Hybrid Metaheuristic Contrast Stretching for Enhanced Segmentation of Skin Cancer Lesions

The World Health Organization (WHO) estimates that one in three individuals will likely get skin cancer in their lifetime, rendering it one of the most common cancer types globally. A study conducted in the UK indicates that although hereditary and lifestyle factors contribute to merely 14% of melanoma incidences, extended sun exposure is the primary cause. The mortality rate may ascend to 41% if treatment is postponed beyond two to three months, in contrast to merely 5% when identified early, underscoring the vital significance of prompt and precise diagnosis. In medical image analysis, preprocessing is a crucial step that optimizes image quality by enhancing contrast and highlighting lesion boundaries. Improved contrast assists dermatologists in visual examination and markedly enhances the efficacy of automated classification and segmentation algorithms. Current preprocessing techniques frequently lack the ability to produce uniform outcomes across varied datasets due to their restricted adaptability. This paper presents a novel metaheuristic preprocessing strategy, DE-BA-ABC, which amalgamates Differential Evolution (DE), Bat Algorithm (BA), and Artificial Bee Colony (ABC) optimization methodologies to tackle this difficulty. The proposed method integrates the exploration and exploitation capabilities of these algorithms to adaptively estimate optimal contrast values for each image, resulting in consistently improved outputs. The efficacy of the method is confirmed using three publicly accessible skin lesion datasets—PH2, ISIC-2016, and ISIC-2017—utilizing the UNet segmentation framework. Performance is assessed using established criteria, such as the Jaccard Index and Dice Coefficient. Experimental results indicate a significant improvement: the Jaccard Index rises from 84.71% to 93.65%, and the Dice Coefficient climbs from 89.01% to 92.57%. The results validate that the suggested DE-BA-ABC preprocessing technique significantly improves lesion segmentation performance, presenting a viable avenue for the advancement of more dependable computer-aided diagnostic systems for skin cancer diagnosis.