High‐Resolution Diffusion Kurtosis Imaging of Hippocampus Subfields Across the Healthy Lifespan

ABSTRACT

Diffusion kurtosis imaging (DKI) is typically applied to white matter, but it may provide insight into age‐related microstructural changes in gray matter tissue like the hippocampus. The goal was to assess neurodevelopment and aging changes in hippocampal subfields using 1 mm isotropic DKI across the healthy lifespan (5–90 years). Multi‐shell, 1 mm³, diffusion imaging focused on the hippocampus, was acquired at 3 T in 363 healthy participants (5–90 years, 206 females). Automatic hippocampal subfield segmentation and unfolded maps were obtained using HippUnfold. Nonlinear lifespan trajectories, sex differences, and age‐corrected residual correlations with cognitive scores and other demographics were assessed for volume, mean diffusivity (MD), fractional anisotropy (FA), mean kurtosis (MK), and kurtosis FA (KFA). Whole hippocampus yielded distinct age trajectories for the volume, diffusion tensor, and diffusion kurtosis parameters: volume—quadratic fit (maximum ~35–42 years); MD—negative Gamma variate (minimum ~35 years); FA—positive Gamma variate fit (maximum ~25 years); MK—exponential fit (steep increase during development that plateaus after ~25 years); and KFA—negative linear across the lifespan. There were sex differences in age trajectories for volume and MK that were not evident with DTI metrics. Males exhibited larger volume and larger MK than females. The subfields showed similar age trajectories as whole hippocampus albeit with regional variations in the values. The subiculum showed dramatically higher MK after age 20, with males exhibiting higher values than females. MK residuals in whole hippocampus correlated positively with body mass index residuals in three age groups (young, middle age, and older). In conclusion, high‐resolution 1 mm isotropic DKI of hippocampus revealed linear and nonlinear patterns with development and aging over a wide age range of 5–90 years that differ from DTI. These results suggest that DKI can provide novel insight into age‐related microstructural changes at the sub‐hippocampal level.

Developmental and aging changes were examined in the hippocampus using 1 mm isotropic diffusion kurtosis imaging in 363 participants aged 5–90 years. HippUnfold enabled automated subfield segmentation and unfolded hippocampal maps. Whole‐hippocampus showed distinct lifespan trajectories: volume, quadratic; MD, negative Gamma Variate; FA, positive Gamma Variate ; MK, exponential; and KFA, negative linear. Subfields exhibited regional variations, with elevated MK and KFA in the subiculum. These results suggest that DKI can provide novel insight into age‐related microstructural changes at sub‐hippocampal level.

1. Introduction

Diffusion kurtosis imaging (DKI) is an extension of diffusion tensor imaging (DTI) that accounts for non‐Gaussian diffusion effects [1] arising at high b‐values (typically b ≥ 2000 s/mm²) due to diffusion barriers such as membranes. DKI estimates the DTI metrics of axial, radial, and mean diffusivities (AD, RD, MD) and fractional anisotropy (FA), as well as axial, radial, and mean kurtosis (AK, RK, MK), and kurtosis fractional anisotropy (KFA) [2, 3, 4]. Histological studies in genetic knockout and cuprizone mouse models have shown stronger associations between DKI and myelin content in white matter (WM) [5, 6], as well as alterations in the motor and somatosensory cortical gray matter (GM) that DTI failed to identify [7]. Additionally, MK better captured the spatial variability of cortical myelin density in primates [8]. Although KFA has been less extensively studied, evidence from an MRI phantom [4] and model simulations [2] indicates that KFA outperforms FA in the presence of heterogeneous fiber distributions. High‐resolution images from a fixed rat brain further demonstrate that while KFA tends to be high where FA is high, it also exhibits elevated values in complex regions where FA is low, as in the hippocampus [4].

Although far less applied than DTI, DKI has been used to investigate microstructural changes in human brain over “healthy” neurodevelopment and aging [9], with most focusing on WM and MK as the primary parameter of interest. In two studies from neonates to children, either to 5 years [10] or to 14 years [11], MK showed a marked increase in WM (splenium and genu of the corpus callosum) and less so for deep GM (putamen and thalamus), particularly within the first 2 years of life, followed by an apparent plateau. Group comparisons of children (9–12 years) to adults (38–64 years) yielded greater MK and KFA in WM in the older cohort with the greatest changes in association fibers like the cingulum [12]. In aging, WM MK drops typically after ~50–60 years [13, 14, 15, 16] although the inclusion of younger participants to 18 years has revealed quadratic inverted U‐shaped age trajectories for MK, AK, and RK that peaks at ~45–50 years [16]. Only one study has analyzed a broader age range from younger children to the elderly (10–79 years, n = 80) showing an MK increase in some major WM regions and deep GM (putamen, pallidum) during development, peaking at ~40 years and then decreasing with age with no sex differences; other WM and deep GM regions (thalamus) showed negative linear changes of MK with age [17]. Although a few of the above studies examined some deep GM regions, there are no such typical neurodevelopment and aging DKI studies of the hippocampus, a key brain structure for memory that is affected in many neurodevelopment and neurological disorders [18].

Hippocampal studies have mainly focused on whole‐structure volume using anatomical scans [19, 20, 21], showing a positive or null association across development/adolescence, followed by accelerated declines later in life [21, 22]. However, volumetric studies cannot provide information to infer microstructural changes at the cellular level. Using the ubiquitous 2 mm isotropic resolution, a DTI study of the hippocampus (n = 1790, 4–93 years) showed a quadratic U‐shaped MD trajectory with a minimum at ~40 years in the anterior portion of the hippocampus that differed in trajectory from the posterior portion, although both had marked MD increases after 40 years [23]. Using a much higher resolution 1 mm isotropic single shell hippocampal DTI protocol [24], MD became lower in childhood/adolescence, which levelled off after 25 years, while FA went up, peaked at ~35 years, and then went down in a typical healthy study over 5–74 years (n = 153) [25]. This latter study used manual segmentation of the hippocampus, which then trained a machine learning segmentation model that showed similar age trends for volume, MD, and FA, but in a larger independent cohort (also with 1 mm³ images) over a wider age range (n = 354, 5–90 years) [26]. An additional b‐shell of 2000 s/mm² was acquired but not used in their DTI analysis. Such higher spatial resolutions are essential for studying small, complex structures, like the hippocampus, particularly if subregions may have different microstructural and hence quantitative MRI properties.

