Exploring radial asymmetry in MR diffusion tensor imaging and its impact on the interpretation of glymphatic mechanisms

Abstract

Background

Diffusion imaging holds great potential for the non-invasive assessment of the glymphatic system in humans. One technique, diffusion tensor imaging along the perivascular space (DTI-ALPS), has introduced the ALPS-index, a novel metric for evaluating diffusivity within the perivascular space. However, it still needs to be established whether the observed reduction in the ALPS-index reflects axonal changes, a common occurrence in neurodegenerative diseases.

Purpose

To determine whether axonal alterations can influence change in the ALPS-index.

Study Type

Retrospective

Population

100 participants (78 cognitively normal and 22 with mild cognitive impairments) aged 50–90 years old.

Field Strength/Sequence

3T; diffusion-weighted single-shot spin-echo echo-planar imaging sequence, T1-weighted images (MP-RAGE).

Assessment

The ratio of two radial diffusivities of the diffusion tensor (i.e., λ2/λ3) across major white matter tracts with distinct venous/peri-venous anatomy that fulfill (ALPS-tracts) and do not fulfill (control tracts) ALPS-index anatomical assumptions were analyzed.

Statistical Tests

To investigate the correlation between λ2/λ3 and age/cognitive function (RAVLT) while accounting for the effect of age, linear regression was implemented to remove the age effect from each variable. Pearson correlation analysis was conducted on the residuals obtained from the linear regression. Statistical significance was set at p < 0.05.

Results

λ2 was ~50% higher than λ3 and demonstrated a consistent pattern across both ALPS and control tracts. Additionally, in both ALPS and control tracts a reduction in the λ2/λ3 ratio was observed with advancing age (r = −0.39, r = −0.29, association and forceps tract, respectively) and decreased memory function (r = 0.24, r = 0.27, association and forceps tract, respectively).

Data Conclusions

The results unveil a widespread radial asymmetry of white matter tracts that changes with aging and neurodegeration. These findings highlight that the ALPS-index may not solely reflect changes in the diffusivity of the perivascular space but may also incorporate axonal contributions.

INTRODUCTION

The perivascular space within the glymphatic system facilitates fluid flow and waste product clearance, making it crucial in maintaining brain health (1–5). However, non-invasive measurements of perivascular fluid dynamics are challenging due to the small dimensions of the perivascular space and proximity to blood and brain tissue (6). While some non-invasive diffusion imaging techniques have emerged to measure peri-arterial fluid dynamics (7–9) or parenchymal perivascular fluid diffusivity (10, 11), assessing the peri-venous space, which is more closely associated with the efflux of the waste products, has remained limited in both human and animal studies.

Diffusion tensor imaging along the perivascular space (DTI-ALPS) has emerged as an innovative metric for evaluating diffusivity along the perivascular space, specifically, the perivascular space of medullary veins within the deep white matter (12). While diffusion tensor imaging (DTI) has been widely used to investigate white matter integrity in healthy and diseased brains (13–15), DTI-ALPS has been proposed to investigate glymphatic system function through a metric called the ALPS-index (12). DTI-ALPS takes advantage of a specific white matter location near the lateral ventricle, which exhibits a unique anatomic layout where the medullary veins run perpendicular to the direction of axonal tracts. This geometric relationship enables nearly independent analysis of diffusivity along the perivascular space because the diffusivity changes in two axonal tracts along the same direction are “deemed” equally affected and can cancel each other out, revealing diffusion changes specific to the medullary veins in this direction (12). However, it is unclear whether the diffusivity in the two axonal tracts is equally affected in aging and neurodegeneration, and the possibility of axonal changes contributing to ALPS remains unknown.

