Radiodermatitis dermatoscope radiomics prediction method and system and application program
By using a dermoscopic radiomics prediction method for radiation dermatitis and extracting radiomics features through a machine learning model to construct a risk prediction model, this approach addresses the lack of accurate prediction tools in existing technologies. It enables accurate prediction and personalized care recommendations for acute radiation dermatitis, thereby improving treatment outcomes and quality of life.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- FUJIAN MEDICAL UNIV
- Filing Date
- 2025-11-19
- Publication Date
- 2026-04-24
AI Technical Summary
Current technology lacks accurate tools for predicting radiation dermatitis, relying mainly on doctors' experience and judgment, and lacking a scientific and systematic prediction method.
The radiomics prediction method for radiation dermatitis is adopted. By collecting patient characteristic data and high-resolution dermoscopic images, machine learning models are used to extract radiomics features and construct a risk prediction model to accurately predict acute radiation dermatitis and provide personalized nursing suggestions.
It enables accurate prediction of acute radiation dermatitis, provides patients with personalized care advice, reduces the risk of radiotherapy interruption, and improves treatment outcomes and quality of life.
Smart Images
Figure CN121922331A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical technology, and in particular to a method, system, and application for predicting radiation-induced dermatitis using dermoscopic imaging. Background Technology
[0002] Cancer is the second leading cause of death worldwide. According to the "Global Cancer Report 2022" published by the International Agency for Research on Cancer (IARC) of the World Health Organization, approximately 20 million new cancer cases and 9.7 million cancer deaths occurred globally in 2022. Radiation therapy is a local treatment method that uses high-energy radiation (such as X-rays, gamma rays, protons, and heavy ions) to focus on tumors, killing or destroying cancer cells and inhibiting their growth and proliferation. Radiation therapy, along with surgery and medical treatments (chemotherapy, targeted therapy, and immunotherapy), is considered one of the three core methods of cancer treatment. The World Health Organization (WHO) indicates that approximately 50%-60% of cancer patients require radiation therapy at some stage of their disease. Radiation therapy contributes to approximately 40% of curable cancers. Radiation dermatitis is a common side effect of radiation therapy. Severe radiation dermatitis can cause discomfort, interrupt radiation therapy, affect treatment efficacy, and reduce the patient's quality of life. Radiation dermatitis does not have good treatment outcomes; therefore, strengthening the management of radiation dermatitis is of great importance. Identifying risk factors and predicting high-risk groups for radiation dermatitis in advance is an important step in its management. Currently, there is a lack of accurate tools for predicting radiation dermatitis. Summary of the Invention
[0003] The purpose of this invention is to address the problem that the current clinical assessment of acute radiation dermatitis mainly relies on doctors' experience and judgment, lacking scientific and systematic prediction methods and tools. This invention provides a dermoscopic radiomics prediction method, system, and application for radiation dermatitis, enabling accurate prediction of acute radiation dermatitis.
[0004] The technical solution adopted in this invention is: A dermomics-based method for predicting radiation-induced dermatitis includes the following steps: Step 1, Data Acquisition: Collect patient characteristic data and obtain high-resolution dermoscopic images of the patient's radiotherapy area through external devices; Step 2, Data Preprocessing and Feature Engineering: The collected raw data is preprocessed, and radiomics features are extracted from the dermoscopy images; Step 3, Model Prediction: Input the preprocessed patient feature data and radiomics features into the pre-built and trained machine learning risk prediction model to obtain the risk prediction result, which is a risk score or the probability value of developing a specific grade of dermatitis. Step 4, Results Feedback: Present the risk prediction results to the user in a visual format.
[0005] Furthermore, the patient characteristic data in step 1 includes basic patient information, clinical information, and radiotherapy parameters, and the external device is a dermoscopic image acquisition device.
[0006] Furthermore, in step 2, the collected raw data is preprocessed locally on a server or mobile terminal, and radiomics features are extracted from the dermoscopy images.
