AUTOMATED SKIN DISEASE CLASSIFICATION USING DEEP LEARNING WITH AN INTEGRATED WEB-BASED CLINICAL DECISION SUPPORT SYSTEM

Abstract

Skin diseases are among the most prevalent health issues worldwide, ranging from benign conditions to life-threatening cancers. Early and accurate diagnosis is essential for effective treatment; however, the shortage of dermatology specialists, particularly in low-resource regions, makes timely diagnosis challenging. To address this issue, this study proposes an automatic skin disease classification system using deep learning integrated with a web-based clinical decision support system (CDSS). The proposed system classifies ten skin disease categories, including Eczema, Melanoma, Atopic Dermatitis, Basal Cell Carcinoma, Melanocytic Nevi, Benign Keratosis, Psoriasis, Seborrheic Keratosis, Tinea Ringworm, and Warts Molluscum. The dataset is enhanced through extensive image preprocessing, including resizing, normalization, contrast enhancement, and data augmentation, which improves feature representation and reduces overfitting. A comparative evaluation is conducted using CNN, VGG16, ResNet50, InceptionV3, and MobileNetV2 models on dermoscopic images. Experimental results show that all models achieve competitive performance; however, MobileNetV2 outperforms others with an accuracy of 92.8%, precision of 92.4%, recall of 92.0%, and F1-score of 92.2%. In comparison, InceptionV3, ResNet50, VGG16, and CNN achieve accuracies of 88.4%, 87.9%, 86.3%, and 79.1%, respectively. The superior performance of MobileNetV2 is due to its efficient lightweight architecture and strong generalization capability. The system is deployed as a web-based clinical decision support tool for real-time and accessible skin disease prediction, assisting dermatologists in improving diagnostic accuracy and supporting early detection in clinical environments.

Keywords: Skin Diseases, Skin Imaging, Deep Learning, Transfer Learning, EfficientNet, Clinical Decision Support System.