Intelligent evaluation and decision-making system for relapse degree of allergic dermatitis based on skin image
By using an intelligent assessment and decision-making system based on skin images, the recurrence rate of atopic dermatitis can be quantitatively assessed, solving the subjectivity problem of traditional assessment methods and achieving more accurate and efficient assessment and monitoring.
Patent Information
- Application Number
- CN202511008622.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-22
- Publication Date
- 2025-10-31
AI Technical Summary
Traditional methods are difficult to accurately quantify and assess the recurrence of atopic dermatitis, resulting in highly subjective assessment results and making it difficult to conduct dynamic monitoring and evaluate treatment effectiveness.
The intelligent assessment and decision-making system based on skin images constructs a linear regression model through image acquisition, segmentation, and extraction of color, texture, and shape features to achieve a quantitative assessment of the degree of recurrence.
This improves the accuracy and efficiency of recurrence assessment, providing a scientific basis for dynamic monitoring of the disease and evaluation of treatment effectiveness.
Smart Images
Figure CN120876511A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, specifically to an intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images. Background Technology
[0002] Atopic dermatitis is a common, chronic, relapsing, inflammatory skin disease characterized by dry skin, itching, and recurring rashes. This disease not only affects the patient's skin health but is also often accompanied by sleep disturbances and psychological stress, severely impacting their quality of life. Because the relapse mechanisms of atopic dermatitis are complex, involving multiple factors such as genetics, environment, and immunity, accurately assessing the degree of relapse and developing personalized treatment plans has become a clinical challenge.
[0003] Traditional assessment methods rely heavily on physicians' clinical experience, and the results may vary between doctors, leading to a high degree of subjectivity. Patient self-reported symptoms are often inaccurate and easily influenced by factors such as emotions and memory. Furthermore, traditional assessment methods struggle to quantify the recurrence rate of atopic dermatitis, hindering dynamic monitoring of the condition and evaluation of treatment effectiveness. Summary of the Invention
[0004] To address the aforementioned technical problems, this technical solution provides an intelligent assessment and decision-making system for the recurrence severity of atopic dermatitis based on skin images, thus resolving the issues raised in the background section.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0006] An intelligent assessment and decision-making system for the recurrence severity of atopic dermatitis based on skin images includes:
[0007] The image acquisition module is used to intelligently acquire skin images of the atopic dermatitis site of the patient using a skin image acquisition device, and to match the acquired skin images with the patient's identity information;
[0008] The image segmentation module is used to segment a skin image into at least one secondary human body region, remove noise from the skin image, and enhance the difference between the lesion region and normal skin based on image enhancement. The secondary human body region includes the head and neck, trunk, upper limb and lower limb regions.
[0009] The color feature extraction module is used to convert the read skin image into a hue-saturation-brightness color space image. For the converted image, it obtains statistical information on the color distribution in each channel and quantifies and extracts the color features of each skin lesion.
[0010] The texture feature extraction module is used to convert the read skin image into a grayscale image, select a sliding window and generate a grayscale co-occurrence matrix for pixel pairs within the window, and extract texture features from the grayscale co-occurrence matrix.
[0011] The shape feature extraction module is used to extract edge information in the skin image by detecting the degree of gray value change, refine the extracted edges, calculate the geometric shape parameters of the skin lesion area, and output the shape features of each skin lesion.
[0012] The linear regression module is used to fit a regression equation based on the linear relationship between various skin lesion characteristics and the degree of recurrence, construct a linear regression evaluation model, perform cross-validation on it, and optimize the model parameters.
[0013] The intelligent decision-making module is used to display the assessment results to doctors and patients in a visual form, dynamically monitor the degree of recurrence of atopic dermatitis in patients, and provide intelligent treatment suggestions.
[0014] Preferably, the step of converting the read skin image into a hue-saturation-brightness color space image, and for the converted image, obtaining statistical information on the color distribution in each channel, and quantifying and extracting the color features of each skin lesion specifically includes:
[0015] The skin image is converted from the red-green-blue color space to the hue-saturation-lightness color space, and enhancement processing is performed on the hue, saturation, and lightness channels respectively.
