FEDERATED DEEP LEARNING FOR PRIVACY-PRESERVING EARLY LUNG CANCER DETECTION IN DISTRIBUTED HOSPITAL NETWORKS

Abstract

Early identification is essential to increasing survival rates for lung cancer, which continues to be a significant global health concern. However, privacy issues frequently make it difficult for healthcare organizations to work together by exchanging patient data. A federated deep learning system for early lung cancer diagnosis utilizing chest radiographs that protects privacy is presented in this paper. The method uses a ResNet-50 classifier for precise nodule identification after isolating lung regions using a U-Net-based segmentation model. This design protects patient privacy by allowing hospitals to train models locally and only communicate weight updates, in contrast to centralized models. Strong diagnostic performance was shown by simulations utilizing the NIH ChestX-ray14 and LIDC-IDRI datasets, with 91.5% accuracy, 93.1% specificity, and 89.6% sensitivity. This study opens the door for privacy-compliant AI implementations in healthcare and validates the viability of federated learning for clinical diagnostics. It demonstrates how decentralized training can produce trustworthy outcomes while upholding moral principles when handling medical data.