Quantitative T1 and T2 MRI signal characteristics in the human brain: different patterns of MR contrasts in normal ageing

Objective The objective of this study was to examine age-dependent changes in both T1-weighted and T2-weighted image contrasts and spin-echo T2 relaxation time in the human brain during healthy ageing. Methods A total of 37 participants between the ages of 49 and 87 years old were scanned with a 3 Tesla system, using T1-weighted, T2 weighted and quantitative spin-echo T2 imaging. Contrast between image intensities and T2 values was calculated for various regions, including between individual hippocampal subfields. Results The T1 contrast-to-noise (CNR) and gray:white signal intensity ratio (GWR) did not change in the hippocampus, but it declined in the cingulate cortex with age. In contrast, T2 CNR and GWR declined in both brain regions. T2 relaxation time was almost constant in gray matter and most (but not all) hippocampal subfields, but increased substantially in white matter, pointing to an age effect on water relaxation in white matter. Conclusions Changes in T1 and T2 MR characteristics influence the appearance of brain images in later life and should be considered in image analyses of aged subjects. It is speculated that alterations at the cell biology level, with concomitant alterations to the local magnetic environment, reduce dephasing and subsequently prolong spin-echo T2 through reduced diffusion effects in later life. Electronic supplementary material The online version of this article (doi:10.1007/s10334-016-0573-0) contains supplementary material, which is available to authorized users.


Introduction
Normal ageing is associated with inevitable loss of both gray matter (GM) and white matter (WM) tissue in the brain. In addition to macroscopic morphological alterations, cerebral metabolism and haemodynamics [1], tissue and cell microstructure [2], macromolecular composition and interactions [3,4], and chemical composition, such as water [5] and iron content [6,7], also undergo changes with age. The deviation of one of these particular structural or functional parameters beyond its expected value for a given age range may be indicative of disease, of which the However, it is not yet entirely clear what dominates the various common MR contrasts, or what physiological or chemical changes are most likely to elicit observable changes. Some work has nonetheless shown that there is likely to be substantial utility in making further investigations. Salat et al. observed a relationship between gray-towhite matter intensity ratio (GWR) and cortical thickness from T 1 -weighted images [9], and noted that this was statistically stronger than the thinning of cortical GM in ageing. The observation led to modification of a previous algorithm for the estimation of cortical thickness, making use of GWR, which led to improved discrimination between Alzheimer's disease (AD) and control subjects [21]. It was subsequently observed [22] that decreases in T 1 -weighted image contrast were apparent in the hippocampus and limbic system as a whole in AD. Increases in GWR (decreased contrast-to-noise ratio = CNR) were also associated with decreased hippocampal volume. Given the role of the hippocampus in various diseases, there is a clear imperative to better understand how those changes can be detected in an MR image at the earliest stage.
The basis for changes in T 1 -weighted MRI signal intensities and relaxation times with age are poorly understood. Attempts have been made to explain the T 1 in terms of iron content [7]. A correlation exists between 1/T 1 and iron content, as deduced from analyses on post-mortem specimens, but the relationship to GWR or CNR was not sought. Similarly, decreases in magnetisation transfer (MT) both in GM and WM with age have been reported, potentially influencing T 1 contrast. Altered MT may be due to alterations in the interactions between water and macromolecules [3,4].
T 2 -weighted MRI is also gaining applications in highresolution imaging for segmentation of deep GM structures, such as basal ganglia, and subsections of the hippocampal formation in aged subjects. A recent review of approaches to hippocampal subfield labelling showed that 17 of the 21 available studies use T 2 -weighted images [23]. T 2 -weighted contrast is also affected by age. Magnaldi et al. [24] first showed that T 2 -weighted contrast declined with age for various regions of the brain, including external capsule, internal capsule, corpus callosum and periventricular WM. Later, measurements of the GWR were made between WM and various subcortical structures as well as cortical GM [25]. Generally decreasing trends were found with age in T 2 -weighted images.
The ability to infer information on the properties of tissues directly from simple measures of image contrast as well as from relaxometry (and other quantitative techniques) can be seen as a way to infer its health status. Moreover, in many pathologies, microscopic or molecular changes are likely to precede macroscopic (volumetric) changes. With this in mind, an understanding of the mechanisms by which ageing determines the appearance of T 1 and T 2 MR images is important in their interpretation. The objective of this study was to examine tissue contrast and quantitative spin-echo T 2 in cognitively normal, aged subjects in order to better understand the changes in MR images in different brain regions with age. We examine the contrast between individual subfields of the hippocampus and their relaxometric properties, the cingulate gyrus, caudate nucleus, and corpus callosum. The hippocampus is one of the most vulnerable brain structures in AD pathology, along with the limbic structures, including the cingulate cortex, and deep GM structures, including the caudate nucleus, also becoming affected during the progression of AD. The purpose of our analysis is to guide interpretation of MR images in a more "pathology-driven" context, aiming for a characterisation of tissue properties to complement volume and shape analyses.