Due to the widespread availability of DTI data and its straightforward quantification approach, the ALPS-index has become the most widely adopted imaging approach for assessing perivascular spaces and has been associated with glymphatic function. It has been employed to study many neurodegenerative pathologies, including Alzheimer’s disease, Parkinson’s disease, and dementia (12, 16, 17). In these studies, a reduction in the ALPS-index was consistently observed in disease groups. Additionally, these studies have found the ALPS-index to be associated with cognitive function, pathology severity, and prediction of disease progression (18, 19). Nevertheless, it has not been firmly established that the observed reduction in the ALPS-index measured in white matter tracts is specific to changes in perivascular diffusivity and that it does not reflect axonal degeneration, a well-known phenomenon in neurodegenerative diseases (20). Moreover, the possibility of white-matter modifications serving as a confounding element in the interpretation of the ALPS-index has been raised in a recent review paper (6). Considering the well-established occurrence of axonal changes in various neurodegenerative diseases, it is crucial to evaluate the potential influence of axonal alterations on the ALPS-index, and to determine whether the ALPS-index is specific to perivascular diffusivity changes.

The key assumptions involved in the interpretation of the ALPS-index are that white matter tracts exhibit symmetric radial diffusivities, where the second and third eigenvalues of diffusion tensors (λ2 and λ3) are approximately equal (λ2≈λ3), and that radial symmetry does not change in neurodegenerative diseases. These assumptions lead to two important interpretations: (1) an ALPS-index > 1, which can be approximated by λ2/λ3 > 1, reflects the increased water diffusivity within perivascular space aligned in the direction of λ2, and (2) the reduction of the ALPS-index observed in neurodegenerative pathologies is attributable to diffusivity changes in the λ2-aligned perivascular space. This study aimed to test two hypotheses that challenge the key assumptions of the ALPS-index by examining λ2/λ3 properties in major white matter tracts.

Hypotheses

Drawing from observed diffusion tensor properties, two hypotheses were developed concerning the radial asymmetry of white matter tracts that bear direct implications for the ALPS-index.

Hypothesis 1:

The first hypothesis postulates that white matter tracts commonly exhibit radial asymmetry, characterized by a ratio of second and third-tensor eigenvalues greater than 1, λ2/λ3 > 1 (Figure 1). Conventionally, white matter tracts are known for their high anisotropy, where λ1 > λ2 & λ3, and it is commonly assumed that radial diffusivities are symmetric, with λ2 ≈ λ3. However, this hypothesis challenges this notion and suggests that white matter tracts commonly demonstrate radial asymmetry, with λ2/λ3 > 1 (Figure 1A). This hypothesis was founded on the observation of the second eigenvector (eigenvector-2) arrangement in these white matter tracts, revealing a notable coherence in eigenvector-2 (Figure 1, Observation 1). For the principal eigenvector (eigenvector-1), the forceps, projection, and association tracts align with the x-, z-, and y-direction, as expected (Figure 1C). Intriguingly, eigenvector-2 also exhibits significant coherence in spatial alignment, where the forceps tract was aligned with the y-direction, and the projection and association tracts were aligned with the x-direction (Figure 1D). This coherence describes a high degree of organization and directional orientation of eigenvector-2 within the white matter tracts, suggesting widespread radial asymmetry with λ2/λ3 > 1 across white matter tracts. This is because if λ2 ≈ λ3, the image noise would cause a highly random spatial distribution of eigenvector-2 (The spatially coherent relationship between eigenvector-2 and eigenvector-3 is illustrated in Supplemental Figure 1).

Figure 1.