The hippocampus can be divided into distinct subfields including the cornu ammonis (CA1–4), dentate gyrus (DG), and the subiculum [18], which can each undergo dynamic changes during development, including neurogenesis, emergence of astrocytes, glial cells, and myelination [27, 28, 29]. Subfields exhibit different development times (from 16 to 27 gestational weeks) with CA1 neurons maturing earlier than in CA3 and DG [27]. After birth, hippocampal neurogenesis declines sharply [30], but persists into older adults [28, 29]. A meta‐analysis of volumetric studies of the hippocampal subfields with age have reported nonlinear changes with DG and CA3–4 that get larger during development (11 studies: age, 4–33 years; mean, 11 ± 3 years) and all subfields get smaller with elderly aging (41 studies: age, 4–94 years; mean, 42 ± 22 years) [31]. DTI of hippocampal subfields has also been studied showing different FA relationships with age over 5–30 years in development [32]. In older adults, MD has been shown to increase linearly with age in CA2–3 from 50 to 75 years, while FA decreases linearly in the subiculum [33]. Similar findings have been reported in individuals aged 60–85 years, with linear age‐related MD increases in the subiculum and CA1 [34]. Using higher shell data with models beyond the conventional DTI representation, NODDI (neurite and orientation and dispersion imaging) has shown nonlinear increases of neurite density index in all hippocampal subfields during development from 8 to 21 years [35], while SANDI (soma and neurite density imaging) has reported linear increases of neurite fraction from 8 to 19 years [36], and reductions of soma fraction (in addition to other changes) in select subfields during aging from 19 to 85 years [37]. However, these hippocampal subfield diffusion MRI studies had limited spatial resolution with either 1.5 mm [32, 35], 2 mm [33, 36, 37], or 3 mm [34] isotropic nominal voxels. There are no such DKI studies of hippocampus subfields over the lifespan, including both development and aging, which would necessitate both high spatial resolution and a high b‐shell.

The purpose here is to use DKI at 1 mm isotropic spatial resolution to investigate age changes, sex differences, and demographics (e.g., memory, body mass index [BMI]) correlates of the whole hippocampus and its subfields across 5–90 years in a single‐site cohort (n = 363). This study builds on the previous whole hippocampus lifespan study [26] by now incorporating the high b‐shell (b = 2000 s/mm²) for DKI, in addition to the subfield subdivision. The DKI metric age trajectories, including KFA, which has not been studied with development and aging, will be compared to that from standard DTI.

2. Methods

2.1. Participants

This normative lifespan MRI study enrolled 363 healthy participants including 206 females aged 5–90 years and 157 males aged 5–85 years who had no self‐reported history of brain injury, neurological disorders, or contraindication to MRI; details of the study can be found in ref. [38]. Participants were recruited via word of mouth and community advertising and provided written informed consent (for participants under 18 years old, both child assent and parent/guardian consent were obtained) prior to participation. Cognitive and other demographic assessments were performed as listed in Section 2.6. The study was approved by the University of Alberta Human Research Ethics Board.

2.2. MRI Acquisition and Processing

Multi‐shell diffusion‐weighted images of the hippocampus were acquired using a 64‐channel head coil on a 3 T Siemens Prisma (single site) using single‐shot 2D EPI and Stejskal–Tanner diffusion encoding, 20 1‐mm axial‐oblique slices aligned along the length of the hippocampus with no gap, 1 × 1 mm² in plane resolution with no interpolation, GRAPPA 2, 6/8 partial Fourier, A/P phase encode, FOV = 220 × 216 mm², coil combine mode: Adaptive combine, prescan normalize: on, TE = 83 ms, TR = 3100 ms, 20 b = 0 s/mm², 64 diffusion directions of b = 500 s/mm² and 64 diffusion directions of b = 2000 s/mm² in 7:50 min as in ref. [38]. Processing included denoising [39], Gibbs‐ringing [40], Rician bias [41], motion and eddy currents correction [42], and bias field inhomogeneity correction [43]. MRtrix3 [44] was used to estimate DTI (using only the b = 500 s/mm² shell without the Rician correction) to calculate MD and FA, and DKI tensors (using all shells with the Rician correction) to calculate MK and KFA.

2.3. Hippocampal Segmentation

Automatic hippocampal subfield (subiculum, CA1, CA2, CA3, CA4/DG) segmentation and unfolded hippocampal maps of DTI (MD, FA) and DKI (MK, KFA) parameters were obtained using HippUnfold, a deep learning–based software with a custom module (hippb500) that was trained to work solely on the mean DWI (b = 500 s/mm²) data [45]. Three segmentation examples of 5, 25, and 60 years old healthy participants are shown in Figure 1. HippUnfold provides segmentations in both the image native space and in a surface standardized common space (where all the participants have the same number of vertices). The volumes for whole hippocampus and its subfields (cm³) were measured as the average volume (left and right) of the segmentations in the image native space. In addition to individual plots versus age, the DTI/DKI parameters were averaged within each subfield over the standardized hippocampal surfaces for six different age bins for males and females separately: 5–10 years (17 males, 23 females), 11–15 years (11 males, 14 females), 16–20 years (14 males, 20 females), 21–40 years (51 males, 68 females), 41–60 years (24 males, 41 females), and > 60 years (40 males, 40 females).

FIGURE 1.

(A) Representative 1 mm isotropic mean DWI images (b = 500 s/mm²), (B) axial, and (C) coronal HippUnfold segmentations from the five subfield regions from three healthy participants aged 5, 25, and 60 years.

2.4. Test/Retest Reproducibility and Left/Right Asymmetry Assessment

To assess the precision of volume and DTI/DKI metrics, a test/retest analysis was conducted on a subset of 24 participants (12 males, aged 20–49 years; 12 females, aged 20–50 years) with repeated test/retest scans with a mean time between scans of 15 ± 28 days. Data was processed as described in Section 2.2 and segmented according to Section 2.3. Absolute mean differences between repeated scans for volume and DTI/DKI metrics for whole hippocampus and the subfields were calculated and are provided in Supplementary Table S1. These differences were used as thresholds, and therefore any left/right or sex differences below these were considered to not be significant.

In the entire cohort of 363 individuals, a paired t‐test was used to examine hemispheric differences in volume, MD, FA, MK, and KFA (Supplementary Table S2). Although there were several significant group differences (Benjamini–Hochberg false discovery rate, FDR‐BH, correction at p < 0.05), all DTI and DKI metrics were averaged across the left and right surfaces to reduce the number of comparisons. The fitting process of diffusion metrics versus age described in Section 2.5.1 of the Methods was applied to the left and right data separately (when there was a significant difference in the paired t‐test). Supplementary Figure S1 shows that both left and right hippocampi fits for all the metrics (including subfields) followed similar age‐related trajectories, supporting the use of combined left and right hippocampal measurements.