Image acquisition and participant cohort
Our participant cohort comprised a total of 37 persons (22 females, age range from 49 to 87 years, mean age 67.3 years). Participants were required to have no known psychiatric or neurological morbidities. Participants gave informed consent and ethical approval was granted by the NHS Research Ethics Committee of North Bristol-Frenchay.

Image processing
In all scans, the manufacturer's procedure for correction of differential coil sensitivity (the "prescan normalize") was used to avoid, to the maximum possible extent, shading in different regions of the image that would compromise estimates of image CNR and GWR. This is performed at image reconstruction time and uses knowledge of coil sensitivity profiles.
Quantitative T 2 maps were generated by a voxel-wise fit by a mono-exponential function in a logarithmic space after exclusion of the first echo to avoid the effects of stimulated echoes. This was performed using software written in-house. Each entire echo train was then summed to create a single T 2 -weighted image complementary to the T 2 map. To further reduce the impact of all sources of image shading (which derive from B 1 transmit and B 0 inhomogeneity as well as the different receiver coil sensitivity profiles in the 32-channel array), bias field corrections were applied to the T 1 -weighted (MPRAGE) and T 2 -weighted (spin-echo) images using FSL [26]. This was not applied for fitting of T 2 or diffusion tensor maps, nor to the resulting quantitative maps. Our available images were therefore a T 1 -weighted MPRAGE, T 2 -weighted spin-echo image, and T 2 map. This was also reciprocated to create an R 2 map for image registration purposes.

Preparation and validation of hippocampal subfield masks
Masks of the various regions used were drawn manually in native space on echo-summed T 2 -weighted images. We selected subfields that could be reliably demarcated whilst providing a basis for contrast comparisons. The masks used were of the central slices of the hippocampal CA1, DG, and SL/SR/SM subfields generated from T 2 -weighted images according to the recent manual protocol [27] (as well as the total hippocampus). The motivation for choosing these temporal lobe structures is that hippocampal atrophy is a well-established finding in AD [8] and that the CA1 subfield has been reported to be affected in the early phase of the disease whereas DG is preserved [28]. The CA1 and DG are predominantly GM whereas the SL/SR/SM is predominantly WM. In our images, we were unable to distinguish between the individual stratum lacunosom (SL), stratum radiatum (SR), and stratum moleculare (SM), for which reason these three layers were collectively masked as the SL/SR/SM subfield [27]. The CA1 and DG subfields border the SL/SR/SM subfield (from opposite sides), such that comparison of the relative contrast between these two selected GM subfields with the (WM) SL/SR/SM subfield are useful comparisons. The subiculum borders the CA1, though this border is the most variable in the literature. There is no subiculum-SL/SR/SM border. The CA2 and CA3 subfields are small and difficult to reproducibly mask. The CA1, DG, and SL/SR/SM therefore fulfil the criteria of providing both GM (CA1, DG) and WM (SL/SR/SM) to contrast, being mutually adjacent and containing sufficient voxels for reliable estimates of signal intensity. A random subset of hippocampal subfield masks was re-prepared both by a second rater and by the original rater in order to determine consistency of boundary placement. Consistency was judged based on the Dice-kappa statistic and intra-class correlation [23].

Preparation of masks of other regions
We also prepared masks of the cingulate gyrus, the WM immediately posterior to the cingulate gyrus, the caudate nucleus, globus pallidus, and corpus callosum. Masks of the cingulate gyrus were defined directly above the hippocampal body and spanned five slices. For use on T 1weighted images, R 2 maps (R 2 = 1/T 2 ) were registered to T 1 -weighted space and the transformations applied to the masks originally prepared in T 2 -weighted image space. Some manual alteration was necessary since image transformations are never perfect. R 2 maps were used for registration to the T 1 -weighted space due to the similarity in contrast between those image types. Example masks are shown in the Supplementary Information.

