Disentangling Sampling from Training Budget in Class-Imbalanced CT Body Composition Segmentation
The paper discusses the challenge of class imbalance in medical image segmentation, particularly in CT body composition. It introduces episodic sampling as a method to achieve class-balanced batch construction and evaluates its effectiveness against traditional sampling strategies. The findings suggest that episodic sampling offers advantages in low-data training scenarios and highlights the importance of considering training iteration budgets in sampling strategies.
- ▪Class imbalance is a significant issue in medical image segmentation, affecting the representation of rare classes.
- ▪Episodic sampling was compared to random and weighted sampling on muscle and adipose tissues from the SAROS dataset.
- ▪Under low-data training conditions, episodic sampling outperformed other methods, indicating its effectiveness in class-imbalanced scenarios.
Opening excerpt (first ~120 words) tap to expand
Electrical Engineering and Systems Science > Image and Video Processing arXiv:2605.20405 (eess) [Submitted on 19 May 2026] Title:Disentangling Sampling from Training Budget in Class-Imbalanced CT Body Composition Segmentation Authors:Iason Skylitsis, Dimitrios Karkalousos, Ivana Išgum View a PDF of the paper titled Disentangling Sampling from Training Budget in Class-Imbalanced CT Body Composition Segmentation, by Iason Skylitsis and 2 other authors View PDF HTML (experimental) Abstract:Class imbalance is a fundamental challenge in medical image segmentation, where frequent classes typically dominate training at the expense of rare classes.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
