Weakly-supervised Medical Image Segmentation with Gaze Annotations

• DOI: 10.48550/arxiv.2407.07406
• Publication Date: 2024-07-10
• Authors: Yuan Zhong; Chenhui Tang; Yumeng Yang; Ruoxi Qi; Kang Zhou; Yuqi Gong; Pheng‐Ann Heng; Janet H. Hsiao; Qi Dou
• Keywords: FOS: Computer and information sciences; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

Eye gaze that reveals human observational patterns has increasingly been incorporated into solutions for vision tasks. Despite recent explorations on leveraging to aid deep networks, few studies exploit as an efficient annotation approach medical image segmentation which typically entails heavy annotating costs. In this paper, we propose collect dense weak supervision with a scheme. To train gaze, multi-level framework trains multiple networks from discriminative attention, simulated set of pseudo-masks derived by applying hierarchical thresholds heatmaps. Furthermore, mitigate noise, cross-level consistency is exploited regularize overfitting noisy labels, steering models toward clean learned peer networks. The proposed method validated two public datasets polyp and prostate We contribute high-quality dataset entitled GazeMedSeg extension the popular datasets. best our knowledge, first segmentation. Our experiments demonstrate outperforms previous label-efficient schemes in terms both performance time. collected data code are available at: https://github.com/med-air/GazeMedSeg.