Class aware thresholding improves semi supervised 3d abdominal organ segmentation
ThreshGuide: Class-Aware Labeled-Guided Thresholding for Semi-Supervised 3D Abdominal Multi-Organ Segmentation
Computer Vision and Pattern Recognition
Summary
Segmenting different organs in 3D medical images can be hard because organs vary a lot. Usually, computers guess labels for unlabeled images based on confidence, but using the same confidence cut-off for all organs is not ideal. The authors propose ThreshGuide, which smartly adjusts thresholds for each organ by learning from labeled images. This approach helps computers better decide when to trust guesses for each organ, especially improving performance on organs that are harder to recognize.
What this means in practice
- •For medical imaging software developers: Enhance abdominal organ segmentation tools by using class-aware thresholds to improve accuracy on difficult organs with minimal labeled data.
- •For radiology workflow integrators: Implement semi-supervised learning pipelines that reduce annotation needs by better selecting pseudo-labels for multi-organ CT segmentation.
Authors
Hongyu Liu, Yinlong Wang, Lusha Li, Hui Meng
Abstract
Pseudo-labeling is a strong paradigm for semi-supervised medical image segmentation, yet its effectiveness is highly sensitive to confidence thresholding. In abdominal multi-organ segmentation, a fixed global threshold is particularly suboptimal because organ classes differ substantially in size, appearance, and learning difficulty. In this work, we propose ThreshGuide, a class-aware threshold adaptation framework that uses labeled data to guide pseudo-label selection on unlabeled data. Built upon a standard teacher-student architecture, the teacher model evaluates labeled samples during training to estimate class-aware threshold targets by maximizing an error-aware F\b{eta} criterion that balances precision and coverage. These targets are then smoothed with an exponential moving average (EMA) and used to filter unlabeled voxels in a class-dependent manner. Experiments on FLARE2022 and AMOS2022 show that ThreshGuide performs competitively overall, yielding clear improvements specifically on hard-to-learn organs.