Introduction

Mild Cognitive Impairment (MCI) and Alzheimer's disease (AD) are associated with neurodegenerative effects upon the white matter structure of the brain¹. Multi-shell DTI analyses such as NODDI (Neurite Orientation Direction and Density Imaging) have increased sensitivity over regular single tensor DTI computed metrics². In this work, we examine the relationship between white matter microstructure metrics calculated by NODDI and a composite measure of memory function.

Methods

Subjects: Data from 79 participants were included in this study; 45 participants met criteria as cognitively intact, 24 as amnestic MCI with a presumed AD etiology, and 10 with a diagnosis of dementia of the Alzheimer’s Type (DAT). All study procedures took place under approval of the local IRBMED. Consensus diagnoses follow those set forth by the National Alzheimer's Coordinating Center (NACC). A composite measure of memory was created using the mean z-scores of the delayed recall portions of the Craft story, Hopkins Verbal Learning Test, and Benson Complex Figure.

MRI data acquisition and analysis: DTI data were acquired on a 3T GE MR750. Multi-shell DTI was acquired using a multiband EPI sequence (TR=3900, Min TE, Flip angle=77, FOV 24 cm, 1.7 mm slice thickness, MB factor=3, 102 total directions: 12 @ b=0, 30@b=711, 60@b=2000). DTI data was preprocessed for eddy-current and distortions in FSL, with calculation of fractional anisotropy (FA) maps. NODDI parameters, Neurite Density (ND) and Orientation Dispersion (OD), were calculated using the AMICO toolbox³. Tract-based spatial statistics (TBSS) operations were then computed using FSL, including FA normalization, mean FA skeletonization, and projection of the NODDI parameter maps to the FA skeleton.

Predictive Modeling: CPM (connectome predictive modelling)⁴ in MATLAB was adapted to examine the predictive relationship of each of the NODDI parameters with the composite memory score. Using a Leave-One-Out prediction framework, each skeletonized parameter map were correlated across training subjects. Feature reduction was done by selecting the voxels that met 0.01 significance. Then, the raw values for all features were summed to give a composite brain score for each subject, which was then fit to a linear model to predict composite memory; this was done separately for positive and negative correlations. The fit model was then applied to the left-out subject to predict their value, and repeated over all leave-one-out folds. Goodness of fit was evaluated by correlating the predicted versus actual metrics.

Results

Significant relationships between DTI measures and the composite memory measure were found for all calculated parameters (Table I). Results for both the positive and negative correlation of Orientation Dispersion were observed to be stronger than the FA results; while the results for the negative correlations of Neurite Density were not significant. Figure 1 plots predicted vs. actual composite memory scores for the OD positive correlations. Figure 2 shows the voxels used for each predictive analysis, overlaid on the average FA map.

Discussion and Conclusions

Significant relationships were found between NODDI parameters and a composite memory index in older subjects. Compared to the FA positive edges, ND and OD corresponded well to the voxels used in the FA predictive analysis, while also showing sensitivity to additional areas. For the FA negative edges, the OD positive edges correspond well, while the ND negative edges do not (and do not have a significant relationship with the composite memory score). This indicates that the underlying microstructural effect driving the FA negative relationship is OD, not ND. Of note, the tracts involved (uncinate and fronto-occipital fasciculi) have been implicated in memory function in MCI/AD^5,6.

Acknowledgements

This project was supported in part by the National Institute on Aging via Michigan Alzheimer’s Disease Research Center (MADRC; 5P30 AG053760) and R01 AG058724 (to BMH) as well as the Department of Veterans Affairs (I01RX001534 to BMH), and the University of Michigan Precision Health Investigators Award.