[0016] The converted skin image is separated into three independent single-channel images to achieve channel separation, resulting in an array of three channels;
[0017] Based on the value range and requirements of the color channels, the number of histograms for each channel is determined: the number of histograms for the hue channel is 180, and the number of histograms for the saturation and brightness channels is 255.
[0018] Mark the lesion areas in the image and generate a mask image of the lesion areas. In the mask image, the pixel value corresponding to the lesion area is 255, and the pixel value corresponding to the non-lesion area is 0.
[0019] Given an image array, channel indices, a mask, the number of histograms and their ranges, plot the corresponding color histogram.
[0020] Calculate the weighted average, variance, median, mode, kurtosis, and skewness of each histogram value in each channel histogram;
[0021] The color features of the lesion area in the three channels are integrated to form a color feature vector, and the extracted color feature vector is stored in the database.
[0022] Preferably, the step of converting the read skin image into a grayscale image, selecting a sliding window and generating a grayscale co-occurrence matrix for pixel pairs within the window, and extracting texture features from the grayscale co-occurrence matrix specifically includes:
[0023] Based on the weighted coefficients of human eye sensitivity for red, green and blue colors, the values of each channel of the color image are weighted and averaged to obtain a grayscale image;
[0024] Based on the grayscale image resolution and texture feature scale, set the sliding window size and sliding step size;
[0025] Starting from the top left corner of the grayscale image, the sliding window is placed in the corresponding position of the grayscale image in sequence according to the set sliding step size. By continuously moving the sliding window, the grayscale image area is completely covered by the sliding window.
[0026] Within the sliding window, grayscale value pairs are obtained consisting of a pixel and its adjacent pixels, where the adjacent relationship includes horizontal, vertical, and diagonal directions;
[0027] Initialize a zero matrix of size equal to the product of the number of gray levels as the gray-level co-occurrence matrix;
[0028] Iterate through all pixel pairs within the sliding window, increment the count based on the corresponding position of the gray value of the pixel pair in the matrix, and count the frequency of each gray value pair.
[0029] The contrast, correlation, energy, and entropy of the gray-level co-occurrence matrix are calculated using the contrast formula, correlation formula, energy calculation formula, and entropy calculation formula, respectively, and then output as texture features.
[0030] The gray-level co-occurrence matrices in the horizontal, vertical and diagonal directions of the sliding window are calculated respectively, and the texture features in each direction are extracted.
[0031] The texture features from all directions are integrated, and the final texture feature vector of the window is obtained by averaging statistical methods.
[0032] Preferably, the step of extracting edge information from the image by detecting the degree of grayscale value change in the skin image, refining the extracted edges, calculating the geometric shape parameters of the lesion area, and outputting the shape features of each lesion specifically includes:
[0033] The grayscale image is convolved using a horizontal differential operator to obtain the gradient of the grayscale image in the horizontal direction.
[0034] The grayscale image is convolved using a vertical differential operator to obtain the gradient of the grayscale image in the vertical direction.
[0035] Calculate the gradient magnitude of each pixel as the radical of the sum of the squares of the horizontal and vertical gradients, and record the gradient direction of each pixel as the arctangent function value of the ratio of the vertical gradient to the horizontal gradient.
[0036] The calculated gradient magnitude is normalized and mapped to the range of 0-255, and an edge threshold is preset based on the degree of gray value change of the grayscale image.
[0037] Determine whether the gradient magnitude of each pixel is greater than the edge threshold. If it is, mark the pixel as an edge point; otherwise, mark the pixel as a non-edge point.
[0038] Non-maximum suppression is applied to the gradient magnitude image, retaining only the local maxima in the gradient direction, thus connecting the broken edges to form complete edges;
[0039] Determine whether the edge lines are not single-pixel wide. If so, use erosion and dilation in morphological operations to process the edge image and remove redundant pixels at the edge. If not, stop the thinning process.
[0040] In the refined edge image, the contour of the lesion area is extracted to obtain a set of points representing the boundary of the lesion;
[0041] Calculate the number of pixels enclosed by the outline of the lesion area and record it as the area of the lesion area. Calculate the number of pixels on the outline of the lesion area and record it as the perimeter of the lesion area. Determine the minimum bounding rectangle of the lesion area and obtain its length and width parameters.