Parameterising changes in signal intensity
We used two metrics of change in relative signal intensity between two regions of interest (ROI), the CNR and GWR. The CNR is given by: with E the expectation (mean) and σ 2 the variance, whilst S G and S W contain the signal intensities in the GM and WM regions of interest, respectively. The GWR is simply: with the terms defined as above. Note that S G and S W can be quantitative T 2 or image intensities. GWR for quantitative T 2 images was calculated using the formula above by replacing S G and S W by respective T 2 values.
In order for these formulae to represent CNR and GWR, rather than spurious aspects of imaging instrumentation, contributions to the image intensities sampled by the ROI must contain negligible contributions arising from B 0 and B 1 inhomogeneity, from parallel imaging reconstruction routines, and in multi-slice 2D imaging from slice profile imperfections (by using the inner slices only where these are equilibrated). The mean and variance terms are otherwise corrupted by contributions that cause CNR and GWR to represent non-local instrument instability, rather than anything of physiological origin or interest. This may be accomplished by ensuring that ROIs are anatomically close and smaller than the size over which one may anticipate the effects of magnetic or radiofrequency field inhomogeneity to be manifest. ROIs must be representative of signal variance arising due to random sources of noise, rather than representative of anatomical shape. This has motivated our choice of ROI. More is said of this in the discussion section. Statistical analysis was performed using Matlab R2013b. Linear fits for all figures presented in the results were performed, with confidence intervals for fitted parameters and function prediction intervals computed by 500 bootstrap simulations. P-values were also computed. A table of statistics and linear fit parameters is provided in the Supplementary Information.

