Overview of the CXR-LT 2026 Challenge: Multi-Center Long-Tailed and Zero Shot Chest X-ray Classification

TL;DR

The CXR-LT 2026 challenge introduces a large, multi-center chest X-ray dataset and tasks focusing on long-tailed and zero-shot classification, highlighting the effectiveness of vision-language pre-training in medical imaging.

Contribution

This paper presents the third iteration of the CXR-LT challenge with a new multi-center dataset and two core tasks addressing long-tailed and open-world classification in chest X-ray analysis.

Findings

Top solutions achieved 0.5854 mAP on known classes

Zero-shot performance improved with vision-language pre-training

Large-scale datasets enhance generalization to unseen diseases

Abstract

Chest X-ray (CXR) interpretation is hindered by the long-tailed distribution of pathologies and the open-world nature of clinical environments. Existing benchmarks often rely on closed-set classes from single institutions, failing to capture the prevalence of rare diseases or the appearance of novel findings. To address this, we present the CXR-LT 2026 challenge. This third iteration of the benchmark introduces a multi-center dataset comprising over 145,000 images from PadChest and NIH Chest X-ray datasets. The challenge defines two core tasks: (1) Robust Multi-Label Classification on 30 known classes and (2) Open-World Generalization to 6 unseen (out-of-distribution) rare disease classes. We report the results of the top-performing teams, evaluating them via mean Average Precision (mAP), AUROC, and F1-score. The winning solutions achieved an mAP of 0.5854 on Task 1 and 0.4315 on Task 2, demonstrating that large-scale vision-language pre-training significantly mitigates the performance drop typically associated with zero-shot diagnosis.