Retinal blood vessel segmentation: methods and implementations
Since the retinal blood vessel has been acknowledged as an indispensable element in both ophthalmological and cardiovascular disease diagnosis, the accurate segmentation of the retinal vessel tree has become the prerequisite step for automatic or computer-aided diagnosis systems. This thesis, therefore, has investigated different works of image segmentation algorithms and techniques, including unsupervised and supervised methods. Further, the thesis has developed and implemented two systems of the accurate retinal vessel segmentation. The methodologies explained and analyzed in this thesis, have been selected as the most efficient approaches to achieve higher precision, better robustness, and faster execution speed, to meet the strict standard of the modern medical imaging. Based on the intensive investigation and experiments, this thesis has proposed two outstanding implementations of the retinal blood vessel segmentation. The first implementation focuses on the fast, accurate and robust extraction of the retinal vessels using unsupervised techniques, by applying morphology-based global thresholding to draw the retinal venule structure and centerline detection to extract the capillaries. Besides, this system has been designed to minimize the computing complexity and to process multiple independent procedures in parallel. The second proposed system has especially focused on robustness and accuracy in regardless of execution time. This method has utilized the full convolutional neural network trained from a pre-trained semantic segmentation model, which is also called the transfer deep learning. This proposed method has simplified the typical retinal vessel segmentation problem from full-size image segmentation to regional vessel element recognition. Both of the implementations have outperformed their related works and have presented a remarkable scientific value for future computer-aided diagnosis applications. What’s more, this thesis is also a research guide which provide readers with the comprehensive knowledge on how to research on the task of retinal vessel segmentation.
Retinal blood vessel segmentation, Retinal images, Deep learning, Fully convolutional network, Transfer learning, Pre-trained model, Morphological processing, DoOG filter, Automated analysis, Thresholding
Master of Science (M.Sc.)
Electrical and Computer Engineering