CXR-LT 2026 Challenge: Projection-Aware Multi-Label and Zero-Shot Chest X-Ray Classification

2026-04-02

AI Summary

This work addresses the dual challenge of multi-label classification for known pathologies and zero-shot classification for unseen pathologies in chest X-ray images by proposing a projection-aware unified framework. The framework integrates view-specific models to handle multi-label tasks and employs a dual-branch architecture that combines contrastive learning, asymmetric loss (ASL), and semantic prompts generated by large language models to enhance generalization to novel pathologies. Strong data augmentation and test-time augmentation strategies are further incorporated to mitigate the long-tailed class distribution and improve robustness. Experimental results demonstrate that the proposed method significantly outperforms existing approaches under both multi-label and zero-shot settings, achieving notable performance gains particularly in recognizing rare and previously unseen pathologies.

Abstract

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.

Problem

• multi-label classification
• zero-shot classification
• chest X-ray
• projection-aware
• long-tail imbalance

Innovation

• projection-aware
• zero-shot classification
• dual-branch architecture
• contrastive learning
• LLM-generated prompts