Semester Master Project
Life Science, EPFL
This project addresses the problem of brain segmentation in 3D MRI images without using an atlas. This is a complex problem because the acquired images normally represent the entire head, thereby including many non-brain tissues. Two additional conditions are that the algorithm should be automated and the time to execute it should be short.
We have developed a pipeline matching all the required conditions through efficient combination of a surface detection and segmentation model. We derived a matched surface detector by means of a 3D steerable filter and show how recently developed 3D spline surface snakes can be used for brain segmentation. For this purpose we present a novel snake energy term for the fast optimization of the spline snake, as well as an entire implementation strategy that matches our problem. To ensure additional robustness of the algorithm several preprocesing steps are also implemented in the pipeline.
The pipeline is developed to use 3D MP2RAGE T1-weighted MRI images and specifically exploits its advantages over traditional MRI protocols. We have validated the proposed algorithm against state-of-the-art brain segmentation software that are in clinical use and show that it is competitive with them. The proposed model is implemented as a plugin that can be executed in an automated or interactive manner.