Adaptive geometric-attention network segments and classifies lung nodules in federated edge environments

A federated-learning-capable geometric-attention network segmented lung nodules (Dice 0.927) and classified malignancy (AUC 0.951) on three public datasets, and ran on IoT edge hardware.

• Retrospective evaluation on benchmark datasets LUNA16, LIDC-IDRI, and NSCLC-Radiomics (exact sample size not reported).
• Computational efficiency validated on NVIDIA Jetson AGX Orin and Jetson Orin Nano edge devices.
• No external or prospective clinical validation; federated learning was tested in a simulated environment.

RadPigeon summaries are original and for information only. They are not clinical advice.