Multivariate models of Alzheimer's disease progression based on resting-state MEG scans

Abstract

Resting-state measures of brain function have provided insight into the pathology of CNS diseases, including Alzheimer's disease (AD). We have reported a preliminary analysis of longitudinal resting-state MEG scans conducted in AD patients over the course of 10 months (Verdoorn et al., Alz.Dem. 6: S523-S524, 2010). These results indicated that region-specific changes in brain function significantly correlated with changes in cognitive function measured by neuropsychological testing. Further refinement of data processing methods and analysis now support the development of a robust model of disease progression based on these scans. Data were collected during two exploratory clinical trials that included patients (n = 117) with a previous diagnosis of Dementia of Alzheimer's Type (DSM-IV-TR) and a group of demographically matched healthy control subjects (n = 123). A total of 31 AD patients were tested twice with identical procedures. The evaluations were separated by approximately 10 months. One minute of resting state MEG data were processed to generate time-domain functional connectivity measures based on the SNI test (Georgopoulos et al., 2007, J. NeuralEng. 4: 349) and frequency-domain measures of spectral power. Linear models using combinations of connectivity and spectral power features were generated and tested using standard regression analysis. Significant cross-sectional differences between AD and healthy controls were observed for both global functional connectivity and the global distribution of spectral power. For example, global correlation values increased significantly from 0.12 ± 0.004 in healthy subjects to 0.14 ± 0.004 in AD patients, and centroid frequency of the power density spectrum decreased from 8.24 ± 0.20 Hz in healthy subjects to 6.78 ± 0.25 Hz in AD patients. Longitudinal changes were evaluated by measuring the correlation between changes in MEG scan features and changes in the ADAS-Cog (n = 31 AD patients). Significant correlations were observed between changes in SNI values (c02) and changes in ADAS-Cog (maximum Pearson's r value = 0.79). Similar, albeit weaker, correlations were observed between the centroid frequency of spectral density plots and the ADAS-Cog (maximum r value = -0.43). Linear combinations of centroid frequency and SNI values based on multivariate modeling accurately tracked ADAS-Cog changes with a highly significant r value of 0.99. It was possible to extract and characterize multiple features of resting-state MEG scans that changed in concert with changes in the ADAS-Cog over the course of 10 months. These results demonstrate the utility of resting-state MEG scans for tracking the progression of AD over periods of less than a year. Additional longitudinal testing of this and additional patient cohorts will support continued improvement of disease severity models based on resting-state MEG.