Frontotemporal dementia characterization using neurite orientation dispersion and density imaging.

Microstructural alterations in brain tissue play a crucial role in the pathophysiology of frontotemporal dementia (FTD). This study assessed brain white matter (WM) and grey-matter (GM) microstructure in FTD variants using neurite orientation dispersion and density imaging (NODDI) diffusion MRI model and developed an exploratory machine-learning algorithm to classify FTD subtypes according to diffusion MRI metrics. Brain MRI including multi-shell diffusion sequences and neuropsychological assessment were obtained in controls and participants with FTD: 35 behavioural variant of FTD (bvFTD), 20 semantic-variant primary progressive aphasia (svPPA), 14 nonfluent-variant primary progressive aphasia (nfvPPA), 9 semantic-bvFTD (sbvFTD). Fractional anisotropy (FA), mean diffusivity (MD), intracellular-volume fraction (ICVF), and orientation-dispersion index (ODI) were analysed using tract-based and GM-based spatial statistic at the voxel-wise level, with nonparametric and permutation-based methods. Support vector machine (SVM) models were trained on different combinations of diffusion MRI and neuropsychological features to classify FTD subtypes. FA and MD showed widespread WM alterations in all variants. ICVF showed reductions in both WM and GM (bilateral frontotemporal for bvFTD, left temporal-frontal for svPPA and nfvPPA and right temporal for sbvFTD). GM ODI reduction exhibited a similar but more diffuse pattern compared with ICVF. WM ODI alterations were also observed, with specific WM alterations in the corpus callosum and long-range WM tracts when comparing FTD syndromes. SVM algorithm, trained on mean FA, ICVF and ODI values from different brain lobes and neuropsychological scores, achieved 98.6% accuracy in classifying different clinical syndromes, outperforming standard diffusion tensor (DT) imaging-based models. NODDI capture subtle microstructural alterations in brain GM and WM, demonstrating advantages over standard DT imaging in capturing disease-relevant alterations. By integrating NODDI with cognitive data, machine-learning models can learn complex patterns and relationships facilitating the differentiation of FTD subtypes.