SDC-UDA: Volumetric Unsupervised Domain Adaptation Framework for Slice-Direction Continuous Cross-Modality Medical Image Segmentation

Recent advances in deep learning-based medical image segmentation studies achieve nearly human-level performance in fully supervised manner. However, acquiring pixel-level expert annotations is extremely expensive and laborious in medical imaging fields. Unsupervised domain adaptation (UDA) can alleviate this problem, which makes it possible to use annotated data in one imaging modality to train a network that can successfully perform segmentation on target imaging modality with no labels. In this work, we propose SDC-UDA, a simple yet effective volumetric UDA framework for Slice-Direction Continuous cross-modality medical image segmentation which combines intra- and inter-slice self-attentive image translation, uncertainty-constrained pseudo-label refinement, and volumetric self-training. Our method is distinguished from previous methods on UDA for medical image segmentation in that it can obtain continuous segmentation in the slice direction, thereby ensuring higher accuracy and potential in clinical practice. We validate SDC-UDA with multiple publicly available cross-modality medical image segmentation datasets and achieve state-of-the-art segmentation performance, not to mention the superior slice-direction continuity of prediction compared to previous studies.

1. Introduction

With the surprising development of deep learning (DL), many studies are now showing remarkable performance in various applications [8], [16], [20]. However, when a DL model faces data from an unseen domain, performance degradation occurs [9], [32]. Resolving this issue is important for the DL techniques to be applied in real world since collecting data from all domains and labeling them is very impractical and inefficient. Unsupervised Domain Adaptation (UDA) aims to alleviate this problem by adapting a model trained on source domain data to target domain, without the necessity of supervision in the target domain. Data dependency is more serious in medical image segmentation field since acquiring pixel-level expert annotation is extremely expensive and time-consuming [3], [22], [37]. Figure 1.

An illustration that describes the comparison between our proposed method with previous methods. (A) Previous UDA for medical image segmentation studies mostly utilize 2D UDA, which leads to inconsistent predictions in the slice direction when the predictions are stacked. (B) The proposed framework (SDC-UDA) considers volumetric information in the translation and segmentation process, respectively, which leads to improved slice-direction continuity of segmentation that is much practical for clinical use.