A probabilistic atlas of the human thalamic nuclei combining ex vivo MRI and histology

Abstract

The human thalamus is a brain structure that comprises numerous, highly specific nuclei. Since these nuclei are known to have different functions and to be connected to different areas of the cerebral cortex, it is of great interest for the neuroimaging community to study their volume, shape and connectivity in vivo with MRI. In this study, we present a probabilistic atlas of the thalamic nuclei built using ex vivo brain MRI scans and histological data, as well as the application of the atlas to in vivo MRI segmentation. The atlas was built using manual delineation of 26 thalamic nuclei on the serial histology of 12 whole thalami from six autopsy samples, combined with manual segmentations of the whole thalamus and surrounding structures (caudate, putamen, hippocampus, etc.) made on in vivo brain MR data from 39 subjects. The 3D structure of the histological data and corresponding manual segmentations was recovered using the ex vivo MRI as reference frame, and stacks of blockface photographs acquired during the sectioning as intermediate target. The atlas, which was encoded as an adaptive tetrahedral mesh, shows a good agreement with previous histological studies of the thalamus in terms of volumes of representative nuclei. When applied to segmentation of in vivo scans using Bayesian inference, the atlas shows excellent test-retest reliability, robustness to changes in input MRI contrast, and ability to detect differential thalamic effects in subjects with Alzheimer's disease. The probabilistic atlas and companion segmentation tool are publicly available as part of the neuroimaging package FreeSurfer.

Keywords: Atlasing; Bayesian inference; Ex-vivo MRI; Histology; Segmentation; Thalamus.

Graphical abstract

Fig. 1

(a) Sample sagittal slice of ex vivo MRI scan of case NHL8_14. (b) Corrected for bias field and slab boundary artifacts. (c) Close-up of left thalami in coronal view, uncorrected. (d) Corrected version of (c).

Fig. 2

Histological analysis: (a) Sample blockface photograph of case HNL4_13. (b) Homography (i.e., perspective). corrected photograph. (c) Automated segmentation with random forest classifier. (d) Digitized histology of section corresponding to (a). (e) Close-up of the region inside the red square in (d).

Fig. 3

Rigid alignment of stacks of blockface photographs to MRI. (a) Initialization: axial view across the thalamus. (b) MRI. (c) Aligned stacks. We have overlaid a grid on subfigures (b) and (c) to ease the visual assessment of the quality of the registration.

Fig. 4

Nonrigid registration of histology and MRI. (a) Sample histological section, same as in Fig. 2. (b) Corresponding blockface photograph, masked by the corresponding automated segmentation. (c) MRI scan resampled to the space of the photograph. (d) Registered histology. The manually placed landmarks are marked with red circles in subfigures (a) and (c).

Fig. 5

Examples of reconstructed histology. (a) Axial view of thalamus in MRI. (b) Corresponding slice through the stack of reconstructed histology. (c) Sagittal slice of thalamus in MRI. (d) Corresponding histology.

Fig. 6

Filling the gaps between blocks and refining the segmentation. (a) Sample sagittal slice of the thalamus. (b) Propagated manual segmentations. (c) Labels estimated by minimizing Equation (1). The color coding is the same as in Table 2.

Fig. 7

Probabilistic atlas, with tetrahedral mesh superimposed. The color of each voxel is a linear combination of the colors in Table 2, weighted by the corresponding label probabilities. For the surrounding structures, we used the standard FreeSurfer color map. (a) Sample coronal slice. (b) Axial slice. (c) Sagittal slice.

Fig. 8

Violin plots (box plots with a rotated kernel density plot on each side) for volumes of representative nuclei, computed with our proposed atlas and with Krauth's. MD, VL and PU represent the whole mediodorsal, ventral lateral and pulvinar nuclei, i.e., MD is the union of MDm and MDl; VL is the union of VLa and VLp; and PU is the union of PuA, PuM, PuL and PuI.

Fig. 9

Sagittal (left), coronal (middle), and axial (right) slices of a brain MRI scan segmented with the proposed atlas and Krauth's. The color map for our atlas is that described in Table 2. For Krauth, we attempted to match the color map as much as possible; we note that their atlas includes the red nucleus (in red, pointed by yellow arrows) and subthalamic nucleus (in green, pointed by blue arrows), in the inferior region, which are not part of our proposed atlas. On the left thalamus of the input scan, we have overlaid a manually delineated boundary between the mediodorsal and pulvinar nuclei, and the rest of the nuclei; this is the main feature we use to fit the internal boundaries of the thalamus.

Fig. 10

Coronal slice from sample subject in multimodal MRI dataset, with corresponding automated segmentations with the proposed atlas.

Fig. 11

Dice overlap (left-right averaged) for the segmentations yielded by different MRI contrasts. The six matrices correspond to the six representative thalamic nuclei and whole thalamus. As in previous figures, MD is the union of MDm and MDl; VL is the union of VLa and VLp; and PU is the union of PuA, PuM, PuL and PuI. The times in ms refer to the synthetic MPRAGE scans. The color bar is the same for all six matrices.

Fig. 12

ROC curves for Alzheimer's vs. controls classification based on left-right averaged thalamic volumes. The red curve is for the whole thalamic volume estimated by FreeSurfer's recon-all stream (AUC = 0.632). The blue curve is for the whole thalamic volume, estimated as the sum of the volumes of the nuclei given by our proposed atlas (AUC = 0.645). The black curve corresponds to a leave-one-out LDA classifier that simultaneously considers the volumes of all nuclei, as estimated by the proposed atlas (AUC = 0.830). The volumes were left-right averaged and corrected by age and intracranial volumes in all cases.