2.5. Statistical Analysis

Statistical analyses were performed in Python using SciPy 1.6.1 [46].

2.5.1. Curve‐Fitting for Age Relationships

Whole hippocampus and its subfields (subiculum, CA1, CA2, CA3, and CA4/DG) volume, MD, FA, MK, and KFA were fitted as functions of age using four different curve‐fitting approaches: linear (p = ax + b), quadratic (p = ax ² + bx + c), exponential (p = be ^−ax  + c) and Gamma variate function with alpha = 1 (p = axe ^−bx ) where x is age and a, b, c, and d are fitting parameters. The best fit was chosen based on the Akaike information criterion (AIC) and significant F test. The curve‐fitting was performed for males and females separately and the coefficient of determination (R ²) was used to demonstrate the quality of the fit. If there was no sex difference (see Section 2.5.2) then one fit was performed for all participants.

For each DTI and DKI metric, the time from age 5 years to the peak/minima or plateau of the curves was analyzed and compared for the whole hippocampus, as well as for each subfield. In cases where an exponential fit was used, the plateau time was defined as the age at which 90% of the maximum value was reached. Additionally, percentage changes were calculated from 5 years old to the peak/minima/plateau and from the peak/minima/plateau to age 90 years.

2.5.2. Permutation Test for Sex and Age Trajectory Differences

A permutation‐based approach was used to assess sex differences in age‐related trajectories. Separate age curves were fit for males and females, and the mean absolute difference between the predicted values was used to quantify the overall observed difference between sexes. To assess statistical significance, 10,000 permutations were performed. In each iteration, sex labels were randomly shuffled, and the data were split into pseudo‐male and pseudo‐female groups. Curve fitting and prediction were repeated for each group, and the mean absolute difference between the resulting curves was calculated to generate a null distribution. The p‐value was defined as the proportion of permutations in which the permuted curve difference was greater than or equal to the observed difference (FDR‐BH, correction at p < 0.05). When sex differences were detected, divergence age was defined as the earliest point where the 95% confidence intervals of the male and female curves no longer overlapped. The corresponding p‐values and mean differences (males/females) of the age trajectories from the permutation test are provided in Supplementary Table S3.

2.6. Cognitive and Demographic Correlations With Whole Hippocampal DTI and DKI Metrics

Five cognitive tests from the NIH Toolbox Cognitive Battery [47] were used to assess different aspects of memory function: the Picture Sequence Memory Test (PSMT) measured the ability to acquire, store, and retrieve new information; the Dimensional Change Card Sort Test (DCCST) assessed executive function; the List Sorting Working Memory Test (LSWMT) measured working memory capacity; the Oral Symbol Digit Test (OSDT) evaluated processing speed; and the Rey Auditory Verbal Learning Test (RAVLT) assessed verbal recall. Additionally, correlations with three different demographics including BMI, heart rate, and waist circumference were investigated since obesity has been linked with memory function [48, 49, 50]. The NIH toolbox provided aged standardized scores for PSMT, DCCST, and LSWMT. For OSDT, RAVLT, BMI, heart rate, and waist circumference, the fitting procedure described in Section 2.5.1 was applied to fit raw scores with age (separate fits for men and women when significant sex differences were found) and to calculate age‐corrected residuals. Pearson's correlations (males and females combined) were then calculated between these age‐corrected cognitive/demographic scores and age‐corrected residuals from DTI/DKI metrics of the whole hippocampus within three age groups: (i) childhood to adolescence (5–25 years, n = 141), (ii) young to middle adulthood (26–55 years, n = 129), and (iii) middle to later adulthood (56–90 years, n = 93). Significant correlations yielded p < 0.05 with FDR‐BH.

3. Results

3.1. Whole and Subfield Hippocampal Volume Trajectories Versus Age

Volume of the whole hippocampus and its subfields versus age best fit a concave (inverted U‐shaped) quadratic model, except for subiculum in males and CA2 for both males and females where no age‐related changes were observed (Figure 2). There were sex differences in the volume lifespan trajectories for whole hippocampus, subiculum, CA1, and CA4/DG where males showed larger volume; whole hippocampus and CA4/DG volumes were similar at young ages between males and females and then diverged at ~14 years for whole hippocampus and ~40 years for CA4 whereas CA1 volumes already differed at 5 years. Whole hippocampus volume showed a later maximum of 42 years for males than 36 years for females and, at its peak, was 13% larger than at 5 years old for males and 10% larger for females, and then it reduced from its peak to 90 years by −20% and −27% for males and females, respectively (Table 1). Different timing for volume peaks were found for subiculum (females, 33 years), CA1 (males, 41 years; females, 38 years), CA3 (all, 36 years) and CA4/DG (males, 42 years; females, 36 years) with females reaching the peak earlier than males in CA1 and CA4/DG (Table 2). The curve fits of the cross‐sectional data show an interesting observation that the hippocampal volumes by age 70 years and beyond become smaller than at 5 years. The best fitting parameters for volume of whole hippocampus and its subfields can be found in Supplementary Table S4.

FIGURE 2.

Age‐related volume changes over 5–90 years in left/right average of (A) whole hippocampus and (B–F) subfields for all 363 healthy volunteers combined when there are no sex differences (black‐dashed line), and males (blue) and females (red) fit separately. All of the fits followed a concave quadratic function (95% confidence interval shown) except for (B) subiculum in males and (D) CA2 in both males and females. The male curves peaked at higher volumes for (A) whole, (C) CA1, and (F) CA4/DG and later at ~42 years than females at ~35 years.

TABLE 1.

Best fits for left/right averaged whole hippocampal volume and DTI/DKI metrics versus age in all 363 volunteers aged 5–90 years when there are no sex differences or separate fits for males (M, n = 157) and females (F, n = 206) when the age trajectories differ.

Whole hippocampus metric	Best fit a	Fit value at 5 years	Age at min/max/plateau	Fit value Min/max/plateau value	% change 5 years to min/max/plateau	Fit value at 90 years	% change min/max/plateau to 90 years	Age of sex divergence
Vol (M)	Quad (inv U)	3.17	42 max	3.60	13	2.88	−20	> 14
Vol (F)	Quad (inv U)	3.05	36 max	3.30	10	2.43	−27
MD (All)	Gamma neg	0.89	37 min	0.80	−11	0.86	7	—
FA (All)	Gamma pos	0.18	25 max	0.20	6	0.18	−8	—
MK (M)	Exp pos	1.04	27 plateau	1.42	26	1.46	3	> 16
MK (F)	Exp pos	1.05	26 plateau	1.34	22	1.38	2
KFA (All)	Lin neg	0.31	—	—	—	0.26	−16 b	—

Note: The values at the youngest (5 years) and oldest (90 years) ages as well as at minima/maxima/plateau, % changes with age, and the age at which the male and female fit parameters diverge are shown. Ages are in years, volumes are in cm³, and MD has units of 10⁻³ mm²/s.

