An atopic dermatitis diagnosis and treatment support system

Abstract

The application belongs to the medical application field, and specifically discloses an atopic dermatitis auxiliary diagnosis and treatment system, which comprises a user layer, a service layer, a capability layer and an infrastructure layer. The atopic dermatitis auxiliary diagnosis and treatment system is constructed by combining the theories of clinical dermatology, computer science, psychology and information science, and can be applied to realize standard treatment of AD, improve the life quality and treatment satisfaction of patients, and thus improve the overall diagnosis and treatment level of AD.

Description

Technical Field

[0001] This invention belongs to the field of medical applications, specifically relating to an auxiliary diagnostic and treatment system for atopic dermatitis. Background Technology

[0002] Atopic dermatitis (AD) is a chronic inflammatory skin disease characterized by recurrent rashes and itching, severely impacting patients' quality of life. In 2019, the number of people with atopic dermatitis in China was approximately 61.5 million, and this number is projected to reach 63.9 million in 2024 and 65.9 million in 2030, placing a heavy economic burden on families and society. The international consensus on "Treat to Target Treatment for Atopic Dermatitis—A Core Decision Framework for Systemic Treatment of Moderate to Severe AD in Adults" proposes that AD treatment goals should be broken down and quantified using multi-dimensional disease assessment tools, emphasizing dynamic adjustments to treatment plans based on regular follow-up assessments. The core of the Treat to Target (T2T) approach to AD diagnosis, treatment, and follow-up is how to maximize the integration and utilization of existing resources to safely and effectively help patients achieve lasting disease control. It involves refining follow-up assessments, applying different assessment indicators and tools at different time points, and emphasizing the combination of patient and physician assessments to improve patient adherence and provide important clinical decision-making support for clinicians. China has a large number of Alzheimer's disease (AD) patients with varying levels of education and a lack of professional knowledge, making self-assessment extremely difficult for them. Limited contact time also hinders clinicians from meeting patients' psychological and social needs. Therefore, enhancing patient participation in AD treatment, promoting scheduled appointments, facilitating more effective doctor-patient communication, and implementing meticulous assessment and follow-up are key challenges in achieving type-to-type (T2T) care for AD.

[0003] my country's management of Alzheimer's disease (AD) is currently limited to methods such as in-person consultations, telephone follow-ups, and instruction manuals, making it difficult to achieve the goal of treatment-to-treatment (T2T) for AD patients. To realize the T2T management goal for AD patients, a convenient, accessible, and economical new AD diagnosis and treatment management system needs to be built. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides an auxiliary diagnosis and treatment system for atopic dermatitis, comprising: a user layer, a business layer, a capability layer, and an infrastructure layer;

[0005] The user layer includes a web interface, mobile app, PC client, or mini-program, which provides an interactive interface for users.

[0006] The business layer includes an upload module, a display module, and an operation module;

[0007] The upload module allows users to upload medical data from their devices to the server; the display module is used to visually display medical information; and the operation module is used to input interactive commands for users.

[0008] The capability layer includes an AI intelligence module; the AI ​​intelligence module includes a deep learning algorithm processing unit and a communication processing unit.

[0009] The deep learning algorithm processing unit is used to analyze and process the rash photos and self-assessment scores uploaded by patient users to the server;

[0010] The communication processing unit is used to transmit the analysis and processing results to the user layer and render and display them to the user on a webpage or APP;

[0011] The infrastructure layer includes server hardware, server operating system, and database. The infrastructure layer is used to provide computing and storage resources for the operation of the capability layer.

[0012] Furthermore, when the user is a patient, the user layer allows the patient to upload photos of the rash through a web browser, mobile app, PC software, or mini-program, complete a self-assessment, and refer to the AI ​​condition assessment and medical advice.

[0013] Furthermore, when the user is a medical professional, the user layer allows the medical professional to view and / or process the patient's condition data and give medical advice through a web browser, mobile APP, PC software, or mini-program; the condition data includes rash photos, self-scores, and AI condition assessments.

[0014] Furthermore, the upload module supports patient users to upload photos of rashes and self-assessments from their devices and transmit them to the server via the network; it also supports medical staff to upload medical orders from their devices and transmit them to the server via the network; the network used can be a wired network or a wireless network.

[0015] Furthermore, the display module is used to visually display the rash photos and self-assessment scores uploaded by patient users, as well as the medical orders uploaded by medical staff.

