Thalamic nuclei atrophy at high and heterogenous rates during cognitively unimpaired human aging

Abstract

The thalamus is a central integration structure in the brain, receiving and distributing information among the cerebral cortex, subcortical structures, and the peripheral nervous system. Prior studies clearly show that the thalamus atrophies in cognitively unimpaired aging. However, the thalamus is comprised of multiple nuclei involved in a wide range of functions, and the age-related atrophy of individual thalamic nuclei remains unknown. Using a recently developed automated method of identifying thalamic nuclei (3T or 7T MRI with white-matter-nulled MPRAGE contrast and THOMAS segmentation) and a cross-sectional design, we evaluated the age-related atrophy rate for 10 thalamic nuclei (AV, CM, VA, VLA, VLP, VPL, pulvinar, LGN, MGN, MD) and an epithalamic nucleus (habenula). We also used T1-weighted images with the FreeSurfer SAMSEG segmentation method to identify and measure age-related atrophy for 11 extra-thalamic structures (cerebral cortex, cerebral white matter, cerebellar cortex, cerebellar white matter, amygdala, hippocampus, caudate, putamen, nucleus accumbens, pallidum, and lateral ventricle). In 198 cognitively unimpaired participants with ages spanning 20–88 years, we found that the whole thalamus atrophied at a rate of 0.45% per year, and that thalamic nuclei had widely varying age-related atrophy rates, ranging from 0.06% to 1.18% per year. A functional grouping analysis revealed that the thalamic nuclei involved in cognitive (AV, MD; 0.53% atrophy per year), visual (LGN, pulvinar; 0.62% atrophy per year), and auditory/vestibular (MGN; 0.64% atrophy per year) functions atrophied at significantly higher rates than those involved in motor (VA, VLA, VLP, and CM; 0.37% atrophy per year) and somatosensory (VPL; 0.32% atrophy per year) functions. A proximity-to-CSF analysis showed that the group of thalamic nuclei situated immediately adjacent to CSF atrophied at a significantly greater atrophy rate (0.59% atrophy per year) than that of the group of nuclei located farther from CSF (0.36% atrophy per year), supporting a growing hypothesis that CSF-mediated factors contribute to neurodegeneration. We did not find any significant hemispheric differences in these rates of change for thalamic nuclei. Only the CM thalamic nucleus showed a sex-specific difference in atrophy rates, atrophying at a greater rate in male versus female participants. Roughly half of the thalamic nuclei showed greater atrophy than all extra-thalamic structures examined (0% to 0.54% per year). These results show the value of white-matter-nulled MPRAGE imaging and THOMAS segmentation for measuring distinct thalamic nuclei and for characterizing the high and heterogeneous atrophy rates of the thalamus and its nuclei across the adult lifespan. Collectively, these methods and results advance our understanding of the role of thalamic substructures in neurocognitive and disease-related changes that occur with aging.

1. Introduction

Atrophy of brain tissue is a classic hallmark of aging. Brain atrophy reflects the shrinkage and loss of neurons and the connections between them (Fjell and Walhovd, 2010). While brain atrophy is a prominent feature in many brain diseases, atrophy can also be seen both broadly and focally in the cognitively unimpaired aging brain (Fjell and Walhovd, 2010). Characterizing brain structural changes during cognitively unimpaired aging will inform efforts to understand the differences between normal and disease-related atrophy, the structural changes that can be tolerated before they significantly impact function, and the resiliency of the cognitively unimpaired aging brain.

Within the brain, the thalamus is a central integration structure, receiving and distributing information among the cerebral cortex, subcortical structures, and the peripheral nervous system via the brainstem (Steriade et al., 1997). This high degree of integration is accomplished in part by thalamic projection neurons that send long-range outputs with collateral projections and large ramifications at their targets (Steriade et al., 1997). This processing is organized by intra-thalamic nuclei dedicated to distinct functions, including cognitive, visual, auditory, vestibular, and sensorimotor functions (Steriade et al., 1997). Lesions of thalamic nuclei lead to profound disruptions in the modulation of behaviors subserved by these functional domains (Carrera and Bogousslavsky, 2006). As a highly connected and integrative structure in the brain, the thalamus is in a critical but also vulnerable position that may show prominent structural changes resulting from age-related changes across the brain, such as degeneration due to reduced connectivity or accumulated oxidative stress from high levels of neuronal activity.

Current knowledge of age-related thalamic volumetric changes in humans comes primarily from neuroimaging studies which show that the thalamus as a whole significantly atrophies over the cognitively unimpaired adult lifespan (Cherubini et al., 2009; Fjell et al., 2013; Goodro et al., 2012; Hughes et al., 2012; Long et al., 2012; Murphy et al., 1996, 1993; Narvacan et al., 2017; Nyberg et al., 2010; Raji et al., 2009; Schippling et al., 2017; Squarzoni et al., 2018; Su et al., 2012; Takahashi et al., 2011; Tullo et al., 2019; Van Der Werf et al., 2001; Walhovd et al., 2005, 2011; Yoo et al., 2016; Zheng et al., 2019). Postmortem histological studies afford a higher resolution characterization of the volumes of thalamic nuclei. However, while there are a number of postmortem human histological studies showing that thalamic nuclear volumes and neuronal counts are reduced in disease (Byne et al., 2002; Popken et al., 2000; Young et al., 2000), to our knowledge, there are no such studies of cognitively unimpaired human aging. Thus, as yet, we do not know the rates of volumetric change of specific thalamic nuclei over the cognitively unimpaired adult lifespan (Fama and Sullivan, 2015). Such studies would identify whether age-related neurodegeneration is occurring heterogeneously within the thalamus, and therefore whether the concept of “regional vulnerability” (Small, 2014) applies to the thalamus.

Recently, our group developed an automated image segmentation method that non-invasively and accurately identifies 10 thalamic nuclei and two adjacent structures. This method uses 3T or 7T MRI with white-matter-nulled magnetization prepared rapid acquisition gradient echo (WMnMPRAGE) contrast, which provides higher contrast in the thalamus than conventional MPRAGE imaging in which the cerebrospinal fluid (CSF) is nulled (Saranathan et al., 2015; Tourdias et al., 2014). WMnMPRAGE image contrast is strong enough to manually identify individual thalamic nuclei and adjacent structures (Tourdias et al., 2014). Our automated thalamic segmentation method, known as THOMAS (THalamus Optimized Multi-Atlas Segmentation), was trained using manual delineations of these identifiable structures in a set of 20 training volumes to create a multi-atlas label-fusion segmentation tool that accurately identifies 10 thalamic nuclei and two adjacent structures (Su et al., 2019). THOMAS has been used to characterize the volumes of thalamic nuclei in cognitively unimpaired versus patient populations of multiple sclerosis (Planche et al., 2020), alcoholism and HIV (Zahr et al., 2020), and essential tremor (Su et al., 2020).