[0042] The ratio of the area to the perimeter of the lesion region is calculated and recorded as the roundness of the lesion region. The aspect ratio of the smallest bounding rectangle of the lesion region is calculated and recorded as the flatness of the lesion region.
[0043] The calculated geometric parameters are summarized and recorded as shape feature vectors for each lesion area. The geometric parameters include area, perimeter, circumscribed rectangle size, roundness, and flatness.
[0044] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0045] By directly analyzing skin images and extracting features such as color, texture, and shape of skin lesions, the actual condition of the skin can be reflected more accurately, thereby improving the accuracy of recurrence assessment. By constructing a linear regression assessment model, the linear relationship between skin lesion features and recurrence degree is quantified, realizing a quantitative assessment of recurrence degree. This provides a scientific basis for dynamic monitoring of the disease and evaluation of treatment effects. Through automated and intelligent image processing and analysis processes, assessment efficiency is greatly improved. Attached Figure Description
[0046] Figure 1 The flowchart of the intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images is shown in the present invention.
[0047] Figure 2 This is a flowchart of the image segmentation module of the present invention;
[0048] Figure 3 This is a flowchart of the color feature extraction module of the present invention;
[0049] Figure 4 This is a flowchart of the texture feature extraction module of the present invention;
[0050] Figure 5 This is a flowchart of the shape feature extraction module of the present invention;
[0051] Figure 6 This is a flowchart of the linear regression module of the present invention;
[0052] Figure 7 This is a flowchart of the intelligent decision-making module of the present invention. Detailed Implementation
[0053] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0054] Reference Figure 1 As shown, the intelligent assessment and decision-making system for the recurrence severity of atopic dermatitis based on skin images includes:
[0055] The image acquisition module is used to intelligently acquire skin images of the atopic dermatitis site of the patient using a skin image acquisition device, and to match the acquired skin images with the patient's identity information;
[0056] The image segmentation module is used to segment a skin image into at least one secondary human body region, remove noise from the skin image, and enhance the difference between the lesion region and normal skin based on image enhancement. The secondary human body region includes the head and neck, trunk, upper limb and lower limb regions.
[0057] The color feature extraction module is used to convert the read skin image into a hue-saturation-brightness color space image. For the converted image, it obtains statistical information on the color distribution in each channel and quantifies and extracts the color features of each skin lesion.
[0058] The texture feature extraction module is used to convert the read skin image into a grayscale image, select a sliding window and generate a grayscale co-occurrence matrix for pixel pairs within the window, and extract texture features from the grayscale co-occurrence matrix.
[0059] The shape feature extraction module is used to extract edge information in the skin image by detecting the degree of gray value change, refine the extracted edges, calculate the geometric shape parameters of the skin lesion area, and output the shape features of each skin lesion.
[0060] The linear regression module is used to fit a regression equation based on the linear relationship between various skin lesion characteristics and the degree of recurrence, construct a linear regression evaluation model, perform cross-validation on it, and optimize the model parameters.
[0061] The intelligent decision-making module is used to display the assessment results to doctors and patients in a visual form, dynamically monitor the degree of recurrence of atopic dermatitis in patients, and provide intelligent treatment suggestions.
[0062] Reference Figure 2 As shown, the skin image is segmented into at least one secondary human body region, noise in the skin image is removed, and the differences between the lesion area and normal skin are enhanced based on image enhancement. Specifically, this includes:
[0063] Based on human anatomy, we define secondary regions of the human body and manually annotate the collected skin images to mark the boundaries of the secondary regions of the human body in the skin images.
[0064] The types of noise present in skin images were analyzed. The pixel values of Gaussian noise follow a Gaussian distribution, while salt-and-pepper noise consists of black and white pixels in the skin image.
[0065] To address Gaussian noise, the center pixel value is replaced by calculating the average value of the pixels in the neighborhood, thus smoothing the skin image and removing Gaussian noise.
