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

Juno Cho; Dohui Kim; Mingeon Kim; Hyunseo Jang; Chang Sun Lee; Jong Chul Ye

cs.CVApr 2, 2026

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

Authors: Juno Cho, Dohui Kim, Mingeon Kim, Hyunseo Jang, Chang Sun Lee, Jong Chul Ye

Organizations: KAIST · KAIST, Daejeon, South Korea · GIST · GIST, Gwangju, South Korea · Korea University · Korea University, Seoul, South Korea · KAIST Graduate School of AI · KAIST Graduate School of AI (GSAI), Seoul, South Korea

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.

Figures & tables

Figure 1: Multi-label classification diagram.

Figure 2: Dual-Branch Hybrid Architecture.

Setting	Proj	Model	mAP	BCE	mAUC	mECE
Projection aware	AP/PA	PCAM	0.4011	0.0871	0.9117	0.0067
		SWIN	0.4028	0.0866	0.9104	0.0052
		CAIT	0.3925	0.0868	0.9051	0.0067
		Ensemble	0.4185	0.0841	0.9182	0.0045
	L	PCAM	0.3336	0.1061	0.8681	0.0111
		SWIN	0.3045	0.1051	0.8621	0.0099

Table 1: Validation results of projection-aware and mixed-projection models.

Method	Group A	Group B	Group C	Avg.
Baseline (Zero-shot)	0.2635	0.3790	0.3134	0.3186
FT (All Labels)	0.2680	0.3852	0.3126	0.3219
FT (Physician Only)	0.2707	0.3852	0.3098	0.3219
Dual-Branch ( $\alpha=1.5$ )	0.2882	0.4139	0.3151	0.3391

Table 2: Impact of data quality and Asymmetric Loss (ASL) on zero-shot mAP. (FT: Standard Fine-Tuning,

\alpha=0

)

Panel of Figure 3 — Figure 3: Validation mAP on Proxy Group A. ASL ( $\alpha=1.5$ ) consistently outperforms $\alpha=0$ and the baseline.

Explore similar work

Overview of the CXR-LT 2026 Challenge: Multi-Center Long-Tailed and Zero Shot Chest X-ray Classification
Feb 25, 2026Hexin Dong, Yi Lin, Pengyu Zhou +7Chest X-Ray Classification Medical Image Benchmarks
Weill Cornell Medicine · Peking Union Medical College · University of Minnesota +2
CXR-LT 2026 Challenge: Multi-Center Long-Tailed and Zero Shot Chest X-ray Classification
Apr 16, 2026Hexin Dong, Yi Lin, Pengyu Zhou +25Zero-Shot Learning Long-Tailed Classification
Department of Population Health Sciences, Weill Cornell Medicine, New York, USA · Department of Radiology, Fuwai Hospital, National Center for Cardiovascular Diseases, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China · Department of Radiology, West China School of Medicine, Sichuan University, Sichuan University Affiliated Chengdu Second People’s Hospital, Chengdu Second People’s Hospital, Chengdu, China +20
ProtoCLIP: Prototype-Aligned Latent Refinement for Robust Zero-Shot Chest X-Ray Classification
Apr 20, 2026Florian Kittler, Sheethal Bhat, Andreas MaierVLM Robustness Chest X-Ray Classification
Friedrich-Alexander University Erlangen-Nuremberg, Erlangen 91054, Germany

Feb 25, 2026J/K move · Enter open · S save

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

Hexin Dong, Yi Lin, Pengyu Zhou, Xuan Zhong Feng+6

Weill Cornell Medicine · Peking Union Medical College · University of Minnesota+2

Chest X-Ray Classification Medical Image Benchmarks Long-Tailed Classification Open-Set Recognition

Open paper AskPDF arXiv

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.

Abstract

Explore similar work

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

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

ProtoCLIP: Prototype-Aligned Latent Refinement for Robust Zero-Shot Chest X-Ray Classification