In the present study, our principal goal was to characterize the structural changes of thalamic nuclei over the adult lifespan in cognitively unimpaired participants using a cross-sectional approach. This was done by applying THOMAS to segment thalamic nuclei within WMnMPRAGE image volumes acquired from a large group of 198 cognitively unimpaired adult participants with ages ranging from 20 to 88 years. Our secondary goal was to compare the atrophy of thalamic nuclei with those of extra-thalamic regions, with the latter segmented using conventional T1-weighted images of the same participants and the Sequence Adaptive Multimodal SEGmentation (SAMSEG) tool within the FreeSurfer suite of neuroimaging analysis tools. We employed an exponential regression model to estimate atrophy rates (annualized percent change in volume) and differences in atrophy rates among thalamic nuclei and extra-thalamic structures.

2. Material and methods

2.1. Participants

198 Cognitively unimpaired adult participants (86 M, 112 F) were recruited, with a mean age of 59.4 years and an age range of 20 to 88 years at the time of imaging (Table 1). Older participants (>=60 years of age, n = 143, 66 M, 77 F) were recruited at Stanford University as part of the Stanford Aging and Memory Study (SAMS) (Trelle et al., 2020, 2021). Younger participants (<60 years of age) were recruited at Stanford University (n = 17, 9 M, 8 F) and at the University of Bordeaux (n = 38, 11 M, 27 F). Written informed consent was obtained for all participants and study procedures were approved by each institution’s Institutional Review Board. Participants were screened against neurological and psychiatric disease. Older participants were additionally screened to have a Clinical Dementia Rating (CDR) (Morris, 1993) score of zero and age-appropriate normal cognitive functioning. The latter was based on a neuropsychological assessment battery that examined multiple domains, including episodic memory, attention, executive function, visuospatial processing, and language. Participants were classified as cognitively unimpaired by a team of neurologists and neuropsychologists, based on test scores falling within 1.5 standard deviations of demographically adjusted means; for details, see Trelle et al. (2020).

Table 1.

Participant demographics.

	Bordeaux Younger Participants (< 60 years of age)	Stanford Younger Participants (<60 years of age)	Stanford Older Participants (>=60 years of age)	All Participants
Number of Participants	38	17	143	198
Sex (male, female)	11 M, 27 F	9 M, 8 F	66 M, 77 F	86 M, 112 F
Age (mean / standard deviation/range / range) (yrs)	33.1 / 10.4 / 20–56	37 / 11.4 / 24–58	69.1 / 6.1 / 60–88	59.4 / 17.4 / 20–88
Education Level (mean / standard deviation / range) (yrs)	21 / 0 / 21–21	20.2 / 1.8 / 16–21	16.7 / 2.0 / 10–20	17.8 / 2.6 / 10–21

2.2. Image acquisition

MRI parameters for image acquisition are summarized in Table 2. Stanford participants were scanned with a 3D T1-weighted inversion-prepared fast spoiled gradient echo protocol at 3T and a WMnMPRAGE protocol at 7T, using 32-channel head coils in both cases. Bordeaux participants were imaged with 3D T1-weighted inversion-prepared fast spoiled gradient echo and WMnMPRAGE sequence protocols, both acquired at 3T using a 32-channel head coil.

Table 2.

Image acquisition parameters.

	Bordeaux Younger Subjects				Stanford Younger Subjects		Stanford Older Subjects
	Tl-weighted Protocol 1	Tl-weighted Protocol 2	WMnMPRAGE Protocol 1	WMnMPRAGE Protocol 2	Tl-weighted	WMnMPRAGE	Tl-weighted	WMnMPRAGE
Number of Subjects	22	16	22	16	17	17	143	143
Scanner	GE MR750	Canon Galan	GE MR750	Canon Galan	GE MR750	GE MR950	GE MR750	GE MR950
Field Strength	3T	3T	3T	3T	3T	7T	3T	7T
Head Coil	32 channel (GE Healthcare)	32 channel (Canon)	32 channel (GE Healthcare)	32 channel (Canon)	32 channel (Nova Medical)	32 channel (Nova Medical)	32 channel (Nova Medical)	32 channel (Nova Medical)
Base Sequence	3D BRAVO	3D MPRAGE	3D MPRAGE	3D MPRAGE	3D MPRAGE	3D MPRAGE	3D BRAVO	3D MPRAGE
Voxel Size (mm)	1.0 × 1.0 × 1.0	0.6 × 0.6 × 0.6	1.0 × 1.0 × 1.0	0.8 × 0.8 × 0.8	0.8 × 0.8 × 0.8	1.0 × 1.0 × 1.0	0.9 × 0.9 × 0.9	1.0 × 1.0 × 1.0
TE (ms)	3.2	3	4.7	3.6	3.9	4.6	2.8	4.6
TR (ms)	8.5	7	11.1	7.8	8.5	10.1	7.3	10.1
TI (ms)	400	950	500	470	1200	680	450	680
TS (ms)	–	–	4500	–	3700	5500	–	5500
Flip Angle (degrees)	11	9	7	7	6	4	12	4
Receive Bandwidth (kHz)	+/− 31.25	+/− 51.3	+/− 11.9	+/− 53.4	+/− 19.2	+/− 11.9	+/− 41.67	+/− 11.9
Slice Number	180	234	220	200	250	220	186	220
Views Per Segment	–	–	240	240	–	240	–	240
Acceleration Factor	–	1.5	1.25×1.25	1.5	3	3	1.75	3
Slice Orientation	sagittal	sagittal	coronal	axial	coronal	coronal	sagittal	coronal
Scan Time (min)	10	4.5	7.2	11.2	4	11.1	4.8	11.1

T1-weighted images were acquired from Stanford older participants using the following scan parameters: 3T GE MR750 scanner, 3D BRAVO base sequence, sagittal acquisition, echo time (TE) 2.8 ms, repetition time (TR) 7.3 ms, inversion time (TI) 450 ms, flip angle (FA) 12°, receive bandwidth (RBW) +/−41.67 kHz, spatial resolution 0.9 mm isotropic, 186 slices per volume, ARC parallel imaging acceleration: 1.75, scan time 4.8 min.

