Disentangling Sampling from Training Budget in Class-Imbalanced CT Body Composition Segmentation

Researchers compared episodic, random, and weighted sampling for class-imbalanced CT body composition segmentation.

With full data, episodic sampling performed about the same as the other methods.

In low-data settings, episodic sampling improved performance and overfit less than random or weighted sampling.

The study argues that training iteration budget is a major confound in sampling comparisons.

Class-balanced episodic batches may offer a simple, low-cost regularization boost for rare-class learning.