M. Ethan MacDonald^{1}, Nils D. Forkert^{1}, Yuhan Ma^{2}, Rebecca J. Williams^{1}, Alexandru Hanganu^{1}, Hongfu Sun^{1}, Randall Stafford^{1}, Cheryl R. McCreary^{1}, Richard Frayne^{1}, and G. Bruce Pike^{1}

Changes in both cortical thickness and cerebral blood flow are observed with age. In this work, we look at how these parameters are modulated across the lifespan. T1-weighted and arterial spin labelling data from 146 subjects were analyzed, with 68 cortical regions selected in each subject to obtain mean cortical thickness and cerebral blood flow. We calculated rates of change, correlation, and laterality for both parameters. Finally, we explored predictive modeling using cortical thickness, CBF and a model combining the two. Predictive modelling was slightly improved when both measures were included.

A total of 146 subjects (58-M, 88-F; 18 to 87 years) were scanned with a 3T MRI (Discovery 750, GE Healthcare). All subjects included in this analysis had a Montreal Cognitive Assessment score >25 (indicating normal cognition). All subjects self-reported normal neurological and vascular medical history. Among others, T1-weighted anatomical and arterial spin labelling (ASL) were acquired and used for this analysis. The T1-weighted MPRAGE scan with an isotropic image resolution of 1 mm was acquired with TI/TR/TE/α of 650/5.84/2.36 ms/8°. The pseudo-continuous ASL was acquired with 5-mm thick slices and a spiral trajectory with an effective in-plane resolution of 2.33 mm, label duration and post label duration were 1.0 s and 2.0 s, respectively.

FreeSurfer was used to calculate cortical thickness
measurements from the anatomical scans for a total of 68 volumes-of-interest
(VOIs) using the cortical parcellation from the DK atlas.^{[8]}
CBF images were registered to the anatomical data and the same VOIs were used
to determine the corresponding mean CBF. CBF data were then registered to the
ICBM atlas,^{[9]}
and group average and aging regression maps were computed.

Linear regression with age was calculated for each region on
both the cortical thickness and CBF data. Average laterality index and regional
correlation matrices were then calculated for each data type. Multiple linear
regression with recursive feature selection using the ReliefF algorithm^{[10]}
was used for age prediction, in three input setups: 1) cortical thickness, 2)
CBF, and 3) both.

A polar plot is shown in Figure 1 to illustrate the
differences in regional regression (R^{2}) with age for the cortical
thickness and CBF data. In general, the R^{2} is lower for the CBF
data.

Figures 2 and 3 show the correlation and laterality indices. CBF measurements are more highly correlated across regions than cortical thickness data, while there is more laterality in the CBF data.

Figure 4 shows the CBF group average and regression maps. Generally,
the rate of change is higher in gray matter. R^{2} was most robust near
the feeding vessels (MCA, ACA).

Figure 5 shows the outcome of the multiple regression age prediction for the cortical thickness data alone, CBF alone, and for the combination.

While results may vary slightly with different population selection or acquisition parameters, some generalizable observations include: 1) CBF data has lower R2 with age than cortical thickness, 2) there is more laterality in CBF than cortical thickness, 3) cortical thickness is less correlated amongst regions than CBF, 4) group average maps indicate that the reduction of CBF with age is constrained primarily to gray matter with similar rates of change across regions, and 5) age prediction was good for either cortical thickness and CBF alone only, and slightly better with the combination.

Compared to previous similar studies,^{[11,12]}
this study adds new information including the laterality analysis. It has been
known that cortical thickness changes are typically bilateral, and often only
the bilateral average cortical thickness values are reported,^{[11]}
while more laterality has been observed in CBF.^{[12]}
Structural cortical thickness correlation matrices are often computed,^{[13]}
but, to our knowledge, this is the first evaluation of cortical CBF correlation.

Cortical thickness and CBF measurements are not necessarily changing linearly with age, particularly in the younger age ranges. Due to the variability in the data and relatively small sample size, the linear approximation is appropriate. Inclusion of more subjects might allow a more precise piecewise linear regression or non-linear modelling. More advanced techniques, such as support vector machine or random forest modelling, could also improve age prediction based on cortical thickness and CBF.

