Towards continual learning in medical imaging

TL;DR

This paper explores continual learning in brain MRI segmentation, evaluating elastic weight consolidation's ability to reduce catastrophic forgetting between normal structure and lesion segmentation tasks.

Contribution

It applies and assesses elastic weight consolidation in medical imaging, highlighting its potential and limitations in sequential brain MRI segmentation tasks.

Findings

EWC reduces catastrophic forgetting in MRI segmentation.

Significant room for improvement remains in continual learning methods.

The study extends EWC evaluation from reinforcement learning to medical imaging.

Abstract

This work investigates continual learning of two segmentation tasks in brain MRI with neural networks. To explore in this context the capabilities of current methods for countering catastrophic forgetting of the first task when a new one is learned, we investigate elastic weight consolidation, a recently proposed method based on Fisher information, originally evaluated on reinforcement learning of Atari games. We use it to sequentially learn segmentation of normal brain structures and then segmentation of white matter lesions. Our findings show this recent method reduces catastrophic forgetting, while large room for improvement exists in these challenging settings for continual learning.