Dimensionality Reduction for Improving Out-of-Distribution Detection in Medical Image Segmentation

TL;DR

This paper proposes a dimensionality reduction approach using PCA on bottleneck features to enhance out-of-distribution detection in medical image segmentation, specifically for liver MRI scans, with high accuracy and low computational cost.

Contribution

It introduces a novel application of PCA combined with Mahalanobis distance for OOD detection in medical segmentation models, improving performance and efficiency.

Findings

Effective OOD detection with PCA and Mahalanobis distance

High detection accuracy on liver MRI data

Minimal additional computational load

Abstract

Clinically deployed segmentation models are known to fail on data outside of their training distribution. As these models perform well on most cases, it is imperative to detect out-of-distribution (OOD) images at inference to protect against automation bias. This work applies the Mahalanobis distance post hoc to the bottleneck features of a Swin UNETR model that segments the liver on T1-weighted magnetic resonance imaging. By reducing the dimensions of the bottleneck features with principal component analysis, OOD images were detected with high performance and minimal computational load.