^a Quad (inv U), quadratic, inverted U; Gamma, gamma variate; Exp. pos, exponential positive; Lin neg, linear negative.

^b Percent change for KFA was measured from 5 years (max) to 90 years (min).

TABLE 2.

Best fits for left/right averaged hippocampal volume and DTI/DKI metrics versus age are given for the subfields subiculum, CA1, CA2, CA3, and CA4/DG in all 363 participants aged 5–90 years when there are no sex differences or separate fits for males (M, n = 157) and females (F, n = 206) when the age trajectories differ.

	Hippocampus subfield metric	Best fit a	Fit value at 5 years	Age (years) at min/max/plateau	Min/max/plateau value	% change from 5 years to min/max/plateau	Value at 90 years	% change from min/max/plateau to 90 years b	Age of fit divergence (years)
Subiculum	Vol (F)	Quad (inv U)	0.50	33 max	0.53	5	0.42	−20	—
	MD (All)	Gamma neg	0.85	39 min	0.73	−13	0.80	8	—
	FA (All)	Gamma pos	0.21	27 max	0.23	8	0.21	−8	—
	MK (M)	Exp pos	1.13	26 plateau	1.66	31	1.72	3	16
	MK (F)	Exp pos	1.13	24 plateau	1.55	27	1.60	3
	KFA (All)	Lin neg	0.37	—	—	—	0.31	−17	—
CA1	Vol (M)	Quad (inv U)	0.91	41 max	1.03	13	0.82	−20	5
	Vol (F)	Quad (inv U)	0.82	38 max	0.94	15	0.66	−30
	MD (All)	Gamma neg	0.87	39 min	0.78	−10	0.83	6	—
	FA (All)	Gamma pos	0.21	27 max	0.23	8	0.21	−8	—
	MK (M)	Exp pos	1.02	30 plateau	1.37	25	1.41	3	15
	MK (F)	Exp pos	1.03	29 plateau	1.30	21	1.33	2
	KFA (All)	Lin neg	0.38	—	—	—	0.32	−16	—
CA2	MD (All)	Gamma neg	1.00	33 min	0.91	−9	0.99	8	—
	FA (All)	Gamma pos	0.17	20 max	0.18	6	0.16	−11	—
	MK (M)	Exp pos	1.00	22 plateau	1.29	22	1.32	2	16
	MK (F)	Exp pos	1.05	26 plateau	1.25	16	1.27	2
	KFA (All)	Lin neg	0.31	—	—	—	0.25	−20	—
CA3	Vol (All)	Quad (inv U)	0.42	36 max	0.46	10	0.34	−27	—
	MD (All)	Gamma neg	0.96	34 min	0.85	−11	0.93	9	—
	FA (All)	Gamma pos	0.18	24 max	0.20	10	0.17	−12	—
	MK (M)	Exp pos	0.99	22 plateau	1.33	25	1.36	3	16
	MK (F)	Exp pos	1.03	24 plateau	1.28	19	1.30	2
	KFA (All)	Lin neg	0.34	—	—	—	0.28	−20	—
CA4/DG	Vol (M)	Quad (inv U)	0.47	42 max	0.58	24	0.4	−31	40
	Vol (F)	Quad (inv U)	0.48	36 max	0.54	13	0.35	−36
	MD (All)	Gamma neg	0.96	28 min	0.87	−9	0.96	10	—
	FA (All)	Gamma pos	0.16	24 max	0.18	12	0.15	−15	—
	MK (M)	Exp pos	1.00	22 plateau	1.30	23	1.33	2	16
	MK (F)	Exp pos	1.00	23 plateau	1.24	17	1.27	2
	KFA (All)	Lin neg	0.34	—	—	—	0.26	−26	—

Note: The values at the youngest (5 years) and oldest (90 years) ages as well as at minima/maxima/plateau, % changes with age, and the age at which the males and female fit parameters diverge are shown. Ages are in years, volumes are in cm³ and MD has units of 10⁻³ mm²/s.

^a Quad (inv U), quadratic, inverted U; gamma, gamma variate; Exp. pos, exponential positive; Lin neg, linear negative.

^b Percent change for KFA was measured from 5 years (max) to 90 years (min).

3.2. Whole Hippocampus DTI/DKI Trajectories Versus Age

The DTI and DKI metrics measured in the whole hippocampus showed four distinct age‐related trajectories (Figure 3). MD followed a U‐shaped Gamma variate fit decreasing steeply during development and then increasing slowly at older ages. FA showed an inverted U‐shaped Gamma variate fit increasing in early ages, peaking in adulthood (earlier than MD) and then decreasing in older ages. MK followed an exponential fit, increasing sharply during childhood and adolescence, then plateauing in adulthood. In contrast, KFA declined linearly across the lifespan. The best fitting parameters for whole hippocampus DTI/DKI trajectories can be found in Supplementary Table S4.

FIGURE 3.

Whole hippocampus (average left and right) curve‐fitting results (including 95% confidence intervals) across 5 to 90 years for (A) MD (U‐shaped Gamma), (B) FA (inverted U‐shaped Gamma), (C) MK (exponential), and (D) KFA (linear) of all 363 healthy volunteers combined (black‐dashed line), and males (blue) and females (red) fit separately. Whole hippocampus (A) MD decreased in childhood, reached a minimum at 37 years old and then increased with aging. The opposite happened with (B) FA, where it increased during childhood until 25 years, and then decreased with aging. There were no statistically differences in the fits of males and females for MD and FA. (C) MK sharply increased during childhood/adolescence and then reached a plateau at ~26 years with higher values in males. (D) KFA decreased across the whole lifespan with no sex differences.

Whole hippocampus MD (Figure 3A) reached a minimum at 37 years at which point MD was 11% lower than at 5 years old, then increased by 7% by age 90 (Table 1). Whole hippocampus FA (Figure 3B) reached a maximum at 26 years at which point FA was 7% higher than at 5 years old and 8% higher than at 90 years old (Table 1). No sex differences were observed in these two DTI metrics. Whole hippocampus MK was similar between sexes in early childhood, but differences emerged during adolescence, from approximately 15 years onward, which then “plateaued” around age 26 for the rest of the lifespan with MK consistently ~5% higher in males than in females, a difference that then persisted throughout the lifespan (Figure 3C). MK in males had a greater percentage increase from age 5 to the plateau (+26%) compared to females (+22%; Table 1). Whole hippocampus KFA (Figure 3D) showed a negative linear slope throughout the lifespan, with no sex differences (Table S3).

