Uncertainty-Aware Extreme Point Tracing for Weakly Supervised Ultrasound Image Segmentation
This work addresses the high annotation burden in medical imaging for clinicians, offering a practical solution with incremental improvements in weakly supervised segmentation.
The paper tackles the problem of reducing annotation costs in medical image segmentation by proposing a weakly supervised framework that uses only four extreme points as annotation, achieving performance comparable to or surpassing fully supervised methods on ultrasound datasets like BUSI and UNS.
Automatic medical image segmentation is a fundamental step in computer-aided diagnosis, yet fully supervised approaches demand extensive pixel-level annotations that are costly and time-consuming. To alleviate this burden, we propose a weakly supervised segmentation framework that leverages only four extreme points as annotation. Specifically, bounding boxes derived from the extreme points are used as prompts for the Segment Anything Model 2 (SAM2) to generate reliable initial pseudo labels. These pseudo labels are progressively refined by an enhanced Feature-Guided Extreme Point Masking (FGEPM) algorithm, which incorporates Monte Carlo dropout-based uncertainty estimation to construct a unified gradient uncertainty cost map for boundary tracing. Furthermore, a dual-branch Uncertainty-aware Scale Consistency (USC) loss and a box alignment loss are introduced to ensure spatial consistency and precise boundary alignment during training. Extensive experiments on two public ultrasound datasets, BUSI and UNS, demonstrate that our method achieves performance comparable to, and even surpassing fully supervised counterparts while significantly reducing annotation cost. These results validate the effectiveness and practicality of the proposed weakly supervised framework for ultrasound image segmentation.