A novel approach for left ventricle segmentation in tagged MRI. (October 2021)

• Record Type: Journal Article
• Authors: Zou, Xijing; Wang, Qian; Luo, Ting
• Is Part Of: Computers & electrical engineering. Volume 95(2021)
• Publication Date: 2021-10
• Subjects: Left ventricle segmentation -- SinMod -- U-Net network -- Curriculum learning; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.107416
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: Motion information tracking is used for automatic location of the heart. U-net and curriculum learning are combined to segment the left ventricle. Curriculum strategies from easy to difficult can capture more detailed TMRI images features. These strategies showed significant performance on the endocardium segmentation. Abstract: Automatic left ventricle (LV) segmentation from tagged cardiac magnetic resonance imaging is significant for evaluating heart function and providing follow-up treatments in clinical medicine. However, due to the complicated cardiac structure and extra interference, it is challenging for traditional methods to delineate the LV automatically and get accurate results. Therefore, we proposed the automatic LV segmentation algorithm combined with deep learning and curriculum learning strategy. The key technologies are described as follows: firstly, local sine-wave modeling (SinMod) is practiced to track cardiac motion information, implement automatic heart location and obtain the region of interest. Secondly, U-Net is utilized as the basic model to segment the LV endocardium and epicardium. Additionally, a new curriculum learning training strategy is adopted to improve segmentation accuracy. Finally, comparative results demonstrate the superior performance of our approach to those resulting from traditional methods. Graphical abstract: Image, graphical abstract