.png){kind=icon}
.png){kind=icon}
Return
GiTNet: A graph-based trajectory-informed network for gaze-supervised medical image segmentation
.png){kind=icon}
PRE
AI
S
X
J
X
Y
J
DOI：10.1016/j.media.2026.104112.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Graph-based trajectory-informed network integrates static fixations and dynamic trajectories for gaze-supervised medical image segmentation. • Trajectory relational alignment constrains the graph topology to capture anatomical relationships guided by gaze behavior. • Neighbor-aware pseudo supervision incorporates semantic context from neighboring nodes to mitigate gaze noise and uncertainty. • Graph representational consistency reinforce supervisory constraints and enhance the model’s ability to understand complex spatial structures. • Our method outperforms state-of-the-art weakly supervised methods on two public datasets.
Keywords:
Graph-based network
Gaze-supervised segmentation
Trajectory relational alignment
Pseudo supervision
Medical image analysis
Journal
IF:
11.8
Papers:
3.7K
Citations:
2.4W
Organization
Cited Papers