Ossama Mahmoud, September 2018--April 2019, "Image processing algorithm for assessing blood perfusion in microscopic animal blood vessel videos", Computer Science Department, University of Western Ontario, Canada.
B.Sc. Thesis Abstract
Context: Studying microcirculation gives scientists vital information regarding disease progression. However, the process of analyzing the microcirculation is tedious and time consuming. Manual analysis involves studying videos of microcirculation to determine if blood is flowing across a subset of blood vessels. Currently, no robust tools or algorithms exist to automate this analysis.
Problem: Currently, no robust tools or algorithms exist to automate this analysis. To analyze the microcirculation, the algorithm must first detect all vessel structures in the video. Then the algorithm must determine if each vessel structure is perfused or not
Principle ideas: Various approaches were examined, then an algorithm utilizing the best approaches is used. To detect blood vessels in videos, we aggregate video frames to one image by calculating the sum of all differences, then we apply a modified Sager Curvilinear Vessel Detector. Then to determine if blood is flowing through each vessel, we use a Convoluted Neural Net. The results of the algorithm show the accuracy of the proposed algorithm to be 77%.
Contribution: The proposed algorithm is an automated accurate tool to analyze the microcirculation, and in practice can save researchers and clinicians valuable time. More so, being one of the first studies to model if a vessel is perfused or not, this study sets the stage for future work in a relatively stagnant field of research.