Conceptual representation of hypothesis 1 and its implication on the ALPS-index. (a) Hypothesis 1 posits that there exists widespread radial asymmetry of white matter tracts, indicated by the homogeneous alignment of the second eigenvector (eigenvector-2). (b-f) Observations supporting hypothesis 1. Illustration of the high coherence and directional alignment of eigenvector-2 of the diffusion tensor in both ALPS fibers (projection and association) and control fibers (forceps). (b) Susceptibility-weighted image (resolution = 0.6×0.6×1.0 mm³) showing medullary veins along the x-direction (parallel with eigenvector-2) in the projection fibers (fulfilling the ALPS-index assumptions). Due to resolution limitations, the medullary veins of the forceps are not visible but are typically anatomically oriented in the z-direction (normal with eigenvector-2, not fulfilling the ALPS-index assumptions) (21). Given this geometric relationship, the forceps tract serves as control fibers. (c, d) Line display of major eigenvector direction (C, eigenvector-1 and D, eigenvector-2) of the diffusion tensors, depicting expected fiber orientations. Notably, eigenvector-2 displays coherence in space, observed in the forceps, projection, and association fibers (yellow circle). Specifically, eigenvector-2 aligned with the x-direction in projection and association fibers, and with the y-direction in forceps fibers. (e) Schematic diagram illustrating the ALPS location that has a unique anatomical layout between the medullary veins and white matter tracts, with medullary veins running in the x-direction (gray cylinders), perpendicular to the direction of projection and association fibers (blue and green, where in projection fibers λ2 = Dx, λ3 = Dy, and in association fibers λ2 = Dx and λ3 = Dz). (f) By considering the orientation of eigenvector-2 in projection and association fibers, the ALPS-index can be simplified as λ2/λ3, where λ2 = mean(Dxproj + Dxassoc) and λ3 = mean(Dyproj + Dzassoc), assuming that projection and association fibers exhibit similar λ2/λ3. (g) If hypothesis 1 holds true, the previously reported ALPS-index > 1 may be influenced by the white matter tract radial asymmetry (λ2/λ3), in addition to perivascular space diffusivity.

The ALPS-index was calculated in the projection and association tracts due to the presence of medullary veins that traverse perpendicular to their major fiber orientations (z- and y-direction) (Figure 1E). Taking into account the orientation of eigenvector-2 in these tracts, the ALPS-index, defined as mean(Dxproj + Dxassoc)/mean(Dyproj + Dzassoc), can be expressed as ALPS-index = λ2/λ3, assuming similar λ2/λ3 ratios for the two fibers types (Figure 1F). It has been previously suggested that the ALPS-index > 1 (λ2/λ3 > 1) in these tracts results from the increased diffusivity due to the perivascular space of the medullary veins aligned with eigenvector-2, or x-direction (Figure 1E gray cylinders) (12). To critically evaluate this assertion, the forceps tract was used as a control tract because the medullary veins in the forceps are oriented inferior to superior (z-direction), which is normal to eigenvector-2 and parallel to eigenvector-3, as opposed to parallel to eigenvector-2 in the ALPS tracts (Figure 1B) (21). Consequently, the λ2/λ3 values of the forceps tract do not fulfill the ALPS-index anatomical assumptions. Therefore, λ2/λ3 > 1 cannot be explained by an increase in diffusivity within the perivascular space. Thus, if hypothesis 1 holds true and the forceps tract displays radial asymmetry with λ2/λ3 > 1, it implies that the ALPS-index measured in projection and association fibers may be influenced by the radial asymmetry of the white matter tracts.

Hypothesis 2:

The second hypothesis is that the level of radial asymmetry, as measured by the λ2/λ3 ratio, decreases ubiquitously in white matter with neurodegeneration regardless of the fibers’ orientations (Figure 2A). This hypothesis stems from the observations comparing a 23-year-old brain to a 72-year-old brain, where the latter exhibited lower λ2/λ3 values not only in the association and projection fibers but also in the forceps tract (Figure 2B–E). Specifically, the 23-year-old participant’s brain had λ2/λ3 of 1.7, 2.1, and 2.0 in the forceps, projection, and association fibers, respectively (Figure 2BD), while the 72-year-old participant’s brain had lower λ2/λ3 values in all three tracts, measuring 1.2, 1.7, and 1.8, respectively (Figure 2DE). Thus, if hypothesis 2 holds true and the radial asymmetry (λ2/λ3) decreases in most degenerative white matter tracts whose orientations are either perpendicular to or parallel to the PVS, it would imply that the ALPS-index reduction reported in previous studies of neurodegenerative diseases should be attributed, at least in large part, to changes in the intrinsic properties of the white matter tracts instead of the PVS.