[0007] Furthermore, step 3 is as follows: Step 3-1, Data Preparation: Collect patient data labeled with radiation dermatitis, including clinical characteristics, radiomics characteristics, and dose parameters; Step 3-2, Feature Filtering: The extracted features are filtered sequentially using the maximum correlation minimum redundancy algorithm and the recursive feature elimination algorithm to obtain the optimal feature subset; Step 3-3, Model Training: Use the gradient boosting algorithm to build and train a risk prediction model on the training set based on the optimal feature subset; Steps 3-4, Model Evaluation: The predictive efficacy of the model is evaluated by the area under the receiver operating characteristic (ROC) curve; the ability of the risk value to identify acute radiation dermatitis in the training set is evaluated by ROC curve analysis; and the cutoff value is obtained according to the Youden index to divide the subjects into low-risk and high-risk groups. Finally, the model prediction results are compared with the actual occurrence of acute radiation dermatitis.
[0008] Furthermore, in step 4, personalized care suggestions, medical reminders, or medication usage guidance are automatically generated and pushed based on the risk level predicted by the risk assessment results. For example, "Your current radiation dermatitis risk score is * points, indicating high risk. It is recommended to use medical radiation protection spray and avoid friction from clothing." Furthermore, in step 4, when the predicted risk of the risk prediction result exceeds a preset threshold, an alarm message is automatically sent to the attending physician or nurse.
[0009] Furthermore, in step 4, the risk level is displayed in the form of a dashboard, progress bar, and color warning (green / yellow / red).
[0010] A dermoscopic radiomics prediction system for radiation dermatitis includes a hardware module and a software module. The hardware module includes a dermoscopic image acquisition device, which is used to acquire high-resolution dermoscopic images of the patient's radiotherapy treatment area. The software module is integrated into the application of the mobile terminal, including: The radiomics feature extraction module is used to extract quantitative features from dermoscopy images. The quantitative features include texture features, morphological features and color features. The parameter calculation module is used to calculate radiomics parameters associated with changes in skin microstructure; The risk prediction module is used to build an interpretable machine learning risk prediction model and input patient characteristic data and radiomics parameters into the risk prediction model to generate a personalized radiation dermatitis risk score for the patient.
[0011] Furthermore, the radiomics feature extraction module uses the Python-based PyRadiomics library for feature calculation, extracting 753 features.
[0012] Furthermore, the radiomics feature extraction module includes a data acquisition and control module, a central processing module, and a light source. The data acquisition and control module has an image resolution of 1920*1080 pixels and a 3.5-inch thin-film transistor liquid crystal display to ensure accurate extraction of microstructural features. The central processing module is used for image optimization processing and data management. The light source is a polarized light source to reduce skin surface reflection interference and improve image contrast.
[0013] Furthermore, image optimization processing includes noise reduction, sharpening, color correction, and edge enhancement; data management includes storing images, associating them with medical record information, and supporting exporting or uploading to the medical system.
[0014] Furthermore, the steps for constructing the risk prediction model in the risk prediction module are as follows: Data preparation: Collect patient data labeled with radiation dermatitis, including clinical characteristics, radiomics characteristics, and dose parameters; Feature selection: The extracted features are selected sequentially using the maximum correlation minimum redundancy algorithm and the recursive feature elimination algorithm to obtain the optimal feature subset; Model training: A risk prediction model is built and trained on the training set based on the optimal feature subset using a gradient boosting algorithm; Model evaluation: The predictive efficacy of the model was evaluated by the area under the receiver operating characteristic (ROC) curve; the ability of the risk value to identify acute radiation dermatitis in the training set was evaluated by ROC curve analysis; and the cutoff value was obtained based on the Youden index to divide the subjects into low-risk and high-risk groups. Finally, the model prediction results were compared with the actual occurrence of acute radiation dermatitis.
[0015] Furthermore, patient clinical characteristics included age, sex, comorbidities, smoking history, alcohol consumption history, body mass index (BMI), chemotherapy, immunotherapy, radiomics parameters, and radiation dose; dermoscopic radiomics characteristics included texture features, morphological features, and color features, which were summarized to form radiomics parameters; the dose parameters of the tested skin were finally obtained by using a radiotherapy planning system 4 mm below the outer surface of the patient's skin and converting the obtained skin surface dose into DVH parameters, including the average dose of the tested skin area; in addition, other clinical characteristic data were obtained from patient cases; Missing values in the radiomics parameter data preprocessing were replaced with the feature median and all continuous variables were standardized. Categorical variables were encoded. The final data were divided into training and test sets in a 7:3 ratio. The feature values of the training set were standardized using z-scores. The validation set was standardized using the mean and standard deviation of the training set.
