How CAD Performance Changes with Sequential Institution Addition in Federated Learning

CAD software performance for cerebral aneurysm and brain metastasis detection improved when institutions were added sequentially in federated learning, with fine-tuning strategies achieving comparable gains to full fine-tuning using fewer parameters.

• Sequential institution addition under federated learning yielded more consistent performance improvements than simultaneous addition for both aneurysm and metastasis detection.
• Fine-tuning a subset of layers required substantially fewer trainable parameters than full fine-tuning while maintaining comparable performance.

Automated summary

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