfully decouples and eliminates the serious interference caused by the complex and variable ambient lighting in the clinical environment on the judgment of mucosal color, ensuring a high degree of consistency and reliability of the analysis results of image data collected at different times and under different lighting conditions.

[0056] 3. In this invention, by statistically quantifying the deviation of abnormal pixels and generating an intuitive risk level heatmap and comprehensive risk score, a clear and quantitative clinical decision support tool is provided for medical staff, enabling them to intuitively grasp the location, area and severity of abnormal areas, thereby achieving early and accurate warning and intervention for potential risks. Attached Figure Description

[0057] Figure 1 This is an architecture diagram of the color comparison-based enterostomy mucosal status assessment system proposed in this invention.

[0058] Figure 2 This is a flowchart of the color comparison-based method for assessing the mucosal condition of an enterostomy proposed in this invention. Detailed Implementation

[0059] The technical solutions in 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.

[0060] Example 1:

[0061] In a first embodiment of the present invention, the present invention provides a color-matching-based enterostomy mucosal status assessment system, such as... Figure 1 As shown, it includes:

[0062] The multimodal data acquisition module is used to acquire and analyze mucosal image sequences, ambient light features, and basic physiological data, and generate timestamp-aligned acquisition data records.

[0063] Furthermore, the acquisition and analysis of mucosal image sequences, ambient light characteristics, and basic physiological data specifically include:

[0064] The acquired raw mucosal images were subjected to standardized preprocessing, including noise suppression based on Gaussian filtering, detail enhancement based on contrast-limited adaptive histogram equalization, and image distortion correction.

[0065] A semantic segmentation network was used to extract the stoma mucosa region from the image and exclude the interference areas of the surrounding skin and stoma instruments.

[0066] The system analyzes ambient light sensor data to obtain the color temperature, illuminance, and spectral characteristics of the current scene, and then spatiotemporally aligns and binds these characteristics with the corresponding image data frames.

[0067] Furthermore, generating timestamp-aligned collected data records specifically includes:

[0068] The acquisition time points of the camera and the ambient light sensor are synchronized by hardware triggering signals;

[0069] The mucosal image sequence is decomposed into frames, and a timestamp is added to each frame.

[0070] The color temperature and illuminance data streams continuously collected by the ambient light sensor are segmented according to the same time base.

[0071] Establish a mapping relationship between image frames and ambient light data segments to form timestamp-aligned acquisition data records.

[0072] Specifically, for each frame of raw mucosal image captured by the camera, it undergoes standardized preprocessing. The initial step is to apply Gaussian filtering for noise suppression, using a two-dimensional Gaussian function. The image is convolved, where G(x,y) is the Gaussian kernel value at coordinates (x,y), and the input parameter σ represents the standard deviation of the Gaussian kernel, typically set between 0.5 and 2.0. This smooths random texture noise in the enterostomy mucosa image caused by sensor noise or low lighting conditions, resulting in a visually smoother mucosa image with significantly reduced noise levels. Next, a contrast-limited adaptive histogram equalization technique is used to enhance mucosa image details. Then, based on a pre-calibrated camera intrinsic matrix and distortion coefficients, geometric distortion correction is performed on the image. This process uses an inverse mapping model to move pixels from the distorted image coordinates (x,y) to the distorted image coordinates (x,y). d ,y d Transform to corrected image coordinates (x) u ,y u The core radial distortion correction can be expressed by formula x. u =x d (1+k1r 2 +k2r 4 ) and y u =y d (1+k1r 2 +k2r 4 ) description, in which It is a pixel (x) d ,y d The square of the radial distance from the image center is given by the input parameters k1 and k2, which are radial distortion coefficients obtained from camera calibration. The output is a distortion-free image that ensures the true geometric restoration of the stoma mucosa image.