[0016] Specifically, the radiomics feature extraction module uses the PyRadiomics library based on Python for feature calculation. The extracted features include, but are not limited to, texture features, morphological features, and color features. Radiomics parameters are obtained through machine learning models and mapped to the acute radiation dermatitis risk score.
[0017] Furthermore, the feature selection steps specifically include: first, using the mRMR algorithm to select the top 30 features that are highly correlated with the outcome of radiation dermatitis and have low mutual redundancy; then, using the RFE algorithm to recursively eliminate the above 30 features, and finally selecting the 3 best features for model construction.
[0018] Furthermore, the risk prediction module also integrates a model interpretation unit, which uses the SHAP tool to interpret the model's prediction results to provide global and local interpretability.
[0019] A radiation dermatitis dermoscopic radiomics prediction application is stored in memory and executed by a processor. When the prediction application is executed by the processor, it implements the steps of a radiation dermatitis dermoscopic radiomics prediction method.
[0020] This invention employs the above technical solution to acquire high-resolution dermoscopic images of the treatment area of radiotherapy patients. Quantitative features, including texture, morphological, and color features, are extracted from the images using radiomics technology to dynamically monitor changes in the skin's microstructure during radiotherapy. The prediction module comprehensively considers three aspects: individual patient differences (clinical characteristics), radiotherapy parameters, and dermoscopic radiomics features. A prediction model is constructed using machine learning algorithms to accurately predict acute radiation dermatitis.
[0021] This invention is the first to apply dermoscopic radiomics to the prediction of radiation-induced dermatitis. Through dermoscopic radiomics, the risk of dermatitis can be predicted before radiotherapy, providing a "time window" for future early intervention strategies. Simultaneously, a portable detection and reporting system will be further developed for routine skin monitoring and early identification, effectively reducing application costs and promoting product upgrades and wider application. Attached Figure Description
[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments; Figure 1 This is a flowchart illustrating a dermomics prediction method for radiation-induced dermatitis according to the present invention. Figure 2 This is a schematic diagram of the RFE feature selection method of the present invention; Figure 3 This diagram illustrates the importance of the features selected in this invention in the gradient enhancement algorithm. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.
[0024] like Figures 1 to 3 As shown in one example, this invention discloses a dermomics prediction method for radiation-induced dermatitis, comprising the following steps: Step 1, Data Acquisition: Collect patient characteristic data and obtain high-resolution dermoscopic images of the patient's radiotherapy area through external devices; Step 2, Data Preprocessing and Feature Engineering: The collected raw data is preprocessed, and radiomics features are extracted from the dermoscopy images; Step 3, Model Prediction: Input the preprocessed patient feature data and radiomics features into the pre-built and trained machine learning risk prediction model to obtain the risk prediction result, which is a risk score or the probability value of developing a specific grade of dermatitis. Step 4, Results Feedback: Present the risk prediction results to the user in a visual format.
[0025] Furthermore, the patient characteristic data in step 1 includes basic patient information, clinical information, and radiotherapy parameters, and the external device is a dermoscopic image acquisition device.
[0026] Furthermore, in step 2, the collected raw data is preprocessed locally on a server or mobile terminal, and radiomics features are extracted from the dermoscopy images.
[0027] Furthermore, step 3 is as follows: Step 3-1, Data Preparation: Collect patient data labeled with radiation dermatitis, including clinical characteristics, radiomics characteristics, and dose parameters; Step 3-2, Feature Filtering: The extracted features are filtered sequentially using the maximum correlation minimum redundancy algorithm and the recursive feature elimination algorithm to obtain the optimal feature subset; Step 3-3, Model Training: Use the gradient boosting algorithm to build and train a risk prediction model on the training set based on the optimal feature subset; Steps 3-4, Model Evaluation: The predictive efficacy of the model is evaluated by the area under the receiver operating characteristic (ROC) curve; the ability of the risk value to identify acute radiation dermatitis in the training set is evaluated by ROC curve analysis; and the cutoff value is obtained according to the Youden index to divide the subjects into low-risk and high-risk groups. Finally, the model prediction results are compared with the actual occurrence of acute radiation dermatitis.