[0016] Furthermore, the interactive instructions are voice instructions or text instructions.

[0017] Furthermore, the AI ​​intelligent module analyzes and processes the rash photos and self-assessment scores uploaded by patient users to the server using deep learning algorithms.

[0018] Furthermore, the processing steps of the deep learning algorithm processing unit include:

[0019] Data preprocessing: The rash photos are subjected to format conversion, size normalization, brightness and contrast adjustment, noise reduction and standardization; for self-assessment data, structured extraction is performed and stored according to the preset format;

[0020] Feature extraction and segmentation of rash images: A segmentation network based on U-Net or its improved variants is used to accurately segment the lesion region in the rash image through an encoder-decoder structure, resulting in a lesion region map and a binary or multi-class mask of the lesion.

[0021] Disease assessment: ① Combine the segmented lesion area map with the self-score, and use convolutional neural networks or other classification / regression models to assess the severity of the lesions or the healing trend;

[0022] ② The severity or healing trend assessment results of the lesions, the binary or multi-class mask of the lesions, and the patient's previous medical data stored in the system are input into the image-level regression / classification network. The encoder is used to output the severity score and rash grade after passing through the Global Average Pooling (GAP) + MLP head.

[0023] Furthermore, the processing steps also include data augmentation of the rash images through rotation, flipping, random cropping, illumination variation and / or generative adversarial networks to improve the model's generalization ability;

[0024] The improved variants of U-Net include U-Net++ and UNext;

[0025] The encoder in the encoder-decoder is a ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset, and the decoder is a deconvolution / upsampling decoder.

[0026] After precise segmentation, a residual network or attention mechanism module is introduced into the segmentation network to enhance the ability to capture multi-scale lesion features and reduce background interference.

[0027] The encoder used for disease assessment is a ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset.

[0028] Furthermore, the self-assessment scores include EASI, POEM, PP-NRS, and / or DLQI.

[0029] The beneficial effects of this invention are:

[0030] This invention discloses an auxiliary diagnostic and treatment system for atopic dermatitis (AD), which is constructed by combining theories from clinical dermatology, computer science, psychology, and informatics. The system utilizes AI intelligent modules built with specific algorithms in an app / mini-program to help achieve target treatment outcomes for AD, improving patients' quality of life and treatment satisfaction, thereby enhancing the overall level of AD diagnosis and treatment.

[0031] Obviously, based on the above description of the present invention, and according to common technical knowledge and conventional methods in the field, various other modifications, substitutions or alterations can be made without departing from the basic technical concept of the present invention.

[0032] The following detailed embodiments further illustrate the above-described content of the present invention. However, this should not be construed as limiting the scope of the present invention to the following examples. All technologies implemented based on the above-described content of the present invention fall within the scope of the present invention. Attached Figure Description

[0033] Figure 1 Atopic dermatitis quality assessment criteria. Detailed Implementation

[0034] Example 1: An auxiliary diagnostic and treatment system for atopic dermatitis according to the present invention

[0035] The system architecture includes a user layer, a business layer, a capability layer, and an infrastructure layer;

[0036] The user layer includes web-based platforms, mobile apps, PC clients, or mini-programs, providing users with an interactive interface. Patients can upload photos of their rashes, complete self-assessments (EASI, POEM, PP-NRS, and DLQI), and view AI-based disease assessments and medical orders via web browsers, mobile apps, PC software, or mini-programs. Healthcare professionals can view and / or process patient data (rashe photos, self-assessments, and AI-based disease assessments) and provide medical orders via web browsers, mobile apps, PC software, or mini-programs. Users only need to install the corresponding software on their mobile phones or PCs to use these functions.

[0037] The business layer includes an upload module, a display module, and an operation module;

[0038] The upload module supports patient users to upload photos of rashes and self-assessments from their devices and transmit them to the server via the network; it also supports healthcare users to upload medical orders from their devices and transmit them to the server via the network; the network used can be a wired network or a wireless network.

[0039] The display module is used to visually show photos of rashes and self-assessments uploaded by patient users, as well as medical orders uploaded by healthcare users.

[0040] The operation module is used to input interactive commands to the user. The commands issued by the user can be voice commands or text commands.

[0041] The capability layer includes an AI intelligence module; the AI ​​intelligence module includes a deep learning algorithm processing unit and a communication processing unit.

