CIV-DG: Conditional Instrumental Variables for Domain Generalization in Medical Imaging

2026-03-26

AI Summary

This work addresses the challenge of selection bias in medical imaging AI when deployed across institutions, where demographic disparities among patient populations induce spurious correlations that degrade model generalization. To mitigate this, the authors propose CIV-DG, a causal domain generalization framework that introduces conditional instrumental variables (CIVs) into medical image analysis for the first time. By relaxing the strict random assignment assumption of classical instrumental variables, CIV-DG accommodates the non-random, demography-driven clinical referral mechanisms prevalent in real-world settings. The method employs Deep Generalized Method of Moments (DeepGMM) to construct a conditional discriminator that enforces orthogonality between the instrument and error term within demographic strata, effectively disentangling pathological semantics from scanner-induced artifacts. Experiments on Camelyon17 and a large-scale chest X-ray dataset demonstrate that CIV-DG substantially outperforms existing approaches, validating its efficacy in removing structural confounding and enhancing out-of-distribution generalization.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

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.

Problem

• selection bias
• domain generalization
• spurious correlations
• medical imaging
• cross-site generalizability

Innovation

• Conditional Instrumental Variables
• Domain Generalization
• Selection Bias
• DeepGMM
• Causal Inference