Blockchain-Based Federated Learning Framework for Melanoma Detection and Classification

Blockchain-Based Federated Learning Framework for Melanoma Detection and Classification Kalaivani M Piramu Preethika SK.

Melanoma is a deadly cancer; the patient’s survival outcomes are based on the early detection and accuracy of the prediction. The data sharing of patient medical information in centralised systems poses privacy risks and regulatory challenges. To encounter this issue, the Secured Federated Learning framework is designed for melanoma detection and classification that ensures data security, integrity, and privacy in a decentralized manner using Blockchain Technology. The local image processing techniques leverage the HAM10000 dataset that comprises 10000 skin lesions of seven types of skin cancer, incorporating the U-Net-based image segmentation, CapsNet-based feature extraction, and VGG16, VGG19, and Inception V3 as pretrained models in the Ensemble transfer learning for classification of melanoma types. The global model shares the Deep learning based training model and datasets with the local model, allowing multiple healthcare institutions to work collaboratively. The smart contracts ensure trust, immutability, and secure aggregation in the block model updates. The proposed framework outperforms the conventional one by 93% accuracy and test error reduction to 0.01% for 100 iterations. Thus, the proposed work highlights a secure decentralised system for melanoma diagnosis using advanced image processing techniques.
03 25 2026 258 271 10.14445/22315381/IJETT-V74I3P119 https://ijettjournal.org/archive/ijett-v74i3p119 https://ijettjournal.org/Volume-74/Issue-3/IJETT-V74I3P119.pdf