Patient-Conditioned Adaptive Offsets for Reliable Diagnosis across Subgroups

2026-01-19

AI Summary

This work addresses the performance disparities of medical AI models across patient subgroups, a challenge exacerbated by conventional fairness approaches that discard sensitive attributes and consequently degrade diagnostic accuracy. To overcome this limitation, the authors propose HyperAdapt, a novel framework that eschews the removal of sensitive attributes and instead leverages clinically relevant variables—such as age and sex—to conditionally adapt a shared diagnostic backbone. Specifically, a lightweight hypernetwork generates low-rank, bottleneck-constrained residual modulation parameters that enable efficient and robust patient-specific fine-tuning while preserving the core medical knowledge encoded in the backbone. Extensive experiments demonstrate that HyperAdapt significantly improves subgroup performance across multiple medical imaging benchmarks, achieving a 4.1% gain in recall and a 4.4% improvement in F1 score over the strongest baseline on the PAD-UFES-20 dataset, with particularly pronounced benefits for underrepresented groups.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

AI models for medical diagnosis often exhibit uneven performance across patient populations due to heterogeneity in disease prevalence, imaging appearance, and clinical risk profiles. Existing algorithmic fairness approaches typically seek to reduce such disparities by suppressing sensitive attributes. However, in medical settings these attributes often carry essential diagnostic information, and removing them can degrade accuracy and reliability, particularly in high-stakes applications. In contrast, clinical decision making explicitly incorporates patient context when interpreting diagnostic evidence, suggesting a different design direction for subgroup-aware models. In this paper, we introduce HyperAdapt, a patient-conditioned adaptation framework that improves subgroup reliability while maintaining a shared diagnostic model. Clinically relevant attributes such as age and sex are encoded into a compact embedding and used to condition a hypernetwork-style module, which generates small residual modulation parameters for selected layers of a shared backbone. This design preserves the general medical knowledge learned by the backbone while enabling targeted adjustments that reflect patient-specific variability. To ensure efficiency and robustness, adaptations are constrained through low-rank and bottlenecked parameterizations, limiting both model complexity and computational overhead. Experiments across multiple public medical imaging benchmarks demonstrate that the proposed approach consistently improves subgroup-level performance without sacrificing overall accuracy. On the PAD-UFES-20 dataset, our method outperforms the strongest competing baseline by 4.1% in recall and 4.4% in F1 score, with larger gains observed for underrepresented patient populations.

Problem

• algorithmic fairness
• medical diagnosis
• subgroup disparity
• patient heterogeneity
• sensitive attributes

Innovation

• patient-conditioned adaptation
• hypernetwork
• subgroup fairness
• medical imaging diagnosis
• residual modulation