Results
T 1 -weighted and T 2 -weighted images highlight the change in global image contrast that accompanies ageing (Fig. 1). A decline in general image contrast across the images, in addition to conspicuous macroscopic structural alterations, for both T 1 -weighted and T 2 -weighted images, is evident with increasing age. Figure 2 shows the age dependence of the CNR and GWR in T 1 and T 2 -weighted images in the hippocampal body, as well as in quantitative T 2 maps, contrasting the (GM) CA1 and DG subfields with the (WM) SL/SR/SM subfield. We observed an effect of age in the T 2 -weighted CNR and GWR, but not in T 1 -weighted images. T 2weighted CNR between either CA1 or DG with the SL/SR/ SM subfield was also generally higher than T 1 -weighted CNR. The measurements presented in Fig. 2 may be corrupted by poor placement of hippocampal subfield boundaries, for which reason we assessed inter-rater and intra-rater reliability. For intra-rater analysis, the average measured intraclass correlation (ICC) was 0.993 with a 95 % confidence interval from 0.986 to 0.996 (F(62, 62) = 168.911, p < 0.001). For inter-rater analysis, the average measured ICC was 0.946 with a 95 % confidence interval from 0.842 to 0.982 (F(14, 14) = 19.150, p < 0.001). The Dice Kappa statistics for a subset of six hippocampi were also calculated after re-segmentation by the original rater. For those subfields used in this paper, we obtained the following Dice Kappa results. CA1: mean 0.76 with a 95 % confidence interval from 0.67 to 0.98. DG: mean 0.77 with a 95 % confidence interval from 0.58 to 0.99. SL/SR/SM: mean 0.77 with a 95 % confidence interval from 0.64 to 0.95. An additional Monte-Carlo analysis, by which means boundaries were perturbed computationally, can be found in the Supplementary Information. Measures of inter-subfield contrast were reasonably stable under perturbations to the boundaries of sizes, consistent with the discrepancies in labelling across the literature [23].
We also compared the effects of age in the CNR and GWR in T 1 -weighted and T 2 -weighted images for the cingulate gyrus (Fig. 3), a structure showing atrophy in AD [29]. Substantial effects of age are quantifiable for both contrasts. A clear normalisation of GWR towards unity is also seen for both MR contrasts with increasing age, with rather stronger correlations than in the hippocampus. The CNR and GWR obtained using quantitative T 2 maps as a function of age are shown (Fig. 3e, f). The T 2 GWR here is simply the ratio of the mean T 2 between GM and WM, whereas the T 2 CNR is the absolute difference normalised by root variance. Thus, the T 2 in the GM of the cingulate It is instructive to consider whether age-related differences in metrics of relative signal intensity, i.e., CNR and GWR, are more dependent on relaxation characteristics of one or the other tissues being contrasted. Therefore, in Fig. 4 we present the T 2 values obtained in various regions, including caudate nucleus and globus pallidus (both belonging to deep GM) and the genu of corpus callosum (presenting myelinated WM). T 2 in the three subfields of the hippocampus was examined (Fig. 4a-c), and no effect of age was detectable. Likewise, the cingulate T 2 and caudate nucleus have consistent values with increasing age.  Fig. 2 CNR and GWR in the hippocampus plotted against age. The left panels show CNR measurements, whilst the right panels show GWR measurements. The uppermost three rows (a-f) contrast the CA1 and SR/SL/SM subfields, whilst the lower three rows (g-l) contrast the DG and SR/SL/SM subfields. Panels a, b show CNR and GWR, respectively, for T 1 -weighted images, whilst panels c, d show CNR and GWR, respectively, for T 2 -weighted images, contrasting the CA1 and SL/SR/SM subfields. Panels e, f show CNR and GWR, respectively, for T 1 -weighted images, whilst panels g, h show CNR and GWR, respectively, for T 2 -weighted images, contrasting the DG and SL/SR/SM subfields. The red line shows a linear fit, the magenta lines are the 95 % confidence bounds for observations, and the cyan lines are the 95 % functional prediction intervals obtained by biascorrected bootstrapping. The red line is solid if p < 0.05, and dashed otherwise. Fitted parameters can be found in the Supplementary  ◂ However, the WM adjacent to the cingulate, as well as the genu of the corpus callosum, show substantial increases in T 2 with age. An increase in that of the globus pallidus is also detected (all judged on the criteria p < 0.05 for a linear fit). Although the correlation between age and T 2 -weighted CNR is clear, it is not necessarily causal. To further analyse whether the GM or WM T 2 had more of an influence on contrast between the two tissue types, we present correlations between T 2 and T 2 -weighted CNR (Fig. 5). As representative examples, contrast between the cingulate and its adjacent WM, the hippocampal CA1 and SL/SR/SM subfields, and the caudate nucleus and corpus callosum were analysed. The general outcome is that where extensively myelinated WM is involved, contrast declines with increasing WM T 2 , with a modest contrast increase (though not significant at the 95 % level) with increasing GM T 2 . In the hippocampus, where myelination is less prevalent, a more substantive increase in contrast with increasing GM T 2 and minimal (p > 0.05) negative correlation with WM T 2 was observed.

Summary of findings
As a summary of our findings, we have observed strong correlations between age and various metrics of MR image contrast in healthy elderly subjects in the cingulate gyrus and surrounding frontal WM, as well as between individual subfields of the hippocampus. This includes the use of   [30], T 2 increased with age whilst GM T 2 , like non-myelinated WM, stayed constant with age.