Figure 2.

Conceptual representation of hypothesis 2 and its implication on the ALPS-index. (a) Hypothesis 2 postulates that white matter tract radial asymmetry (λ2/λ3) decreases ubiquitously with neurodegeneration. (b-e) Observations supporting hypothesis 2, were that lower radial asymmetry (λ2/λ3) was observed in the older brain compared to the younger brain. (b, c) Line representation of eigenvector-2 in a 23-year-old brain and a 72-year-old brain showed consistent eigenvector-2 orientation in white matter tracts. (d, e) Tensor metrics were quantified in manually delineated central regions of the forceps (blue), projection (green), and association (red) fibers. Notably, while diffusivity is increased in λ1, λ2, and λ3, λ2/λ3 is reduced in all three tracts of the older brain. (f) If hypothesis 2 holds true, it implies that the ALPS-index, as represented by λ2/λ3, can be influenced by white matter tract changes associated with neurodegeneration, in addition to changes in diffusivity within the perivascular spaces.

MATERIALS AND METHODS

Participants, Cognitive Testing, and MRI Acquisition

A retrospective analysis of λ2/λ3 values in major white matter tracts delineated using fiber tracking in a cohort of 100 participants was conducted. This cohort consisted of 78 cognitively normal (CN) participants and 22 participants with mild cognitive impairments (MCI), recruited through the Indiana Alzheimer’s Disease Research Center (IADRC) and was previously analyzed to study white matter alterations in early-stage Alzheimer’s disease (22). All participants in the IADRC cohort underwent cognitive assessment using the Uniform Data Set version 3 battery, which is a standardized cognitive assessment tool used in all National Institute of Aging (NIA) AD Centers (23). MCI subjects were identified based on a multidisciplinary clinical consensus panel review aligning with NIA-AA criteria (24). Exclusion criteria for neuroimaging included significant cerebrovascular disease or malformations, a history of chemotherapy or radiation therapy, current diagnosis of major depressive disorder, a history of schizophrenia, bipolar disorder, developmental disability, Parkinson’s disease, brain surgery, brain infection, or significant head injury with loss of consciousness exceeding 30 min, or excessive alcohol consumption.

As part of the cognitive assessment, the Rey Auditory Verbal Learning Test (RAVLT) was administered to assess memory function. The RAVLT is a widely used test that involves the presentation of word lists to participants, who are then asked to recall them immediately (25).

All brain MRIs were performed on a 3T MR scanner (Prisma, Siemens Healthcare, Erlangen, Germany) using a Siemens 64-channel RF receiver head-neck coil. The imaging sequences and parameters for the anatomical scans followed the Alzheimer’s Disease Neuroimaging Initiative 2 protocols (http://adni.loni.usc.edu/methods/documents/mri-protocols/) and were reviewed by neuroradiologists for incidental findings. As previously completed by Wen et al., the diffusion MRI protocol (dMRI) employed a single-shot spin-echo echo-planar imaging sequence that contained three zero diffusion-weighting (i.e., b-value = 0 s/mm²) and five concentric diffusion-weighting shells (b-values = 0, 250, 1000, 2000, 3250, 5000 s/mm²) with 142 diffusion-weighting gradient directions (FOV = 240×240 mm, matrix size = 120×120, slice thickness = 2 mm, in-plane resolution = 2×2 mm, TE/TR = 83.6/2690 msec) (22). An additional b = 0 s/mm² with reversed-phase encoding was acquired for geometric distortion correction. Written informed consent was obtained from all participants following procedures approved by the Institutional Committee for the Protection of Human Subjects at Indiana University School of Medicine (IRB #: 1604443276).

