DCAMIL: Eye-tracking guided dual-cross-attention multi-instance learning for refining fundus disease detection

Hongyang Jiang, Mengdi Gao, Jingqi Huang, Chen Tang, Xiaoqing Zhang, Jiang Liu

Abstract: Highlights•We propose an eye-tracking-based HITL CAD system for fundus disease detection.•We propose a novel DCAMIL model with the contrast learning regularization.•We introduce the SA and DAN modules into the DCAMIL model.•We construct DR-Gaze and AMD-Gaze datasets according to clinical settings.