Data-driven histological mapping of the human amygdala.

(A) The amygdala was segmented from the 100-micron resolution BigBrain dataset using an existing subcortical parcellation (45). Slice orientation of subpanels containing amygdala images is consistent across all panels in this figure. (B) Leveraging the pyRadiomics package v3.0.1 (47), we built a multiscale histological feature bank of the amygdala capturing fine-to-coarse intensity variations within this structure. Feature values were all normalized to better visualize relative intensity differences. (C) Matrix representation of the normalized feature bank shown in A. (D) We applied UMAP to this feature bank to derive a low dimensional embedding of amygdala cytoarchitecture, defining a 2-dimensional coordinate space (scatter plot, middle). Colors of the scatter plot represent proximity to axis limits. (E) Reordering the feature bank according to each eigenvector (U1 and U2) highlights the underlying variance in each feature captured by UMAP. (F) Coloring each amygdala voxel according to its corresponding location in the UMAP embedding space partially recovered its anatomical organization. (G) U1 and U2 were correlated to the three spatial axes and variogram matching tests assessed the statistical significance of each correlation. (H) Coloring the embedding space with openly available probabilistic map labels of the three main amygdala subregions showed that this region’s microstructural architecture could be recovered by UMAP. (I) Ridge plots of the probability values per subregion also illustrate a characterization of the subregions in U1.

Translating amygdala histological space to in vivo, ultra high-resolution, myelin-sensitive MRI.

(A) We segmented the left and right amygdalae of individual subjects from quantitative T1 (qT1) scans, and applied the same framework as developed in post mortem imaging to derive subject-specific, in vivo representations of amygdala microstructure. (B) Correlation values between the UMAP components (U1 and U2) and the three coordinate axes of the 10 MRI subjects were computed in MNI152 space and then contrasted with the correlation values found in the histological data (BigBrain transformed to MNI152 space).

Functional network mapping of amygdala microstructural subregions.

(A) We isolated the rsfMRI timeseries of two amygdala subregions, defined from subject-specific U1 topography, as well as the whole amygdala. (B) We computed the functional connectivity of both amygdala subregions, and project resulting correlations to the cortex. We further demonstrate the differences in connectivity patterns between both subregions (t-value) and highlight the regions with significant differences (pFWE<0.05). (C) Left: The activation patterns illustrated in (B, top) were averaged within intrinsic functional communities defined by Yeo, Krienen, et al. (2011). Right: Meta-analytic decoding of functional connectivity patterns of both amygdala subregions and the whole amygdala dissociated cognitive and affective functional affiliations of this region.