Enhancing melanoma identification and classification in dermoscopic images using optimal deep ensemble learning

Skin cancer (SC) is a major global health concern, with melanoma being one of the most aggressive and life-threatening forms. Accurate and early diagnosis remains challenging due to the visual similarity between benign and malignant lesions and the dependence on expert interpretation. To address these limitations, this study proposes an advanced computer-aided diagnostic (CAD) framework for melanoma identification and classification using dermoscopic images. The proposed approach integrates a deep ensemble learning architecture with robust pre-processing, segmentation, feature extraction, and classification techniques. Initially, image quality is enhanced through Wiener filtering (WF), contrast-limited adaptive histogram equalization (CLAHE), and hair artifact removal. Subsequently, precise lesion segmentation is achieved using the U 2 Net model, followed by discriminative feature extraction using the NASNet architecture. For classification, an ensemble of recurrent deep learning (DL) models Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Gated Recurrent Unit (GRU) is employed to capture complex patterns and temporal dependencies. The Root mean square propagation (RMSProp) optimizer is utilized for efficient hyperparameter tuning and faster convergence. Experimental results on dermoscopic datasets demonstrate that the proposed framework achieves superior performance compared to existing deep learning models across multiple evaluation metrics, highlighting its robustness and effectiveness. The integration of segmentation-driven feature learning and optimized ensemble classification significantly enhances diagnostic accuracy, making the proposed system a promising tool for supplementary dermatologists in early and reliable melanoma detection.