[0042] The deep learning algorithm processing unit is used to analyze and process the rash photos and self-reported scores uploaded to the server by patient users. The specific analysis and processing steps are as follows:

[0043] 1. Data preprocessing: The rash photos are converted in format, normalized in size, adjusted in brightness / contrast, denoised and standardized; the self-assessment data are extracted in a structured manner and stored in a preset format;

[0044] 2. Data Augmentation (Optional): Augment the rash images by rotating, flipping, randomly cropping, changing the lighting, or using a Generative Adversarial Network (GAN) to improve the model's generalization ability.

[0045] 3. Feature extraction and segmentation of the rash image:

[0046] A U-Net-based segmentation network is used to accurately segment lesion regions in rash images through an encoder-decoder structure (i.e., a ResNet-50 encoder with weights initialized by pre-training on the ImageNet dataset, and a deconvolution / upsampling decoder). A residual network (ResNet) is introduced into the segmentation network to enhance the ability to capture multi-scale lesion features, reduce background interference, and obtain lesion region maps and binary or multi-class masks of lesions.

[0047] 4. Disease assessment:

[0048] ① Combine the segmented lesion area map with self-scoring (EASI, POEM, PP-NRS and DLQI) and use a convolutional neural network (CNN) to assess the severity of the lesion or the healing trend;

[0049] ② Input the lesion severity or healing trend assessment results obtained in step ①, the binary or multi-class mask of the lesion obtained in step 3, and the patient's previous medical data stored in the system into the image-level regression / classification network. Use the ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset, and pass it through the Global Average Pooling (GAP) + MLP head to output the rash severity score (EASI) and rash grade (mild / moderate / severe).

[0050] The communication processing unit is used to transmit the analysis results to the user layer and render and display them to the user on a webpage or APP.

[0051] The infrastructure layer includes server hardware, server operating system, and database. The infrastructure layer is used to provide computing and storage resources for the operation of the capability layer.

[0052] Methods for auxiliary diagnosis and treatment of atopic dermatitis using an auxiliary diagnostic and treatment system include:

[0053] 1) Patients upload photos of their rashes and complete self-assessment (EASI, POEM, PP-NRS, and DLQI) through a web browser, mobile app, PC software, or mini-program user interface.

[0054] 2) AI assists in assessing the condition based on photos of the rash uploaded by the patient and their self-reported scores;

[0055] 3) Medical staff can view the data uploaded by patients through the user interface on a web browser, mobile APP or PC software, and the AI-assisted assessment results can be used to provide and upload medical orders.

[0056] Example 2: An auxiliary diagnostic and treatment system for atopic dermatitis according to the present invention

[0057] The system architecture includes a user layer, a business layer, a capability layer, and an infrastructure layer;

[0058] The user layer includes web-based platforms, mobile apps, PC clients, or mini-programs, providing users with an interactive interface. Patients can upload photos of their rashes, complete self-assessments (EASI, POEM, PP-NRS, and DLQI), and view AI-based disease assessments and medical orders via web browsers, mobile apps, PC software, or mini-programs. Healthcare professionals can view and / or process patient data (rashe photos, self-assessments, and AI-based disease assessments) and provide medical orders via web browsers, mobile apps, PC software, or mini-programs. Users only need to install the corresponding software on their mobile phones or PCs to use these functions.

[0059] The business layer includes an upload module, a display module, and an operation module;

[0060] The upload module supports patient users to upload photos of rashes and self-assessments from their devices and transmit them to the server via the network; it also supports healthcare users to upload medical orders from their devices and transmit them to the server via the network; the network used can be a wired network or a wireless network.

[0061] The display module is used to visually show photos of rashes and self-assessments uploaded by patient users, as well as medical orders uploaded by healthcare users.

[0062] The operation module is used to input interactive commands to the user. The commands issued by the user can be voice commands or text commands.

[0063] The capability layer includes an AI intelligence module; the AI ​​intelligence module includes a deep learning algorithm processing unit and a communication processing unit.

[0064] The deep learning algorithm processing unit is used to analyze and process the rash photos and self-reported scores uploaded to the server by patient users. The specific analysis and processing steps are as follows:

[0065] 1. Data preprocessing: The rash photos are converted in format, normalized in size, adjusted in brightness / contrast, denoised and standardized; the self-assessment data are extracted in a structured manner and stored in a preset format;

[0066] 2. Data Augmentation (Optional): Augment the rash images by rotating, flipping, randomly cropping, changing the lighting, or using a Generative Adversarial Network (GAN) to improve the model's generalization ability.