3.3. Unfolded Hippocampal Maps Reveal Spatial Heterogeneity of Diffusion Changes With Age

Unfolded hippocampal maps of all participants, averaged within six age bins and separated by sex, showed subfield variation across all ages for both DTI and DKI metrics (Figure 4). MD showed similar values for males and females across all age bins with MD higher in CA2/CA3/CA4/DG; the subiculum and CA1 MD was progressively lower going from 5–10 to 11–15 to 16–20 years old age bins (Figure 4A). FA was highest in the subiculum, while lower FA values were observed in CA1 and CA2 with similar values between sexes (Figure 4B). Similar to FA, MK was also elevated in the subiculum, which became greater with age becoming noticeable after age 20 years, following similar patterns in both sexes, although males had higher MK (Figure 4C). KFA was higher in CA1 and declined steadily with age (Figure 4D).

FIGURE 4.

Subfields show DTI/DKI parameter heterogeneity and different changes with age as shown in unfolded hippocampal maps averaged in six age bins (left and right combined) with similar age patterns for males (top rows) and females (bottom rows). (A) MD was higher in CA2/CA3/CA4/DG while (C) MK was greater in the subiculum and CA1, especially after 20 years. (B) FA was highest in the subiculum and lower in CA1 and CA2. (D) KFA was highest in CA1 across all age bins and values decreased consistently across the lifespan.

3.4. Hippocampal Subfields DTI/DKI Trajectories Versus Age

All four DTI/DKI metrics of the five hippocampal subfields (Figure 5) followed the same general age‐related trajectories as in those observed in the whole hippocampus (Figure 3). The best fitting parameters for all the hippocampal subfields DTI/DKI trajectories can be found in Supplementary Table S4. There were no significant sex differences for the two DTI metrics (Figure 5A–E and Figure 5K–O), nor KFA (Figure 5P–T). This is in notable contrast to MK, which showed sex differences for all five subfields, with males demonstrating higher MK than females (Figure 5F–J). The corresponding p‐values and mean differences (males/females) of the age trajectories from the permutation test are provided in Supplementary Table S3. For all the DTI/DKI metrics, different subfields reached their min/max/plateau at different ages (Table 2).

FIGURE 5.

Hippocampal subfields (average left and right) curve fitting results (including 95% confidence intervals) across 5 to 90 years for MD, MK, FA, and KFA in subiculum, CA1, CA2, CA3, and CA4/DG of all 363 healthy volunteers combined (black‐dashed line), and males (blue) and females (red) fit separately when there is a significant sex difference. (A–E) MD followed a U‐shaped Gamma variate fit with no sex differences and highest MD values in (C) CA2. (F–J) MK followed an exponential fit with higher MK in males for all subfields and had highest values in (F) subiculum. (K–O) FA followed an inverted‐U‐shaped Gamma variate fit with no differences between sexes for all subfields with the highest value in (K) subiculum. (P–T) KFA followed a negative linear fit for all subfields with no sex differences and highest values in (P) subiculum and (Q) CA1.

Subiculum showed the lowest MD values (MD ~0.76 × 10⁻³ mm²/s, over 5–90 years), while CA2 had the highest MD (~0.94 × 10⁻³ mm²/s, over 5–90 years). Greater percentage changes in MD were observed from age 5 years to the minimum point, except in CA4/DG, which showed more pronounced changes later in life (Table 2). The subiculum also had the highest FA (~0.22, over 5–90 years) and MK (males, ~1.60; females, ~1.52; over 5–90 years) with greater FA percentage change from the maximum to 90 years old (Table 2). Males exhibited higher MK than females after ~16 years across all of the subfields, with CA2 showing the lowest MK values (males MK ~1.28, females MK ~1.23, over 5–90 years). The lowest KFA was observed in CA2 (KFA ~0.29, over 5–90 years), while the highest KFA were in subiculum and CA1 (KFA ~0.35, over 5–90 years). For all subfields, MD reached its minimum later than when FA/MK reached their peak/plateau (Table 2). Over all metrics, MK had the greatest % changes from 5 years to min/peak/plateau relative to MD, FA, and KFA for all subfields, yet MK has minimal change with aging, unlike the other three parameters.

3.5. Whole Hippocampus Correlations With Cognitive Scores and Demographics

Of the 120 linear correlations (8 cognitive/demographics × 5 whole hippocampus volume/DTI/DKI metrics × 3 age groups) tested between age‐corrected cognitive scores, age‐corrected demographics, and DTI/DKI/volume residuals, 15 remained significant after FDR‐BH correction (Figure 6A). All significant associations involved demographic residuals, while cognitive scores showed no surviving correlations. Among the significant findings, hippocampal volume residuals were only correlated with BMI in the younger group (young R = 0.23, p = 0.04). BMI residuals showed consistent associations with multiple hippocampus parameters across age groups. In the young and middle age groups, BMI residuals were negatively correlated with MD residuals (young R = −0.48, p < 0.001; middle R = −0.34, p < 0.001) and positively correlated with FA residuals (young R = 0.25, p = 0.03; middle R = 0.34, p < 0.001) and MK residuals (young R = 0.50, p < 0.001; middle R = 0.46, p < 0.001). Interestingly, in the older group, only MK residuals were positively correlated with BMI residuals (old R = 0.37, p = 0.02), which is the only correlation seen in all three separate age groups (Figure 6B). Heart rate residuals showed a positive correlation in the younger group with MK residuals (young R = 0.24, p = 0.03). Waist circumference residuals in the young and middle groups showed the same associations seen with BMI residuals, which were negatively correlated with MD residuals (young R = −0.47, p < 0.001; middle R = −0.34, p < 0.001) and positively correlated with FA residuals (young R = 0.24, p < 0.04; middle R = 0.38, p < 0.001) and MK residuals (young R = 0.47, p < 0.001; middle R = 0.47, p < 0.001).

FIGURE 6.

