Efficient Melanoma Detection Using Pixel Intensity-Based Masking and Intensity-Weighted Binary Cross-Entropy

Abstract

Melanoma, the deadliest form of skin cancer, requires accurate and timely detection to improve survival rates and treatment outcomes. Deep learning has shown significant potential in automating melanoma detection; however, existing methods face challenges such as irrelevant background information in dermoscopic images and class imbalance in melanoma datasets, which hinder diagnostic performance. To address these challenges, this paper introduces two complementary contributions: Pixel Intensity-Based Masking (PIBM) and Intensity-Weighted Binary Cross-Entropy (IW-BCE). PIBM is a novel preprocessing technique that dynamically identifies and masks low-priority regions in dermoscopic images based on pixel intensity values. By preserving high-intensity lesion regions and suppressing irrelevant background artifacts, PIBM reduces computational complexity and enhances the model's focus on diagnostically critical features, all without requiring ground truth annotations or pixel-level labeling. Additionally, IW-BCE, a custom loss function, is designed to handle class imbalance by dynamically adjusting the contribution of each class during training. By assigning higher weights to the minority class (malignant lesions), IW-BCE enhances the model's sensitivity, reduces false negatives, and improves recall, an essential metric in medical diagnostics. The proposed framework integrates PIBM and IW-BCE into a deep-learning pipeline for melanoma detection. Evaluations on benchmark datasets demonstrate that the combined approach achieves superior performance compared to traditional methods in terms of accuracy, sensitivity, and computational efficiency. Specifically, the proposed method achieves a higher recall and F1-score, highlighting its ability to address the critical limitations of existing systems. This work offers a robust and clinically relevant solution for real-time melanoma detection, paving the way for improved early diagnosis and patient outcomes.