[0066] To address salt-and-pepper noise, the gray values of pixels in the neighborhood are sorted, and the median value is taken as the gray value of the center pixel, thus preserving the edge shape of the skin image and removing the salt-and-pepper noise.
[0067] Global histogram equalization is performed on the denoised skin image. By redistributing the gray values of the skin image, the histogram of the skin image is made more uniform, thereby enhancing the contrast of the skin image.
[0068] A nonlinear function with logarithmic transformation is used to perform contrast stretching on the skin image, compressing the high grayscale region and expanding the low grayscale region.
[0069] Gaussian noise manifests as random fluctuations in pixel values in skin images, with pixel values following a Gaussian normal distribution. Gaussian noise is usually caused by thermal noise from electronic devices such as camera sensors. Salt and pepper noise manifests as randomly appearing black and white pixels in skin images, i.e., extremely bright or extremely dark pixels, usually caused by errors in image transmission or processing.
[0070] Reference Figure 3 As shown, the read skin image is converted into a hue-saturation-lightness color space image. For the converted image, statistical information on the color distribution in each channel is obtained, and the color features of each skin lesion are quantitatively extracted, specifically including:
[0071] The skin image is converted from the red-green-blue color space to the hue-saturation-lightness color space, and enhancement processing is performed on the hue, saturation, and lightness channels respectively.
[0072] The converted skin image is separated into three independent single-channel images to achieve channel separation, resulting in an array of three channels;
[0073] Based on the value range and requirements of the color channels, the number of histograms for each channel is determined: the number of histograms for the hue channel is 180, and the number of histograms for the saturation and brightness channels is 255.
[0074] Mark the lesion areas in the image and generate a mask image of the lesion areas. In the mask image, the pixel value corresponding to the lesion area is 255, and the pixel value corresponding to the non-lesion area is 0.
[0075] Given an image array, channel indices, a mask, the number of histograms and their ranges, plot the corresponding color histogram.
[0076] Calculate the weighted average, variance, median, mode, kurtosis, and skewness of each histogram value in each channel histogram;
[0077] The color features of the lesion area in the three channels are integrated to form a color feature vector, and the extracted color feature vector is stored in the database.
[0078] Kurtosis measures the "peak" or "tail" thickness of a data distribution, that is, the degree of concentration of the data distribution and the probability of extreme values. Kurtosis reflects the degree of concentration of the data distribution around the mean and the degree of dispersion at the tails compared to a normal distribution. Kurtosis is usually calculated by subtracting 3 from the square of the variance of the data distribution. Skewness measures the asymmetry of a data distribution, that is, whether the data distribution deviates from its mean, and the direction and extent of the deviation. Skewness is usually calculated by the cube of the standard deviation of the data distribution.
[0079] Reference Figure 4As shown, the read skin image is converted into a grayscale image, a sliding window is selected, and a gray-level co-occurrence matrix is generated for each pixel pair within the window. Extracting texture features from the gray-level co-occurrence matrix specifically includes:
[0080] Based on the weighted coefficients of human eye sensitivity for red, green and blue colors, the values of each channel of the color image are weighted and averaged to obtain a grayscale image;
[0081] Based on the grayscale image resolution and texture feature scale, set the sliding window size and sliding step size;
[0082] Starting from the top left corner of the grayscale image, the sliding window is placed in the corresponding position of the grayscale image in sequence according to the set sliding step size. By continuously moving the sliding window, the grayscale image area is completely covered by the sliding window.
[0083] Within the sliding window, grayscale value pairs are obtained consisting of a pixel and its adjacent pixels, where the adjacent relationship includes horizontal, vertical, and diagonal directions;
[0084] Initialize a zero matrix of size equal to the product of the number of gray levels as the gray-level co-occurrence matrix;
[0085] Iterate through all pixel pairs within the sliding window, increment the count based on the corresponding position of the gray value of the pixel pair in the matrix, and count the frequency of each gray value pair.
[0086] The contrast, correlation, energy, and entropy of the gray-level co-occurrence matrix are calculated using the contrast formula, correlation formula, energy calculation formula, and entropy calculation formula, respectively, and then output as texture features.