[0067] 3. Feature extraction and segmentation of the rash image:

[0068] A segmentation network based on an improved variant of U-Net (U-Net++) is used to accurately segment lesion regions in rash images through an encoder-decoder structure (i.e., a ResNet-50 encoder with weights initialized by pre-training on the ImageNet dataset, and a deconvolution / upsampling decoder). An attention mechanism (Transformer) module is introduced into the segmentation network to enhance the ability to capture multi-scale lesion features, reduce background interference, and obtain lesion region maps and binary or multi-class masks of lesions.

[0069] 4. Disease assessment:

[0070] ① Combine the segmented lesion area map with self-scores (EASI, POEM, PP-NRS and DLQI), and use a classification / regression model (attention mechanism network Transformert) to assess the severity of lesions or healing trends;

[0071] ② Input the lesion severity or healing trend assessment results obtained in step ①, the binary or multi-class mask of the lesion obtained in step 3, and the patient's previous medical data stored in the system into the image-level regression / classification network. Use the ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset, and pass it through the Global Average Pooling (GAP) + MLP head to output the rash severity score (EASI) and rash grade (mild / moderate / severe).

[0072] The communication processing unit is used to transmit the analysis results to the user layer and render and display them to the user on a webpage or APP.

[0073] The infrastructure layer includes server hardware, server operating system, and database. The infrastructure layer is used to provide computing and storage resources for the operation of the capability layer.

[0074] The method for auxiliary diagnosis and treatment of atopic dermatitis using the auxiliary diagnosis and treatment system for atopic dermatitis is the same as in Example 1.

[0075] Example 3: An auxiliary diagnostic and treatment system for atopic dermatitis according to the present invention

[0076] The system architecture includes a user layer, a business layer, a capability layer, and an infrastructure layer;

[0077] The user layer includes web-based platforms, mobile apps, PC clients, or mini-programs, providing users with an interactive interface. Patients can upload photos of their rashes, complete self-assessments (EASI, POEM, PP-NRS, and DLQI), and view AI-based disease assessments and medical orders via web browsers, mobile apps, PC software, or mini-programs. Healthcare professionals can view and / or process patient data (rashe photos, self-assessments, and AI-based disease assessments) and provide medical orders via web browsers, mobile apps, PC software, or mini-programs. Users only need to install the corresponding software on their mobile phones or PCs to use these functions.

[0078] The business layer includes an upload module, a display module, and an operation module;

[0079] The upload module supports patient users to upload photos of rashes and self-assessments from their devices and transmit them to the server via the network; it also supports healthcare users to upload medical orders from their devices and transmit them to the server via the network; the network used can be a wired network or a wireless network.

[0080] The display module is used to visually show photos of rashes and self-assessments uploaded by patient users, as well as medical orders uploaded by healthcare users.

[0081] The operation module is used to input interactive commands to the user. The commands issued by the user can be voice commands or text commands.

[0082] The capability layer includes an AI intelligence module; the AI ​​intelligence module includes a deep learning algorithm processing unit and a communication processing unit.

[0083] The deep learning algorithm processing unit is used to analyze and process the rash photos and self-reported scores uploaded to the server by patient users. The specific analysis and processing steps are as follows:

[0084] 1. Data preprocessing: The rash photos are converted in format, normalized in size, adjusted in brightness / contrast, denoised and standardized; the self-assessment data are extracted in a structured manner and stored in a preset format;

[0085] 2. Data Augmentation (Optional): Augment the rash images by rotating, flipping, randomly cropping, changing the lighting, or using a Generative Adversarial Network (GAN) to improve the model's generalization ability.

[0086] 3. Feature extraction and segmentation of the rash image:

[0087] A segmentation network based on an improved variant of U-Net (UNext) is used to accurately segment lesion regions in rash images through an encoder-decoder structure (i.e., a ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset, and a deconvolution / upsampling decoder). An attention mechanism (Transformer) module is introduced into the segmentation network to enhance the ability to capture multi-scale lesion features, reduce background interference, and obtain lesion region maps and binary or multi-class masks of lesions.

[0088] 4. Disease assessment:

[0089] ① Combine the segmented lesion area map with self-scoring (EASI, POEM, PP-NRS and DLQI), and use a classification / regression model (Residual Network ResNet) to assess the severity of the lesion or the healing trend;

[0090] ② Input the lesion severity or healing trend assessment results obtained in step ①, the binary or multi-class mask of the lesion obtained in step 3, and the patient's previous medical data stored in the system into the image-level regression / classification network. Use the ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset, and pass it through the Global Average Pooling (GAP) + MLP head to output the rash severity score (EASI) and rash grade (mild / moderate / severe).

