An Improved ResNet-Based U-Net Model for Small Blood Vessel Segmentation

Abstract

Accurate segmentation of cerebral blood vessels remains challenging due to the slender morphology, low contrast, and susceptibility to structural discontinuities of small vessels in cerebral vascular images. To address this issue, this paper proposes an improved ResNet-based U-Net model. By introducing residual connections into the classic U-Net encoder-decoder architecture, the model enhances feature reuse and gradient propagation, thereby improving the recognition of fine-grained vascular structures. Experimental results on the DIAS cerebral DSA dataset demonstrate that the proposed method outperforms the traditional U-Net in terms of both regional overlap and topological continuity, effectively improving the completeness and continuity of small vessel segmentation, and providing a feasible technical pathway for clinical cerebrovascular analysis.

Ähnliche Arbeiten