[0028] Furthermore, in step 4, personalized care suggestions, medical reminders, or medication usage guidance are automatically generated and pushed based on the risk level predicted by the risk assessment results. For example, "Your current radiation dermatitis risk score is * points, indicating high risk. It is recommended to use medical radiation protection spray and avoid friction from clothing." Furthermore, in step 4, when the predicted risk of the risk prediction result exceeds a preset threshold, an alarm message is automatically sent to the attending physician or nurse.
[0029] Furthermore, in step 4, the risk level is displayed in the form of a dashboard, progress bar, and color warning (green / yellow / red).
[0030] A dermoscopic radiomics prediction system for radiation dermatitis includes a hardware module and a software module. The hardware module includes a dermoscopic image acquisition device, which is used to acquire high-resolution dermoscopic images of the patient's radiotherapy treatment area. The software module is integrated into the application of the mobile terminal, including: The radiomics feature extraction module is used to extract quantitative features from dermoscopy images. The quantitative features include texture features, morphological features and color features. The parameter calculation module is used to calculate radiomics parameters associated with changes in skin microstructure; The risk prediction module is used to build an interpretable machine learning risk prediction model and input patient characteristic data and radiomics parameters into the risk prediction model to generate a personalized radiation dermatitis risk score for the patient.
[0031] Furthermore, the radiomics feature extraction module uses the Python-based PyRadiomics library for feature calculation, extracting 753 features.
[0032] Furthermore, the radiomics feature extraction module includes a data acquisition and control module, a central processing module, and a light source. The data acquisition and control module has an image resolution of 1920*1080 pixels and a 3.5-inch thin-film transistor liquid crystal display to ensure accurate extraction of microstructural features. The central processing module is used for image optimization processing and data management. The light source is a polarized light source to reduce skin surface reflection interference and improve image contrast.
[0033] Furthermore, image optimization processing includes noise reduction, sharpening, color correction, and edge enhancement; data management includes storing images, associating them with medical record information, and supporting exporting or uploading to the medical system.
[0034] Furthermore, the steps for constructing the risk prediction model in the risk prediction module are as follows: Data preparation: Collect patient data labeled with radiation dermatitis, including clinical characteristics, radiomics characteristics, and dose parameters; Feature selection: The extracted features are selected sequentially using the maximum correlation minimum redundancy algorithm and the recursive feature elimination algorithm to obtain the optimal feature subset; Model training: A risk prediction model is built and trained on the training set based on the optimal feature subset using a gradient boosting algorithm; Model evaluation: The predictive efficacy of the model was evaluated by the area under the receiver operating characteristic (ROC) curve; the ability of the risk value to identify acute radiation dermatitis in the training set was evaluated by ROC curve analysis; and the cutoff value was obtained based on the Youden index to divide the subjects into low-risk and high-risk groups. Finally, the model prediction results were compared with the actual occurrence of acute radiation dermatitis.
[0035] Furthermore, patient clinical characteristics included age, sex, comorbidities, smoking history, alcohol consumption history, body mass index (BMI), chemotherapy, immunotherapy, radiomics parameters, and radiation dose; dermoscopic radiomics characteristics included texture features, morphological features, and color features, which were summarized to form radiomics parameters; the dose parameters of the tested skin were finally obtained by using a radiotherapy planning system 4 mm below the outer surface of the patient's skin and converting the obtained skin surface dose into DVH parameters, including the average dose of the tested skin area; in addition, other clinical characteristic data were obtained from patient cases; Missing values in the radiomics parameter data preprocessing were replaced with the feature median and all continuous variables were standardized. Categorical variables were encoded. The final data were divided into training and test sets in a 7:3 ratio. The feature values of the training set were standardized using z-scores. The validation set was standardized using the mean and standard deviation of the training set.
[0036] Specifically, the radiomics feature extraction module uses the PyRadiomics library based on Python for feature calculation. The extracted features include, but are not limited to, texture features, morphological features, and color features. Radiomics parameters are obtained through machine learning models and mapped to the acute radiation dermatitis risk score.