[0091] The communication processing unit is used to transmit the analysis results to the user layer and render and display them to the user on a webpage or APP.

[0092] The infrastructure layer includes server hardware, server operating system, and database. The infrastructure layer is used to provide computing and storage resources for the operation of the capability layer.

[0093] The method for auxiliary diagnosis and treatment of atopic dermatitis using the auxiliary diagnosis and treatment system for atopic dermatitis is the same as in Example 1.

[0094] The following experimental examples further illustrate the beneficial effects of the present invention.

[0095] Experimental Example 1

[0096] 1. Clinical research protocol

[0097] 1.1 Multicenter Survey Study

[0098] In collaboration with more than five medical institutions, the "AD Diagnosis and Treatment App," an auxiliary diagnosis and treatment system constructed in Example 1, was evaluated on its usage. Medical staff and patients were asked to rate their satisfaction on a scale of 0 to 100, with 0 representing complete dissatisfaction and 100 representing high satisfaction. This yielded a comprehensive assessment of the satisfaction of both medical staff and patients with the App. Questionnaires were also distributed periodically to solicit opinions and suggestions from both medical staff and patients. The Information Center of West China Hospital then upgraded and improved the App to further meet user needs.

[0099] 1.2 Randomized controlled clinical trials

[0100] 1.2.1 Study Subjects: AD patients aged 1–75 years who visited the Department of Dermatology and Venereology, West China Hospital of Sichuan University, met the diagnostic criteria for AD, and whose own or their guardians were proficient in using mobile smart terminal devices such as smartphones and tablets, could communicate in Chinese, and had basic reading and writing abilities, were included in the study. Patients with comorbid mental illness, language, hearing, or communication impairments, serious comorbidities such as cardiopulmonary insufficiency, liver or kidney dysfunction, malignant tumors, or other conditions unsuitable for participation in clinical research were excluded.

[0101] 1.2.2 Calculation of Sample Size

[0102] According to the "Core Decision-Making Framework for Drug-Based Treatment of Moderate to Severe Alzheimer's Disease" Figure 1The endpoints specified in the treatment plan, namely 3 months and 6 months post-treatment, are the Eczema Area and Severity Index (EASI), Patient Oriented Eczema Measure (POEM), Peak Pruritus Numerical Rating Scale (PP-NRS), Scoring Atopic Dermatitis (SCORAD), and Dermatology Life Quality Index. The overall achievement rate of treatment goals (DLQI) was used as the primary indicator in this study. The calculation method was: (Total number of patients achieving the five treatment goals / Total number of patients achieving the five treatment goals) * 100%. Based on previous experience, the overall achievement rate of treatment goals was 35%. Assuming that an overall achievement rate of 50% after implementing T2T management would be clinically significant, an experimental group and a control group were established, with α = 0.05, 1-β = 90%, and a two-sided differential test. Each group required 55 patients. Considering a 20% loss to follow-up rate, 35 patients were included in each group, for a total of 70 AD patients in both groups.

[0103] 1.2.3 Intervention Program

[0104] Using SPSS 20.0 to generate a random number table, AD patients who met the inclusion criteria were randomly divided into the App group and the control group according to the order of their visits. Researchers explained the purpose and significance of the trial to the subjects or their guardians and obtained their informed consent.

[0105] App Group: On the day of their visit, participants were assisted by researchers in installing, registering, and trying out the "AD Treatment App." They logged into the App using their ID number and set their own password. Healthcare staff used a fixed account and password to log into the "AD Treatment App" installed on a tablet or mobile phone for the diagnosis, follow-up, and management of AD patients. One week prior to treatment, at 2, 4, 8, 12 weeks, 6 months, and 12 months, push notifications were sent to remind patients to upload photos of their rashes, complete various assessments, and upload them. The system automatically interpreted the assessments and sent push notifications to healthcare staff for manual evaluation and follow-up management. In addition to completing necessary assessments at the designated time points, patients could initiate self-assessments at any time.

[0106] Control group: AD patients were diagnosed, followed up, and managed according to current standard treatment protocols. At each patient visit, various assessments were conducted, treatment recommendations were given on-site, and the corresponding clinical data were entered into a computer for storage.