[0087] The gray-level co-occurrence matrices in the horizontal, vertical and diagonal directions of the sliding window are calculated respectively, and the texture features in each direction are extracted.
[0088] The texture features from all directions are integrated, and the final texture feature vector of the window is obtained by averaging statistical methods.
[0089] The contrast ratio formula is:
[0090]
[0091] In the formula, C1 is the contrast of the gray-level co-occurrence matrix, and P(i,j) is the value of the gray-level co-occurrence matrix at position (i,j).
[0092] The correlation formula is:
[0093]
[0094] In the formula, C2 represents the correlation of the gray-level co-occurrence matrix, and μ i μ jσ represents the mean of the rows and columns of the gray-level co-occurrence matrix, respectively. i σ j These are the standard deviations of the rows and columns of the gray-level co-occurrence matrix, respectively.
[0095] The energy calculation formula is as follows:
[0096]
[0097] In the formula, E1 is the energy of the gray-level co-occurrence matrix;
[0098] The entropy calculation formula is as follows:
[0099]
[0100] In the formula, is the entropy of the gray-level co-occurrence matrix;
[0101] Contrast reflects the sharpness and depth of texture grooves in an image; correlation reflects the linear dependence of pixel pairs in an image; energy reflects the uniformity and regularity of image texture; and entropy reflects the complexity and non-uniformity of image texture.
[0102] Reference Figure 5 As shown, by detecting the degree of grayscale value change in skin images, edge information in the image is extracted, and the extracted edges are refined. The geometric shape parameters of the lesion area are calculated, and the shape features of each lesion are output. Specifically, these include:
[0103] The grayscale image is convolved using a horizontal differential operator to obtain the gradient of the grayscale image in the horizontal direction.
[0104] The grayscale image is convolved using a vertical differential operator to obtain the gradient of the grayscale image in the vertical direction.
[0105] Calculate the gradient magnitude of each pixel as the radical of the sum of the squares of the horizontal and vertical gradients, and record the gradient direction of each pixel as the arctangent function value of the ratio of the vertical gradient to the horizontal gradient.
[0106] The calculated gradient magnitude is normalized and mapped to the range of 0-255, and an edge threshold is preset based on the degree of gray value change of the grayscale image.
[0107] Determine whether the gradient magnitude of each pixel is greater than the edge threshold. If it is, mark the pixel as an edge point; otherwise, mark the pixel as a non-edge point.
[0108] Non-maximum suppression is applied to the gradient magnitude image, retaining only the local maxima in the gradient direction, thus connecting the broken edges to form complete edges;
[0109] Determine whether the edge lines are not single-pixel wide. If so, use erosion and dilation in morphological operations to process the edge image and remove redundant pixels at the edge. If not, stop the thinning process.
[0110] In the refined edge image, the contour of the lesion area is extracted to obtain a set of points representing the boundary of the lesion;
[0111] Calculate the number of pixels enclosed by the outline of the lesion area and record it as the area of the lesion area. Calculate the number of pixels on the outline of the lesion area and record it as the perimeter of the lesion area. Determine the minimum bounding rectangle of the lesion area and obtain its length and width parameters.
[0112] The ratio of the area to the perimeter of the lesion region is calculated and recorded as the roundness of the lesion region. The aspect ratio of the smallest bounding rectangle of the lesion region is calculated and recorded as the flatness of the lesion region.
[0113] The calculated geometric parameters are summarized and recorded as shape feature vectors for each lesion area. The geometric parameters include area, perimeter, circumscribed rectangle size, roundness, and flatness.
[0114] Multiple skin images of patients with atopic dermatitis were collected, and the edges of the actual lesion areas were marked. The gradient amplitude of each image was calculated. Based on the statistical characteristics of the gradient amplitude, an initial threshold range was set, such as 10%-50% of the maximum gradient amplitude. Within the set threshold range, all possible thresholds were traversed with a certain step size, such as 5%. For each threshold, edge detection was performed, and the similarity index between the detection results and the actual lesion area edges was calculated. The threshold with the highest similarity index was selected as the optimal threshold.
