Which white matter measures best predict processing speed and intelligence in healthy aging?
Poster No:
1150
Submission Type:
Abstract Submission
Authors:
Petar Raykov1, Rafael Henriques2, Kamen Tsvetanov3, . Cam-CAN3, Marta Correia1, Richard Henson4
Institutions:
1MRC Cognition and Brain Sciences Unit, Cambridge, United Kingdom, 2Champalimaud Research,, Lisbon, Portugal, 3University of Cambridge, Cambridge, Cambridgeshire, 4MRC Cognition and Brain Sciences Unit, Great Shelford, Cambridgeshire
First Author:
Co-Author(s):
Introduction:
Healthy aging is accompanied by decline in cognitive functions such as processing speed (PS) and fluid intelligence (gF). White matter (WM) integrity is important for both of these functions, and declines with age. However, it remains unclear which WM measures are most closely related to variation in cognitive performance. Furthermore, it's unclear whether information from different imaging modalities (e.g., T1/T2) adds complimentary information beyond the multitude of metrics derived from diffusion magnetic resonance imaging (dMRI). Here we use adult life-span data from the Cam-CAN cohort to extract commonalities across 8 different WM measures using factor analysis, and examine how many latent factors are needed. Furthermore, we compare the factors scores, as well as the original WM metrics, in their ability to predict PS and gF, after accounting for effects of age and sex.
Methods:
We used multimodal neuroimaging data from 651 adults from the Cam-CAN cohort, aged between 18-88 years (www.cam-can.org). We took six dMRI metrics from a previous analysis of dMRI data across 27 WM ROIs (Henriques et al, 2023): 1) Fractional Anisotropy (FA); 2) Mean Signal Diffusion (MSD); 3) Mean Signal Kurtosis (MSK); 4) Neurite Density Index (NDI); 5) Orientation Dispersion Index (ODI); 6) Free water volume fraction (Fiso). We then added two further WM metrics: 7) the ratio of two magnetisation-weighted images (MTR) and 8) the T1/T2 ratio from T1- and T2-weighted images. PS was calculated from a factor analysis across the mean and standard deviation of reaction times on a simple and a choice reaction task; gF was calculated from a factor analysis on four Cattell subtests.
We performed factor analysis on the 8 WM measures, concatenating across participants and ROIs. We then performed regression analyses to examine how much of the variance in participants' PS or gF was explained by each of the individual factor scores, or by each of the individual WM measures (averaged over all ROIs), after accounting for sex, linear and quadratic age, and their interactions. Additionally, we included a regressor to account for motion artefacts in the dMRI measures (see Henriques et al., 2023). Finally, we also regressed PS and gF on a whole-brain measure of 9) Peak Width of Skeletonized Mean Diffusivity (PSMD).
We performed factor analysis on the 8 WM measures, concatenating across participants and ROIs. We then performed regression analyses to examine how much of the variance in participants' PS or gF was explained by each of the individual factor scores, or by each of the individual WM measures (averaged over all ROIs), after accounting for sex, linear and quadratic age, and their interactions. Additionally, we included a regressor to account for motion artefacts in the dMRI measures (see Henriques et al., 2023). Finally, we also regressed PS and gF on a whole-brain measure of 9) Peak Width of Skeletonized Mean Diffusivity (PSMD).
Results:
We found that 3 factors captured 83% of the variance in the 8 WM measures. Given that Henriques et al. (2023) found that 3 factors also captured the majority of the variance in the 6 dMRI measures, this suggests that the addition of MTR and T1/T2-ratio measures did not add complimentary information about WM integrity. The three factors seemed to reflect 1) tissue microscopic properties; 2) free-water contamination, and 3) fibre complexity (e.g., crossing or dispersing, fanning fibres), and explained respectively 43%, 23%, 16% of the variance in data (see Fig. 1).
Regression analyses showed that, of the three factors, Factor 1 explained the most variation in gF and in PS. Including all 3 factors scores in the same analysis showed that only Factor 1 explained unique variance in cognition. In terms of single metrics, MSD explained most variation in PS, but not so much variation in gF. The best individual predictor of both gF and PS was MSKI, which performed better than Factor 1 (see Table 1). The global PSMD metric showed the strongest association with age.
Regression analyses showed that, of the three factors, Factor 1 explained the most variation in gF and in PS. Including all 3 factors scores in the same analysis showed that only Factor 1 explained unique variance in cognition. In terms of single metrics, MSD explained most variation in PS, but not so much variation in gF. The best individual predictor of both gF and PS was MSKI, which performed better than Factor 1 (see Table 1). The global PSMD metric showed the strongest association with age.
·Figure 1. Shows factor loadings across different WM measures, and how factors scores related to age.
·Table 1.
Conclusions:
Our factor analysis suggest that MTR and T1/T2 measures do not add complimentary information about WM track integrity (in healthy adults) over the information already present in other dMRI metrics. Factor 1 from this analysis, reflecting age-related microscopic alterations, explained appreciable and unique variance in processing speed and fluid intelligence, after adjusting for age and sex. However, if one had to choose one MR sequence and measure, then MSKI estimated from a (multi-shell) dMRI sequence showed comparable associations across both processing speed and fluid intelligence.
Higher Cognitive Functions:
Higher Cognitive Functions Other 2
Lifespan Development:
Aging 1
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis
Keywords:
Aging
Cognition
Myelin
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review