[0037] Furthermore, the feature selection steps specifically include: first, using the mRMR algorithm to select the top 30 features that are highly correlated with the outcome of radiation dermatitis and have low mutual redundancy; then, using the RFE algorithm to recursively eliminate the above 30 features, and finally selecting the 3 best features for model construction.
[0038] Furthermore, the risk prediction module also integrates a model interpretation unit, which uses the SHAP tool to interpret the model's prediction results to provide global and local interpretability.
[0039] A dermomics-based radiomics prediction application for radiation-induced dermatitis is stored in memory and executed by a processor. When executed by the processor, the prediction application implements the steps of a dermomics-based radiomics prediction method for radiation-induced dermatitis. The specific principles of this invention are described in detail below: Feature Extraction: Patient data was derived from dermoscopy imaging. Each patient was divided into left and right sides, with one image taken from each side. Features were extracted from pre-radiotherapy (RT) data to construct a model for predicting radiation dermatitis. Features were extracted from both left and right images of each patient using the same method, and the average of the left and right features was used for modeling. Features are extracted using radiomics features. These are high-throughput quantitative data, i.e., non-semantic features, automatically extracted from medical images using computer algorithms. These features can include multiple aspects such as image texture, shape, and intensity. Feature Extraction Method: The "pyradiomics" package was used, extracting a total of 753 features.
[0040] Data Processing: The radiation dermatitis outcome label is defined as follows: 1 for the occurrence of the outcome, and 0 for the absence of the outcome. The radiation dermatitis outcome label is a quantification of the indicator. The outcome is a binary variable, with 1 defined as the occurrence of the outcome. The outcome is defined by two doctors making separate dermoscopic diagnoses; a diagnosis of radiation dermatitis is assigned as outcome 1. The collected dermoscopic images were standardized using methods such as resampling, discretization, and normalization to avoid software compatibility issues. Fields containing Chinese characters were encoded.
[0041] Dataset partitioning: The data was randomly divided into training and validation sets in a 7:3 ratio. The features of the training set were standardized using z-scores, and the validation set was standardized using the mean and standard deviation of the training set. Core parameters: R's caret and CBCgrps packages. Dataset partitioning: The data was randomly divided into training and validation sets in a 7:3 ratio; and a difference analysis was performed between the training and validation sets. The purpose of the difference analysis was to validate and test whether the difference in the predictive models between the two groups (those who experienced the outcome event and those who did not) was statistically significant.
[0042] mRMR_RFE Feature Selection: Includes the mRMR algorithm and the RFE algorithm. Core parameters are provided in the "mRMRe" and "caret" packages in R. The Maximum Relevance, Minimum Redundancy (mRMR) algorithm selects features by considering not only the correlation between a feature and the variable to be predicted, but also the correlation between features themselves. The metric used is mutual information. For the mRMR method, the correlation between a feature subset and the class is calculated by the mean of the information gain of each feature relative to the class, while the redundancy between features is calculated by summing the mutual information between features and dividing by the square of the number of features in the subset.
[0043] Recursive feature elimination (RFE) involves ranking predictors before modeling and removing less important ones. The goal is to find a subset of predictors that can be used to generate an accurate model. The model is trained repeatedly, removing n low-importance features after each training iteration, then training again with new features to determine feature importance, and repeating this process until the optimal feature subset is obtained.
[0044] The top 30 features selected by the mRMR method are then further filtered using RFE to select the optimal feature subset.
[0045] Both mRMR and RFE algorithms are machine learning algorithms used for feature selection in radiomics. This is an important step in machine learning and data mining, referring to the selection of the most useful subset of features for the target variable from all features (variables) of the original data. This improves model performance, reduces computational complexity, and lowers the risk of overfitting. In general, mRMR is a commonly used feature selection method. This algorithm considers both the correlation between features and the target variable (maximum correlation) and the redundancy between features (minimum redundancy). This improves the model's generalization ability and reduces the risk of overfitting. mRMR is a filtering method.
[0046] like Figure 2 The diagram illustrates RFE feature selection, yielding three features. RFE is an algorithm, detailed below. RFE is a wrapper method. Its purpose is to iteratively eliminate unimportant features, ultimately retaining the most predictive subset. By integrating these two algorithms, feature selection is performed, focusing on high-pathway features and dimensionality reduction. The output of the prediction model is the predicted probability, which is the probability of whether the outcome event has occurred.
