Recent Advances in Transfer Learning for Enhanced Melanoma Detection Using Hybrid Texture Features: A Systematic Review

Abstract

Melanoma skin cancer remains one of the most aggressive and life-threatening dermatological conditions, necessitating early and accurate diagnosis to improve survival rates. In recent years, transfer learning has emerged as a powerful paradigm in medical image processing, enabling the adaptation of pre-trained deep learning models for melanoma detection tasks. This systematic review explores recent advancements in transfer learning architectures integrated with hybrid texture feature extraction techniques for enhanced melanoma classification. The study critically examines how combining deep convolutional neural networks with handcrafted texture descriptors such as Local Binary Patterns, Gray-Level Co-occurrence Matrix, and wavelet transforms improves diagnostic accuracy and robustness. By synthesizing findings from contemporary research, the review highlights significant improvements in classification performance, reduced computational complexity, and enhanced generalization across diverse datasets. Furthermore, the integration of hybrid feature representations addresses limitations associated with purely data-driven models, particularly in scenarios with limited annotated medical data. The review also identifies emerging trends, challenges, and future directions in developing efficient, scalable, and clinically viable melanoma detection systems. Overall, this work provides a comprehensive understanding of how transfer learning and hybrid texture-based approaches contribute to advancing automated melanoma diagnosis in modern healthcare systems.