A Secure Federated Hybrid Deep Learning Model with Blockchain-Assisted Aggregation for Skin Cancer Diagnosis: A Review

Skin cancer is one of the most common and fastest-growing cancers worldwide. When diagnosed late, can a serious health threat, whether it’s melanoma or non-melanoma skin cancer. In the medical imaging domain, deep learning has revolutionized the way medical images are analysed, with the ability to automatically extract discriminative features from dermoscopic images with impressive accuracy. The existing centralised solutions, however, demand a large number of annotated datasets, which are generally hard to access due to strict data privacy laws (HIPAA, GDPR) and the distribution of Data across geographically widely spread healthcare institutions. The current models are Ill-Generalizable in the presence of Variations in imaging conditions, biased towards common lesion types due to class imbalance, and susceptible to adversarial and poisoning attacks in shared computing infrastructures. The complementary and synergistic solutions are recent advances in Federated learning (FL), Hybrid deep learning architectures and Blockchain Technology. FL provides privacy-preserving collaborative training without moving data; the hybrid architectures benefit from the strengths of various architectures for better and more robust feature representation; blockchain ensures tamper-resistance and auditable aggregation of distributed model updates. This review systematically summarises the current state of the art in each of these areas, critically appraises the strengths and weaknesses, identifies research gaps and proposes an integrated conceptual framework for an accurate, scalable, real-world, multi-institutional secure skin cancer diagnosis.