Learning a Sampling-Free Variational DNN Plugin from Tiny Training Sets to Refine OOD Segmentation With Uncertainty Estimation

2026-06-14

AI Summary

This work addresses the limited generalization of deep neural networks on out-of-distribution (OOD) medical images, which often suffer from domain shifts caused by variations in imaging devices and protocols, coupled with the high cost of retraining. To tackle this challenge, the authors propose VarDeepPCA—a lightweight variational deep learning framework that learns anatomical geometric priors from only a small amount of source-domain data, without requiring any target-domain data or large-scale pretraining. It effectively restores degraded segmentations and provides uncertainty estimates by reinterpreting the implicit probabilities of softmax outputs, enabling efficient, sampling-free variational modeling and inference. Extensive experiments across four clinical tasks and fourteen public datasets demonstrate that VarDeepPCA significantly improves anatomical plausibility and clinical usability of OOD segmentations, substantially reducing errors and outperforming fifteen state-of-the-art methods.

Abstract

Deep neural networks (DNNs) frequently fail to generalize to out-of-distribution (OOD) medical images because of variations in scanners and acquisition protocols. Retraining DNN models to address these distribution shifts is often impractical due to the high cost of acquiring and annotating new medical datasets. To address this, we introduce VarDeepPCA, a novel lightweight variational DNN framework designed to restore/refine degraded segmentation maps by leveraging intrinsic geometric priors. Unlike existing approaches that require target-domain data or extensive pre-training, our VarDeepPCA explicitly learns a distribution of valid anatomical geometries using only small in-distribution (ID) datasets. Theoretically, our novel variational learning framework leverages a reinterpretation of the softmax mapping to implicitly perform exact distribution modeling, thereby enabling computationally efficient, sampling-free learning and inference. This also enables VarDeepPCA to provide uncertainty estimates associated with its restored segmentation maps. We empirically validate our framework across 4 distinct clinical applications, using 14 publicly available datasets, involving segmentation of the myocardium, neuroretinal rim, prostate, and fetal head. Comparisons against 15 existing methods demonstrate that VarDeepPCA consistently restores segmentation maps produced by the existing methods on OOD data to (i) significantly improve anatomical plausibility of geometries and clinical utility of the segmentations, and (ii) significantly reduce errors, without needing any more training data than that used by existing methods.

Problem

• out-of-distribution
• medical image segmentation
• distribution shift
• uncertainty estimation
• anatomical plausibility

Innovation

• sampling-free variational learning
• out-of-distribution segmentation
• geometric priors
• uncertainty estimation
• lightweight DNN plugin