Towards continual learning in medical imaging
This work addresses incremental improvements in continual learning for medical imaging, specifically for brain MRI segmentation tasks.
The study tackled catastrophic forgetting in continual learning for brain MRI segmentation by applying elastic weight consolidation, originally from reinforcement learning, to sequentially learn normal brain structures and white matter lesions, showing it reduces forgetting but with significant room for improvement.
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.