Brain regional volume analysis to differentiate Alzheimer’s disease (AD) and vascular dementia (VD) from healthy control (HC): Machine learning approach

Background In previous decades, structural MRI has been widely adopted in the diagnosis of AD, MCI and other forms of dementia[1]. Through quantitative measurement of volume of regions‐of‐interest (ROI), it is suggested that hippocampal volume reduction and the rate of reduction can be use in dementia diagnosis and predict the disease progression[2]. However, there is no such application to VD. Method 81 participants (25 HC, 16AD, 28MCI, 12VD) were recruited to undergo structural MRI scanning. T1W Images MRI images were acquired using a Philips Achieva 3T. Volume of brain structures were assessed using Freesurfer (v. 7.1). MANCOVA test was adopted to determine the significant difference between diseased groups and HC, with GM volume, age and gender as covariates. R decision tree was adopted to identify major parameter and its threshold to classified AD, MCI, VD from HC. Results In brain structural volume analysis, AD demonstrated significant atrophy in thalamus, hippocampus, amygdala, accumbens areas when compared to HC. VD showed similar results as AD, with ventral diencephalon as an addition region of atrophy when compared to HC group. For MCI, less severe atrophy was observed when compare to AD and VD toward HC. Furthermore, VD showed significantly enlarged in lateral ventricle, third ventricle, fourth ventricle, CSF, choroid plexus and white matter hypointenisty (WMH). While AD showed significantly enlarged lateral ventricle, inferior lateral ventricle and CSF when compared to HC. There is no statistically significant difference for MCI. In the decision tree model, WMH volume > 5826cm3 was used to classify VD from the rest. The left hippocampus volume < 3287cm3 was used to classify AD from HC. However, the MCI group could not be accurately identified. The accuracy of the model to predict AD, VD, HC from others were 87.5%, 99% and 96% respectively. Conclusion The findings demonstrated that AD and VD had significant reduction in brain regional volume when compared to HC, but not MCI. The WMH and left hippocampus volume were included in the decision tree model as predictive parameters to classify AD and VD from HC. Though the sample size is small, further validation may be required.