MedSelect: Selective Labeling for Medical Image Classification Combining Meta-Learning with Deep Reinforcement Learning
This work addresses the challenge of expensive label acquisition in medical imaging, offering a selective labeling approach that is incremental, building on existing methods like meta-learning and reinforcement learning.
The authors tackled the problem of limited labeling resources in medical image classification by proposing MedSelect, a selective labeling method combining meta-learning and deep reinforcement learning, which outperformed baseline strategies across seen and unseen medical conditions for chest X-ray interpretation, with significant differences in latent embeddings and clinical features compared to the strongest baseline.
We propose a selective learning method using meta-learning and deep reinforcement learning for medical image interpretation in the setting of limited labeling resources. Our method, MedSelect, consists of a trainable deep learning selector that uses image embeddings obtained from contrastive pretraining for determining which images to label, and a non-parametric selector that uses cosine similarity to classify unseen images. We demonstrate that MedSelect learns an effective selection strategy outperforming baseline selection strategies across seen and unseen medical conditions for chest X-ray interpretation. We also perform an analysis of the selections performed by MedSelect comparing the distribution of latent embeddings and clinical features, and find significant differences compared to the strongest performing baseline. We believe that our method may be broadly applicable across medical imaging settings where labels are expensive to acquire.