Skin Diseases Classification Framework On Lesion Images Using Deep Learning

Abstract

The utilization of automated classification frameworks leveraging deep learning methodologies presents a promising approach for the early identification of skin diseases, thereby enhancing diagnostic accuracy and treatment effectiveness. Traditional diagnostic methods often suffer from time constraints, error susceptibility, and resource limitations, underscoring the need for more efficient diagnostic solutions. In this context, AI-driven automated systems offer an attractive alternative by expediting diagnosis processes and minimizing the risk of disease progression. However, existing frameworks tend to focus predominantly on individual skin conditions, neglecting the possibility of concurrent manifestation of multiple diseases in a single patient. Hence, there is a pressing need to develop integrated skin disease classification frameworks capable of simultaneously detecting various dermatological conditions. Such advancements are crucial for improving diagnostic precision and facilitating prompt treatment interventions. By leveraging advanced techniques in deep learning, including YOLO V5, SSD MobileNet v3, and Faster RNN, the framework achieves high performance in lesion detection and classification. Through data annotation and model training, the framework is capable of identifying patterns and features indicative of various skin conditions. The evaluation metrics, including mAP = 0.8140, precision = 0.8590, and recall = 0.7700, serve as benchmarks for assessing the optimal performance of the YOLOv5 framework across various skin disease categories. These results underscore the potential of deep learning to enhance dermatological diagnostics, providing a promising pathway for improving clinical decision-making and patient care within dermatology practices.