[0073] After standardization preprocessing, the image is input into a pre-trained semantic segmentation network based on the U-Nt architecture. The training objective is to maximize the overlap between the predicted results and the ground truth labels. Typically, the Dice loss function is used for optimization, and its formula is as follows: The input parameter P represents the set of stoma mucosal pixels predicted by the network, T represents the set of real mucosal region pixels manually labeled in the training data, the symbol |P∩T| represents the number of the intersection of the two sets P and T (i.e., the correctly segmented mucosal pixels), and |P| and |T| represent the total number of pixels in the sets P and T, respectively.

[0074] Simultaneously with image acquisition and processing, the hardware trigger signal also drives the integrated non-contact physiological monitoring unit to acquire data. This unit includes an infrared temperature sensor and a reflective pulse oximetry sensor. It analyzes the signal returned by the infrared temperature sensor and converts it into the patient's body surface temperature value. At the same time, it analyzes the ratio of photoelectric signals returned by the reflective pulse oximetry sensor in the red light and infrared light bands. Where the input parameter AC REDand AC IR These represent the AC components of red and infrared light signals generated by arterial pulsation, respectively, while DC... RED and DC IR These represent the direct current (DC) component generated by the absorption of light by static tissues. Based on this ratio R and the empirical calibration formula, the patient's blood oxygen saturation percentage SpO2 is calculated as a key physiological indicator reflecting the perfusion status of mucosal tissues.

[0075] Ultimately, the system logically maps and stores the segmented stoma mucosal region image, the precisely corresponding ambient light data segment, the synchronously acquired physiological data of body temperature and blood oxygen saturation, and the shared timestamp, together forming a structured, timestamp-aligned acquisition data record containing image, illumination, and physiological multimodal information. This provides high-quality input data for subsequent individualized baseline modeling and environmental noise decoupling modules.

[0076] The individualized baseline modeling module, based on the collected data records and preoperative physiological indicators, establishes a patient-specific color reference baseline through HSL color space analysis and generates an initial adaptive color gamut threshold model.

[0077] Furthermore, based on the collected data records and preoperative physiological indicators, a patient-specific color reference benchmark is established through HSL color space analysis, specifically including:

[0078] During the postoperative stable period of patients, multiple sets of standard mucosal images were collected under standard lighting conditions and processed by the environmental noise decoupling module to construct an initial sample set for each individual.

[0079] The standard mucosal image was converted from the RGB color space to the HSL color space, and the hue and saturation component values ​​of all pixels in the stoma mucosal area were extracted.

[0080] Based on the patient's skin color characteristics in the preoperative physiological indicators, the extracted hue and saturation components were initially calibrated for cluster centers, and a clustering algorithm was applied to construct a color gamut cluster representing the patient's healthy mucosal state.

[0081] Based on the distribution of color gamut clusters in the HSL color space, the statistical boundaries of their hue and saturation are calculated to establish a patient-specific initial adaptive color gamut threshold model.

[0082] Specifically, after the patient enters a stable period post-surgery, data is continuously acquired using a multimodal data acquisition module and processed by an environmental noise decoupling module to generate multiple sets of standard stoma mucosa images corrected under standard lighting conditions. These images collectively constitute the patient's individual initial sample set. For each standard mucosa image in the individual initial sample set, it is first converted from the RGB color space to the HSL color space. The conversion process follows standard color model formulas, such as mapping RGB values ​​(R,G,B) to hue (H), saturation (S), and lightness (L) values. Subsequently, the hue (H) and saturation (S) component values ​​of all pixels within the stoma mucosa region determined during the semantic segmentation stage are extracted to form a two-dimensional data point set.

[0083] To establish a patient-specific color reference, skin color characteristics recorded in preoperative physiological indicators were incorporated. For example, HSL values ​​of standard preoperative skin areas were measured. The HSL ranges corresponding to these skin color characteristics were used as initial heuristics to calibrate the initial centers of the clustering algorithm, making them closer to the patient's physiological intrinsic color. Specifically, the K-means clustering algorithm was applied to perform unsupervised classification of the extracted mucosal pixel hue and saturation data, aiming to divide these data points into K distinct clusters. The K-means algorithm iteratively optimizes the cluster centers by minimizing the sum of the squared distances from each data point to its cluster center. Its objective function is... In this formula, N represents the total number of pixels within the mucosal region, K is the preset cluster size, and x i μ is the (H,S) two-dimensional vector value representing the i-th pixel. j It is the center vector of the j-th cluster, and the input parameter w ij It is an indicator variable, when pixel x i Its value is 1 when assigned to cluster j, and 0 otherwise, while ||x i -μ j || 2 This represents pixel x. i To its cluster center μ j The square of the Euclidean distance. After clustering, these clusters collectively represent the color gamut distribution of the stoma mucosa in the patient under healthy conditions.

