Medical image segmentation improves with adaptive label noise correction
Sam-Correction: Fully Adaptive Label Noise Reduction for Medical Image Segmentation
Computer Vision and Pattern Recognition
Summary
Medical images are often labeled by experts to teach computers how to identify important areas, but these labels can be inconsistent or noisy. The authors propose a method called SAM-Correction that automatically cleans up these noisy labels without needing extra expert-corrected data. Their method uses a large pre-trained segmentation model together with a new segmentation network that works alongside it to gradually improve the label quality. Tests show this approach reduces label errors and leads to better computer segmentation results even when clean labels are not available.
What this means in practice
- •For medical image software developers: Improve segmentation tools by integrating adaptive label correction that reduces annotation errors without extra expert data.
- •For radiology ai teams: Train more reliable medical image segmentation models when perfectly labeled data is unavailable, improving diagnostic support systems.
Authors
Takuro Shimaya, Masahiro Saiko
Abstract
The quality of the teacher label has a dramatic impact on the outcome of the machine learning process. Medical image segmentation often suffers from a noisy ground truth (GT) mask due to difficulties in consistent annotation. Although a recent study has proposed a promising method to curate the noisy GT mask, this requires additional well-curated GT masks that are not always accessible without the aid of multiple medical experts. As an alternative, we propose SAM-Correction, which utilizes a foundation model to correct the noisy GT mask in a fully adaptive manner instead of manually preparing well-curated masks. In our approach, we progressively mitigate label noise via the segment anything model (SAM), where prompts are given from another segmentation network that we train concurrently. Experiments with various types of label noise derived from actual human annotations demonstrate that SAM-C achieves precise noise reduction and high-performing segmentation models trained thereby. Our method paves the way for robust model training even when clean labels are entirely unavailable.