Manual Annotation, 3-D Shape Reconstruction,and Traumatic Brain Injury Analysis

摘要

Bitmask drawing is still the established standard for manual annotation of brain structures by experts. To alleviate problems such as bitmask inconsistencies between slices that lead to jagged contours in corresponding orthogonal cross-sections, we propose a 2-D spline-based contour editing tool in combination with a new algorithm for surface reconstruction from 3-D point clouds. This approach uses a new implicit surface formulation that adapts to the local density of points. We show that manual segmentation of the brainstem, cerebellum, corpus callo-sum, caudate, putamen, hippocampus and thalamus can be performed with high reproducibility in Magnetic Resonance (MR) data and sufficient accuracy to analyze volume changes for mild Traumatic Brain Injury (TBI) patients. In addition, we show that the new surface reconstruction method allows to reconstruct the shape of brain structures such as the brainstem better than other established surface reconstruction approaches. Our tool can, therefore, not only be used for volume measurements, but may also be used to assess local shape changes of brain structures going along with the progression of neuro-degenerative diseases such as TBI.