Breaking Diagnostic Barriers: NODDI-Machine learning Fusion for FTD Differential Diagnosis
Federica Agosta1, Silvia Basaia4, Stefano Pisano6, Elisa Canu4, Edoardo Spinelli1, Giordano Cecchetti2, Alma Ghirelli2, Elisa Sibilla4, Giuseppe Magnani5, Francesca Caso5, Paola Caroppo7, Sara Prioni7, Cristina Villa7, Lucio Tremolizzo8, Ildebrando Appollonio8, Federico Verde9, Nicola Ticozzi10, Vincenzo Silani10, Massimo Filippi3
1Neuroimaging Research Unit, Division of Neuroscience; and Neurology Unit, 2Neuroimaging Research Unit, Division of Neuroscience; Neurology Unit; and Neurophysiology Service, 3Neuroimaging Research Unit, Division of Neuroscience; Neurology Unit; Neurorehabilitation Unit,and Neurophysiology Service, IRCCS San Raffaele Scientific Institute; and Vita-Salute San Raffaele University, 4Neuroimaging Research Unit, Division of Neuroscience, 5Neurology Unit, IRCCS San Raffaele Scientific Institute, 6Neuroimaging Research Unit, Division of Neuroscience, IRCCS San Raffaele Scientific Institute; and University of Cagliari, 7Unit of Neurology, Fondazione IRCCS Istituto Neurologico Carlo Besta, 8Neurology Unit, San Gerardo" Hospital and University of Milano-Bicocca, 9Department of Neurology and Laboratory of Neuroscience, IRCCS Istituto Auxologico Italiano, 10Department of Neurology and Laboratory of Neuroscience; and "Dino Ferrari" Center, Department of Pathophysiology and Transplantation, IRCCS Istituto Auxologico Italiano; and Università degli Studi di Milano
Objective:
The investigation of in vivo microstructural white (WM) and grey-matter (GM) alterations through diffusion-MRI imaging in frontotemporal lobe degeneration (FTLD) patients holds potential in better understanding neuropathological changes.
Background:
In this study, we identified differences in patterns of WM and GM alterations through the application of the NODDI diffusion model in different FTLD variants and developed a machine learning (ML) algorithm that accurately classified FTLD subtypes using diffusion MRI results and neuropsychological data.
Design/Methods:
Multi-shell diffusion sequences and a neuropsychological assessment were obtained in controls and FTLD patients: 35 behavioral-variant frontotemporal dementia (bvFTD), 20 semantic variant primary progressive aphasia (svPPA), 14 nonfluent variant primary progressive aphasia (nfvPPA), 9 semantic bvFTD (sbvFTD).
Results:
The analysis of diffusion metrics in all FTLD syndromes compared to controls, including fractional anisotropy (FA), intracellular volume fraction (ICVF), and orientation dispersion index (ODI), performed using tract-based (TBSS) and gray-matter based spatial statistic (GBSS), revealed specific patterns of WM and GM alterations for each variant. FA ahowed widespread WM alterations. ICVF showed reduction both in WM and GM (bilateral frontotemporal for bvFTD, left temporal-frontal for svPPA and nfvPPA, right temporal for sbvFTD). GM ODI reduction was present with a similar, but more diffuse pattern compared to ICVF. WM ODI alterations were also observed: (i) reduction in corpus callosum and corona radiata (bvFTD, svPPA, nfvPPA) and right corona radiata (sbvFTD) and (ii) increase in temporo-occipital WM bundles (bvFTD) and stria-terminalis (svPPA). A Support vector machine (SVM) algorithm trained on mean ICVF and ODI values from different lobes and a set on neuropsychological test achieved a 97.3% accuracy in classifying different clinical syndromes.
Conclusions:
This approach holds great potential for advancing our understanding of FTLD pathology, facilitating diagnosis, personalized management and treatment strategies at individual-level.