![[2602.15811] Task-Agnostic Continual Learning for Chest Radiograph Classification](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15811] Task-Agnostic Continual Learning for Chest Radiograph Classification
Summary
This article presents CARL-XRay, a novel continual learning framework for chest radiograph classification that adapts to new datasets without retraining on previous data, ensuring robust performance in clinical settings.
Why It Matters
As medical imaging evolves, the ability to continuously learn from new data is crucial for maintaining diagnostic accuracy. CARL-XRay addresses the challenge of updating models without degrading performance, making it significant for clinical applications where timely adaptation is essential.
Key Takeaways
- CARL-XRay enables continual learning in chest radiograph classification without retraining on previous datasets.
- The framework maintains high performance with fewer trainable parameters compared to traditional methods.
- It uses a latent task selector for stable task identification and adaptation during sequential updates.
- The proposed method outperforms joint training in task-unknown scenarios, achieving higher routing accuracy.
- This approach provides a practical alternative for clinical deployment, enhancing diagnostic reliability.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.15811 (cs) [Submitted on 17 Feb 2026] Title:Task-Agnostic Continual Learning for Chest Radiograph Classification Authors:Muthu Subash Kavitha, Anas Zafar, Amgad Muneer, Jia Wu View a PDF of the paper titled Task-Agnostic Continual Learning for Chest Radiograph Classification, by Muthu Subash Kavitha and 3 other authors View PDF HTML (experimental) Abstract:Clinical deployment of chest radiograph classifiers requires models that can be updated as new datasets become available without retraining on previously ob- served data or degrading validated performance. We study, for the first time, a task-incremental continual learning setting for chest radiograph classification, in which heterogeneous chest X-ray datasets arrive sequentially and task identifiers are unavailable at inference. We propose a continual adapter-based routing learning strategy for Chest X-rays (CARL-XRay) that maintains a fixed high-capacity backbone and incrementally allocates lightweight task-specific adapters and classifier heads. A latent task selector operates on task-adapted features and leverages both current and historical context preserved through compact prototypes and feature-level experience replay. This design supports stable task identification and adaptation across sequential updates while avoiding raw-image storage. Experiments on large-scale public chest radiograph datasets demonstrate robust performance retention and re...