[0115] Reference Figure 6 As shown, based on the linear relationship between various skin lesion characteristics and recurrence severity, a regression equation is fitted, a linear regression assessment model is constructed, and cross-validation is performed. The optimization of model parameters specifically includes:
[0116] Plot a scatter plot of each skin lesion feature and the target recurrence degree variable, observe the linear trend, and calculate the correlation coefficient matrix;
[0117] Fit regression equations on the training set and output the feature coefficients and intercepts of each skin lesion feature;
[0118] The optimal regularization coefficient is selected through grid search optimization, and key features are screened based on the regression coefficient to remove redundant variables;
[0119] The standardized regression coefficients are output as the direction and intensity of the influence of each skin lesion feature on the degree of recurrence, generating a linear regression assessment model.
[0120] The standardized regression coefficients are output to reflect the influence of each skin lesion feature on the degree of recurrence, with the direction being positive or negative and the intensity being the absolute value of the coefficient. The standardized coefficients facilitate the comparison of the importance of different features. The final linear regression assessment model is generated, including the selected key features, the optimized regularization coefficients, and the standardized regression coefficients. This model can be used to assess the degree of recurrence of skin lesion features in new patients.
[0121] Reference Figure 7 As shown, the assessment results are presented to doctors and patients in a visual format, dynamically monitoring the recurrence rate of atopic dermatitis and providing intelligent treatment recommendations, specifically including:
[0122] A line graph showing the relapse trend was plotted to illustrate how the patient's relapse severity changed over time, with treatment intervention points overlaid and the impact of each event on the relapse marked.
[0123] A risk warning radar chart is drawn to display multidimensional indicators of the patient's current recurrence risk, including the area of skin lesions, the degree of itching, and environmental humidity.
[0124] A feature contribution heatmap was drawn, and the contribution weight of each skin lesion feature to the degree of recurrence was visualized through regression analysis results. The intensity of the influence was represented by the color intensity.
[0125] The dashboard displays the current relapse severity score, overlaid with trend arrows to show recent increases, stabilization, and decreases in the condition.
[0126] Based on the prediction results of the linear regression evaluation model, actionable treatment recommendations are generated.
[0127] Establish a treatment suggestion library containing standardized treatment suggestions for different recurrence levels and characteristic combinations, such as drug treatment plans, moisturizing suggestions, and environmental adjustment suggestions. Based on the model prediction results and the treatment suggestion library, generate actionable treatment suggestions, which may include medication guidance, lifestyle adjustment suggestions such as avoiding allergens and keeping the skin moisturized, and follow-up visit reminders.
[0128] Furthermore, this solution also proposes a computer-readable storage medium storing a computer-readable program, which, when invoked, executes the aforementioned intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images.
[0129] It is understandable that the storage medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid-state drive (SSD).
[0130] In summary, the advantages of this invention are as follows: by directly analyzing skin images and extracting features such as color, texture, and shape of skin lesions, it can more accurately reflect the actual state of the skin, thereby improving the accuracy of recurrence assessment. By constructing a linear regression assessment model, the linear relationship between skin lesion features and recurrence degree is quantified, realizing a quantitative assessment of recurrence degree, providing a scientific basis for dynamic monitoring of the disease and evaluation of treatment effects. Through automated and intelligent image processing and analysis processes, the assessment efficiency is greatly improved.
[0131] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. An intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images, characterized in that, include: The image acquisition module is used to intelligently acquire skin images of the atopic dermatitis site of the patient using a skin image acquisition device, and to match the acquired skin images with the patient's identity information; The image segmentation module is used to segment a skin image into at least one secondary human body region, remove noise from the skin image, and enhance the difference between the lesion region and normal skin based on image enhancement. The secondary human body region includes the head and neck, trunk, upper limb and lower limb regions. The color feature extraction module is used to convert the read skin image into a hue-saturation-brightness color space image. For the converted image, it obtains statistical information on the color distribution in each channel and quantifies and extracts the color features of each skin lesion. The texture feature extraction module is used to convert the read skin image into a grayscale image, select a sliding window and generate a grayscale co-occurrence matrix for pixel pairs within the window, and extract texture features from the grayscale co-occurrence matrix. The shape feature extraction module is used to extract edge information in the skin image by detecting the degree of gray value change, refine the extracted edges, calculate the geometric shape parameters of the skin lesion area, and output the shape features of each skin lesion. The linear regression module is used to fit a regression equation based on the linear relationship between various skin lesion characteristics and the degree of recurrence, construct a linear regression evaluation model, perform cross-validation on it, and optimize the model parameters. The intelligent decision-making module is used to display the assessment results to doctors and patients in a visual form, dynamically monitor the degree of recurrence of atopic dermatitis in patients, and provide intelligent treatment suggestions.
