Curriculum Knowledge Switching for Pancreas Segmentation

TL;DR

This paper introduces a novel curriculum learning framework called Curriculum Knowledge Switching (CKS) for pancreas segmentation, which progressively trains models through phases of increasing difficulty, leading to state-of-the-art results.

Contribution

The paper proposes a new curriculum learning approach with a phased training strategy and momentum update mechanism for improved pancreas segmentation.

Findings

Achieved state-of-the-art DSC scores on NIH dataset

Effective phased training improves segmentation accuracy

Momentum update stabilizes training across dataset changes

Abstract

Pancreas segmentation is challenging due to the small proportion and highly changeable anatomical structure. It motivates us to propose a novel segmentation framework, namely Curriculum Knowledge Switching (CKS) framework, which decomposes detecting pancreas into three phases with different difficulty extent: straightforward, difficult, and challenging. The framework switches from straightforward to challenging phases and thereby gradually learns to detect pancreas. In addition, we adopt the momentum update parameter updating mechanism during switching, ensuring the loss converges gradually when the input dataset changes. Experimental results show that different neural network backbones with the CKS framework achieved state-of-the-art performance on the NIH dataset as measured by the DSC metric.