Federated Learning Approach Using Transfer Learning Architectures for Lung Cancer Detection

There have been many advancements in the field of medical imaging but even then, accurate cancer detection remains a challenge because of limited labelled data. In the research field, a lot of work is already done for this task, using pre-trained features from prominent architectures like VGG16, ResNet50, MobileNetV2, InceptionV3, and DensNet121. These approaches face the issue of privacy of patients’ sensitive information and unnecessary latency of exchange of data from nodes to sever, so in this paper, we use Federated Learning that enables collaborative learning across geographically distributed medical institutions. Additionally, we implement differential privacy techniques to obscure the patients’ identities which would further enhance privacy protection. This paper also presents the evaluation of effectiveness of different transfer learning architectures within the FL setting, comparing their performance with centralized learning and standalone transfer learning approaches. This work adds a new direction to cancer detection with improved privacy protection, leading to earlier intervention for cancer detection.