(A) Linear correlations of whole hippocampus volume, MD, FA, MK, and KFA (age‐corrected residuals) versus five age‐normalized cognitive scores and three demographic variables (age‐corrected residuals) in three age groups of younger (5–25 years, n = 141), middle (26–55 years, n = 129), and older (56–90 years, n = 93) with significant correlations shown in red (positive) or green (negative; p < 0.05 with FDR‐BH correction). Eight of the correlations were with BMI residuals across all three age groups including with volume residuals (positive, younger only), MD residuals (negative, younger and middle), FA residuals (positive, younger and middle), and (B–D) MK residuals (positive, younger, middle, and older). Another six correlations were similar for waist circumference residuals in the younger and middle‐aged groups and another one with heart rate residuals versus MK residuals in the younger group.

4. Discussion

This study applied high‐resolution DTI and DKI to the whole hippocampus and its subfields (subiculum, CA1, CA2, CA3, and CA4/DG) to investigate healthy development and aging across 5–90 years. Volumes of the whole hippocampus and each subfield were automatically measured on the 1 mm isotropic DWI (b = 500 s/mm²) using HippUnfold [45], which showed quadratic inverted U‐shaped age trajectories that varied across subfields with sex differences in whole hippocampus, subiculum, CA1, and CA4/DG. These volume age trajectories differed markedly from the DTI/DKI metric age trajectories with Gamma variate (negative—MD, positive—FA), exponential positive (MK), and linear negative (KFA) across all subfields, with only MK showing sex differences (males higher with age). The quadratic peaks for volume were at ~35–42 years, where the volume drops at older ages were greater than the increases in children, in agreement with earlier work [25, 26, 31]. In contrast, the MD/FA/MK fits showed steeper changes from childhood/adolescence to young adults as compared to the more gradual reversal (or flat for MK) towards old age. All three age ranges examined showed linear correlations of BMI with total hippocampal volume (+), MD (−), FA (+), and MK (+) in the younger ages (5–25 years), MD, FA, and MK in middle ages (26–55 years), and only MK for the older ages (56–90 years). MK showed the strongest correlations and was evident across all three independent age groups.

Given prior histological links of cerebral MK with myelin causing greater restrictive barriers to water diffusion [5, 6, 7, 8], the marked rise of MK (and reduction of MD and increase of FA) during childhood/adolescent development towards young adulthood is consistent with increased myelination within the hippocampus [51]. Other multi‐shell, high b‐value diffusion studies of healthy individuals have used the NODDI model showing increases of the model parameter “neurite density” in hippocampus either linearly over 8 to 14 years [52] or nonlinearly from 8 to 21 years leveling off after ~16 years [35], although both had limited spatial resolutions of 2.2 or 1.5 mm isotropic resolution, respectively. Furthermore, NODDI‐derived metrics should be interpreted cautiously, particularly in GM structures like the hippocampus, where the assumptions of the model are not fully satisfied, and where the parameter estimation may be biased by model degeneracy [53]. In agreement with this latter observation, MK here stays stable for both males and females after ~30 years of age whereas MD goes up and FA down, thereby arguing against demyelination in aging, but perhaps is suggestive of neuron/axon/synaptic loss [54] or alterations in the abundance of astrocytes, oligodendrocytes, and microglia [55], as opposed to inflammation, given the concurrent reduction of hippocampus volume. This is at odds with the conclusions of a HippUnfold study showing reductions with aging of other quantitative MRI metrics like magnetization transfer (MT_sat) and longitudinal relaxation rate (R₁) in healthy older adults (57–87 years) at risk of Alzheimer's disease. These findings were interpreted as a loss of macromolecular tissue content, presumably demyelination in the aging hippocampus (although the authors point out that factors other than demyelination could be at play) [56]. A loss of hippocampal cell density with typical aging has also been suggested by the SANDI diffusion model over 19–85 years, where the model parameter “soma fraction” decreased with age along with increases of “extracellular diffusion coefficient” and “extracellular volume fraction” in subiculum, CA4, and DG, although the 2 mm isotropic data was rather low resolution for such subfield‐specific analysis [37]. Notably, the supplementary data of their paper showed no age changes in MK over 19–85 years, in agreement with our data, but only one subregion (subiculum) with positive linear correlation of MD versus age, and none with FA, both in contrast with our work.

The directional variance of the kurtosis tensor, given by KFA, shows a negative linear correlation over the entire age range from 5 to 90 years and is clearly sensitive to a different aspect of hippocampal development, presumably related to changes in the complex tissue architecture such as crossing fibers (in development) and/or additional isotropic compartments (in aging) [2, 3, 4]. A theoretical situation was highlighted in the original KFA article where a decrease in the number of crossing fibers could result in an increase of FA with a decrease in KFA [2], a pattern which was observed in the younger individuals here. The hippocampus KFA being greater than FA in the human brain here is expected and consistent with a prior fixed rat brain (see Figure 1 in Hansen and Jespersen [4]). The elevated MK and reduced KFA with development could be consistent with prior histological CA1 observations where older rats (24 months old relative to younger 2‐month‐old rats) had longer basal dendrites and entire neurons and increased complexity of CA1 pyramidal neurons [57]. The plateau in MK suggests these hippocampal neurons have reached their final mature form. MK had similar values between males and females in young children, which then diverged during childhood/adolescence with males staying higher for the rest of the lifespan; this is consistent with sex differences in dendritic structures (more Sholl intersections and spine density in males) of CA1 and CA3 pyramidal neurons that vary by age in rat hippocampus [58]. This latter study also showed that young adult female rats had greater dendritic complexity compared to adolescent female rats. One of the primary limitations of this work is the nonspecific nature of the DTI/DKI metrics, which limits the interpretation of the underlying microstructural changes. Although the plateau observed in MK with age may suggest that hippocampal neurons have reached their final mature form, it could also reflect the integrative nature of the DKI representation, where opposing microstructural effects counterbalance one another. Future work could explore the application of biophysical models to further investigate hippocampal microstructure.

The hippocampus exhibits variation in cell types/sizes/connectivity across its subregions [59, 60]. This is reflected in the 1 mm isotropic diffusion DTI/DKI HippUnfold maps that enable the visualization of marked heterogeneity of these metrics across the unfolded hippocampus with elevated MD in CA2, elevated FA in the subiculum and a posterior region of CA2/3, elevated MK in the subiculum, and elevated KFA in CA1 primarily although also in the subiculum. The DTI‐derived MD and FA patterns agree with Human Connectome Project (HCP) young adult data (1.25 mm isotropic) using HippUnfold [61]. Two NODDI studies have shown the highest neurite density index (NDI) in the subiculum across the hippocampus as well as the highest orientation dispersion index (ODI) in CA1 [35, 61], both notably in line with the regions identified in the present study with the highest MK and KFA, respectively. Both kurtosis and NDI rely on restriction leading to a levelling off of signal at high b‐values, while KFA and ODI are driven by directional variance. The interpretation of elevated MK in the subiculum attributable to myelin is consistent with HippUnfold studies of other image contrasts using high‐resolution (≤ 1 mm isotropic) that also yielded their highest values in the subiculum such as T1w/T2w signal [45, 61], as well as R1, MT_sat, and R2* [56], although all these values are nonspecific for myelin. Prior work using HippUnfold and the SANDI model on low‐resolution 2 mm isotropic diffusion data reported marked heterogeneity in the SANDI metrics across hippocampal subfields, such as elevated soma fraction in CA1, increased soma radius and extracellular volume fraction in CA2, and higher neurite fraction in the subiculum [37]. Their supplementary analyses further demonstrated elevated MK in the subiculum and no age‐related changes in MK from 19 to 85 years, findings that are consistent with the observations of the present study.