Diffusion Tensor Fitting, Fiber Tracking, and Tract Analysis

Image processing involved a series of steps, starting with preprocessing, followed by diffusion tensor fitting, fiber tracking, and tract analysis. The diffusion-weighted images were first denoised using the local principal component analysis approach (26). Subsequently, the denoised diffusion-weighted images were corrected for static-field geometric distortion, motion, and eddy current artifacts using the TOPUP and EDDY tool from FMRIB software library (FSL) (http://fsl.fmrib.ox.ac.uk/fsl).

Maps of DTI metrics were computed using the FSL DTIFIT command, utilizing the first (b-value = 250 s/mm², 6 directions) and second shells (b-value = 1000 s/mm², 21 directions) of the dMRI data. The current study focused on the three eigenvalues of the diffusion tensor, namely λ1, λ2, and λ3. λ1 represents the diffusivity along the principal water diffusion direction, commonly referred to as the axial direction. λ2 and λ3 corresponded to the two diffusivities measured in the transverse radial plane, and their ratio λ2/λ3 was used to assess the radial asymmetry. The fractional anisotropy (FA) maps were calculated using three eigenvalues of the diffusion tensor maps.

For tractography and mapping of major white matter tracts, all five-shell diffusion-weighted images were used, employing the autoPtx plugin within FSL (27). AutoPtx involved modeling a within-voxel multifiber tract orientation structure using BEDPOSTx, followed by probabilistic tractography using PROBTRACKx (28). The start/stop region-of-interest masks for the 24 major white matter bundles were first rendered and identified in a coregistered standard Montreal Neurological Institute (MNI) space before being reverse-transformed to the native diffusion space (as implemented in the AutoPtx plugin in FSL) (29). Tract-specific measures of diffusion metrics, such as λ2/λ3, were derived for each tract, including the median, 75^th percentile, and 90^th percentile. The 24 tracts were categorized into four fibers categories: association fibers (superior, inferior longitudinal, and inferior fronto-occipital fasciculi (slf, ilf, ifo)), projection fibers (forceps major and minor (fma, fmi); corticospinal tract (cst); acoustic radiation (ar)), thalamic radiations (anterior, superior, and posterior thalamic radiation (atr, str, ptr)), and limbic fibers (uncinate fasciculus (unc); cingulate gyrus, and parahippocampal portions of the cingulum bundle (cgc, cgh).

Statistical Analysis

Student t-tests were used to compare demographic variables between the CN and MCI groups. Linear regression was used to investigate the relationship between λ2/λ3 and age, as well as the association between λ2/λ3 and cognitive function (RAVLT) while accounting for the effect of age. Subsequently, Pearson correlation analysis was conducted on the residuals obtained from the linear regression. Statistical significance was set at a level of p < 0.05. R version 4.0.4 was used for all statistical analyses.

RESULTS

Subject Characteristics and Clinical Outcomes

There were no significant differences observed between the participant groups in terms of age, sex, or education (p=0.13, p=0.21, p=0.94, respectively, Table 1). The MCI group exhibited significantly lower performance on the RAVLT compared to the CN group.

Table 1.

Subject demographics and memory function. Demographic and cognitive characteristics include the mean (standard deviation) for each group. Abbreviations: CN, cognitively normal; MCI, mild cognitive impairment; RAVLT, Rey Auditory Verbal Learning Test.

	CN	MCI	t-value	p-value
# of subjects	78	22
Age	68.5 (7.6)	72.2 (10.3)	−1.567	0.13
Biological Sex (M:F)	20:58	9:13	1.291	0.21
Education	16.5 (2.5)	16.5 (2.7)	−0.07	0.94
RAVLT	46.8 (8.3)	32.9 (7.7)	6.752	<0.001

Results Supporting Hypothesis 1: Broad Existence of Radial Asymmetry across White Matter Tracts

Brain maps of FA and λ2/λ3 measurements for two representative subjects aged 50 and 90 years old are shown in Figure 3. The maps reveal widespread radial asymmetry (λ2/λ3 > 1) in the white matter tracts of both subjects. Notably, white matter exhibited elevated measures of both FA and λ2/λ3, whereas the gray matter and ventricular CSF displayed lower measures of FA and λ2/λ3.

