DSFormer: Dynamic size attention with enhanced long-range dependency modeling for artery/vein classification

Open Webpage

Zeyuan Ju, Chouyu Chen, Zhipeng Liu, Lijun Guo, Zhenyu Lei, Masaaki Omura, Shangce Gao

Published: 2025, Last Modified: 03 Feb 2026Knowl. Based Syst. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•A network tailored for artery/vein classification.•A dynamic block to enhance feature extraction and reduce irrelevant details.•An enhanced attention mechanism to generate discriminative global features.•A mixed shallow-deep context bridge to strengthen critical feature relationships.•DSFormer achieves state-of-the-art in artery/vein classification.