A possible explanation for the findings
The current data indicate cross-sectional, age-related differences for CNR and GWR between the cingulate and its adjacent WM in T 1 and T 2 -weighted images (Fig. 3), but in the hippocampus such effects are weaker and only apparent in T 2 -weighted images (Fig. 2). This is mirrored by the observation that no age-related differences in T 2 could be observed in the hippocampus, though a distinct value for each subfield was obtained, the (mainly WM) SL/SR/SM subfield having the shortest. On the other hand, in more extensively myelinated and macroscopically ordered WM such as the corpus callosum (and adjacent to cingulate cortex) the T 2 increases cross-sectionally with age (Fig. 4).
We have used mono-exponential fits to our spin-echo dephasing data to obtain effective T 2 maps. Since a single voxel of an MR image contains a dynamically mixed ensemble of spins (many different molecules interacting with one another across many timescales), we can expect that even where the criteria for mono-exponential loss of signal are met (which, on data-driven grounds, they are here), we observe a mixture of relaxation (T 2 processes) and dynamic dephasing processes. With effective refocussing pulses, we can legitimately assume that static dephasing is negligible. With that in mind, the heuristic monoexponential time constants used here to describe the loss of refocussable spin-phase coherence at rather long echo times (24-120 ms, spaced by 12 ms) can only somewhat colloquially be referred to as "T 2 ". We can anticipate that dynamic dephasing is a major contributor to loss of signal in a spin echo experiment (and hence, to our effective T 2 ). It is driven by interactions of water with macromolecules and by diffusion of water through anisotropic magnetic fields such as those created by the myelin sheath owing to its diamagnetic susceptibility being different from its surroundings. As such, a change in myelination is likely to yield a change in effective T 2 , even at long echo times, through dynamic dephasing. This may be the cause of the age-dominated changes in myelinated WM T 2 changes with age, as myelination is known to decrease with age [31].
The argument is supported by the existing literature demonstrating that T 1 -weighted image contrast is dominated by myelination [32][33][34], which we have found here also to be strongly age-dependent only where myelinated axons are involved. Likewise, T 2 -weighted imaging and quantitative T 2 can be used as a proxy of myelin-associated water content in the brain [35,36]. Image contrast can therefore be interpreted directly in terms of myelination. There is a high degree of order in WM, both at the molecular level where oriented layers of myelin surround axons, and at the macroscopic level where large groups of axons may maintain a common orientation in the major fiber bundles. This is not so in GM, and may also be relevant to the changes in T 2 in (myelinated) WM as compared to conserved GM and non-myelinated WM T 2 .
Since ageing is a multi-factorial process, we must consider that many other factors are likely to play a role, but focussing on those that are likely to make the largest observable differences to MR parameters is a pragmatic and testable way to proceed.

Relation to DTI literature
There has been considerable work on the changes of diffusion parameters in the brain with age using DTI. In particular, the fractional anisotropy (FA) of many WM tracts declines after a certain age is passed (~30) [37], whereas mean diffusivity (MD) increases with age [38]. This technique is highly sensitive to tissue microstructure and water content, changes in which are also determinants of signals in other MR modalities since most MRI approaches are determined by hydrodynamics. The decline in FA and increase in MD commensurate with age implies an increased tissue water content or degradation of anisotropic barriers to diffusion on length scales √ MDt where MD is mean diffusivity and t is the time over which diffusion is sampled (the time for which diffusion-sensitising gradients are applied). The latter mechanism implies an alteration of tissue microstructure. However, in post-mortem brains no significant change in water content was observed in subjects without neurological disorders [38]. This being so, the hydrodynamic properties of GM and WM are made more similar with increasing age, so a decline in their distinction by either T 1 -weighted or T 2 -weighted imaging, both of which are highly sensitive to hydrodynamics, is inevitable. It is easy to imagine that changes to tissue microstructure on a length scale √ MDt would also result in changes to dynamic dephasing. It is more difficult to imagine that tissue microstructure could be changed without altering diamagnetic susceptibilities of the various materials comprising tissues (and hence, T 2 ). In this regard, the diffusion literature vindicates (and motivates) our suggestions in the preceding section: that T 2 increases in WM due to the degradation of ordered structures, of which the myelin sheath is a prime, though not unique, candidate.

Relation to other qMRI research
Work combining MT and quantitative T 1 and T 2 * mapping has also shown that MT decreases with age (on a crosssectional basis) in various GM structures, but not in the hippocampus or cingulate gyrus [4]. T 1 was also reported to be less sensitive to age than the other MR parameters used. Similar findings have also been reported in a study extending such methods to a larger cohort and employing statistical parametric mapping [3]. In both the latter studies [3,4], T 2 * was used, in distinction to ours, which uses spin-echo T 2 and is less affected by static dephasing. Thus, the information content of the two parameters is rather different. Recent work has shown that T 2 * is highly sensitive to iron content in deep GM structures, significantly more so than T 2 , and that there exist different age effects on T 2 and T 2 * for different GM structures and WM regions [39]. Our observation of a modest increase of the T 2 in the globus pallidus with age is in agreement with the latter study, though ours is a somewhat different cohort. The former study demonstrates a substantial decrease in T 2 * in the iron-rich globus pallidus with increasing age. The increased T 2 we have found with age in iron-rich regions is unlikely to be reflective of iron accumulation but rather of accompanying changes that decrease the dynamic dephasing contributions to decoherence of the nuclear spin phase.
Arterial spin labelling (ASL) has also revealed agerelated changes in cross-sectional studies of cerebral perfusion [40,41]. We may therefore speculate that given cerebral perfusion changes with age, and that by virtue of alterations in the amount of (paramagnetic) deoxyhaemoglobin present in the capillary bed and venules, effective T 2 may be perturbed by such a mechanism (the capillary bed will cause decoherence by creating an inhomogeneous local field). However, cerebral perfusion changes to a greater extent in GM than WM, whereas our age-related T 2 changes were larger in WM than GM. Whilst such a mechanism is certainly relevant to ageing, it may be less relevant to changes in T 2 , and to T 1 -weighted CNR and GWR.

