SDA-Detection Melanoma: Deep Approach System for Detection and Segmentation in Melanoma Images using Fine-Tuning

Melanoma is a pathology that poses a risk to the health of the global population, leading to various complications from skin lesions that can result in patient death if not detected early. Medical imaging diagnosis is an increasingly used procedure in current literature, making the clinical investigation process more effective for high-precision imaging diagnosis. Computational methods have been employed to assist medical diagnosis through Computer-Aided Diagnosis (CAD) systems. This study introduces a new approach called SDA-Detection Melanoma for fully automatic detection and segmentation of melanomas in dermoscopy examination images. In this work, different deep-learning networks were used for melanoma detection, combined with the use of fine-tuning and computational methods based on Parzen windowing, clustering, and region growth for melanoma region segmentation. The results were quite satisfactory, achieving a high accuracy rate of 96.39% for melanoma region detection and 96.50% for melanoma segmentation, showing great precision in identifying and segmenting the pathology region. Thus surpassing renowned works found in the state of the art, both in automatic and non-automatic methods, classic methods, and those using deep learning.