T1-weighted images were acquired from Stanford younger participants using the following scan parameters: 3T GE MR750 scanner, 3D MPRAGE base sequence, coronal acquisition, TE 3.9 ms, TR 8.5 ms, TI 1200 ms, sequence time (time between inversion pulses: TS) 3700 ms, FA 6°, RBW +/−19.2 kHz, spatial resolution 0.8 mm isotropic, 250 slices per volume, ARC parallel imaging acceleration: 3, scan time 4 min.

T1-weighted images were acquired from the Bordeaux cohort using one of two scan protocols: 1) for 22 participants of this cohort: 3T GE MR750w scanner, 3D BRAVO base sequence, sagittal acquisition, TE 3.2 ms, TR 8.5 ms, TI 400 ms, FA 11°, RBW +/−31.25 kHz, spatial resolution 1 mm isotropic, 180 slices per volume, scan time 10 min; or 2) for 16 participants of this cohort: 3T Canon Galan scanner, 3D MPRAGE base sequence, sagittal acquisition, TE 3.0 ms, TR 7.0 ms, TI 950 ms, FA 9°, RBW +/− 51.3 kHz, spatial resolution 0.6 mm isotropic, 234 slices per volume, SPEEDER parallel imaging acceleration: 1.5; scan time 4.5 min.

WMnMPRAGE images were acquired from all Stanford participants using the following parameters: 7T GE MR950 scanner, 3D MPRAGE base sequence, coronal acquisition, TE 4.6 ms, TR 10.1 ms, TI 680 ms, TS 5500 ms, views per segment 240, FA: 4°, RBW +/−11.9 kHz, spatial resolution 1 mm isotropic, 220 slices per volume, k-space ordering; 2D radial fanbeam, scan time 11.1 min.

WMnMPRAGE images were acquired from the Bordeaux cohort using one of two scan protocols: (1) for 22 participants of this cohort: 3T GE MR750w scanner, MPRAGE base sequence, coronal acquisition, TE 4.7 ms, TR 11.1 ms, TI 500 ms, TS 4500 ms, views per segment 240, FA: 7°, RBW +/− 11.9 kHz, spatial resolution 1 mm isotropic, 220 slices per volume k-space ordering; 2D radial fanbeam, ARC parallel imaging acceleration: 1.25×1.25, scan time 7.2 min; or (2) for 16 participants of this cohort: 3T Canon Galan scanner, 3D MPRAGE base sequence, axial acquisition, TE 3.6 ms, TR 7.8 ms, TI 470 ms, TS 4500 ms, FA 7°, RBW +/− 53.4 kHz, spatial resolution 0.8 mm isotropic, 200 slices per volume, SPEEDER parallel imaging acceleration: 1.3 × 1.6; scan time 11.3 min. See Saranathan et al. (2015) for further details.

T1-weighted and WMnMPRAGE image volumes were visually inspected to ensure that scans were of sufficient quality for analysis and were not corrupted by motion artifact.

2.3. Segmentations

Whole-brain WMnMPRAGE volumes were processed with the THOMAS thalamic segmentation tool with no preprocessing. The THOMAS algorithm applied to WMnMPRAGE images has been validated against manual segmentation (Su et al., 2019). Using version v0 of this tool (https://github.com/sujason/thomas), we segmented and extracted the volumes of 12 lateralized structures in each hemisphere of the brain: whole thalamus, ten thalamic nuclei (anteroventral [AV], centromedian [CM], lateral geniculate nucleus [LGN], mediodorsal [MD], medial geniculate nucleus [MGN], pulvinar [Pul], ventral anterior [VA], ventral lateral anterior [VLA], ventral lateral posterior [VLP], and ventral posterolateral [VPL]), and one adjacent epithalamic structure, the habenula (Hb) (Fig. 1A). Note that THOMAS segments the whole thalamus separately from the thalamic nuclei; this whole thalamus encompasses all these preceding structures, as well as the mammillothalamic tract and some additional unlabeled thalamic areas (i.e., between segmented thalamic nuclei). See Tourdias et al. (2014), Saranathan et al. (2015), and Su et al. (2019) for additional details.

Fig. 1.

Segmentations in representative younger and older subjects. (A) Whole thalamus, thalamic nuclei, and habenula segmentations using THOMAS and white matter-nulled MPRAGE imaging. (B) Extra-thalamic segmentations using FreeSurfer’s SAMSEG tool. (C) Intracranial cavity segmentation using SPM12’s Segment and Tissue Volume Utility modules.

Whole-brain T1-weighted gradient echo volumes were processed with the FreeSurfer 7.1.1 pipeline with no preprocessing. Using the FreeSurfer tool SAMSEG (https://surfer.nmr.mgh.harvard.edu/fswiki/Samseg), we segmented and extracted the volumes of 11 structures on each side of the brain: nucleus accumbens, amygdala, caudate, cerebellar cortex, cerebellar white matter, cerebral cortex, cerebral white matter, hippocampus, lateral ventricle, pallidum, and putamen (Fig. 1B). SAMSEG produces volumes via both probabilistic and discrete voxel segmentation summations; we used the volumes calculated via the probabilistic method. The SAMSEG algorithm has been validated using T1-weighted images from different scanners and acquired with different scan parameters, using expert manual segmentation of these images as ground truth. SAMSEG was shown to have the highest average Dice score of the five segmentation methods studied (SAMSEG, FreeSurfer, PICSL MALF, BrainFuse, and majority voting) (Puonti et al., 2016). Note that SAMSEG produces a whole thalamus segmentation, but we used the THOMAS segmentation of the whole thalamus because it includes the LGN and MGN, two major thalamic nuclei, whereas the SAMSEG whole thalamus does not. In addition, the thalamic contrast is higher in WMnMPRAGE images than in conventional T1-weighted images; thus we expect the THOMAS segmentation of thalamic boundaries to be more accurate.

Intracranial volume (ICV) estimates were obtained from 3T whole-brain T1-weighted gradient echo volumes using SPM12’s Segment and Tissue Volume Utility modules (https://www.fil.ion.ucl.ac.uk/spm/software/spm12) (Malone et al., 2015) (Fig. 1C). The accuracy of all segmentations was checked visually for each subject.

