Active contour model for medical image segmentation.
Project summary
The purpose of this work is to learn active contour models (snakes) and to research the way they can be utilized for segmentation of medical images like MRI, CT-images. Then to implement the model that works out best for the task and to improve and develop the implementation in several ways. Snake is chosen due to two great advantages over other segmentation techniques. First, snake outputs an ordered set of connected vertices, thus making further analysis easier by far. Second, it not only accounts low-level image details (color, texture), but involves some high-level image knowledge (example: segments regions are salient) and works with regions as they are solid objects, eliminating punctures.
Participants
Results
Application, with following features was developed: implements series of snakes and various image-processing filters; provides user with performance counters to ease performance comparison of different snakes. Several improvements were made to the original snake to make it better in terms of performance and convergence for medical image segmentation. Parallel and vectorized GLSL snake was developed, that achieves 40x times better performance on GPU over CPU of the same generation. The application is a part of a big system to be developed. The system is solving the task of building complete and detailed 3D model of the brain out of 3D raster image, represented by the set of MRI-images, arranged by third coordinate.
