Domain Incremental Learning for Pandemic-Resilient Chest X-Ray Analysis

In three linesReplay-based continual learning method for adapting pneumonia detection models across clinical domain variations without catastrophic forgetting. Incorporates class-aware balanced replay and dynamically reweighted class-imbalance loss. Achieves 88.66% accuracy on PneumoniaMNIST with 5 simulated domains, outperforming Experience Replay and Fine-Tuning baselines.

Summary generated by Claude — human-verified