M3F: Multi-Field-of-View Feature Fusion Network for Aortic Vessel Tree Segmentation in CT Angiography

Yunsu Byeon, Hyeseong Kim, Kyungwon Kim, Doohyun Park, Dosik Hwang

Abstract: Accurate segmentation of the aortic vessel tree (AVT) in computed tomography angiography (CTA) is crucial for diagnosing and monitoring vascular conditions. However, achieving automated and precise segmentation remains a challenging task due to the intricate structure of the AVT. To address this challenge, we introduce the Multi-Field-of-View Feature Fusion Network (M3F) for the AVT segmentation. M3F processes two distinct 3D patches: a large field-of-view patch for context information and a small field-of-view patch for detailed information. A key aspect of M3F is its fusion mechanism, which integrates the context from the coarse branch with the detail from the fine branch to improve segmentation performance. The M3F (proposed by team name ATB) achieves the 1st place on the second phase of the 2023 MICCAI Seg.A challenge leaderboard. Such remarkable performance highlights M3F’s potential for both clinical applications and further research in aortic vessel segmentation.