2.4. Statistical analyses

For each structure in the left and right hemispheres, we employed an exponential regression model to quantify the association between brain structural volume and age, adjusting for education, cohort, sex, and ICV, yielding the rate of age-related atrophy. In particular, we estimated the volume of each structure as a function of age, years of education, cohort, sex, and ICV using the following model:

$$\begin{gathered} E(Y\mid age,education,cohort,sex,ICV)=\exp ({\beta }_0+{\beta }_{1}age+{\beta }_{2}education \\ +{\beta }_{3}cohort+{\beta }_{4}sex+{\gamma }_1\log ICV)=exp({\beta }_0+{\beta }_{1}age+{\beta }_{2}education \\ +{\beta }_{3}cohort+{\beta }_{4}sex)\times I{\text{CV}}^{{\gamma }_1}, \end{gathered}$$

which is equivalent to

$$\begin{array}{l}E\left(\frac{Y}{IC{V}^{{\gamma }_1}}\mid age,education,cohort,sex\right)\\ =exp\left({\beta }_0+{\beta }_{1}age+{\beta }_{2}education+{\beta }_{3}cohort+{\beta }_{4}sex\right),\end{array}$$

where Y is the volume; education is in years; cohort = 0 for Stanford and 1 for Bordeaux, sex = 0 for female and 1 for male; and ICV is the total intracranial volume. Note: our principal analysis did not include education and cohort as covariates, but we did examine the influence of including these variables on our primary conclusions, described in more detail in Results and Discussion sections. An advantage of this exponential regression model is that the structure’s relative (percent) volumetric change per year (defined as atrophy rate) is constant over the lifespan and represented by exp(β₁) − 1. A Wald test was performed to test against the null hypothesis that there is no association between volumetric change and age (i.e., H₀ : β₁ = 0). In the Wald test, the standard error of the estimated coefficient was assessed by nonparametric bootstrap. We compared the atrophy rates of two structures by estimating the difference in the β₁ coefficients from the regression models of the two structures. The standard error of the estimated difference was assessed by nonparametric bootstrap, accounting for correlations between the volumes of structures from the same individual. Significant differences between the atrophy rates of structures were assessed with paired t-tests, adjusted for multiple comparisons. Hemisphere- and sex-specific differences in atrophy rates were similarly assessed by comparing the corresponding regression coefficients of age.

For the analysis assessing the functional specificity of thalamic nuclei atrophy, we grouped thalamic nuclei into six broad functional groups, based on their known anatomical connectivity (Steriade et al., 1997); that is, based on the known functions of the brain regions connected to each nucleus. These broad functional groups are cognitive (left+right AV and MD), visual (left+right LGN and pulvinar), auditory and vestibular (left+right MGN), motor (left+right VA, VLA, VLP, and CM), and somatosensory (left+right VPL). For the proximity-to-CSF analysis, we grouped thalamic nuclei into two groups based on their proximity to the nearest ventricular surface: an “adjacent” group (left+right AV, MD, MGN, LGN, and pulvinar) and a “remote” group (left+right CM, VA, VLA, VLP, and VPL). For each of these group analyses, we summed the volumes of all nuclei within a group and across both hemispheres, to create a single “composite” volume, and fit this composite volume with our regression model to obtain its atrophy rate. We then performed the Wald test against the null hypothesis that the atrophy rate was zero, and pairwise comparisons between groups to test for differences in their atrophy rates.

To adjust for multiple comparisons, the simultaneous confidence region for parameters of interest (e.g., regression coefficient of age corresponding to a group of structural volumes of interest) was constructed, where the covariance between relevant estimates was estimated by nonparametric bootstrap, and the corresponding p-values controlling the family-wise error rate via the supremum Wald test statistic were reported. The statistical analyses were conducted using R 4.1.1 (The R Foundation for Statistical Computing).

2.5. Data/code availability statement

Data and code are available upon request.

3. Results

3.1. Different thalamic nuclei show widely different age-related atrophy rates across the cognitively unimpaired adult lifespan

We found significant age-related atrophy of most of the thalamic nuclei (Fig. 2). Absolute thalamic nuclear volumes varied approximately linearly with age and with different atrophy rates for different nuclei. The following lists the THOMAS structures in descending order of annualized atrophy rate (Table 3). Left hemisphere: AV (1.18%), LGN (0.72%), MGN (0.67%), pulvinar (0.65%), CM (0.62%) VA (0.51%), MD (0.43%), VPL (0.36%), and VLP (0.30%). Right hemisphere: AV (1.08%), LGN (0.77%), CM (0.62%), MGN (0.61%), pulvinar (0.58%), VA (0.50%), MD (0.40%), VLP (0.34%), and VPL (0.28%). Atrophy rates of left and right VLA and habenula were not significantly different from zero. AV was notable in having a significantly greater atrophy rate than all other ipsilateral thalamic nuclei, except for the LGN in the right hemisphere (Supp. Fig. 1). These atrophy rates were not significantly changed after adjusting for years of education or cohort (Supp. Fig. 2). We also qualitatively observed that the older cohort had higher atrophy rates compared to those of the younger cohort (Fig. 2). We therefore assessed whether the atrophy rates for younger (Bordeaux and Stanford) or older cohorts differed from those of the whole group. We found that the younger cohorts together showed low, mostly non-significant atrophy rates, while the older cohort showed mostly significant atrophy rates that were similar to those of the whole group (Supp. Fig. 4 and Supp. Table 1).

Fig. 2.

Absolute volumes of THOMAS-derived thalamic structures plotted versus age, showing heterogeneous atrophy between nuclei. Curves show best fits to the defined exponential regression model. Red=right hemisphere. Black=left hemisphere.

Table 3.

Annualized atrophy rates (percent change in volume per year) of thalamic and extra-thalamic structures with 95% confidence bounds and p-values adjusted for multiple comparisons. Bold numbers indicate change that is significantly different from zero.

