CXR-LT 2026 Challenge: Projection-Aware Multi-Label and Zero-Shot Chest X-Ray Classification
This work addresses classification challenges in medical imaging for known and unseen lesions, though it appears incremental by building on existing methods like CheXzero.
The paper tackles multi-label and zero-shot classification for chest X-ray lesions, integrating projection-specific models and a novel dual-branch architecture with contrastive learning and LLM prompts to handle long-tail imbalances, achieving robust generalization across tasks.
This challenge tackles multi-label classification for known chest X-ray (CXR) lesions and zero-shot classification for unseen ones. To handle diverse CXR projections, we integrate projection-specific models via a classification network into a unified framework. For zero-shot classification (Task 2), we extend CheXzero with a novel dual-branch architecture that combines contrastive learning, Asymmetric Loss (ASL), and LLM-generated descriptive prompts. This effectively mitigates severe long-tail imbalances and maximizes zero-shot generalization. Additionally, strong data and test-time augmentations (TTA) ensure robustness across both tasks.