Across development and aging from 5 to 90 years, all five hippocampal subfields showed DTI and DKI changes similar to those observed for the whole hippocampus, although the actual values were offset across subfields. This is in concordance with a NODDI development study over 8–21 years where all subfields showed nonlinear age increases of NDI with the same timing; however, the ODI metric only changed with development in subiculum and CA1, which does not agree with our observations of marked DTI/DKI changes in CA2/3/4/DG [35]. The SANDI aging study over 19–85 years did not see age‐related changes in all the subfields with various metrics showing linear changes primarily in subiculum, CA4, and DG; note that CA2 and CA3 showed no age changes unlike our study [37]. Our observation of elevated FA in all subfields during development is in opposition to a prior study reporting lower FA in subfields over 6–30 years [32]. One aging study from 50 to 75 years showed limited regional age changes with an MD increase in CA2–3 and FA decrease in subiculum [33]. Another study from 60 to 85 years showed elevated MD with age mainly confined to the hippocampal body and tail in CA1, CA3, and the subiculum [34]. The most obvious hippocampal subfield change with age in the present study was MK in the subiculum, which showed marked increases during development that leveled off after the 20s (higher in males). This may reflect changes due to the increase of fibers (possibly also myelination) in the perforant path that passes through the subiculum, as seen on polarized light imaging [62], reflecting enhanced circuitry changes of this critical hippocampal hub [63].

Most DKI studies of development and aging have focused on WM showing MK increases with development in young children from birth [10, 11] and children to adulthood [12] followed by MK decreases during adult aging [13, 14, 15, 16, 64, 65], although one study reported 4 WM regions with MK increases with aging, with the other 13 regions showing no change [66]. One small study on 80 individuals covered more of the “lifespan” from 10 to 79 years (although only 12 participants under 19 years) and showed a mix of MK versus age relationships such as WM regions with quadratic inverted U‐shaped trajectories peaking around 40 years and other regions decreasing linearly [17]. It is notable that a large sample of 651 typical adults from ages 18 to 88 years showed marked nonlinear MK reductions in WM mainly after ~60 years, which looked nearly identical to the NODDI‐derived NDI [15]. Fewer studies have examined deep GM (including thalamus, putamen, caudate, and globus pallidus) showing MK increases during development [10, 11, 17] and reductions with aging [17, 64], although one study showed three out of four deep GM regions with no MK change with aging (the other was putamen with MK increases) [65]. Our MK findings on hippocampus agree with the general observations of MK increasing in WM and other deep GM regions during development, but the aging results with lack of hippocampal MK changes here agree somewhat with the latter study but are not in line with the rest. Few DKI studies have examined sex differences, and those that have focused primarily on WM, reporting higher MK in females than males between ages 13–15 years [67] and across 18–94 years [16], in opposition to the hippocampal findings here with males exhibiting higher MK than females after ~16 years old. Interestingly, a study using the SANDI model on 2 mm isotropic resolution diffusion MRI in healthy adolescents aged 8–19 years reported similar neurite fractions between males and females at younger ages, with differences emerging after ~12 years old, at which time males showed higher neurite fractions than females [36]. KFA followed a negative linear fit over 5–90 years decreasing across the lifespan for whole hippocampus as well as its subfields with no sex differences. KFA is rarely reported, yet our findings are in disagreement with two studies albeit in WM: one with KFA increases in 20 WM regions between children (9–12 years) and adults (38–64 years) [12] and another with no age associations from 17 to 70 years for KFA in 16 out of 17 WM regions (one linear decrease in genu of corpus callosum) [66].

Few studies have reported DKI metrics in the hippocampus, with most studies typically using a whole hippocampus segmentation and focused on MK rather than KFA, and none of them analyzing DKI changes across development and aging (5–90 years). Across existing studies, reported MK values in the hippocampus have been relatively consistent: MK ~0.8 in healthy adolescents (n = 30) aged 12–18 years [68], MK ~0.8 in healthy adults (n = 20) with a mean age of 70 ± 5 years [69], and MK ~0.75 and KFA ~0.31 in adults (n = 34) aged 50–78 years [70], and MK ~0.72 in healthy adults aged 73± years [71]. A subfield analysis reported MK ~0.65 in the subiculum, MK ~0.6 for CA1 and MK ~0.57 for the rest of the subfields across 19–85 years [37]. All these values are lower than those reported in our study for the same age ranges. However, prior studies have substantially lower spatial resolution (voxel volumes > 8 mm³) than the 1 mm³ used here. Although MK has been shown to be relatively insensitive to partial volume effects [72], partial volume could still be an issue in ROI analysis of voxels adjacent to low MK CSF, such as in the hippocampus. Nevertheless, the most plausible explanation for the elevated MK in the present study is the noise floor and Rician noise observed when SNR is low at high spatial resolution, which can potentially bias MK measurements to higher values. Efforts to minimize these effects were done by using a conservative b‐value of 2000 s/mm² (instead of a higher b value) and by the application of a postprocessing correction strategy [41]. While Rician bias correction is included in the processing pipeline, it may not fully eliminate noise‐related bias, which could contribute to the MK overestimation. However, there is no evidence to suggest that this bias varies as a function of age or sex. Supplementary Figure S2 shows MK and KFA mean hippocampal values across different resolutions, before and after correction in a 29‐year‐old healthy subject. Whole hippocampus MK values increased, going from 3/2 mm isotropic (~0.8) to 1.5 mm isotropic (~0.9). The uncorrected 1 mm isotropic showed considerable elevated hippocampal MK values (~1.6), whereas the corrected map yielded a lower value (~1.3). KFA values increased progressively with higher resolution.