1. Salat DH, Buckner RL, Snyder AZ, et al. Thinning of the Cerebral Cortex in Aging. Cereb Cortex 2004;14(7):721-730.

2. Amin-Hanjani S, Du X, Pandey DK, Thulborn KR, Charbel FT. Effect of age and vascular anatomy on blood flow in major cerebral vessels. J Cereb Blood Flow Metab 2014.

3. Valizadeh SA, Hänggi J, Mérillat S, Jäncke L. Age prediction on the basis of brain anatomical measures. Hum Brain Mapp 2017;38(2):997-1008.

4. Tustison NJ, Cook PA, Klein A, et al. Large-scale evaluation of ANTs and FreeSurfer cortical thickness measurements. Neuroimage 2014;99:166-179.

5. Gaser C, Franke K, Klöppel S, Koutsouleris N, Sauer H, Alzheimer's Disease Neuroimaging I. BrainAGE in Mild Cognitive Impaired Patients: Predicting the Conversion to Alzheimer’s Disease. PLoS ONE 2013;8(6):e67346.

6. Franke K, Ziegler G, Klöppel S, Gaser C. Estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: Exploring the influence of various parameters. Neuroimage 2010;50(3):883-892.

7. Chen Y, Wang DJ, Detre JA. Test-retest reliability of arterial spin labeling with common labeling strategies. Journal of magnetic resonance imaging : JMRI 2011;33(4):940-949.

8. Desikan RS, Ségonne F, Fischl B, et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 2006;31(3):968-980.

9. Mazziotta J, Toga A, Evans A, et al. A probabilistic atlas and reference system for the human brain: International Consortium for Brain Mapping (ICBM). Philos Trans R Soc Lond B Biol Sci 2001;356(1412):1293-1322.

10. Robnik-Šikonja M, Kononenko I. Theoretical and Empirical Analysis of ReliefF and RReliefF. Machine Learning 2003;53(1):23-69.

11. Lemaitre H, Goldman AL, Sambataro F, et al. Normal age-related brain morphometric changes: nonuniformity across cortical thickness, surface area and gray matter volume? Neurobiol Aging 2012;33(3):617.e611-617.e619.

12. MacDonald ME, Frayne R. Phase contrast MR imaging measurements of blood flow in healthy human cerebral vessel segments. Physiol Meas 2015;36(7):1517–1527.

13. Chen ZJ, He Y, Rosa-Neto P, Gong G, Evans AC. Age-related alterations in the modular organization of structural cortical network by using cortical thickness from MRI. Neuroimage 2011;56(1):235-245.

Figure 1: Polar plot indicating the R2 regression
with age parameter of the cortical thickness and cerebral blood flow in 68 cortex
regions. For the vast majority of the regions, the cortical thickness
measurements have higher R2 than the cerebral blood flow
measurements.

Figure 2: Laterality indices. Hemispheric laterality index for
cortical thickness and CBF. Laterality index = ( L – R ) / ( L + R ). Plotted
are mean (grey bar) and standard deviation (error bars).

Figure 3: Correlation matrices of cortical thickness (left)
and cerebral blood flow (right). The left hemisphere is at the top and right
hemisphere is at the bottom. The cerebral blood flow data is more highly correlated
than the cortical thickness.

Figure 4: Group analysis of CBF vs. age map. In addition to
calculating the average CBF, a regression was performed on each pixel to obtain
the slope, R^{2}, and intercept. The gray matter cortical regions have
a decrease of 0.4 to 0.6 mL / 100 g / min / year.

Figure 5: Age prediction performance using the cortical
thickness, cerebral blood flow, and the combined data. Recursive feature
selection using the ReliefF algorithm was used to reduce the number of features,
and the mean age difference versus the number of features reduced, and is shown
on the left. The minimum mean absolute differences for the cortical thickness,
CBF and combined data are 9.22, 10.03, and 8.67, years respectively. The leave-one-out
regression analysis is shown on the right.