Joana Moreira^{1}, Joana Pinto^{1}, Ana Fouto^{1}, Luísa Alves^{2,3}, Sofia Calado^{2,3}, Carina Gonçalves^{2,3}, Margarida Rebolo^{4}, Pedro Vilela^{5}, Miguel Viana Baptista^{2,3}, and Patrícia Figueiredo^{1}

Cerebral small vessel disease (SVD) is a major vascular contributor to dementia and stroke, being associated with age-related cognitive decline. In this work, we aim to assess the potential of spontaneous BOLD fluctuations metrics to predict cognitive impairment in a group of SVD patients, therefore providing sensitive SVD biomarkers. The amplitude of low-frequency fluctuations (ALFF) and fractional ALFF (fALFF) were computed in four different frequency bands as metrics of spontaneous BOLD signal fluctuations. Results from multiple linear regression analyses demonstrated that spontaneous BOLD fluctuations metrics contribute to the prediction of cognitive impairment in executive function, processing speed and working memory.

Materials and Methods

A group of 11 patients with sporadic SVD (sSVD) (52±7 yrs) and 6 patients with CADASIL (47±11 yrs) was studied on a 3T Siemens scanner, including: T1-weigthed MPRAGE (1mm isotropic), T2-weighted FLAIR (0.7x0.7x3.3mmResults

Fig.1 shows the ALFF, fALFF1, fALFF2 and fALFF3 maps, averaged across all patients, displaying the distribution of each metric across the brain. Fig.2 presents the Pearson correlation analysis between the neuropsychological scores in each of the four cognitive domains and the eight different metrics. Only processing speed was predicted with significance using single predictors, namely: fALFF2 and fALFF3 in NAWM (p=0.033 and p=0.015, respectively) and ALFF in GM and NAWM (p=0.014 and p=0.017, respectively). Correlation analyses between metrics and covariates are displayed in Fig.3. Significant correlations (p<0.05) between metrics can be observed, with the exception of fALFF1, which only correlates with fALFF3 in GM. In contrast, covariates generally did not present significant correlations with metrics and among themselves. Fig.4 displays results from MLR analyses, including the covariates and the corresponding significant processing speed predictors (ALFF in GM and NAWM, and fALFF2 and fALFF3 in NAWM). The model that best explained processing speed scores (49,74% of variance, p=0.018) included the ALFF metric in NAWM (predictor with the lowest p-value, p=0.005) in addition to the covariates: group, nBV and nLV (p=0.047, p=0.044 and p=0.026, respectively). Further MLR analyses of all four cognitive domains were also performed using only the covariates, the covariates and the composite scores (principal components explaining at least 80% of the variance) of GM metrics, NAWM metrics, and the combination of GM and NAWM metrics. These results are displayed in Fig.5.1. Pantoni L. Cerebral small vessel disease: from pathogenesis and clinical characteristics to therapeutic challenges. The Lancet Neurology, 1;9(7):689-701, 2010.

2. Sam K, Conklin J, Holmes KR, et al. Impaired dynamic cerebrovascular response to hypercapnia predicts development of white matter hyperintensities. NeuroImage: Clinical, 11:796-801, 2016

3. Makedonov, S. E. Black, and B. J. MacIntosh. BOLD fMRI in the White Matter as a Marker of Aging and Small Vessel Disease. PLoS ONE, 8(7):1–9, 2013.

4. X. N. Zuo, A. Di Martino, C. Kelly, Z. E. Shehzad, D. G. Gee, D. F. Klein, F. X. Castellanos, B. B. Biswal, and M. P. Milham. The oscillating brain: Complex and reliable. NeuroImage, 49(2):1432–1445, 2010

Figure 1. Maps of the amplitude of low-frequency
fluctuations metrics ALFF, fALFF1,
fALFF2 and fALFF3, averaged across all SVD patients. The different spatial
distributions of the different metrics can be appreciated.

Figure 2. Pearson correlation plots between scores in each
of the 4 cognitive domains (rows) and the 8 metrics of interest: ALFF, fALFF1,
fALFF2 and fALFF3, in GM and NAWM (columns). Significant correlations were
found between processing speed scores and the following metrics: fALFF2 and fALFF3
in NAWM (p=0.033 and p=0.015, respectively) and ALFF in GM and NAWM (p=0.014
and p=0.017, respectively).

Figure 3. Matrices of Pearson correlation coefficients
(top) and corresponding p-values (bottom) between the 8 metrics of interest (mean
values of ALFF, fALFF1, fALFF2 and fALFF3 within GM and NAWM), and also the
demographic and structural imaging covariates (age, nBV and nLV). Significant
correlations were considered for p-value<0.05. Significant correlations
between metrics can be observed, with the exception of fALFF1, which only
correlates with fALFF3 in GM. However, the covariates present a small number of
significant correlations with the metrics of interest (only fALFF2 and fALFF3
in GM are correlated with age) and with other covariates (only between age and
nBV).

Figure 4. Multiple linear regression model of processing
speed cognitive scores, including the four metrics of interest that were found
to have significant correlations with this score, as well as the covariates
group, gender, age, nBV and nLV. The p-value of the model indicates that
processing speed was significantly predicted by: group,
nLV, nBV and ALFF in NAWM. The coefficient estimate, β, is shown for each
predictor together with the respective p-value. The adjusted coefficient of
determination (R^{2}) of the model is also shown, indicating that
almost 50 % of processing speed scores’ variance was explained by this model.

Figure 5. Comparison
between adjusted R^{2} values for the four cognitive domains obtained
from models using as predictors only covariates and covariates plus: GM metrics’ composite scores; NAWM metrics’ composite scores; and GM and NAWM
metrics’ composite scores. Models with a significant
p-value are marked with an asterisk. The number of composite scores is
presented in brackets in the legend. The model with only
covariates was significant only for executive function. The model including composite scores derived from both GM and NAWM
metrics exhibited the best results. Overall, NAWM metrics performed better than
GM metrics in the prediction of cognitive functions.