Machine Learning for Quantification of Small Vessel Disease Imaging Biomarkers

This thesis is devoted to developing fully automated methods for quantification of small vessel disease imaging bio-markers, namely WMHs and lacunes, using vari- ous machine learning/deep learning and computer vision techniques. The rest of the thesis is organized as follows: Chapter 2 describes a conventional machine learning method for automated detection of WMHs. It should be noted that this method is optimized to detect WMHs of all size, including small lesions which are much more difficult to spot, rather than accurately delineating the WMH boundaries. Chap- ter 3 describes a customized deep learning method for automated segmentation of WMHs. In Chapter 4, we develop and experiment with a biologically inspired sam- pling method combined with deep neural networks. Chapter 5 is devoted for delv- ing deep into transfer learning of the trained deep networks on different domains for the WMH segmentation task. Finally, in Chapter 6, we describe a two-stage deep learning method for detection of lacunes.