2D and 3D multiphase active contours without edges based algorithms for simultaneous segmentation of retinal layers from OCT images

iv Acknowledgements I would like to thank my supervisor, Dr. Grady Rylander, for his guidance and support. I would also like to thank Bingqing Wang for his help throughout my research. Thanks as well to Dr. Mia Markey for serving as a reader for this thesis. A special thanks to my family and friends for their support throughout my college experience. Glaucoma is a common disease that is difficult to diagnosis early using only visual field tests. Current research indicates that determination of retinal nerve fiber layer thickness (RNFLT) can serve as an early indicator of glaucoma [1]. RNFLT is measured by segmenting non-invasive optical coherence tomography images. However, high speckle noise and presence of artifacts in the images cause traditional layer detection and segmentation methods to fail. Multiphase active contours segmentation methods utilize region intensity and shape terms to produce multiple continuous boundaries simultaneously in noisy environments. A 2D and 3D multiphase active contours based algorithm was created to segment synthetic and real human retina OCT images. The 2D multiphase algorithm segmented eight simultaneous layers with a 3.14% mean A-scan error rate per layer. The 3D approach performed qualitatively accurate segmentation of a 20 image stack vi simultaneously. In an artificial, high-noise image stack the incorporation of more pixels per layer allowed improved segmentation using the 3D algorithm over the 2D. These results indicate that 2D and 3D multiphase active contours algorithms can be used to accurately segment retina layers. With further development to reduce computation time and automate initialization, these algorithms could be used to provide close to real-time clinical retinal image segmentation.