[0047] Model Building: Using features selected through ICC, mRMR, and RFE, a model is built on the training set using a gradient boosting algorithm. The gradient boosting machine algorithm uses a set of weak classifiers (usually decision trees) to train newly added weak classifiers based on the negative gradient information of the current model's loss function. These trained weak classifiers are then cumulatively combined into the existing model to build and train the risk prediction model. The gradient boosting model is used to model the selected dermoscopic image features to predict gene expression.
[0048] Model Evaluation: The model's performance was evaluated. The receiver operating characteristic (ROC) curve shows the false positive rate (FPR) on the X-axis and the true positive rate (TPR) on the Y-axis. A larger ROC-AUC, a larger area under the curve, and a more convex curve towards the upper left corner indicate a better model performance. This model demonstrates good predictive performance: as shown by the ROC curve, the AUC value on the training set is 0.943, and on the validation set it is 0.750.
[0049] Analysis of differences between model groups: A graph showing the difference in predicted values between groups with high and low gene expression. The model outputs a probability score (Score) for predicting gene expression levels; the Wilcoxon test is used to compare whether there are differences in Score between the high and low gene expression groups. Significance indicators: ns, p≥0.05; *, p<0.05; **, p<0.01; ***, p<0.001. The training set Score distribution shows a significant difference between the high and low gene expression groups (p<0.05); groups with outcome variables have higher Score values; the validation set has only 2 cases with outcome variables, and due to sample size limitations, there are no positive results.
[0050] This invention employs the above-mentioned technical solution to provide a method and system for predicting the risk level of radiation dermatitis in cancer patients during radiotherapy via a mobile terminal (such as a smartphone or tablet) application (APP). The method includes a dermoscopic image acquisition system and a machine learning prediction model. The dermoscopic image acquisition system acquires high-resolution dermoscopic images of the treatment area of the radiotherapy patient. Quantitative features, including texture, morphological, and color features, are extracted from the images using radiomics technology to dynamically monitor changes in the skin's microstructure during radiotherapy. The prediction module comprehensively considers three aspects: individual patient differences (clinical characteristics), radiotherapy parameters, and dermoscopic radiomics features. It uses machine learning algorithms to construct and train a risk prediction model to achieve accurate prediction of acute radiation dermatitis.
[0051] Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. The components of the embodiments of this application described and illustrated herein can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
Claims
1. A dermomics-based predictive method for radiation-induced dermatitis, characterized in that: Includes the following steps: Step 1, Data Acquisition: Collect patient characteristic data and obtain high-resolution dermoscopic images of the patient's radiotherapy area through external devices; Patient characteristic data includes basic patient information, clinical information, and radiotherapy parameters; Step 2, Data Preprocessing and Feature Engineering: The collected raw data is preprocessed, and radiomics features are extracted from the dermoscopy images; Step 3, Model Prediction: Input the preprocessed patient feature data and radiomics features into the pre-built and trained machine learning risk prediction model to obtain the risk prediction result, which is a risk score or the probability value of developing a specific grade of dermatitis. Step 4, Results Feedback: Present the risk prediction results to the user in a visual format.
2. The method for predicting radiation-induced dermatitis using dermomics according to claim 1, characterized in that: In step 2, the raw data collected is preprocessed locally on a server or mobile terminal, and radiomics features are extracted from the dermoscopy images.
3. The method for predicting radiation-induced dermatitis using dermomics according to claim 1, characterized in that: The steps for building and training the risk prediction model in step 3 are as follows: Step 3-1, Data Preparation: Collect patient data labeled with radiation dermatitis, including clinical characteristics, radiomics characteristics, and dose parameters; Step 3-2, Feature Filtering: The extracted features are filtered sequentially using the maximum correlation minimum redundancy algorithm and the recursive feature elimination algorithm to obtain the optimal feature subset; Step 3-3, Model Training: Use the gradient boosting algorithm to build and train a risk prediction model on the training set based on the optimal feature subset; Steps 3-4, Model Evaluation: The ability of the risk value to identify the occurrence of acute radiation dermatitis in the training set is evaluated by using receiver operating characteristic (ROC) curve analysis. The cutoff value is obtained based on the Youden index, and the subjects are divided into low-risk and high-risk groups. Finally, the model prediction results are compared with the actual occurrence of acute radiation dermatitis.
