MMOTU: A Multi-Modality Ovarian Tumor Ultrasound Image Dataset for Unsupervised Cross-Domain Semantic Segmentation

TL;DR

This work introduces MMOTU, a multi-modality ovarian tumor ultrasound dataset combining 2d ultrasound and CEUS images with pixel-wise and global annotations to support unsupervised cross-domain semantic segmentation. It proposes DS2Net, a feature-alignment framework with dual encoders and two domain-selected modules (DDSM and DUSM) to bridge domain gaps between modalities, aided by adversarial training. Extensive experiments show DS2Net outperforms existing UDA methods for bidirectional cross-domain segmentation, with strong ablations and generalization to CT/MR datasets. The dataset and code enable further research on cross-modality ovarian tumor analysis and broader cross-domain medical image segmentation tasks.

Abstract

Ovarian cancer is one of the most harmful gynecological diseases. Detecting ovarian tumors in early stage with computer-aided techniques can efficiently decrease the mortality rate. With the improvement of medical treatment standard, ultrasound images are widely applied in clinical treatment. However, recent notable methods mainly focus on single-modality ultrasound ovarian tumor segmentation or recognition, which means there still lacks researches on exploring the representation capability of multi-modality ultrasound ovarian tumor images. To solve this problem, we propose a Multi-Modality Ovarian Tumor Ultrasound (MMOTU) image dataset containing 1469 2d ultrasound images and 170 contrast enhanced ultrasonography (CEUS) images with pixel-wise and global-wise annotations. Based on MMOTU, we mainly focus on unsupervised cross-domain semantic segmentation task. To solve the domain shift problem, we propose a feature alignment based architecture named Dual-Scheme Domain-Selected Network (DS2Net). Specifically, we first design source-encoder and target-encoder to extract two-style features of source and target images. Then, we propose Domain-Distinct Selected Module (DDSM) and Domain-Universal Selected Module (DUSM) to extract the distinct and universal features in two styles (source-style or target-style). Finally, we fuse these two kinds of features and feed them into the source-decoder and target-decoder to generate final predictions. Extensive comparison experiments and analysis on MMOTU image dataset show that DS2Net can boost the segmentation performance for bidirectional cross-domain adaptation of 2d ultrasound images and CEUS images. Our proposed dataset and code are all available at https://github.com/cv516Buaa/MMOTU_DS2Net.

Figures (12)

• Figure 1: Task description on MMOTU image dataset. Task1: single-modality semantic segmentation. Task2: bidirectional unsupervised domain adaptation between OTU_2d and OTU_CEUS for semantic segmentation. Task3: single-modality image recognition. Here, Task2 is our main research focus. Task1 is the prior research of Task2. Task3 is an independent task, which is also meaningful in clinical treatment.
• Figure 2: The number of samples containing in each category.
• Figure 3: Typical samples in MMOTU image dataset. Images in first and second row are respectively 2d ultrasound and CEUS image samples.
• Figure 4: The scatter plot showing the distribution of image scale.
• Figure 5: The diagram of recent notable architectures on semantic segmentation. top-left: CNN-based "Encoder-Decoder", top-right: Transformer-based "Encoder-Decoder", bottom-left: U-shape networks, bottom-right: Spatial-context based two-branch networks.