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

• DOI: 10.48550/arxiv.2207.06799
• Publication Date: 2022-01-01
• Authors: Qi Zhao; Shuchang Lyu; Wenpei Bai; Linghan Cai; Binghao Liu; Meijing Wu; Xiu‐Bo Sang; Min Yang; Lijiang Chen
• Keywords: Modality (human–computer interaction)

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