Open Collections

Real-time interactive retinal vessel segmentation and analysis

University of British Columbia

Vessel analysis is important for a wide range of clinical diagnoses and disease research such as diabetes and malignant brain tumours. Vessel segmentation is a crucial first step in such analysis but is often complicated by structural diversity and pathology. Existing automated techniques have mixed results and difficulties with non-idealities such as imaging artifacts, tiny vessel structures, and regions with bifurcations. Live-Vessel is a novel and intuitive semi-automatic vessel segmentation technique that extends the classic Live-Wire technique from user-guided contours to user guided paths along vessel centre-lines with automated boundary detection. Live-Vessel achieves this by globally optimizing vessel filter responses over both spatial (x,y) and non-spatial (radius) variables simultaneously. In this thesis I provide three main contributions. First, I bring Live-Vessel into the domain of real-time interactivity. Second, I enhance the objective function for improved contrast and graph search performance by incorporating colour image information, adding penalty terms, utilizing a smaller data type, and increasing the contrast between desirable and undesirable paths. Third, I gather and retain vessel connectivity information and provide post-segmentation analysis tools. I validated this technique using real medical data from the DRIVE, STARE, and REVIEW retina vessel databases. Quantitative results show that, on average, Live-Vessel resulted in an 7.28 times reduction in overall manual segmentation task time at a 95% accuracy level with most radial and medial errors being under 1 pixel in distance.

Text

2011-01-10

Attribution-NonCommercial-ShareAlike 2.5 Canada

http://hdl.handle.net/2429/30550

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-sa/2.5/ca/