Saved in:
Bibliographic Details
Main Authors: Bai, Shaojin, Su, Yuting, Nie, Weizhi
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.25202
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912983646470144
author Bai, Shaojin
Su, Yuting
Nie, Weizhi
author_facet Bai, Shaojin
Su, Yuting
Nie, Weizhi
contents Cross-site generalizability in medical AI is fundamentally compromised by selection bias, a structural mechanism where patient demographics (e.g., age, severity) non-randomly dictate hospital assignment. Conventional Domain Generalization (DG) paradigms, which predominantly target image-level distribution shifts, fail to address the resulting spurious correlations between site-specific variations and diagnostic labels. To surmount this identifiability barrier, we propose CIV-DG, a causal framework that leverages Conditional Instrumental Variables to disentangle pathological semantics from scanner-induced artifacts. By relaxing the strict random assignment assumption of standard IV methods, CIV-DG accommodates complex clinical scenarios where hospital selection is endogenously driven by patient demographics. We instantiate this theory via a Deep Generalized Method of Moments (DeepGMM) architecture, employing a conditional critic to minimize moment violations and enforce instrument-error orthogonality within demographic strata. Extensive experiments on the Camelyon17 benchmark and large-scale Chest X-Ray datasets demonstrate that CIV-DG significantly outperforms leading baselines, validating the efficacy of conditional causal mechanisms in resolving structural confounding for robust medical AI.
format Preprint
id arxiv_https___arxiv_org_abs_2603_25202
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle CIV-DG: Conditional Instrumental Variables for Domain Generalization in Medical Imaging
Bai, Shaojin
Su, Yuting
Nie, Weizhi
Computer Vision and Pattern Recognition
Multimedia
Cross-site generalizability in medical AI is fundamentally compromised by selection bias, a structural mechanism where patient demographics (e.g., age, severity) non-randomly dictate hospital assignment. Conventional Domain Generalization (DG) paradigms, which predominantly target image-level distribution shifts, fail to address the resulting spurious correlations between site-specific variations and diagnostic labels. To surmount this identifiability barrier, we propose CIV-DG, a causal framework that leverages Conditional Instrumental Variables to disentangle pathological semantics from scanner-induced artifacts. By relaxing the strict random assignment assumption of standard IV methods, CIV-DG accommodates complex clinical scenarios where hospital selection is endogenously driven by patient demographics. We instantiate this theory via a Deep Generalized Method of Moments (DeepGMM) architecture, employing a conditional critic to minimize moment violations and enforce instrument-error orthogonality within demographic strata. Extensive experiments on the Camelyon17 benchmark and large-scale Chest X-Ray datasets demonstrate that CIV-DG significantly outperforms leading baselines, validating the efficacy of conditional causal mechanisms in resolving structural confounding for robust medical AI.
title CIV-DG: Conditional Instrumental Variables for Domain Generalization in Medical Imaging
topic Computer Vision and Pattern Recognition
Multimedia
url https://arxiv.org/abs/2603.25202