Diffusion-weighted imaging predicts cognitive impairment in multiple sclerosis.

Following a previous study with diffusion tensor imaging, we investigated the correlation between diffusion-weighted imaging (DWI) and cognitive dysfunction in multiple sclerosis (MS). We studied 60 MS patients (mean age 45.8+/-9.0 years) using 1.5-T MRI. Disease course was RR=40 and SP = 20. Mean disease duration was 12.8+/-8.7 years. Mean EDSS was 3.4+/-1.7. Whole brain, gray and white matter normalized volumes were calculated on 3D SPGR T1-WI using a fully automated Hybrid SIENAX method. Parenchymal mean diffusivity (PMD) maps were created after automated segmentation of the brain parenchyma and cerebrospinal fluid using T2-WI and DW images. Histogram analysis was performed and DWI indices of peak position (PP), peak height (PH), mean parenchymal diffusivity (MPD) and entropy were obtained. Neuropsychological (NP) evaluation emphasized auditory/verbal and visual/spatial memory, as well as processing speed and executive function. We found significant correlations between DWI and performance in all cognitive domains. Overall, stronger correlations emerged for MPD and entropy than other DWI measures, although all correlations were in the expected direction. The strongest association was between DWI entropy and performance on the Symbol Digit Modalities Test, which assesses processing speed and working memory (r = -0.54). Fisher r to z transformations revealed that DWI, gray matter (GMF) and whole brain (BPF) atrophy, T1-lesion volume (LV) and T2-LV all accounted for similar amounts of variance in NP testing. Stepwise regression models determined whether multiple MRI measures predicted unique additive variance in test performance. GMF (R2 = 0.35, F =30.82, P <0.01) and entropy (DeltaR2 =0.06, DeltaF=5.47, P <0.05) both accounted for unique variance in processing speed. Our data make a stronger case for the clinical validity of DWI in MS than heretofore reported. DWI has very short acquisition times, and the segmentation method applied in the present study is reliable and fully automated. Given its overall simplicity and moderate correlation with cognition, DWI may offer several logistic advantages over more traditional MRI measures when predicting the presence of NP impairment.