Federated Learning on Brain Disease Research: Segmentation of Cerebral Small Vessel Diseases (CSVD) using Multi-scale Hybrid Spatial Deep Learning Approach

Federated learning is an emerging approach that enables large-scale decentralized learning without the need to share data among different data owners.This approach is particularly valuable in addressing data privacy concerns in medical image analysis.However, existing methods often impose a strict requirement for label consistency across clients, which significantly limits its applicability.Various clinical sites may only provide annotations for specific organs of interest, and there may be limited or no overlap in the labeled data among different sites.The human brain receives nutrients and oxygen through blood vessels in the brain.The pathology of small vessels, i.e. mesoscopic scale, is a vulnerable component of the cerebral blood supply and can result in major complications such as Cerebral Small Vessel Diseases (CSVD).In this paper, we propose a hybrid architecture for medical image segmentation to produce efficient representations from global and local features and adaptively aggregate them, aiming to fully exploit their strengths to obtain better segmentation performance in federated learning.Furthermore, we propose a multiscale feature extraction module embedded at the bottom of the proposed model, which can efficiently extract hidden multi-scale contextual information and aggregate multi-scale features.Experiments on segmentation over threedimensional rotational angiography of internal Carotid Artery with aneurysm (SHINY-ICARUS) challenge dataset show the effectiveness of the proposed multiscale framework.