Figure 3:

FA and λ2/λ3 maps for two representative subjects. The (a) FA map and (b) radial asymmetry map of a 50-year-old female. The (c) FA map and (d) radial asymmetry map of a 90-year-old female. Abbreviations: FA, fractional anisotropy.

Figure 4 illustrates the distribution of the median, 75^th percentile, and 90^th percentile of λ2/λ3 values quantified for each of the 24 white matter tracts. The box plots represent data obtained from 78 cognitively normal participants aged 50–90 years. Similar distributions were observed in the mild cognitive impaired participants (see Supplmental Figure 2). The median value of λ2/λ3 consistently hovered around or exceeded 1.5 for all tracts, indicating a notable discrepancy in diffusivity between λ2 and λ3 with λ2 being approximately 50% greater than λ3. Moreover, the 75^th percentile of λ2/λ3 fell within the range of 1.5 and 2, while the 90^th percentile λ2/λ3 was between 2 and 2.5.

Figure 4.

Widespread radial asymmetry (λ2/λ3) in white matter tracts, which provides support for hypothesis 1. (a) Boxplots illustrating the distribution of the median, 75^th percentile, and 90^th percentile of λ2/λ3 values of a white matter tract in 78 cognitively normal participants. All tracts exhibit λ2/λ3 ratios well above 1, providing evidence for widespread radial asymmetry. Among these tracts, the corticospinal tract shows the highest median λ2/λ3 ratio, surpassing 1.5. (b) Rendered representations of the 24 tracts displayed in sagittal (left), coronal (center), and axial (right) views, categorized into four groups. Abbreviations: prc, percentile; l, left; r, right.

Results Supporting Hypothesis 2: Decreased Radial Asymmetry with Advancing Age and Impaired Memory Function

The association between λ2/λ3, age, memory function within ALPS-tracts, and control tracts are illustrated in Figure 5. Among the cohort of cognitively normal participants, a decrease in radial asymmetry (λ2/λ3) was observed with advancing age in the association tract (Pearson’s r = −0.39) and in the forceps tracts (r = −0.29). The relationship between advancing age and radial asymmetry within the projection tract did not reach statistical significance (r = −0.15, p = 0.21). Furthermore, when considering memory function across all participants (N=100), while correcting for age, a reduced radial asymmetry (λ2/λ3) was associated with lower performance on the RAVLT in the association fibers (r = 0.24) and in the forceps tracts (r = 0.27). Within the projection tract, the relationship between radial asymmetry and memory function did not reach statistical significance (r = 0.12, p = 0.27).

Figure 5.

Radial asymmetry (λ2/λ3) is decreased with advancing age and decreased memory function, which provides support for hypothesis 2. (a) Three specific tracts were selected for analysis: the association (green) and projection (blue) tracts, which are the two tracts used to calculate the ALPS-index and are characterized by the presence of medullary veins oriented parallel with eigenvector-2, and the control tract represented by the forceps (red) with medullary veins oriented normal to eigenvector-2. (b) The analysis revealed a decrease in radial asymmetry (λ2/λ3) with advancing age in both the association and forceps tracts, with a trend towards lower asymmetry in the projection tract. (c) The findings demonstrate a reduction in radial asymmetry (λ2/λ3) associated with worse memory function, as measured by lower scores on the RAVLT, in the association and forceps tracts. Similarly, there is a noticeable trend towards decreased asymmetry in the projection tract. The significance level is labeled as: * p < .05, ** p < .01, and *** p < .001. Abbreviations: CN, cognitively normal; MCI, mild cognitive impairment.