There was no correlation between whole hippocampus DKI/DTI metrics and cognitive scores such as memory in any of the three independent age groups (young, middle, and older). In contrast, there were significant correlations (accounting for age and sex) between BMI residuals and several hippocampal measure residuals including volume (young—positive), MD (young, middle—negative), FA (young, middle—positive), and MK (young, middle, older—positive). A prior study reported reduced left hippocampal volume in obese children aged 8–12 years when compared with healthy‐weight children [49]. Similarly, a weak negative association (r = −0.12, p < 0.01) between BMI and left hippocampal volume has been observed in children and adolescents aged 8–19 years [73]. Furthermore, a review article found that 11 of 28 studies reported a negative association between adiposity and hippocampal volume [74]. All these findings contrast with the positive relationship observed in the young group in our study. Other studies have found that BMI was associated with reduced hippocampal T2‐weighted intensity (both left and right separately), which was interpreted as an accumulation of macromolecules such as lipids, but there were no relationships with volume [75]. A previous study analyzed two independent datasets and reported contrasting associations between BMI and DTI metrics where the Human Connectome Project (HCP) database in healthy participants aged 22–37 years (n = 819) showed a negative association between BMI and MD/FA of the hippocampus whereas the UK Biobank cohort in healthy participants aged 45–82 years (n = 26,242) revealed a positive association between BMI and FA [76]; their HCP MD finding and their FA UK Biobank finding is in agreement with our results in the young and middle aged cohorts. Similarly, an HCP‐Aging study from 36 to 100 years (n = 494) showed relationships of imaging (including higher T1w/T2w, lower MD, higher FA) in both left and right hippocampus with increased BMI [77]. The observed DTI/DKI associations in the present work may reflect obesity‐related hippocampal neuroinflammation and gliosis, which are known to restrict water diffusion, in both human [78, 79] and animal models [80, 81]. Additionally, a “western diet” and benzodiazepine exposure rat model of obesity showed an elevation of diffusion kurtosis in the hippocampus, which was associated with negative effects on memory [82]; notably, MK was higher with greater BMI for all three age groups here.

In summary, high‐resolution 1 mm isotropic DKI of the hippocampus provides sensitive and complementary information about hippocampal microstructural development and aging across 5–90 years. Distinct whole hippocampus age trajectories were observed for DTI, DKI, and volume, with sex differences evident only for MK. Hippocampal subfields followed similar age trajectories to the whole hippocampus, and the unfolded maps revealed marked heterogeneity that highlights the importance of high spatial resolution. Although it is not possible to be certain regarding the underlying microstructural changes, these results suggest that DKI may provide novel insight into age‐related microstructural changes at a sub‐hippocampal level. These results establish a normative baseline for future comparisons to neurodevelopmental and neurodegenerative disorders affecting the hippocampus, such as temporal lobe epilepsy and Alzheimer's disease.

Author Contributions

P.S.‐S.: methodology, formal analysis, investigation, writing – original draft, visualization. A.R.K.: software, writing – review and editing. D.W.G.: conceptualization, supervision, writing – review and editing, funding acquisition. C.B.: conceptualization, project administration, supervision, writing – review and editing, funding acquisition.

Funding

University Hospital Foundation and Women's & Children's Health Research Institute for operating funds (coauthor C.B.), Canada Research Chairs (coauthors C.B. and A.R.K.), Canadian Institutes of Health Research (CIHR) (coauthors C.B. and D.W.G.), and Secretaria de Ciencia Humanidades Tecnologia e Inovacion (SECIHTI Mexico) (coauthor P.S‐S.).

Conflicts of Interest

The authors declare no conflicts of interest.

Supplementary Material

Supplementary material includes detailed tables on test/retest analysis, hemispheric and sex differences, and parameters of the best fit trajectories for all reported metrics. Figures show left/right trajectories and MK and KFA maps at different spatial resolutions, with and without Rician bias correction.

Supporting information

Table S1: nbm_70355‐sup‐0001‐Supplementary_Material.docx. Mean absolute differences for volume and DTI/DKI metrics for whole hippocampus and the subfields (subiculum, CA1, CA2, CA3, and CA4/DG) from the test/retest reproducibility analysis of 24 healthy participants (12 males, aged 20–49 years; 12 females, aged 20–50 years). Any left/right or subfield differences below these thresholds were considered not significant.

Table S2: Analysis of hemispheric asymmetry for volume and DTI/DKI metrics for whole hippocampus and the subfields (subiculum, CA1, CA2, CA3, and CA4/DG) in all 363 participants aged 5–90 years. Significant left/right hippocampal differences were observed for 5 of 6 volume measurements, 5 of 6 MD measurements, 1 of 6 FA measurements, 6 of 6 MK measurements, and 3 of 6 KFA measurements.

Table S3: Permutation test for sex differences in volume and DTI/DKI metrics for whole hippocampus and the subfields (subiculum, CA1, CA2, CA3, and CA4/DG) where the mean difference between males and females, age trajectories, and p values are shown.

Table S4: Best fits with relevant parameters for left/right averaged whole hippocampal volume and DTI/DKI metrics versus age in all 363 volunteers aged 5–90 years when there are no sex differences or separate fits for males (M, n = 157) and females (F, n = 206) when the age trajectories differ.

Figure S1: Right (green) and left (orange) hippocampal curve fitting results (including 95% confidence intervals) across 5 to 90 years for volume, MD, FA, MK, and KFA in whole hippocampus, subiculum, CA1, CA2, CA3, and CA4/DG versus age over all 363 healthy volunteers either combined (black‐dashed line) or fit separately when there was an overall significant group difference as in Supplementary Table 2. The magnitude of the asymmetries was generally small and the whole hippocampus and its subfields exhibited similar age‐related trajectories in the left and right hippocampi (e.g., MK increasing during childhood and adolescence before reaching a plateau).

Figure S2: Mean DWI (b = 500 s/mm² and b = 2000 s/mm²; scaled independently), MK and KFA maps from a 29‐year‐old healthy subject with different image resolutions ranging from 3/2/1.5 mm isotropic (no Rician noise bias correction) to 1 mm isotropic (before and after Rician noise bias correction). Improved visualization of the hippocampus (necessary to measure subfields) as well as contrast between gray matter and white matter is observed in MK when increasing the resolution. Whole hippocampus MK values increased going from 3/2 mm isotropic (~0.8) to 1.5 mm isotropic (~0.9). The uncorrected 1 mm isotropic showed considerable elevated hippocampal MK values (~1.6), whereas the corrected map yielded a lower value (~1.3). KFA values increased progressively with higher resolution.

Data Availability Statement

The analysis scripts of this study are available from the corresponding author upon request. Participant MRI images and maps are not publicly available due to ethical considerations.