Effective Skin Cancer Diagnosis Through Federated Learning and Deep Convolutional Neural Networks

Abstract

Skin cancer is a prevalent type of cancer that affects millions of people globally. However, detecting it can be a challenging task, even for specialized dermatologists. Early detection is crucial for successful treatment, and deep learning techniques, particularly deep convolutional neural networks (DCNNs), have shown tremendous potential in this area. However, achieving high accuracy results requires large volumes of data for training these DCNNs. Since medical organizations and institutions, individually, do not usually have such amounts of information available, and due to the current regulations regarding intellectual property and privacy of medical patient data, it is difficult to share data in a direct way. The primary objective of this work is to overcome this issue through a federated learning approach. We created a privacy-preserving and accurate skin cancer classification system that can assist dermatologists and specialists in making informed patient care decisions. The federated learning DCNNs architecture uses a combination of convolutional and pooling layers to extract relevant features from skin lesion images. It also includes a fully connected layer for classification. To evaluate the proposed architecture, we tested it on three datasets of varying complexity and size. The results demonstrate the applicability of the proposed solution and its efficiency for skin cancer classification.