[0084] Based on the distribution of the constructed color gamut clusters in the HSL color space, the statistical boundaries of hue (H) and saturation (S) for each cluster are calculated. Specifically, the mean and standard deviation of the hue and saturation values ​​of all pixels within the cluster are calculated, and the healthy color gamut range represented by the cluster is defined as the hue interval. and saturation range The union of . In this definition, and These are the mean hue and saturation values ​​of the j-th cluster, respectively. and These are the corresponding standard deviations, while α and β are preset confidence interval coefficients. Ultimately, these calculated statistical boundaries collectively construct a patient-specific initial adaptive color gamut threshold model. This model outputs a multidimensional set of healthy color ranges, serving as a benchmark for subsequent mucosal state assessment, ensuring the individualization and objectivity of the evaluation.

[0085] The environmental noise decoupling module, based on the collected data recording and the adaptive color gamut threshold model, analyzes the ambient light characteristics and performs physical calibration to generate a standard mucosal image after environmental noise decoupling;

[0086] Furthermore, based on the collected data records and the adaptive color gamut threshold model, the ambient light characteristics are analyzed and physical calibration is performed to generate a standard mucosal image after environmental noise decoupling. Specifically, this includes:

[0087] The color temperature and illuminance data captured synchronously with the environmental image frames are extracted from the collected data records.

[0088] Based on the reference values ​​of the standard color check card, a mapping model is constructed from the device-dependent color space under the current ambient light conditions to the device-independent color space under standard lighting conditions;

[0089] A mapping model is applied to perform pixel-level color correction on the original mucosal image;

[0090] The corrected image is converted to a color representation under standard reference lighting conditions, and the output is a standard mucosal image after environmental noise decoupling.

[0091] Specifically, upon startup, the environmental noise decoupling module receives data records from the multimodal data acquisition module and extracts the ambient light features captured in strict synchronization with each frame of the original mucosal image, primarily including color temperature and illuminance data. To achieve accurate physical color calibration, before each acquisition task begins, a standard color check card containing multiple known color patches, such as the X-RiteColorCheckerClassic, is briefly placed within the camera's field of view to capture a calibration image. The actual measured RGB values ​​of each color patch in the image are then paired with the device-independent color space measured under standard lighting conditions (typically referring to the CIED65 standard illuminator defined by the International Commission on Illumination), such as the CIELab reference value. This constructs a mapping model from the device-dependent color space under current ambient light conditions to the device-independent color space under standard lighting conditions. This mapping model is built based on the root multinomial regression algorithm, the core of which is to solve for a transformation that minimizes the color difference between all measured color patch values ​​and the reference value. The transformation formula can be expressed as C... std =M·f(C raw In this formula, the input parameter Craw It is a raw device-associated color vector representing a single pixel read from a calibration image or subsequent mucosal image, specifically a column vector [R, G, B] containing the intensity values ​​of the red, green, and blue channels. T The function f(C) raw ) is a polynomial expansion function applied to the original color vector to capture the nonlinear characteristics of the camera sensor response. For example, for a second-order model, it transforms the three-dimensional input vector into a higher-dimensional feature vector, which may contain elements such as [1, R, G, B, RG, R]. 2 G 2 B 2 ,...] T The terms are as follows. M is a transformation matrix calculated by applying the least squares method to the data of all color patches on the color check card. The goal of solving this matrix is ​​to find an optimal M that minimizes the sum of the Euclidean distances between the transformed measurements of all known color patches and their standard reference values. The final output result is C. std This is the device-independent color vector of the pixel under standard lighting conditions, such as its corresponding CIELab value. This represents the true color of the mucosal tissue, eliminating the color shift caused by the current ambient light. The transformation matrix M for the current lighting environment is solved, and this mapping model is stably applied to all subsequently acquired raw mucosal images. Point-by-point color correction is performed on each pixel in the image, ultimately generating and outputting a standard mucosal image with true color, unaffected by ambient light interference and decoupled from environmental noise. This image provides an objective and highly consistent analytical basis for subsequent dynamic threshold updates and mucosal state decision-making modules.

