Manual and automated segmentation of the human hippocampus in cerebral magnetic resonance images

In vivo measurement of the human hippocampus in MR images is an important field of neuroimaging research. But current anatomical protocols underlying manual and automated segmentation procedures differ significantly and limit the comparability of research results. Here we aim at quantifying the differences in hippocampal morphometry due to differing anatomical definitions and describe consequences for hippocampus research. Based on these findings, we developed a protocol for segmentation of the hippocampus, termed anatomically-based hippocampal segmentation (ABHS). Tracing of the hippocampal borders was performed by a trained researcher on T1 weighted 3 T-MRI data of 20 patients with major depressive disorder and 17 age- and gender-matched healthy controls. Reliability of current segmentation protocols (Bartzokis, Jack, Watson) and comparability among them were assessed and compared with ABHS. The automated hippocampal segmentation implemented in FreeSurfer was compared to all manual protocols. Inter- and intra-rater-reliability of ABHS were high (Cronbachs Alpha: 0.976, Intra-class coefficient:0.976). Between ABHS and FreeSurfer, there was a good correlation (r = 0.382, p < 0.000 left; r = 0.382, p = 0.020 right) and the smallest percental volume difference (pdiff left = 1.77 (16.84); pdiff right = 2.48 (13.61)) in comparison to the other manual protocols. In conclusion, ABHS represents a valid hippocampal definition as a basis for manual as well as automated segmentation procedures.