Retinal Vessel Segmentation Using Frangi Filter and U-Net

Can a light neural network, trained on a laptop CPU, map the blood vessels of the retina as well as a human expert? This project segments the vessels in real fundus photographs of adults (STARE) and children (CHASE_DB1) with the classical Frangi vesselness filter and Otsu threshold, a Frangi filter and a matched filter tuned on training images, a pixel classifier on 17 filter features and a light U-Net, and evaluates every method inside the camera's field of view against manual labels and against a second human observer. It tests what happens on a new camera and population, measures vessel width, tortuosity and branch points as clinical outputs, and offers a web application with overlays for graders. ImageJ and GNU Octave cross-check the classical part; a complete MATLAB version is included.

Category: Biomedical Engineering

Biomedical Engineering Project BM01 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.