[0092] The dynamic threshold update module, based on the standard mucosal image sequence and the current adaptive color gamut threshold model, monitors color gamut drift through the sliding window algorithm. When the drift exceeds the preset baseline ratio, it triggers model iteration optimization and generates an updated version of the adaptive color gamut threshold model.

[0093] Furthermore, based on standard mucosal image sequences and the current adaptive color gamut threshold model, the monitoring of color gamut drift using a sliding window algorithm specifically includes:

[0094] During system operation, recently generated standard mucosal images are cached in chronological order to form a fixed-capacity sliding window data set;

[0095] For each standard mucosal image in the sliding window dataset, extract the hue and saturation values ​​of all pixels in the stoma mucosal region in the HSL color space;

[0096] Calculate the hue and saturation distribution characteristics of all mucosal pixels within the sliding window dataset;

[0097] The distribution characteristics are compared with the baseline distribution characteristics defined by the current adaptive color gamut threshold model, and the feature drift is calculated.

[0098] Furthermore, when the drift exceeds the baseline ratio, iterative optimization of the model is triggered to generate an updated adaptive color gamut threshold model, specifically including:

[0099] The calculated feature drift is compared with a preset baseline ratio threshold calculated based on historical system operation data.

[0100] When the feature drift exceeds a preset baseline ratio threshold, the model optimization process is triggered.

[0101] The statistical distribution parameters of the color gamut cluster established by the individualized benchmark modeling module are updated using a sliding window dataset.

[0102] Based on the updated color gamut cluster distribution, the statistical boundaries of hue and saturation are recalculated.

[0103] Generate an updated adaptive color gamut threshold model and replace the original model.

[0104] Specifically, the dynamic threshold update module continuously receives and processes time-ordered standard mucosal image sequences output by the environmental noise decoupling module. To ensure that the adaptive color gamut thresholding model can accurately capture the slow changes in the patient's physiological state, a fixed-capacity sliding window dataset is maintained, which continuously caches the latest N using a first-in-first-out strategy. w A standard mucosal image, of which N w This is a preset window size used to dynamically reflect the average color state of the patient's mucosa in recent times. For each standard mucosal image in this sliding window dataset, the hue (H) and saturation (S) components of all pixels within the stoma mucosa region determined during the semantic segmentation stage are first extracted in the HSL color space. These values ​​are then aggregated to calculate the statistical characteristics representing the mucosal color distribution within the current window period, specifically represented as the mean vector μ of a two-dimensional Gaussian distribution composed of hue and saturation. c The covariance matrix Σ c .

[0105] To quantify the difference between the current color distribution and the established health baseline, the distribution characteristics (μ) calculated within the current sliding window period will be used. c ,Σ c ) and the baseline distribution characteristics (μ) defined by the current adaptive color gamut threshold model b ,Σ b The comparison is performed by calculating the characteristic drift between the two. This drift is calculated using the corrected Bach distance D. B This distance is used to measure the similarity between two Gaussian probability distributions, and its specific calculation formula is as follows: In this formula, the input parameter μ c is a two-dimensional vector whose elements are the mean values ​​of hue H and saturation S of all mucosal pixels in the sliding window dataset, representing the recent mucosal color center of the patient; μ b This is the baseline mean vector stored in the current adaptive color gamut threshold model, representing the color center of the healthy mucosa. Σ c and Σ b These are the covariance matrices of the relationship between hue and saturation in the current sliding window data and the baseline data, respectively, which reflect the shape and dispersion of the color distribution. It is the average of the two covariance matrices, while det(·) represents the determinant of the matrix. The output of this formula is D. B It is a dimensionless scalar value, namely the feature drift. The larger the value, the more significant the deviation of the current mucosal color state from the healthy baseline.