4. The method for predicting radiation-induced dermatitis using dermomics according to claim 3, characterized in that: Patient clinical characteristics include age, sex, comorbidities, smoking history, alcohol consumption history, body mass index, chemotherapy, immunotherapy, radiomics parameters, and radiation dose; dermoscopic radiomics characteristics include texture features, morphological features, and color features, which are summarized to form radiomics parameters; the dose parameters of the tested skin are finally obtained by using a radiotherapy planning system 4 mm below the outer surface of the patient's skin and converting the obtained skin surface dose into DVH parameters, including the average dose of the tested skin area; In the preprocessing of radiomics parameters, missing values were replaced with the median of features and all continuous variables were standardized. Categorical variables were encoded. The final data were divided into training and test sets in a 7:3 ratio. The feature values of the training set were standardized using z-scores, and the mean and standard deviation of the training set were used to standardize the validation set.
5. The method for predicting radiation-induced dermatitis using dermomics according to claim 1, characterized in that: In step 4, personalized nursing suggestions, medical reminders, or medication usage guidance are automatically generated and pushed based on the risk level of the risk prediction results; when the predicted risk exceeds the preset threshold, an alarm message is automatically sent to the attending physician or nurse.
6. A dermomics prediction system for radiation-induced dermatitis, employing the dermomics prediction method for radiation-induced dermatitis as described in any one of claims 1 to 5, characterized in that: The system includes hardware modules and software modules. The hardware modules include a dermoscopic image acquisition device, which is used to acquire high-resolution dermoscopic images of the patient's radiotherapy treatment area. The software modules are integrated into the mobile terminal's applications, including: The radiomics feature extraction module is used to extract quantitative features from dermoscopy images. The quantitative features include texture features, morphological features and color features. The parameter calculation module is used to calculate radiomics parameters associated with changes in skin microstructure; The risk prediction module is used to build an interpretable machine learning risk prediction model and input patient characteristic data and radiomics parameters into the risk prediction model to generate a personalized radiation dermatitis risk score for the patient.
7. The dermomics prediction system for radiation-induced dermatitis according to claim 6, characterized in that: The radiomics feature extraction module includes a data acquisition and control module, a central processing module, and a light source. The data acquisition and control module has an image resolution of 1920*1080 pixels and a 3.5-inch thin-film transistor liquid crystal display to ensure accurate extraction of microstructural features. The central processing module is used for image optimization and data management. The light source is a polarized light source to reduce skin surface reflection interference and improve image contrast. The radiomics feature extraction module uses the Python-based PyRadiomics library for feature calculation, extracting 753 features.
8. The dermomics prediction system for radiation-induced dermatitis according to claim 6, characterized in that: The steps for constructing a risk prediction model in the risk prediction module are as follows: Data preparation: Collect patient data labeled with radiation dermatitis, including clinical characteristics, radiomics characteristics, and dose parameters; Feature selection: The extracted features are selected sequentially using the maximum correlation minimum redundancy algorithm and the recursive feature elimination algorithm to obtain the optimal feature subset; Model training: A risk prediction model is built and trained on the training set based on the optimal feature subset using a gradient boosting algorithm; Model evaluation: The predictive efficacy of the model was evaluated by the area under the receiver operating characteristic (ROC) curve; the ability of the risk value to distinguish the occurrence of acute radiation dermatitis in the training set was evaluated by ROC curve analysis, and the cutoff value was obtained according to the Youden index to divide the subjects into low-risk and high-risk groups. Finally, the model prediction results were compared with the actual occurrence of acute radiation dermatitis.
9. The dermomics prediction system for radiation-induced dermatitis according to claim 6, characterized in that: The risk prediction module also integrates a model interpretation unit, which uses the SHAP tool to interpret the model's prediction results to provide global and local interpretability.
10. A dermoscopic radiomics prediction application for radiation-induced dermatitis, stored in memory and executed by a processor, characterized in that: When the prediction application is executed by the processor, it implements the steps of the radiodiagnostic dermatologic imaging prediction method for radiation dermatitis as described in any one of claims 1 to 5.