Unsupervised Domain Adaptation with Target-Only Margin Disparity Discrepancy
2026-03-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors focus on improving liver image segmentation for a special type of scan called Cone-Beam CT (CBCT), which is used during minimally invasive procedures but has fewer annotated examples than regular CT scans. They use a technique called unsupervised domain adaptation to teach a model trained on regular CT images to work better on CBCT images, even without many labeled CBCT examples. Their approach involves refining an existing method called Margin Disparity Discrepancy (MDD) to make the adaptation more effective. Tests show their method performs better than previous ones in both zero-annotation and few-annotation settings.
Cone-Beam Computed Tomography (CBCT)Computed Tomography (CT)Liver SegmentationUnsupervised Domain Adaptation (UDA)Margin Disparity Discrepancy (MDD)Interventional RadiologyDomain GapFew-Shot LearningImage Segmentation
Authors
Gauthier Miralles, Loïc Le Folgoc, Vincent Jugnon, Pietro Gori
Abstract
In interventional radiology, Cone-Beam Computed Tomography (CBCT) is a helpful imaging modality that provides guidance to practicians during minimally invasive procedures. CBCT differs from traditional Computed Tomography (CT) due to its limited reconstructed field of view, specific artefacts, and the intra-arterial administration of contrast medium. While CT benefits from abundant publicly available annotated datasets, interventional CBCT data remain scarce and largely unannotated, with existing datasets focused primarily on radiotherapy applications. To address this limitation, we leverage a proprietary collection of unannotated interventional CBCT scans in conjunction with annotated CT data, employing domain adaptation techniques to bridge the modality gap and enhance liver segmentation performance on CBCT. We propose a novel unsupervised domain adaptation (UDA) framework based on the formalism of Margin Disparity Discrepancy (MDD), which improves target domain performance through a reformulation of the original MDD optimization framework. Experimental results on CT and CBCT datasets for liver segmentation demonstrate that our method achieves state-of-the-art performance in UDA, as well as in the few-shot setting.