2. The intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images according to claim 1, characterized in that, The process of segmenting the skin image into at least one secondary human body region, removing noise from the skin image, and enhancing the difference between the lesion area and normal skin based on image enhancement specifically includes: Based on human anatomy, we define secondary regions of the human body and manually annotate the collected skin images to mark the boundaries of the secondary regions of the human body in the skin images. The types of noise present in skin images were analyzed. The pixel values of Gaussian noise follow a Gaussian distribution, while salt-and-pepper noise consists of black and white pixels in the skin image. To address Gaussian noise, the center pixel value is replaced by calculating the average value of the pixels in the neighborhood, thus smoothing the skin image and removing Gaussian noise. To address salt-and-pepper noise, the gray values of pixels in the neighborhood are sorted, and the median value is taken as the gray value of the center pixel, thus preserving the edge shape of the skin image and removing the salt-and-pepper noise. Global histogram equalization is performed on the denoised skin image. By redistributing the gray values of the skin image, the histogram of the skin image is made more uniform, thereby enhancing the contrast of the skin image. A nonlinear function with logarithmic transformation is used to perform contrast stretching on the skin image, compressing the high grayscale region and expanding the low grayscale region.
3. The intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images according to claim 2, characterized in that, The process of converting the read skin image into a hue-saturation-brightness color space image, and then obtaining statistical information on the color distribution in each channel of the converted image, and quantifying and extracting the color features of each skin lesion specifically includes: The skin image is converted from the red-green-blue color space to the hue-saturation-lightness color space, and enhancement processing is performed on the hue, saturation, and lightness channels respectively. The converted skin image is separated into three independent single-channel images to achieve channel separation, resulting in an array of three channels; Based on the value range and requirements of the color channels, the number of histograms for each channel is determined: the number of histograms for the hue channel is 180, and the number of histograms for the saturation and brightness channels is 255. Mark the lesion areas in the image and generate a mask image of the lesion areas. In the mask image, the pixel value corresponding to the lesion area is 255, and the pixel value corresponding to the non-lesion area is 0. Given an image array, channel indices, a mask, the number of histograms and their ranges, plot the corresponding color histogram. Calculate the weighted average, variance, median, mode, kurtosis, and skewness of each histogram value in each channel histogram; The color features of the lesion area in the three channels are integrated to form a color feature vector, and the extracted color feature vector is stored in the database.
4. The intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images according to claim 3, characterized in that, The process of converting the read skin image into a grayscale image, selecting a sliding window and generating a grayscale co-occurrence matrix for pixel pairs within the window, and extracting texture features from the grayscale co-occurrence matrix specifically includes: Based on the weighted coefficients of human eye sensitivity for red, green and blue colors, the values of each channel of the color image are weighted and averaged to obtain a grayscale image; Based on the grayscale image resolution and texture feature scale, set the sliding window size and sliding step size; Starting from the top left corner of the grayscale image, the sliding window is placed in the corresponding position of the grayscale image in sequence according to the set sliding step size. By continuously moving the sliding window, the grayscale image area is completely covered by the sliding window. Within the sliding window, grayscale value pairs are obtained consisting of a pixel and its adjacent pixels, where the adjacent relationship includes horizontal, vertical, and diagonal directions; Initialize a zero matrix of size equal to the product of the number of gray levels as the gray-level co-occurrence matrix; Iterate through all pixel pairs within the sliding window, increment the count based on the corresponding position of the gray value of the pixel pair in the matrix, and count the frequency of each gray value pair. The contrast, correlation, energy, and entropy of the gray-level co-occurrence matrix are calculated using the contrast formula, correlation formula, energy calculation formula, and entropy calculation formula, respectively, and then output as texture features. The gray-level co-occurrence matrices in the horizontal, vertical and diagonal directions of the sliding window are calculated respectively, and the texture features in each direction are extracted. The texture features from all directions are integrated, and the final texture feature vector of the window is obtained by averaging statistical methods.