[0106] The calculated characteristic drift D B Compare with a preset baseline proportion threshold τ derived from the patient's historical operational data. When D B When the value exceeds τ, it indicates that the baseline color of the mucosa has undergone a persistent rather than temporary change, at which point the model optimization process will be automatically triggered. This optimization process employs an incremental learning strategy, using a weighted average of the current sliding window dataset and the original baseline data to update the statistical distribution parameters of the color gamut cluster established by the individualized baseline modeling module. Specifically, the new cluster mean μ... new The covariance matrix Σ new (μ) can be calculated from the sliding window data c ,Σ c ) and the original model baseline (μ b ,Σ b The weighted average is used to update, i.e., μ new =(1-w)μ b +wμ c and Σ new =(1-w)Σ b +wΣ c , where w is a learning rate or weighting coefficient between 0 and 1, used to balance the impact of historical and latest data on model updates. Based on these updated color gamut cluster distributions, healthy hue and saturation statistical boundaries are recalculated and set, ultimately generating an updated adaptive color gamut threshold model. This new model then replaces the original model, serving as the latest and most accurate individualized benchmark for subsequent mucosal state assessments.

[0107] The mucosal condition decision module, based on standard mucosal images and an adaptive color gamut threshold model, generates a risk level heatmap by comparing the distribution of abnormal color spots, and outputs clinical warning signals and quantitative assessment reports.

[0108] Furthermore, based on standard mucosal images and an adaptive color gamut threshold model, a risk level heatmap is generated by comparing the distribution of abnormal color spots, specifically including:

[0109] Pixel-level analysis was performed on standard mucosal images, and the hue and saturation values ​​in the HSL color space of each pixel were compared with the normal color gamut range defined by the adaptive color gamut threshold model.

[0110] Identify and label abnormal pixels that are outside the normal color gamut, and classify them into different risk levels according to the degree of deviation;

[0111] Abnormal pixels of different risk levels are aggregated according to their spatial location to generate a risk level heatmap superimposed on the original image.

[0112] Based on the area proportion and spatial distribution characteristics of each level region in the risk level heatmap, a quantitative assessment report containing risk scores and abnormal area location information is generated.

[0113] When the risk score exceeds a preset threshold, a corresponding clinical warning signal is generated.

[0114] Specifically, the mucosal state decision module receives the latest standard mucosal image from the environmental noise decoupling module and the latest version of the adaptive color gamut threshold model provided by the dynamic threshold update module. It then performs pixel-by-pixel analysis on the segmented stoma mucosal region in the standard mucosal image, converting the color value of each pixel from the RGB color space to the HSL color space, and comparing its hue and saturation values ​​with the statistical boundary of the healthy mucosal color gamut defined by the adaptive color gamut threshold model. Any pixel whose hue or saturation value exceeds the healthy range is identified and marked as an abnormal pixel, and the severity of its deviation is further quantified by calculating its Mahalanobis distance from the healthy color center. The calculation formula is as follows: In this formula, the input parameter x is a two-dimensional column vector [H, S] representing the hue and saturation values ​​of the current pixel to be analyzed. T μ is the two-dimensional mean vector representing the color gamut distribution of healthy mucosa stored in the adaptive color gamut threshold model. H ,μ S ] T Σ is the 2×2 covariance matrix stored in the model, describing the shape and orientation of the healthy color gamut data distribution; the symbol T represents the vector transpose, Σ -1 The inverse matrix of the covariance matrix is ​​represented by D; the output of this formula is D. M(x) is a non-negative scalar value, namely Mahalanobis distance, which represents the statistical distance between the current pixel color and the healthy color center. The larger the value, the higher the degree of abnormality.