Hemisphere	Structure	Mean Volumetric Change Per Year	95% Confidence Bound-Lower Limit	95% Confidence Bound-Upper Limit	Adjusted P-Value
Left	Cerebral Cortex	−0.15%	−0.23%	−0.08%	<0.001
Left	Cerebral White Matter	−0.10%	−0.19%	0.00%	0.052
Left	Cerebellar Cortex	−0.24%	−0.35%	−0.14%	<0.001
Left	Cerebellar White Matter	−0.06%	−0.25%	0.14%	1
Left	Amygdala	−0.03%	−0.14%	0.09%	1
Left	Hippocampus	0.00%	−0.10%	0.10%	1
Left	Caudate	−0.11%	−0.24%	0.03%	0.31
Left	Putamen	−0.19%	−0.29%	−0.09%	<0.001
Left	Nucleus Accumbens	−0.44%	−0.58%	−0.30%	<0.001
Left	Pallidum	0.01%	−0.12%	0.14%	1
Left	Lateral Ventricle	1.74%	1.10%	2.42%	<0.001
Left	Thalamus	−0.45%	−0.55%	−0.35%	<0.001
Left	AV	−1.18%	−1.43%	−0.93%	<0.001
Left	MD	−0.43%	−0.55%	−0.30%	<0.001
Left	Pulvinar	−0.65%	−0.79%	−0.50%	<0.001
Left	LGN	−0.72%	−1.05%	−0.38%	<0.001
Left	MGN	−0.67%	−0.83%	−0.50%	<0.001
Left	VA	−0.51%	−0.71%	−0.30%	<0.001
Left	VLA	−0.18%	−0.67%	0.34%	0.999
Left	VLP	−0.30%	−0.45%	−0.14%	<0.001
Left	CM	−0.62%	−0.84%	−0.39%	<0.001
Left	VPL	−0.36%	−0.57%	−0.15%	<0.001
Left	Habenula	−0.15%	−0.38%	0.10%	0.704
Right	Cerebral Cortex	−0.15%	−0.23%	−0.07%	<0.001
Right	Cerebral White Matter	−0.09%	−0.19%	0.00%	0.075
Right	Cerebellar Cortex	−0.25%	−0.36%	−0.14%	<0.001
Right	Cerebellar White Matter	0.00%	−0.20%	0.20%	1
Right	Amygdala	−0.03%	−0.15%	0.09%	1
Right	Hippocampus	−0.01%	−0.12%	0.09%	1
Right	Caudate	−0.21%	−0.32%	−0.10%	<0.001
Right	Putamen	−0.29%	−0.38%	−0.19%	<0.001
Right	Nucleus Accumbens	−0.54%	−0.68%	−0.39%	<0.001
Right	Pallidum	0.03%	−0.07%	0.14%	1
Right	Lateral Ventricle	1.61%	1.03%	2.22%	<0.001
Right	Thalamus	−0.45%	−0.54%	−0.36%	<0.001
Right	AV	−1.08%	−1.31%	−0.84%	<0.001
Right	MD	−0.40%	−0.52%	−0.27%	<0.001
Right	Pulvinar	−0.58%	−0.71%	−0.43%	<0.001
Right	LGN	−0.77%	−1.03%	−0.49%	<0.001
Right	MGN	−0.61%	−0.75%	−0.46%	<0.001
Right	VA	−0.50%	−0.69%	−0.31%	<0.001
Right	VLA	0.13%	−0.29%	0.56%	0.999
Right	VLP	−0.34%	−0.50%	−0.18%	<0.001
Right	CM	−0.62%	−0.83%	−0.41%	<0.001
Right	VPL	−0.28%	−0.47%	−0.09%	<0.001
Right	Habenula	−0.06%	−0.32%	0.20%	1

3.2. The atrophy of thalamic nuclei segregates by function

We next examined whether the atrophy rates of thalamic nuclei were specific to function. This was hypothesized based on literature demonstrating that age-related atrophy occurs heterogeneously across functionally distinct cortical areas, with the greatest atrophy seen in frontal and temporal cortices and the least atrophy seen in sensorimotor and occipital cortices (Fjell and Walhovd, 2010; Fjell et al., 2009a; Raz et al., 1997). Thalamic nuclei were grouped into broad functional groups: cognitive (AV and MD), visual (LGN and pulvinar), auditory and vestibular (MGN), motor (VA, VLA, VLP, and CM), and somatosensory (VPL). We found a significant segregation of atrophy rates by function. The cognitive, visual, and auditory/vestibular groups showed the strongest atrophy rates, which ranged from 0.53% to 0.64% (Table 4). These were significantly different (adjusted p<0.006; Table 5) from the atrophy rates of the motor and somatosensory groups, which were 0.37% and 0.32%, respectively.

Table 4.

Annualized atrophy rates of thalamic nuclei grouped by function. Group atrophy rates were tested with t-tests for significant difference from zero. 95% confidence bands and p-values were adjusted for multiple comparisons. Nuclei are color-coded by functional group: yellow = cognitive; green = visual; pink = auditory/vestibular; blue = motor; and purple = somatosensory.

Group	Function	Annualized Atrophy Rate	95% Confidence Bound-Lower Limit	95% Confidence Bound-Upper Limit	Adjusted P-Value
AV, MD	Cognitive	0.53%	0.62%	0.44%	<0.001
LGN, Pulvinar	Visual	0.62%	0.73%	0.51%	<0.001
MGN	Auditory/Vestibular	0.64%	0.75%	0.52%	<0.001
VA, VLA, VLP, CM	Motor	0.37%	0.48%	0.26%	<0.001
VPL	Somatosensory	0.32%	0.46%	0.18%	<0.001

Table 5.

Functional specificity in the atrophy of thalamic nuclei with age. Significant between-group differences in atrophy rates were tested with paired t-tests. P-values were adjusted for multiple comparisons. Bold numbers indicate statistical significance.

Between Group Comparison	Adjusted P-Value
Cognitive > Visual	0.039
Cognitive > Auditory/Vestibular	0.026
Cognitive > Motor	<0.001
Cognitive > Somatosensory	0.006
Visual > Auditory/Vestibular	0.998
Visual > Motor	<0.001
Visual > Somatosensory	<0.001
Auditory/Vestibular > Motor	<0.001
Auditory/Vestibular > Somatosensory	<0.001
Motor > Somatosensory	0.929

3.3. Thalamic nuclei that are adjacent to the ventricles show greater age-related atrophy

Prior work has found evidence in multiple sclerosis that thalamic nuclei adjacent to the ventricles show greater atrophy than nuclei that are located farther away from the ventricles (Lassmann, 2018; Planche et al., 2020). Here, we examined whether there was evidence that a similar effect occurs in cognitively unimpaired aging. To test this, thalamic nuclei were grouped into two groups – those that were adjacent to the ventricles (AV, MD, LGN, pulvinar, and MGN) and those that were non-adjacent (i.e., remote) to the ventricles (CM, VA, VLA, VLP, and VPL). We found a significant difference of atrophy rates based on ventricular proximity: the adjacent group showed a higher atrophy rate (0.59% per year), which was significantly different (adjusted p<0.001) from that of the remote group (0.36% per year).