Relation to volumetry
It is well appreciated that the hippocampus and cingulate suffer atrophy in MCI and AD [42]. Furthermore, atrophy rates in the hippocampus [43][44][45] and cingulate [29,46] are able to distinguish between healthy ageing, MCI and AD, but an understanding of the ageing process beyond volume changes may be pertinent to dementia research and clinical diagnosis. The distinct functions of the different hippocampal subfields, combined with the progression of AD pathology through those subfields, suggests that our approach to quantifying tissue characteristics, such as CNR and GWR, at an individual subfield level may in the future be useful in detecting early dementia risk or distinguishing between different dementia pathologies. This adds to an expanding literature examining the hippocampus by MRI at the subfield level, rather than considering its entire volume [23,[47][48][49]. Recent data point to localised changes in WM of the hippocampus in the SL/SR/SM subfield group, analysed here in a tauopathy mouse model for AD [50]. This accelerates the need for a detailed characterisation of healthy hippocampal WM aging at the level of individual subfields, and the identification of the MRI parameters sensitive to deviations from such a pathway at early stages.

Methodological challenges and limitations
When using CNR, as stated in the methods section, it is important to ensure that the variance in the regions of interest derives from sources of random (stochastic) noise, and not from slowly varying (deterministic) magnetic field inhomogeneities, manifest as gradual shading across regions of an image, hence the use of receiver coil sensitivity corrections ('prescan normalize' on the particular scanner used) and bias field corrections. This is far less an issue when using quantitative T 2 maps, though to some extent is still present (in particular due to transmitter B 1 inhomogeneity, which causes RF-pulse flip angle variation leading to incomplete refocussing). Obtaining meaningful results for CNR estimates of image intensities therefore requires the use of complementary ROIs that are close in image space and over which "shading" is irrelevant, but which nevertheless sample sufficient voxels to determine the mean and variance with high precision. This is in contrast to volume determination, where an ROI must fully represent the structure of interest, irrespective of image intensity. CNR and GWR estimates have the advantage, however, that ROIs need only sample sufficient voxels to be representative, rather than capturing the shape and thus volume of a structure. Provided the first conditions of this section are met, the result is that substantial freedom in ROI preparation is permissible whilst still generating a stable CNR or GWR estimate that would not be acceptable if attempting volume determination.
Our careful approach to data analysis with conservative ROIs and use of high-performance phase-arrayed coils, as well as the B 0 field strength of 3 T in concert is believed to explain the discrepancy between our findings and earlier studies; the work of Magnaldi et al. [24] reported a negative correlation between WM T 2 and age, which is rather contradicted here. Work by Kim et al. to determine GWR in T 2 -weighted images using ROIs covering all subcortical structures (and thus not obeying the conditions for GWR representative of physiology rather than instrumental performance) obtained considerably smaller effects than reported here [25].
Our study, like others using different qMRI parameters [3,9,51], also suffers from the limitation of being crosssectional rather than longitudinal, though in covering an age range of 38 years this may have been difficult to overcome practically. The consequence is that factors other than age may spuriously contribute to our findings. The future extension to a larger cohort or a longitudinal study over a respectable duration may ameliorate these limitations.

Conclusions
Overall, it is likely that several mechanisms are simultaneously responsible for changes in MR contrast, present to different extents in different regions of the brain that need to be considered in design of volumetric protocols in aged brain. Tissue contrast can and should be interpreted directly in terms of microscopic and/or physico-chemical aspects of the tissues that comprise the systems being imaged, with quantitative imaging providing a further window into their properties. We anticipate the detailed examination presented here as having applications in dementia imaging research, as well as contributing to an understanding of healthy aging.