Currently, the early detection of lung cancer remains elusive in hospitals that have a scarcity of high-performance hardware and large-annotated CT datasets. Most modern deep-learning systems require GPUs and centralized data access, which are mostly impractical in rural or resource-limited settings. Patient privacy regulations further limit data sharing. This work presents a lightweight convolutional neural network integrated with a federated learning framework, which is designed to operate on basic CPU-only machines while keeping all patient scans local to each hospital. Only model weight updates are transmitted to a central server, which will aggregate them to form an improved global model. Experiments conducted on standard medical datasets such as LIDC-IDRI and LUNA16 demonstrate that the global model achieves stable performance gains within a few training rounds, despite limited data availability. The outcome of this study is that compact federated learning can enable accessible, privacy-preserving, and hardware-efficient lung-cancer screening in scenarios where the application of a conventional centralized deep-learning pipeline may not be feasible.