Disentangling Sampling from Training Budget in Class-Imbalanced CT Body Composition Segmentation
Researchers have developed an episodic sampling method to improve class-balanced batch construction for medical image segmentation, particularly in scenarios with imbalanced datasets. This technique, adapted from few-shot learning, was evaluated on CT body composition segmentation and showed superior performance over random and weighted sampling under low-data conditions. The study highlights the importance of considering training iteration budgets when comparing sampling strategies, suggesting episodic sampling offers a low-cost, model-agnostic approach for addressing class imbalance in medical imaging. AI
IMPACT Offers a novel, low-cost method to improve AI model performance on imbalanced medical imaging datasets.