[0115] Based on the calculated Mahalanobis distance D M (x) Classify each abnormal pixel into multiple preset risk levels, for example by setting two threshold levels τ. low and τ high Abnormal pixels were categorized into low-risk, medium-risk, and high-risk classes. Then, based on the spatial location of these pixels in the original image, a risk level heatmap was generated and overlaid on the mucosal image. This heatmap visually indicated the abnormal color patches of different risk levels using a continuous color spectrum from yellow to orange to red. Next, the proportion P of each risk level region to the total mucosal area was calculated based on this heatmap. low ,P mid ,P high And through a weighted summation model S risk =w low P low +w mid P mid +w high P high To generate a comprehensive quantitative risk score, where the weighting coefficient w low ,w mid ,w high This is a constant set based on clinical experience, reflecting the severity of each risk level. Ultimately, a risk score S is output, incorporating this risk level. risk The system generates a quantitative assessment report on the area proportion of each risk level and the risk level heat map. When the calculated risk score exceeds a preset clinical safety threshold, it will automatically generate and issue a clear clinical warning signal to prompt medical staff to take timely professional intervention.

[0116] Example 2:

[0117] Relying on fixed, universally applicable healthy color thresholds derived from population statistics ignores significant individual differences in the baseline color of healthy mucosa caused by factors such as skin color and hemoglobin levels among different patients. This easily leads to a large number of false positive or false negative warnings because it is difficult to distinguish between normal individual differences or physiological changes and true pathological abnormalities. To solve the above problems, the color comparison-based method for assessing the state of enterostomy mucosa provided in this invention is adopted, such as... Figure 2 The method includes the following steps:

[0118] S1. Collect and analyze mucosal image sequences, ambient light characteristics, and basic physiological data to generate timestamp-aligned data records;

[0119] S2. Based on the collected data records and preoperative physiological indicators, establish a patient-specific color reference benchmark through HSL color space analysis and generate an initial adaptive color gamut threshold model.

[0120] S3. Based on the collected data records and the adaptive color gamut threshold model, analyze the ambient light characteristics and perform physical calibration to generate a standard mucosal image after environmental noise decoupling.

[0121] S4. Based on the standard mucosal image sequence and the current adaptive color gamut threshold model, monitor the color gamut drift through the sliding window algorithm. When the drift exceeds the preset baseline ratio, trigger the model iterative optimization to generate an updated adaptive color gamut threshold model.

[0122] S5. Based on the standard mucosal image and the adaptive color gamut threshold model, a risk level heatmap is generated by comparing the distribution of abnormal color spots, and a clinical early warning signal and quantitative assessment report are output.

[0123] Specifically, step S1 uses a hardware trigger signal to precisely and synchronously activate the camera, ambient light sensor, and physiological monitoring unit at a physical time starting point. The system simultaneously captures image frames of the enterostomy mucosa, ambient light color temperature and illuminance data, and basic physiological data such as the patient's body temperature and blood oxygen saturation. Subsequently, the system logically maps and binds these multimodal information elements—images, illumination, and physiological data—acquired at the same time, forming a structured record of acquired data with a shared timestamp, providing precisely aligned, high-quality input data for subsequent analysis.

[0124] In step S2, after the patient enters a stable period post-surgery, the system analyzes the pre-processed mucosal images using multiple sets of collected data. The system converts the pixels of the mucosal region from RGB to HSL color space and applies clustering algorithms such as K-means to analyze the hue and saturation distribution of all pixels, forming several "color gamut clusters" representing the patient's healthy mucosal state. Finally, by calculating the statistical boundaries of these clusters in HSL space, such as the mean and standard deviation, a patient-specific, quantified initial adaptive color gamut threshold model is constructed as an individualized health benchmark for subsequent color comparisons.

[0125] In step S3, during continuous monitoring, for each newly generated data record, the system extracts the ambient light characteristics recorded synchronously with the mucosal image. By photographing a standard color check card before acquisition, the system constructs a color mapping model from the current ambient light to standard lighting conditions. This model is applied to perform pixel-by-pixel physical color calibration on the original mucosal image. The purpose is to eliminate color deviations introduced by changes in ambient light, ultimately generating and outputting a standard mucosal image with true color, unaffected by ambient light noise, and capable of consistent comparison at different time points and under different environments.

