Enhancing AI-Based Diagnostic Accuracy for Multi-Class Skin Lesion Detection

Prompt and accurate diagnosis of skin lesions is crucial for effective dermatological care, directly impacting patient outcomes and reducing healthcare costs. Recognizing neoplastic conditions associated with cancer, such as Melanoma and Squamous Cell Carcinoma, is particularly important. This prompt recognition dramatically enhances survival rates, while correct identification of benign processes prevents unnecessary interventions. Despite the promise of artificial intelligence (AI) in this area, automated systems have significant drawbacks. Most systems fail to systematically differentiate between a wide range of skin lesions, particularly those that are uncommon or visually unidentifiable. This limitation reduces the clinical utility of these systems. Notably, there are currently no measures to combat ethnic bias in the literature, which further hinders the development of equitable AI products. This paper addresses these gaps by proposing a novel approach focused on improving multiclass diagnostic validity, thereby broadening the applicability of AIpowered dermatology devices in clinical practice. The framework prioritizes accuracy and stability as the central pillars of future AI dermatology systems.