[0107] 1.2.4 Evaluation Indicators

[0108] 1.2.4.1 Main indicator: The average weekly medication dosage for the two groups of patients after 12 weeks of treatment.

[0109] 1.2.4.2 Secondary Indicators:

[0110] 1) Percentage of patients with an IGA score ≤2 or an improvement of ≥1 point from baseline after 12 weeks of treatment in both groups. The achievement rate of Atopic Dermatitis Control Tool (ADCT) <7 after 12 weeks of treatment was calculated as (number of patients achieving ADCT <7 / total number of patients) * 100%.

[0111] 2) The proportion of patients with EASI-50, EASI-75 and EASI≤7 scores in the two groups after 12 weeks of treatment.

[0112] 3) The proportion of patients with SCORAD-50, SCORAD-75 and SCORAD ≤24 points in both groups after 12 weeks of treatment.

[0113] 4) The proportion of patients with NRS improvement ≥3 points and score ≤4 points in both groups after 12 weeks of treatment.

[0114] 5) The proportion of patients with DLQI improvement of ≥4 points and ≤5 points in both groups after 12 weeks of treatment.

[0115] 6) The proportion of patients with POEM improvement ≥4 points and scores ≤7 points in both groups after 12 weeks of treatment.

[0116] 7) Treatment target achievement rate of the two groups of patients at 3 months and 6 months after treatment.

[0117] 8) Economic comparison: mainly the economic costs from the patient's perspective, such as travel expenses, lost wages, accommodation fees, and medical expenses for the patient and their companion.

[0118] 9) Time lost due to medical treatment is defined as work and leisure time lost due to medical treatment for AD, such as work leave, weekend or holiday leisure time sacrificed for medical treatment, in hours.

[0119] 1.3 Statistical Methods

[0120] SPSS 20.0 software was used for statistical analysis. Data were described using mean, standard deviation, frequency, median, and percentage. t-tests were used for continuously distributed variables that conformed to a normal distribution, chi-square tests were used for categorical variables, and nonparametric tests were used for variables that did not conform to a normal distribution. Repeated measures ANOVA was used to assess the achievement rates of treatment goals at each time point after treatment. P < 0.05 was considered statistically significant.

[0121] 2. Results

[0122] 2.1 Key Indicators

[0123] After 12 weeks of treatment, the average medication dosage of the subjects was calculated. After analysis using an independent samples t-test, it was found that the average weekly medication dosage of the App group was significantly lower than that of the control group, and the difference was statistically significant (P<0.05).

[0124] 2.2 Secondary Indicators

[0125] The proportion of subjects with SCORAD-50, SCORAD-75, and SCORAD scores ≤24 at 12 weeks post-treatment was calculated. Results showed that the percentage of patients in the experimental group was significantly higher than that in the control group (P < 0.05).

[0126] The proportions of patients with EASI-50, EASI-75, and EASI ≤7 scores at 12 weeks post-treatment were calculated. Results showed that the percentage of patients in the experimental group was significantly higher than that in the control group (P < 0.05).

[0127] The proportion of patients with NRS improvement ≥3 and ≤4, DLQI improvement ≥4 and ≤5, and POEM improvement ≥4 and ≤7 at 12 weeks post-treatment were calculated. Results showed that the proportion of patients in the experimental group was significantly higher than that in the control group (P < 0.05).

[0128] Treatment target achievement rates were calculated at 3 months and 1 year post-treatment. The results showed that the treatment target achievement rates at 3 months and 1 year were significantly higher in the experimental group than in the control group, with statistically significant differences (P < 0.05).

[0129] Data from all follow-up time points were used to dynamically capture the intervention effect. A linear mixed-effects model was employed, incorporating group, time, and group × time interaction terms, while controlling for individual random effects, to analyze the dynamic changes in the intervention effect. The results showed a significant interaction effect between group and time (P < 0.05), indicating that the trial intervention had a significant effect on improving patients' conditions.

[0130] The comparative economic indicators mainly focused on the economic costs from the patient's perspective, such as travel expenses, lost wages, accommodation costs, and medical expenses for the patient and their companion. The results showed that the average economic costs for patients in the experimental group were lower than those in the control group, and the economic burden on the experimental group was significantly reduced. The difference was statistically significant (P<0.05), indicating that "AD Diagnosis and Treatment Pass" can reduce the medical costs for AD patients.