[0126] In step S4, the system maintains a sliding window dataset consisting of multiple recent standard mucosal images and continuously calculates the overall statistical distribution characteristics of mucosal color within this window. By calculating the statistical distance between the current distribution and the baseline distribution in the adaptive color gamut threshold model, the system can monitor whether the patient's mucosal baseline color undergoes a slow and continuous physiological drift. Once the calculated drift exceeds a preset threshold, the system automatically triggers an optimization process, using the data within the sliding window to update the statistical parameters of the healthy color gamut cluster, generating an updated version of the adaptive color gamut threshold model that better reflects the patient's current true physiological state.

[0127] In step S5, the system compares the latest single-image standard mucosal image with the latest version of the adaptive color gamut threshold model. For each pixel within the mucosal region, the system calculates the statistical distance between its color and the center of the healthy color gamut defined by the model, thereby quantifying its degree of abnormality. Based on this distance, the system classifies abnormal pixels into different risk levels and generates an intuitive risk level heatmap on the image. Finally, the system combines the area and severity of each risk level region to calculate a quantitative risk score and generates an assessment report containing the heatmap and score. When the risk score exceeds a preset clinical safety threshold, the system automatically issues a clinical warning signal.

[0128] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A color-based assessment system for the state of the mucosa of an enterostomy, characterized in that, The method comprises the following steps: a multi-modal data acquisition module for acquiring and analyzing mucosal image sequences, ambient light features and basic physiological data, generating timestamp-aligned acquisition data records; an individualized reference modeling module for establishing a patient-specific color reference based on the acquisition data records and preoperative physiological indicators through HSL color space analysis, generating an initial adaptive color gamut threshold model; an ambient noise decoupling module for analyzing ambient light features and performing physical calibration based on the acquisition data records and the adaptive color gamut threshold model, generating standard mucosal images after ambient noise decoupling; a dynamic threshold updating module for monitoring color gamut drift through a sliding window algorithm based on the standard mucosal image sequences and the current adaptive color gamut threshold model, triggering model iterative optimization when the drift exceeds a preset baseline proportion, and generating an updated adaptive color gamut threshold model; a mucosal state decision module for generating a risk level heat map by comparing abnormal stain distribution based on the standard mucosal images and the adaptive color gamut threshold model, and outputting a clinical early warning signal and a quantitative evaluation report.

2. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, characterized by, The acquisition and analysis of mucosal image sequences, ambient light features and basic physiological data specifically includes: standardizing the raw mucosal images, including noise suppression based on Gaussian filtering, detail enhancement based on limited contrast adaptive histogram equalization, and image distortion correction; applying a semantic segmentation network to extract the stomal mucosa region in the image and exclude the interference area of the surrounding skin and stomal equipment; analyzing ambient light sensor data to obtain the color temperature, illumination and spectral features of the current scene, and temporally and spatially aligning and binding the color temperature, illumination and spectral features with the corresponding image data frames.

3. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, characterized by, The generation of timestamp-aligned acquisition data records specifically includes: synchronizing the acquisition time points of the camera and the ambient light sensor through a hardware trigger signal; decomposing the mucosal image sequence by frame and adding a timestamp to each frame of image; segmenting the color temperature and illumination data stream continuously collected by the ambient light sensor according to the same time reference; establishing a mapping relationship between the image frames and the ambient light data segments to form a timestamp-aligned acquisition data record.

4. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, characterized by, The establishment of a patient-specific color reference based on the acquisition data records and preoperative physiological indicators through HSL color space analysis specifically includes: During the postoperative stable period of the patient, multiple groups of standard mucosal images processed by the ambient noise decoupling module under standard lighting conditions are collected to construct an individual initial sample set; convert the standard mucosal images from the RGB color space to the HSL color space and extract the hue and saturation component values of all pixels in the stomal mucosa region; based on the patient's skin color features in the preoperative physiological indicators, calibrate the extracted hue and saturation components with the initial cluster center, and apply a clustering algorithm to construct a color gamut cluster representing the patient's healthy mucosa state; According to the distribution of the color gamut cluster in the HSL color space, calculate the statistical boundary of its hue and saturation, and establish an initial adaptive color gamut threshold model specific to the patient.

5. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, characterized by, The generating of the standard mucosa image decoupled from the ambient noise based on the acquisition data record and the adaptive color gamut threshold model specifically comprises: extracting the color temperature and illumination data synchronously captured with the ambient image frames from the acquisition data record; constructing a mapping model from the device-dependent color space under the current ambient light condition to the device-independent color space under the standard lighting condition based on the reference values of the standard color check card; applying the mapping model to the original mucosa image for pixel-level color correction; converting the corrected image to the color representation under the standard reference lighting condition and outputting as the standard mucosa image decoupled from the ambient noise.

6. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, wherein The monitoring of the color gamut drift based on the standard mucosa image sequence and the current adaptive color gamut threshold model through the sliding window algorithm specifically comprises: buffering the standard mucosa images generated in the recent period in time sequence to form a sliding window data set with a fixed capacity during the system operation; extracting the hue and saturation values of all pixels in the stoma mucosa region in the HSL color space for each standard mucosa image in the sliding window data set; calculating the distribution characteristics of the hue and saturation of all mucosa pixels in the sliding window data set; comparing the distribution characteristics with the reference distribution characteristics defined by the current adaptive color gamut threshold model to calculate the feature drift amount.

7. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, characterized by, The triggering of the model iteration optimization when the drift amount exceeds the baseline proportion to generate the updated adaptive color gamut threshold model specifically comprises: comparing the calculated feature drift amount with the preset baseline proportion threshold calculated based on the historical operation data of the system; triggering the model optimization process when the feature drift amount exceeds the preset baseline proportion threshold; updating the statistical distribution parameters of the color gamut cluster established by the individualized reference modeling module using the sliding window data set; recomputing the statistical boundaries of the hue and saturation according to the updated color gamut cluster distribution; generating the updated adaptive color gamut threshold model and replacing the original model.

8. The color-comparison-based enterostomy mucosa state evaluation system according to claim 1, characterized by, The generation of the risk level heat map through the comparison of the abnormal stain distribution based on the standard mucosa image and the adaptive color gamut threshold model specifically comprises: performing pixel-level analysis on the standard mucosa image, comparing the hue and saturation values of each pixel in the HSL color space with the normal color gamut range defined by the adaptive color gamut threshold model; identifying and marking the abnormal pixels that exceed the normal color gamut range and classifying them into different risk levels according to their deviation degree; aggregating the abnormal pixels of different risk levels according to their spatial positions to generate a risk level heat map superimposed on the original image; generating a quantitative evaluation report containing the risk score and abnormal region positioning information based on the area proportion and spatial distribution characteristics of each level region in the risk level heat map; generating the corresponding clinical warning signal when the risk score exceeds the preset threshold.

9. A method for assessing the state of the mucosa of an enterostomy based on color comparison, characterized in that, The method for the color comparison-based intestinal stoma mucosa state evaluation system of any one of claims 1-8 comprises the following steps: S1, collect and analyze mucosal image sequences, ambient light features and basic physiological data to generate timestamp-aligned collection data records; S2, based on the collection data records and preoperative physiological indicators, establish a patient-specific color reference benchmark through HSL color space analysis to generate an initial adaptive color gamut threshold model; S3, based on the collection data records and the adaptive color gamut threshold model, analyze the ambient light features and perform physical calibration to generate standard mucosal images after ambient noise decoupling; S4, based on the standard mucosal image sequences and the current adaptive color gamut threshold model, monitor color gamut drift through a sliding window algorithm, and trigger model iterative optimization when the drift exceeds a preset baseline proportion to generate an updated adaptive color gamut threshold model; S5, based on the standard mucosal images and the adaptive color gamut threshold model, generate a risk level heat map by comparing abnormal stain distribution to output a clinical early warning signal and a quantitative evaluation report.

Images