3.4. The thalamus and its nuclei have a high rate of atrophy in comparison to other brain structures across the cognitively unimpaired adult lifespan

For the extra-thalamic structures, we found that the absolute volumes of certain gray matter structures decreased approximately linearly with age, those of white matter structures did not significantly change with age, and those of the lateral ventricles increased nonlinearly with age (Fig. 3). Fig. 4 shows the percent change in volume per year for each structure; Table 3 shows these atrophy rates and their confidence bounds and p-values, adjusted for multiple comparisons. These atrophy rates were not significantly changed after adjusting for years of education or cohort (Supp. Fig. 2).

Fig. 3.

Absolute volumes of the whole thalamus and extra-thalamic structures plotted versus age. Curves show best fits to the defined exponential regression model. Red=right hemisphere. Black=left hemisphere.

Fig. 4.

Comparison of annualized atrophy rates of extra-thalamic and thalamic structures. Lines indicate 95% confidence bands, adjusted for multiple comparisons. Note that thalamic nuclei atrophy at rates higher than most other structures in the brain. Red=right hemisphere. Black=left hemisphere.

Significant differences between atrophy rates were assessed with paired t-tests, adjusted for multiple comparisons (Supp. Fig. 3). Notably, the left and right whole thalamic atrophy rates were higher than those of all extra-thalamic structures with the exception of the nucleus accumbens. Specifically, the atrophy rate of the left thalamus (0.45%) was significantly greater than that of all other significantly atrophying structures in the left hemisphere (Table 3), those being cerebellar cortex (0.24%), putamen (0.19%), and cerebral cortex (0.15%). The exception to this was the left nucleus accumbens (0.44%), which had an atrophy rate similar to (but not significantly different from) that of the left thalamus. The following left hemispheric structures had no statistically significant atrophy with age: cerebral white matter, cerebellar white matter, amygdala, hippocampus, caudate, and pallidum. We note that we do observe significant atrophy of these structures in the older cohort alone (Supp. Table 1). Similarly, the atrophy rate of the right thalamus (0.45%) was significantly greater than all other significantly atrophying right hemispheric structures, those being putamen (0.29%), cerebellar cortex (0.25%), caudate (0.21%), and cerebral cortex (0.15%). The exception to this was the right nucleus accumbens, whose atrophy rate (0.54%) was similar to that of the right thalamus. The following right hemispheric structures had no statistically significant atrophy with age: cerebral white matter, cerebellar white matter, amygdala, hippocampus, and pallidum; again, we note that the atrophy rates of these structures are significantly greater than zero when analyzing the older cohort alone (Supp. Table 1). Roughly half of the thalamic nuclei had higher rates of atrophy than all of the extra-thalamic structures examined in both the left and right hemispheres (Fig. 4).

We also found that the left and right lateral ventricles significantly increased in volume with age (left: 1.74% volume increase per year; right: 1.61%). The volumetric expansion of the lateral ventricle was nonlinear with age, increasing strongly for ages greater than 60 years.

3.5. Virtually no hemispheric or sex differences in the age-dependent changes of thalamic and extra-thalamic structures

Based on pairwise t-tests between the left and right volumes for each nucleus, adjusted for multiple comparisons, we did not observe any significant hemisphere-specific differences in atrophy rates for any thalamic or extra-thalamic structures.

Sex was entered as an independent variable in the exponential regression models described above for both intra- and extra-thalamic structures. Only one structure showed a sex-specific difference in atrophy rate. For the left CM nucleus of the thalamus, male participants showed a significantly greater atrophy rate than female participants (male, 0.86%; female, 0.41%; p = 0.034, adjusted for multiple comparisons).

4. Discussion

Prior literature examining the volumetric changes of brain structures during aging has consistently reported that the thalamus, along with other brain structures, significantly atrophies with age (Cherubini et al., 2009; Coupé et al., 2017; Narvacan et al., 2017; Raji et al., 2009; Walhovd et al., 2005, 2011; Zheng et al., 2019). Our principal finding is that thalamic nuclei have high and heterogeneous atrophy rates during cognitively unimpaired aging. Our results further suggest that a number of thalamic nuclei atrophy at greater rates than most of the extra-thalamic structures examined here. The high atrophy rates of thalamic nuclei may be the result of several underlying causes. One involves the “metabolism hypothesis”, proposed in the context of cortical hubs, which are brain regions with high degrees of connectivity and integration of information. Neuroimaging work has shown that these cortical hubs overlap with disease loci in Alzheimer’s disease – sites of amyloid beta deposition and hypometabolism (Buckner et al., 2009; Drzezga et al., 2011). The metabolism hypothesis proposes that cortical hubs are susceptible to disease due to their high levels of neuronal activity, which demand high levels of metabolism and create an excess of harmful free radical byproducts as well as pathological changes in gene and protein expression (Buckner et al., 2008, 2005). Similarly, the highly connected and integrative role of thalamic nuclei may make them susceptible to metabolic and molecular changes that disrupt normal neuronal processes and ultimately lead to structural changes.

Another hypothesis is that nuclei adjacent to ventricles may be more susceptible or accessible to CSF-mediated factors or immune cells that cause tissue damage and subsequent structural atrophy. Such CSF-mediated damage has been reported in multiple sclerosis (De Meo et al., 2022; Lassmann, 2018; Planche et al., 2020; Blyau et al., 2022) and recently in traumatic brain injury (Sandry and Dobryakova, 2021). We speculate that the same may occur in healthy aging. Indeed, our analysis shows that thalamic nuclei that are adjacent to ventricular surfaces (AV, MD, MGN, LGN, and pulvinar) atrophy at significantly higher rates than those that are more remote from CSF (CM, VA, VLA, VLP, and VPL), supporting the CSF-mediated damage hypothesis and providing the first evidence of this proximity-to-CSF hypothesis in healthy aging.

We note an alternative hypothesis to the proximity-to-CSF hypothesis that could also explain our observations. Our analysis of the functional specificity of thalamic atrophy rates showed that cognitive, visual, and auditory/vestibular nuclei exhibited higher atrophy rates than the sensorimotor nuclei. However, these functional groupings are also consistent with the proximity-to-CSF groupings, with the cognitive, visual, and auditory/vestibular nuclei in the adjacent group and the sensorimotor nuclei in the remote group. With the present data and analyses, we are unable to lend support for one hypothesis over the other; indeed, the biology underlying volumetric atrophy of thalamic nuclei is undoubtedly complex and may allow for both hypotheses.

