Evaluation of Automatic and Manual Segmentations in Neonate Subcortical Structures (2016)
Undergraduates: Kirsten Consing, Jeffrey Young Claudia Buss, Pathik Wadwha, Mehdi Sarkeshi, Cheryl Dietrich, Jerod Ramussen, Martin Styner
Faculty Advisor: Martin Styner
Department: Psychology & Neuroscience
The segmentation of the hippocampus and amygdala structures are crucial in understanding their association in neurodevelopmental disorders and neurodegenerative diseases. While automatic segmentations have been proposed, manual segmentation is still a common method in many neuroimaging studies particularly due to the lack of significant contrast in the regions of these subcortical structures. We evaluate the differences between automatic and manual segmentations in neonate subjects that uses an automatic segmentation setting of a multi-modality, multi-atlas approach through coefficient of variance (COV), intra-class and inter-class correlations, volumetric overlap and surface distance as well as shape analysis in the hippocampus. We found that within and across raters have reliable manual editing of the automatic segmentations. Also, the left hippocampus underwent a significantly higher amount of editing as compared to the right. Overall, of the structures show a significant asymmetric bias due to presentation and all automatic segmentations of the hippocampus and amygdala are appropriately reliable. In addition to that, we find that the both left and right amygdale and left and right hippocampus display an almost complete volumetric overlap. Shape analysis of the hippocampus show mainly regions of reduction as well as some regions of enlargement. Our results overall indicate that manual correction is beneficial but not necessary.