Unsupervised Domain Adaptation for Small Bowel Segmentation using Disentangled Representation

TL;DR

This paper introduces a novel unsupervised domain adaptation approach for small bowel segmentation that leverages feature disentanglement to improve transferability across different CT scan domains, enhancing clinical applicability.

Contribution

The paper proposes a unique two-stream auto-encoding architecture that disentangles intensity and non-intensity features for more controllable domain adaptation in medical image segmentation.

Findings

Significant improvement over existing methods in three metrics.

Effective transfer of non-intensity features across domains.

Closer to clinical application in small bowel segmentation.

Abstract

We present a novel unsupervised domain adaptation method for small bowel segmentation based on feature disentanglement. To make the domain adaptation more controllable, we disentangle intensity and non-intensity features within a unique two-stream auto-encoding architecture, and selectively adapt the non-intensity features that are believed to be more transferable across domains. The segmentation prediction is performed by aggregating the disentangled features. We evaluated our method using intravenous contrast-enhanced abdominal CT scans with and without oral contrast, which are used as source and target domains, respectively. The proposed method showed clear improvements in terms of three different metrics compared to other domain adaptation methods that are without the feature disentanglement. The method brings small bowel segmentation closer to clinical application.