Despite the observed significant decline in thalamic volumes with age, all participants in our study had normal cognitive function. What might explain this finding? One possibility is that the brain can compensate for structural/volumetric changes via alternative circuitry, at least up to a point, such that cognition is largely preserved (Park and Reuter-Lorenz, 2009). Postmortem histological and patient neuroimaging studies indicate that thalamic volumes, as well as those of other brain structures, are smaller for patients with a variety of neurological conditions than for cognitively unimpaired age-matched control participants (Azevedo et al., 2018; Bishop et al., 2017; Callen et al., 2001; Cash et al., 2018; Fischl et al., 2002; Jernigan et al., 1991; Lee et al., 2020; Li et al., 2013; Schönecker et al., 2018). Thus, while there is significant observable atrophy in the cognitively unimpaired aged brain, an even greater amount of atrophy occurs in the diseased brain and may be necessary before observable clinical signs occur (Fjell and Walhovd, 2010). Alternatively, the atrophy observed in the cognitively unimpaired aged versus diseased brains may reflect different underlying biological processes. For example, neurodegenerative disease-related atrophy typically involves neurotoxic proteins and neuronal death. However, studies of cognitively unimpaired aging suggest that age-related atrophy generally does not involve neuronal death, but rather a shrinkage in neuronal size and reductions of the neuropil (see Fjell and Walhovd 2010 for review).

With the recent development of advanced thalamic segmentation methods (Iglesias et al., 2018; Su et al., 2019), studies have begun to show that specific thalamic nuclei are altered in disease (Aleman-Gomez et al., 2020; Bocchetta et al., 2020; Chipika et al., 2020; Huang et al., 2020; Jonak et al., 2020; Lee et al., 2020; Low et al., 2019; Planche et al., 2020; Shin et al., 2019). Several studies have also found evidence of volumetric differences in thalamic nuclei between asymptomatic genetic carriers and non-carriers in frontotemporal dementia (Bocchetta et al., 2021) and their association with PET-based levels of amyloid and tau in Alzheimer’s disease (Pardilla-Delgado et al., 2021), highlighting the possibility that the volumes of thalamic nuclei may represent preclinical markers for disease. In our study, AV was the brain structure with the highest atrophy rate. We note that while the THOMAS method segments AV, this segmentation likely also includes some or all of the small, adjacent anterodorsal (AD) nucleus. The strong age-related atrophy of the AV nucleus observed in our study is consistent with prior work showing that anterior thalamic structures are implicated in Alzheimer’s disease (Aggleton et al., 2016; Braak and Braak, 1991a). Postmortem histology shows that the AD nucleus is abnormal relatively early in Alzheimer’s disease, with dense neurofibrillary tangles observed in stages II-III of the disease (Braak et al., 1996). AV is spared early in the disease, but sustains amyloid deposition in later stages of the disease (Braak and Braak, 1991a, b). Thus, our observation of AV’s strong atrophy in this cognitively unimpaired sample may be an early indicator of processes contributing to the development of Alzheimer’s disease. Future studies are needed to longitudinally correlate the atrophy of AV and other thalamic nuclei with functional and disease markers, such as neuropsychological test scores, fMRI or PET activity, or cerebrospinal fluid levels of disease-related proteins, such as amyloid beta and tau.

We observed no hemisphere-specific differences and almost no sex-specific differences in age-related thalamic atrophy. There has been disagreement in the literature concerning the existence of hemisphere-specific (Hughes et al., 2012; Pieperhoff et al., 2008; Serbruyns et al., 2015; Squarzoni et al., 2018; Walhovd et al., 2011; Xu et al., 2000) or sex-specific (Coupé et al., 2017; Fjell et al., 2009b; Goodro et al., 2012; Király et al., 2016; Li et al., 2014; Murphy et al., 1996; Sullivan et al., 2004; Wang et al., 2019; Xu et al., 2000) differences in whole thalamic atrophy rates. Discrepancies may arise due to methodological differences between the studies, such as the sample size, ethnic background of the participants, and differing accuracies of the segmentation methods. In the present study, using a comparatively large sample size and the most accurate existing method for automatically segmenting thalamic nuclei, our data show virtually no hemisphere- or sex-specific differences in age-dependent atrophy rates for either the whole thalamus or individual thalamic nuclei. Future studies with larger sample sizes and state-of-the-art thalamic segmentation methods are needed to systematically tease apart thalamic atrophy in the context of different demographic factors, genetic backgrounds, and cognitive capabilities.

Although the focus of our study was on thalamic nuclei, we note several differences in the atrophy of some extra-thalamic structures compared to prior literature. We observed low or no significant atrophy of cerebral and cerebellar white matter and the hippocampus over the adult lifespan. Coupé et al. (2017) examined volumetric changes in a survey of brain structures (including the whole thalamus, but not thalamic nuclei) during development and over the adult lifespan and found robust cerebral and cerebellar white matter atrophy and moderate hippocampal atrophy over the older age range (>60 years), while for these brain regions the average atrophy rate over the middle age range (20–60) was near zero. Our data showed similar trends, with near-zero atrophy rates when the entire age range was analyzed, but increased atrophy rates (cerebral and cerebellar white matter: 0.14% to 0.31% per year; hippocampus: 0.48% to 0.57% per year; Supp. Fig. 4B; Supp. Table 1) when only the older age range was analyzed. Similar (though not as extreme) observations of increased atrophy rates in the older cohort compared to the full cohort were seen for other brain structures (Supp. Fig. 4B). Notwithstanding these nonlinear effects, the thalamus and its nuclei remain among the highest atrophying structures with age, even in the older cohort. Future studies will need to examine this issue using a larger number of participants analyzed with both the SAMSEG and volBrain pipelines.

4.1. Caveats and limitations

In the present study, we combined an older cohort with two younger cohorts to create a large sample spanning seven decades of adult age. As a result, there were several group-specific differences in scan acquisition parameters, such as scan protocol, field strength, and institution. For example, the WMnMPRAGE imaging for the Stanford younger and older adults was acquired at 7T, while for the Bordeaux younger adults the WMnMPRAGE imaging was obtained at 3T. We first note that our analysis of brain volumes made exclusive use of fractional (or relative) volumes; this means that we analyzed only brain volumes normalized by ICV. We also adjusted for ICV in all regression analyses. Brain volumes and ICVs were obtained using the same segmentation tools for all participants. Therefore, if there is a multiplicative scaling bias as a result of a site-specific field strength, scanner, or software effect, then this scaling bias would be canceled in the formation of the fractional volumes. To further rule out a significant influence of field strength or institution, we tested for thalamic volume differences between Stanford (7T) and Bordeaux (3T) participants within the same age group and found no significant differences. We also found a similar pattern of atrophy after adjusting for cohort (Supp. Fig. 2B) or after analyzing solely the Stanford cohort (single site, same field strength) (Supp. Fig. 2C).