DISCUSSION

This study revealed the presence of widespread radial asymmetry within the white matter tracts, which has direct implications for the interpretation of the ALPS-index. Specifically, these results provided two lines of evidence supporting the contributions of white matter tract intrinsic properties to the ALPS-index. Firstly, the consistent findings of λ2/λ3 > 1 across white matter tracts offer an alternative perspective to the assumption in the ALPS-index, which attributes ALPS-index > 1 to the presence of an increased diffusivity within the perivascular spaces. Given λ2/λ3 > 1 is observed in white matter tracts without perivascular spaces parallel with λ2, our observations suggest that white matter radial asymmetry may be a significant factor contributing to the observed ALPS-index > 1. Secondly, a decrease in radial asymmetry (λ2/λ3) with aging and neurodegeneration, was observed, mirroring the changes in the ALPS-index observed in neurodegenerative diseases. Considering that λ2/λ3 decreases with age in white matter tracts, both with and without perivascular spaces parallel to λ2, this suggests that the observed decrease in the ALPS-index seen with age may not be solely attributable to perivascular space diffusivity. Taken together, these results indicate the potential influence of white matter tract properties on the ALPS-index and underscore the importance of considering it as a potential confounding factor when interpreting the DTI-ALPS-index. Therefore, these results underscore the importance of cautious interpretations of the ALPS-index in investigations involving glymphatic function and neurodegenerative diseases.

While previous studies have generally investigated axial diffusivity (λ1) and mean radial diffusivities (mean(λ2, λ3)) of DTI in aging (30, 31), as well as various neurodegenerative pathologies (32–34), the ratio of λ2/λ3 has received limited attention. Previously, the extreme disparity between λ2 and λ3 within a diffusion tensor has been described using tensor mode (35, 36). Tensor mode was introduced to complement FA, aiming to provide a more nuanced description of tensor shape variance that cannot be fully captured by FA alone. Notably, a high FA may indicate a linear anisotropy (λ1>>λ2≈λ3, mode=1), or a planner anisotropy (λ1≈λ2>>λ3, mode=−1). However, due to the tensor mode computational normalization to λ1, it lacks sensitivity in detecting moderate differences between λ2 and λ3 when a larger λ1 is present. As a result, the contrast between λ2 and λ3 has remained relatively unexplored and largely unconsidered in white matter studies, despite electron microscopy studies revealing heterogeneous axon cross-section geometries (37) and DTI reports indicating λ2>λ3 in areas of white matter fiber dispersion (38). This study examined λ2/λ3 and revealed widespread radial asymmetry in the axonal organization of white matter tracts. Specifically, the median λ2/λ3≈1.5, and a 90^th percentile λ2/λ3≈2.

The follow-up question to ask is what causes the radial diffusivity to be asymmetric? In the white matter, the diffusion signal reflects water diffusivity within the axons of the white matter tracts. The organization of these axons provides insights into the observed diffusivity and its directionality. Although the spaghetti noodle-like axons themselves may possess a circular cross-section, their arrangement within the white matter tract has been described by Wedeen and colleagues, as a 2D structured sheet, where axons align in parallel along a preferred direction, akin to aligning up noodle strands on a sheet of paper (39). This 2D sheet is further stacked in parallel to adjacent sheets, creating a 3D arrangement akin to sheets of paper stacked on top of one another. This geometric configuration may provide a plausible explanation for the observed results. Given the sheet-like arrangement of axons stacked in parallel, the water diffusivity between the parallel sheets (between-sheet diffusivity) would be different compared to within the sheet (within-sheet diffusivity). This scenario would result in white matter radial asymmetry, consistent with the findings λ2 > λ3. While the sheet-like axonal arrangement represents one plausible explanation for the observed radial asymmetry λ2/λ3 > 1, other contributing factors may also exist, necessitating further studies dedicated to the geometric influence of white matter tracts on DTI measures.

