BackgroundSporadic cerebral small vessel disease is an important cause of vascular dementia, and is a syndrome of cognitive impairment with evidence of vascular brain damage. At post-mortem examination pure vascular dementia is rare, with coexisting Alzheimer's disease pathology in 95% of cases. We aimed to use MRI to characterise the structural abnormalities during the preclinical phase of vascular dementia in symptomatic small vessel disease, and use these characteristics to accurately predict the development of future dementia. Methods121 adults with symptomatic small vessel disease were initially recruited to the St George's Cognition and Neuroimaging in Stroke (SCANS) study and followed up longitudinally for 5 years, with 22 individuals converting to dementia. Baseline T1-weighted MRI data were acquired for all 121 partcipants. Voxel-based morphometry was used to identify differences in patients with preclinical vascular dementia. Support vector machines were then used to predict future dementia from the baseline scans. Anatomical endophenotypes were defined using cluster ward linkage. FindingsWe found reduced grey matter density in the left striatum and hippocampus, and more white matter hyperintensities in the frontal white matter, in preclinical dementia. Future dementia could be predicted with a balanced accuracy of 73%. Four anatomical subtypes were identified. In one of them, patients were younger than those in the other three groups and had the highest levels of vascular damage; they also had milder cognitive impairment but rapid deterioration in processing speed and executive function, consistent with primary vascular dementia. The other groups had progressively less vascular damage and increasing memory impairment consistent with more Alzheimer's like pathology. The rates of hippocampal atrophy supported these groupings, with the vascular group resembling the cohort that did not develop dementia, and the Alzheimer's like group demonstrating more global hippocampal atrophy. InterpretationWe show that baseline MRI can reliably predict preclinical vascular dementia, with 73% of patients converting to dementia within 5 years. MRI can identify distinct anatomical endophenotypes representing a spectrum between vascular and Alzheimer's like pathology. This work provides a way to accurately stratify patients by use of a baseline MRI scan, and has utility in future clinical trials designed to slow or prevent the onset of dementia in these high-risk cohorts. FundingThe SCANS study was supported by the Wellcome Trust (grant 081589). Patient recruitment was supported by the National Institute for Health Research (NIHR) Clinical Stroke Research Network.