Iglesias, J. E. 1, 2 , Van Leemput, K. 2, 3, 4 , Augustinack, J. 2 , Fischl, B. 2 , Lerma-Usabiaga, G. 1 , Paz-Alonso, P. 1 & Carreiras, M. 1
1 BCBL
2 Massachusetts General Hospital - Harvard Medical School
3 Aalto University, Finland
4 Technical University of Denmark
Automated analysis of brain MRI at the substructure level requires computational atlases built at a higher resolution than those that are typically used in current neuroimaging studies. Here we present a method to construct a computational atlas of the hippocampal subfields and amygdaloid nuclei from heterogeneous ex vivo and in vivo MRI data.
The ex vivo data consist of autopsy samples of 16 human hippocampi and 6 amygdalae. The samples were scanned at 0.13 mm isotropic resolution (on average) using customized hardware. The hippocampal samples were manually segmented into 13 different subfields, whereas the amygdaloid samples were divided into 11 nuclei. The in vivo data consist of 39 T1-weighted MRI scans of the whole brain (1 mm resolution). These scans were manually segmented into 36 structures, and provide the computational atlas with contextual information about the hippocampus and amygdala that is (physically) not present in the ex vivo data.
The manual labels from the in vivo and ex vivo data were combined into a single statistical atlas using Bayesian inference. The atlas can be used to automatically segment the hippocampal subfields and amygdaloid nuclei in MRI scans acquired with any MRI contrast and resolution. We will show segmentation results on heterogeneous MRI data, and also present ongoing work on histology and ex vivo MRI to add the thalamus to the atlas.