Atlas-Based Segmentation Pipelines on 3D Brain MR Images: A Preliminary Study

Abstract

Three dimensional structural MR imaging is a high-resolution imaging technique used in the
detection and follow up of neurological disorders. Rigid changes in the brain are usually interpreted and
reported manually by radiologists using MR images. The results of manual interpretation may vary with
respect to the experts. At the same time, measurement and segmentation of the brain regions and the
manual evaluation of the volume changes are a difficult process. With the increase of numerical
methods, automated and semi-automated package programs have been developed for the analysis of
brain measurements. These programs use electronic brain atlases or tissue probability maps. However,
since the package programs have a lot of analysis time and give only certain outputs, they may be
disadvantaged in the use of segmentation and measurement of brain regions. Hence, special pipelines
are needed especially to obtain valuable features for artificial intelligence and classification studies. In
this study, we propose pipelines to segment 3D certain brain regions, which will help to find the basic
features such as volume changes, intensity variations, symmetry deteriorations, and tissue changes. With
these pipelines, 3D segmentation of the brain regions defined in the atlas can be performed and
normalized. It is aimed to use these studies as a preliminary study in order to quantitatively determine
the basic changes in the brain by performing the volume of interest methods and to formulate a decision
support system.