Evaluation of glymphatic system activity using diffusion tensor image analysis along the perivascular space and amyloid PET in older adults with objectively normal cognition: a preliminary study

Objectives Diffusion tensor image analysis along the perivascular space (DTI-ALPS) is a recently introduced method for the assessment of the glymphatic system without the need for contrast injection. The purpose of our study was to assess the glymphatic system in cognitively normal older adults with or without subjective cognitive decline (SCD) using DTI-ALPS, and correlating with amyloid PET. Design and participants To evaluate the glymphatic system in cognitively normal older adults using DTI-ALPS, we built a prospective cohort including a total of 123 objectively cognitively normal older adults with or without SCD. The ALPS index was calculated from DTI MRI and was assessed by correlating it with standardized uptake value ratios (SUVRs) from amyloid PET and clinically relevant variables. The study subjects were also divided into amyloid “positive” and “negative” groups based on the result of amyloid PET, and the ALPS indices between those two groups were compared. Results The ALPS index was not significantly different between the normal and SCD groups (P = 0.897). The mean ALPS index from the amyloid positive and amyloid negative group was 1.31 and 1.35, respectively, which showed no significant difference (P = 0.308). Among the SUVRs from variable cortices, that of the paracentral cortex was negatively correlated with the ALPS index (r = −0.218, P = 0.016). Multivariate linear regression revealed that older age (coefficient, −0.007) and higher SUVR from the paracentral cortex (coefficient, −0.101) were two independent variables with a significant association with a lower ALPS index (P = 0.015 and 0.045, respectively). Conclusion DTI-ALPS may not be useful for evaluation of the glymphatic system in subjects with SCD. Older age was significantly associated with lower ALPS index. Greater amyloid deposition in the paracentral cortex was significantly associated with lower glymphatic activity in cognitively normal older adults. These results should be validated in future studies on the relationships between ALPS index and other fundamental compartments in glymphatic system, such as perivenous space and the meningeal lymphatic vessels.


MRI preprocessing
The preprocessing of anatomical 3D T1WI data started with bias field correction using ANTs N4 algorithm followed by registration-based brain extraction (1). Subsequently, T1WI was aligned to the MNI space template using rigid-body registration, which enabled 2 the image to be aligned at the same orientation as the template (i.e., AC-PC alignment) without changing the original brain size and shape. Then, the AC-PC-aligned T1WI was registered to MNI space with symmetric image normalization (SyN) for maximizing the cross-correlation within the space of diffeomorphic maps (2). The output transforms were combined with the one obtained from AC-PC alignment to produce the final registration warp field. Last, the final warp field was applied to bias field-corrected T1WI for registration.
The DTI data were preprocessed with a 3D non-local mean filter to remove the Rician noise. Next, the b=0 image was used for brain extraction to obtain a brain mask. A noiseremoved DTI dataset was corrected for eddy-current-induced distortions and subject movements using the FSL eddy tool (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/eddy) (3). However, we were not able to correct susceptibility-induced geometric distortion as phase-encoding reversed DTI data were not available in routine clinical trials. Eddy-current-corrected DTI data were then registered to an AC-PC-aligned T1W image resampled to 1.5 mm isotropic resolution. Finally, 6 elements of a diffusion tensor matrix were obtained by fitting DTI data using a Python library for analysis of diffusion MRI (DIPY, https://dipy.org/) (4), and maps of fractional anisotropy (FA), mean diffusivity (MD), and axial and radial diffusivity (AD and RD) were generated from the eigenvalues of the tensor.

MRI image processing
MRI scans were automatically segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) partitions using the segmentation function of PVEPET12 adopted from the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm/). The partitions of each subject in native space were registered to an MNI-152 T1-weighted template provided by the PETPVE12 toolbox. In this process, two files containing individual reverse normalization parameters (deformity fields) were produced for imaging analysis.

PET image processing
Each subject's amyloid PET scans were co-registered to a bias-corrected image of the corresponding structured MRI scan using the PETPVE12 function, and visual inspection was performed. Correction for PVE followed the algorithm proposed by the Muller-Gartner method (PVEc-MG method) (5), which was implemented in the PETPVE12 toolbox. The PVEc-MG method is a three-compartment PVEc method that discriminates signals from brain GM, WM, and CSF and is one of the most widely used MRI-based methods for PET image analysis (6). In brief, this method assumes that the observed PET signal of a GM voxel is a spatially weighted average of the actual tracer uptake signal at the GM voxel and the 5 signal from surrounding WM and CSF. Spatial weights are determined by the point spread function (PSF) of the PET scanner. The proposed PVEc algorithm consists of correction for the spill-out effect of signal leakage from the GM to the surrounding tissue as well as the spill-in effect from the surrounding tissue to the GM compartment. Tracer activities in WM and CSF are assumed to be homogeneous in each compartment.

Extraction of regional SUVR
Regional amyloid PET uptake was sampled from 82 brain regions defined in the Desikan-Killiany atlas (7) (atlas included in the PETPVE12 toolbox, the original brain atlas was propagated to the MNI space). The Desikan-Killiany atlas covers the whole cerebral cortex and widely is used for amyloid PET studies including staging (8). The atlas labels were multiplied with the reference template's binary GM mask thresholded at 50% GM probability. The atlas in reference space was transformed into each subject's native space using inverse deformity fields. Mean uptake value of the whole cerebellum was extracted from the PVE-uncorrected PET image and used as a reference region, like in previous studies (9)(10)(11)(12). The uptake value of voxels was converted to standard uptake value ratios by scaling to the mean uptake of the whole cerebellum in non-PVE-corrected data. The regional mean standardized uptake ratios of subjects were obtained using the function implemented in the PETPVE12 toolbox.

Standardization and validation of a local Centiloid standard pipeline
The standardization and validation of the PET imaging analysis methods are based on a dataset and standardized cortical and whole cerebellar volume of interest (VOI) templates, freely available on the Global Alzheimer Association Interactive Network website (GAAIN; http://www.gaain.org). This dataset was used for flutemetamol Centiloid scaling (13) and 6 consisted of a total of 74 subjects comprising 24 young controls, 20 AD patients, 20 psychotic MCI patients, and 10 elderly normal controls. All subjects underwent both Pittsburgh compound-B (PiB) and flutemetamol (Flute) scans.

Validation of the local Centiloid standard pipeline
We created the local standard Centiloid pipeline based on the details of a standard processing system provided by Klunket et al., (14) using SPM8 (Statistical Parametric Mapping, Version We obtained the correlation between local standard Centiloid SUVRPiB and local standard Centiloid SUVRFlute, and the result (y = 0.77x + 0.22, R 2 of 0.96) was identical to the published data ( Figure 1D).