Print Email Facebook Twitter Automatic Left Atrial Wall Segmentation from Space MRI via Advanced Two-Layer Level Sets with Distance Constraints Title Automatic Left Atrial Wall Segmentation from Space MRI via Advanced Two-Layer Level Sets with Distance Constraints Author Ji, Yuanbo (TU Delft Electrical Engineering, Mathematics and Computer Science) Contributor Lelieveldt, Boudewijn (mentor) van der Geest, R.J. (mentor) Tao, Qian (mentor) van Gemert, Jan (graduation committee) Degree granting institution Delft University of Technology Programme Electrical Engineering Date 2017-08-17 Abstract Segmentation of medical images is always a challenge due to complicated anatomical structures and poor image quality. In this paper, aiming to solve dual surfaces segmentation problems, we propose a two-layer levelset method with an efficient distance constraint. With the distance constraint, physical information of wall thickness can be imposed. In our method, a soft distance constraint is designed to promise that unclear boundaries in locations with poor image quality can be detected. Due to our distance constraint, evolution speed of level set function around boundaries can be limited effectively, so that it is much easier for unclear boundaries to be detected. Our method is applied to segment MRI CINE left ventricle (LV) data and MRI SPACE left atrium (LA) data. The accuracy of our method is analyzed by quantitative evaluation methods, which reports that Endocardial average perpendicular distance (APD) and Epicardial APD for LA and LV data are both below or around 1 mm, and the dice similarity coefficients (DSC) are all around 0.95. The evaluation results are within a desirable range. Our method has an excellent performance in both segmenting LA and LV data. Even in locations with poor image quality, our method still performs very well, which proves that our method is flexible and robust. Subject SegmentationTwo-layer levelsetDistance constraints To reference this document use: http://resolver.tudelft.nl/uuid:ccb9696f-9f6d-4894-b579-34a7fcaaa157 Part of collection Student theses Document type master thesis Rights © 2017 Yuanbo Ji Files PDF Thesis_YuanboJi_4475402.pdf 5.57 MB Close viewer /islandora/object/uuid:ccb9696f-9f6d-4894-b579-34a7fcaaa157/datastream/OBJ/view