5. The intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images according to claim 4, characterized in that, The process of detecting changes in grayscale values in skin images, extracting edge information from the image, refining the extracted edges, calculating the geometric shape parameters of the lesion region, and outputting the shape features of each lesion specifically includes: The grayscale image is convolved using a horizontal differential operator to obtain the gradient of the grayscale image in the horizontal direction. The grayscale image is convolved using a vertical differential operator to obtain the gradient of the grayscale image in the vertical direction. Calculate the gradient magnitude of each pixel as the radical of the sum of the squares of the horizontal and vertical gradients, and record the gradient direction of each pixel as the arctangent function value of the ratio of the vertical gradient to the horizontal gradient. The calculated gradient magnitude is normalized and mapped to the range of 0-255, and an edge threshold is preset based on the degree of gray value change of the grayscale image. Determine whether the gradient magnitude of each pixel is greater than the edge threshold. If it is, mark the pixel as an edge point; otherwise, mark the pixel as a non-edge point. Non-maximum suppression is applied to the gradient magnitude image, retaining only the local maxima in the gradient direction, thus connecting the broken edges to form complete edges; Determine whether the edge lines are not single-pixel wide. If so, use erosion and dilation in morphological operations to process the edge image and remove redundant pixels at the edge. If not, stop the thinning process. In the refined edge image, the contour of the lesion area is extracted to obtain a set of points representing the boundary of the lesion; Calculate the number of pixels enclosed by the outline of the lesion area and record it as the area of the lesion area. Calculate the number of pixels on the outline of the lesion area and record it as the perimeter of the lesion area. Determine the minimum bounding rectangle of the lesion area and obtain its length and width parameters. The ratio of the area to the perimeter of the lesion region is calculated and recorded as the roundness of the lesion region. The aspect ratio of the smallest bounding rectangle of the lesion region is calculated and recorded as the flatness of the lesion region. The calculated geometric parameters are summarized and recorded as shape feature vectors for each lesion area. The geometric parameters include area, perimeter, circumscribed rectangle size, roundness, and flatness.
6. The intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images according to claim 5, characterized in that, Based on the linear relationship between various skin lesion characteristics and recurrence severity, a regression equation is fitted, a linear regression evaluation model is constructed, and cross-validation is performed on it. The optimization of model parameters specifically includes: Plot a scatter plot of each skin lesion feature and the target recurrence degree variable, observe the linear trend, and calculate the correlation coefficient matrix; Fit regression equations on the training set and output the feature coefficients and intercepts of each skin lesion feature; The optimal regularization coefficient is selected through grid search optimization, and key features are screened based on the regression coefficient to remove redundant variables; The standardized regression coefficients are output as the direction and intensity of the influence of each skin lesion feature on the degree of recurrence, generating a linear regression assessment model.
7. The intelligent assessment and decision-making system for the degree of recurrence of atopic dermatitis based on skin images according to claim 6, characterized in that, The process of visually presenting the assessment results to doctors and patients, dynamically monitoring the recurrence rate of atopic dermatitis, and providing intelligent treatment recommendations specifically includes: A line graph showing the relapse trend was plotted to illustrate how the patient's relapse severity changed over time, with treatment intervention points overlaid and the impact of each event on the relapse marked. A risk warning radar chart is drawn to display multidimensional indicators of the patient's current recurrence risk, including the area of skin lesions, the degree of itching, and environmental humidity. A feature contribution heatmap was drawn, and the contribution weight of each skin lesion feature to the degree of recurrence was visualized through regression analysis results. The intensity of the influence was represented by the color intensity. The dashboard displays the current relapse severity score, overlaid with trend arrows to show recent increases, stabilization, and decreases in the condition. Based on the prediction results of the linear regression evaluation model, actionable treatment recommendations are generated.