The unreasonable effectiveness of Batch-Norm statistics in addressing catastrophic forgetting across medical institutions
This addresses the problem of model brittleness for medical practitioners deploying AI in clinical settings, but it is incremental as it builds on existing methods like EWC.
The paper tackled catastrophic forgetting in medical deep learning models when fine-tuning across institutions by adapting Elastic Weight Consolidation using global batch normalization statistics from the original dataset, achieving improved performance with a 15% reduction in forgetting on mammographic breast density assessment.
Model brittleness is a primary concern when deploying deep learning models in medical settings owing to inter-institution variations, like patient demographics and intra-institution variation, such as multiple scanner types. While simply training on the combined datasets is fraught with data privacy limitations, fine-tuning the model on subsequent institutions after training it on the original institution results in a decrease in performance on the original dataset, a phenomenon called catastrophic forgetting. In this paper, we investigate trade-off between model refinement and retention of previously learned knowledge and subsequently address catastrophic forgetting for the assessment of mammographic breast density. More specifically, we propose a simple yet effective approach, adapting Elastic weight consolidation (EWC) using the global batch normalization (BN) statistics of the original dataset. The results of this study provide guidance for the deployment of clinical deep learning models where continuous learning is needed for domain expansion.