Transferable Low-Rank Convolutional Bases for Onboarding Unseen Medical Imaging Modalities

2026-07-18

AI Summary

This work addresses the challenge of efficiently adapting medical imaging models to unseen modalities post-deployment while avoiding catastrophic forgetting. The authors propose a parameter-efficient adaptation method that, under a strict leave-one-domain-out setting, freezes the pre-trained convolutional backbone and leverages transferable low-rank convolutional bases learned from source modalities. Adaptation is achieved solely through projection parameters atop these bases, constituting only 0.78% of the total model parameters. By integrating convolutional LoRA, low-rank decomposition, and Mahalanobis anomaly detection, the approach improves adaptation accuracy on new modalities by 6.11 percentage points over random bases, while incurring zero performance loss on source modalities (Δ = 0.00 pp), substantially outperforming full fine-tuning and decision-level adaptation strategies.

Abstract

Deploying a medical imaging model that must later accommodate a modality it has never seen is a recurring practical problem: retraining the shared representation is expensive and destroys performance on the modalities already in service. We study this \emph{onboarding} problem under a strict leave-one-domain-out protocol, in which a convolutional backbone is pre-trained on source modalities (Kidney CT and Brain MRI), frozen permanently, and then required to accommodate an unseen modality (Chest X-ray). Under this protocol we establish three findings. First, decision-layer parameter-efficient fine-tuning is insufficient when the backbone has never observed the target modality: a linear probe and fully-connected LoRA both fall well short, whereas convolutional LoRA recovers most of the achievable accuracy, showing that adaptation must reach the convolutional features. Second, and centrally, the low-rank convolutional \emph{basis} learned on the source modalities \emph{transfers}: freezing that basis and training only its up-projections onboards the unseen modality using just $0.78\%$ of full fine-tuning's parameters, at an accuracy $6.11$ percentage points above a random basis of identical size, while an equivalent decision-layer basis exhibits no reliable transfer. Third, adapter-based onboarding leaves source-modality accuracy exactly unchanged ($Δ= 0.00$ pp), whereas full fine-tuning reaches the highest target accuracy only by catastrophically degrading the source modalities. A Mahalanobis score on frozen backbone features detects the unseen modality with high sensitivity at a strict source-retention threshold, providing a practical trigger for when onboarding is required. All results are reported over three seeds with paired bootstrap confidence intervals.

Problem

• medical imaging
• unseen modality
• onboarding
• transfer learning
• catastrophic forgetting

Innovation

• transferable low-rank convolutional bases
• parameter-efficient fine-tuning
• modality onboarding
• frozen backbone adaptation
• medical imaging generalization