In addition, while the education level of our study population was uniformly high (mean 17.8 years, standard deviation 2.6 years), the older cohort had a somewhat lower group education level (mean 16.7 years) compared to the younger cohorts (mean 20.8 years). It is known that education level correlates with brain health and performance of certain cognitive abilities (Capitani et al., 2009; Mortiner, 1988). Indeed, Trelle et al. (2020) reported poorer episodic memory retrieval with lower education level in the present older cohort. Thus, we may have observed lower age-related atrophy rates if our average education levels had been more similar between the younger and older age cohorts. However, we found a similar pattern of atrophy across brain structures when we adjusted for years of education in our regression analysis (Supp. Fig. 2A and B). Thus, the confound of varying education level did not significantly impact the observed atrophy rates across the adult lifespan. Nonetheless, future work should include longitudinal studies following the same participants as they age to obtain more sensitive and controlled data on the atrophy of the thalamus and its nuclei with aging.

As an alternative to THOMAS, Iglesias et al. (2018) have created a thalamic nuclear segmentation method that is available in the FreeSurfer package. While the FreeSurfer thalamic segmentation method has a number of advantages, such as operating on conventional T1 images and offering segmentations of 25 nuclei, we used WMnMPRAGE and THOMAS in the present study, motivated by prior work (Majdi et al., 2020; Su et al., 2019) showing that the combination of WMnMPRAGE imaging and THOMAS segmentation results in higher accuracy in defining the boundaries of thalamic nuclei (as measured by Dice and VSI overlap metrics) compared to that obtained with the FreeSurfer method.

In the present study, we used two different methods to segment structural volumes –THOMAS for thalamic nuclei using WMnMPRAGE images and SAMSEG for extra-thalamic structures using conventional T1-weighted images. This approach was unavoidable, as no single method produces both sets of segmentations. Consequently, we note that there may be systematic software-specific effects that could affect comparisons between the atrophy rates of thalamic versus extra-thalamic structures. We note, however, that both THOMAS-derived and SAMSEG-derived whole thalamus volumes showed significant non-zero atrophy with age (0.45% per year for THOMAS, 0.23% per year for SAMSEG), indicating comparable volumetry for these two segmentation methods.

Another caveat is that we compared brain structures of very differently sized volumes: the smallest structures we analyzed (habenula with an average volume of 25 mm³) were almost 10,000 times smaller in volume than the largest structures (cerebral cortex with an average volume of nearly 250,000 mm³). This huge dynamic range is the reason that we used an exponential model whose fundamental characteristic is a constant percent change in volume per unit time over the lifespan, allowing us to directly compare the relative atrophy rates of small and large structures; however, this did constrain our analysis to one particular nonlinear model. That said, our principal findings about the age-related atrophy rates of thalamic nuclei would not be affected by this issue, as the thalamic nuclei are relatively similar in volume. Future studies that more closely compare thalamic and extra-thalamic atrophy would likely benefit from examining subregions of large structures, such as cortical areas within the cerebral cortex.

Prior studies examining linear and nonlinear models of thalamic volume changes versus age have generally found linear relationships (Fjell et al., 2013; Goodro et al., 2012; Hasan et al., 2011; Sullivan et al., 2004). A few recent studies have reported nonlinear relationships (Coupé et al., 2017; Narvacan et al., 2017; Tullo et al., 2019; Wang et al., 2019). Using an exponential model with the capability of modeling linear and nonlinear atrophy, we also observed a consistently near-linear decrease throughout the adult lifespan. However, we note that our sample was not uniformly distributed over the age range – we had fewer participants in the 50–60 year age range, which is an age at which certain cognitive abilities such as verbal memory and visuospatial processing begin to decline (Salthouse, 2009; Schaie, 2005). Similarly, we note that while certain structures (e.g., whole thalamus, AV, and nucleus accumbens) clearly showed a near-linear reduction of volume throughout the age-span, this linearity is not so clear for other structures (e.g., cerebral and cerebellar cortices) as a result of a higher degree of scatter in our cross-sectional dataset, combined with the non-uniform sample distribution. These limitations could be addressed in future longitudinal studies with more participants, ideally distributed uniformly over the lifespan, to better detect any potential nonlinearities in the age-related atrophy of thalamic nuclear volumes.

Finally, the cross-sectional design adopted here is susceptible to variation in the volumes of these structures between individuals and does not provide a direct measure of within-individual age-related atrophy. Nonetheless, the present findings provide important initial evidence for heterogenous rates of volume loss among thalamic nuclei across the lifespan, a pattern that should be confirmed in future longitudinal studies.

5. Conclusions

Here, we showed that the thalamic nuclei have high and heterogeneous age-related atrophy in the cognitively unimpaired human brain. Thalamic nuclei involved in cognition (AV, MD), visual/visual association (LGN, pulvinar), and auditory/vestibular (MGN) function showed the highest atrophy rates. These five nuclei also happen to be adjacent to a ventricular surface, which may make them susceptible to CSF-mediated factors that affect brain structure. The next tier of atrophy rates involved the nuclei related to motor (VA, VLA, VLP, and CM) and somatosensory (VPL) functions, and these five nuclei happen to be located at some distance from the nearest ventricular surface. We found no significant hemispheric, sex-specific, or education-specific differences in these rates of change for thalamic nuclei. These results provide a baseline measure of atrophy rates occurring in thalamic nuclei across the cognitively unimpaired adult lifespan. Given that our participants were all cognitively unimpaired, our results indicate the range of structural atrophy that can be tolerated by the resilient cognitively unimpaired aged brain while preserving cognitive functioning. These findings indicate that the thalamus and its nuclei are a component of age-related brain atrophy in the cognitively unimpaired human brain. A detailed understanding of such atrophy may provide insights into the transition from cognitively unimpaired brain aging to pathological neurodegeneration and other neurological and psychiatric disease.

Abbreviations:

THOMAS

THalamus Optimized MultiAtlas Segmentation

Data/code availability statement

Data and code are available upon request.

Data Availability

Data will be made available on request.