[0131] The study compared the time lost due to medical visits between the two groups of patients, defined as the work and leisure time lost due to AD medical visits, such as work leave, weekend or holiday leisure time sacrificed for medical visits, expressed in hours. Results showed that the average time lost due to medical visits in the experimental group was less than that in the control group, with a significant reduction in time loss in the experimental group (P<0.05). This indicates that the "AD Treatment App" can save patients' time and improve medical efficiency.

[0132] The above results indicate that "AD Diagnosis and Treatment Tool" has significant advantages in assisting the diagnosis, treatment, and follow-up management of AD patients. It can improve treatment outcomes, reduce economic burden, and save time costs, providing AD patients with a convenient and efficient management tool with promising application prospects.

Claims

1. An auxiliary diagnostic and treatment system for atopic dermatitis, characterized in that, It includes: user layer, business layer, capability layer and infrastructure layer; The user layer includes a web interface, mobile app, PC client, or mini-program, which provides an interactive interface for users. The business layer includes an upload module, a display module, and an operation module; The upload module allows users to upload medical data from their devices to the server; the display module is used to visually display medical information; and the operation module is used to input interactive commands for users. The capability layer includes an AI intelligence module; the AI ​​intelligence module includes a deep learning algorithm processing unit and a communication processing unit. The deep learning algorithm processing unit is used to analyze and process the rash photos and self-assessment scores uploaded by patient users to the server; The communication processing unit is used to transmit the analysis and processing results to the user layer and render and display them to the user on a webpage or APP; The infrastructure layer includes server hardware, server operating system, and database. The infrastructure layer is used to provide computing and storage resources for the operation of the capability layer.

2. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: When the user is a patient, the user layer allows the patient to upload photos of the rash through a web browser, mobile app, PC software, or mini-program, complete a self-assessment, and refer to the AI ​​condition assessment and medical advice.

3. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: When the user is a medical professional, the user layer allows the medical professional to view and / or process the patient's condition data and give medical advice through a web browser, mobile APP, PC software or mini-program; the condition data includes rash photos, self-scores and AI condition assessments.

4. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: The upload module supports patient users to upload photos of rashes and self-assessments from their devices and transmit them to the server via the network; it also supports medical staff to upload medical orders from their devices and transmit them to the server via the network; the network used can be a wired network or a wireless network.

5. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: The display module is used to visually display photos of rashes and self-assessments uploaded by patient users, as well as medical orders uploaded by healthcare users.

6. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: The interactive commands are either voice commands or text commands.

7. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: The AI ​​module uses deep learning algorithms to analyze and process the rash photos and self-assessment scores uploaded by patients to the server.

8. The auxiliary diagnostic and treatment system according to claim 1, characterized in that: The processing steps of the deep learning algorithm processing unit include: Data preprocessing: The rash photos are subjected to format conversion, size normalization, brightness and contrast adjustment, noise reduction and standardization; for self-assessment data, structured extraction is performed and stored according to the preset format; Feature extraction and segmentation of rash images: A segmentation network based on U-Net or its improved variants is used to accurately segment the lesion region in the rash image through an encoder-decoder structure, resulting in a lesion region map and a binary or multi-class mask of the lesion. Disease assessment: ① Combine the segmented lesion area map with the self-score, and use convolutional neural networks or other classification / regression models to assess the severity of the lesions or the healing trend; ② Input the assessment results of lesion severity or healing trend, the binary or multi-class mask of the lesion, and the patient's previous medical data stored in the system into the image-level regression / classification network. Use the encoder to pass through the Global AveragePooling+MLP head to output the rash severity score and rash grade.

9. The auxiliary diagnostic and treatment system according to claim 8, characterized in that: The processing steps also include data augmentation of the rash images through rotation, flipping, random cropping, illumination changes and / or generative adversarial networks to improve the model's generalization ability; The improved variants of U-Net include U-Net++ and UNext; The encoder in the encoder-decoder is a ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset, and the decoder is a deconvolution / upsampling decoder. After precise segmentation, a residual network or attention mechanism module is introduced into the segmentation network to enhance the ability to capture multi-scale lesion features and reduce background interference. The encoder used for disease assessment is a ResNet-50 encoder initialized with weights pre-trained on the ImageNet dataset.

10. The auxiliary diagnostic and treatment system according to claims 1-5, 7-9, characterized in that: The self-assessment scores include EASI, POEM, PP-NRS, and / or DLQI.

Images