This study further revealed a decrease in radial asymmetry (reduced λ2/λ3) with aging and neurodegeneration. The decline in radial asymmetry suggests that aging and neurodegeneration processes may be linked to a reduction in the structure and organization of the stacking sheets within the white matter tracts. Such disorganization may reflect alterations in the integrity and coherence of axonal tracts, potentially contributing to the observed changes in diffusion properties.

While the precise mechanisms underlying the observed radial asymmetry remains to be fully elucidated, the widespread radial asymmetry (λ2/λ3 > 1) observed in the current study has immediate implications for the interpretation of the DTI-ALPS-index. The results suggest that the ALPS-index is influenced, to a large extent, by the intrinsic structure of the white matter tracts. Specifically, the observed ALPS-index > 1 can be attributed to the radial asymmetry of projection and association tracts, rather than solely originating from the diffusivity within the perivascular space. Nevertheless, the index’s potential sensitivity to perivascular diffusion should be acknowledged. Notably, in this analysis the two tracts used in the calculation of the ALPS-index demonstrated the highest level of radial asymmetry (λ2/λ3 > 1.5): the corticospinal tract (a projection fiber) and the superior longitudinal fasciculus (an association fiber). Additionally, in this study the λ2/λ3 in the cortical spinal tract and superior longitudinal fasciculus align closely with reported values of the ALPS-index in healthy controls (see Supplemental Table 1). The pronounced λ2/λ3 ratio in this location is likely due to a combination of white matter tract radial asymmetry and the presence of medullary veins aligned along the λ2 direction.

Moreover, the results reveal that changes in the axonal organization that occur with aging and neurodegeneration can result in a reduction in the ALPS-index, mirroring the ALPS-index changes that have been reported in previous studies (12, 16, 17). Hence, the observed decrease in the ALPS-index in neurodegenerative diseases may be influenced by alterations in white matter tract properties associated with the disease. The combined impact of changes in both white-matter tract organization and the diffusivity within the perivascular spaces underscores the intricate nature of the ALPS-index and highlights the need for comprehensive interpretations, avoiding the exclusive attribution of its changes to the diffusivity within the perivascular space or glymphatic function.

Limitations

The measure of interest in this study, radial asymmetry (λ2/λ3), is partially dependent on the accuracy of DTI fiber tracking. DTI fiber tracking technology errors are increased in regions with high levels of white matter fiber crossings which reduces the accuracy of eigenvector and eigenvalue calculations (38). As a general limitation in DTI fiber tracking studies, it will propagate into the measures of radial asymmetry and could lead to an overestimation or underestimation of radial asymmetry. Additionally, the coarse voxel size of DTI results in voxels that are comprised of mixed tissues (i.e. white matter, gray matter), resulting in contamination of the white matter tract diffusivities with gray matter, especially at the white-gray boundary. This could underestimate diffusion anisotropy and lead to λ2/λ3 inaccuracies (40). Despite these limitations, the voxel-wise λ2/λ3 coherence is high, suggesting that partial volume effects did not severely affect tract-wise λ2/λ3 quantification. Additional limitations include that the study was limited to a single center, single vendor, and single field strength, and that it was cross-sectional for examining the age effect. The observations in this study need further verification with multi-center and multi-vendor datasets, and a longitudinal study is required to further validate the effect of age on radial asymmetry.

Conclusion

This study provides evidence of widespread radial asymmetry within white matter tracts, a phenomenon that has remained largely overlooked in previous diffusion imaging studies. Additionally, this study demonstrates that this radial asymmetry decreases with aging and neurodegeneration, mirroring the effects that would occur to the ALPS-index due to alterations in the diffusivity within the perivascular space. Collectively, these findings suggest a potential white matter contribution to the ALPS-index, underscoring the importance of considering white matter radial asymmetry in diffusion studies.

Data and code availability:

Due to the ethics and privacy issues of clinical data, the original imaging data will not be made openly available to the public. No new code was developed throughout the methodology of this manuscript. Data processing was completed using open-source code described in the methods.