Abstract
Chest X-ray (CXR) classification aims to automatically identify abnormalities and generate reliable diagnostic results, a critical process facilitating disease screening while alleviating doctors’ workload pressure. Despite deep learning advancements, existing concept bottleneck models (CBMs) for CXR classification suffer from two main limitations: they rely on general-purpose large language models (LLMs) lacking domain-specific medical knowledge to establish diagnostic standards, and they rely solely on single-view images, failing to leverage the complementary diagnostic information provided by multiple views in clinical practice. Inspired by actual clinical workflows, we introduce a novel classification model integrating disentangled multi-view learning enhanced with structured concepts, namely SCD-CXR. First, a diagnostic standard construction module incorporates specialized CXR guidelines, medical domain knowledge from clinical texts, and LLM reasoning capabilities to construct accurate, medically reliable structured diagnostic standards. Furthermore, a multi-view disentangled visual representation learning module simulates doctors’ diagnostic process to disentangle and fuse complementary multi-view representations, enhancing the model’s diagnostic performance. Experimental evaluations on the CheXpert and IU X-RAY benchmarks demonstrate that SCD-CXR exhibits competitive performance compared to state-of-the-art methods, highlighting its strong potential to advance automatic CXR classification.The code is available at the anonymized repository: https://anonymous.4open.science/r/Anonymous-C9A8
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MWM_002.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MWM_002_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=59tKHdnnzA
BibTex
@InProceedings{JiaXin_Towards_MICCAISAT2026,
author = { Jia, Xing AND Liu, Yichen AND Pei, Songwen AND Xiong, Yun AND Shen, Ailing AND Lin, Siqi AND Dai, Anqi AND Tang, Chunlei},
title = { { Towards Reliable CXR Diagnosis through Disentangled Multi-View Learning Enhanced with Structured Concepts } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17262},
month = {pending},
page = {pending}
}