An unbiased algorithm for objective separation of Alzheimer's, Alzheimer's mixed with cerebrovascular symptomology, and healthy controls from one another using electrovestibulography (EVestG).

Diagnosis of Alzheimer's disease (AD) from AD with cerebrovascular disease pathology (AD-CVD) is a rising challenge. Using electrovestibulography (EVestG) measured signals, we develop an automated feature extraction and selection algorithm for an unbiased identification of AD and AD-CVD from healthy controls as well as their separation from each other. EVestG signals of 24 healthy controls, 16 individuals with AD, and 13 with AD-CVD were analyzed within two separate groupings: One-versus-One and One-versus-All. A multistage feature selection process was conducted over the training dataset using linear support vector machine (SVM) classification with 10-fold cross-validation, k nearest neighbors/averaging imputation, and exhaustive search. The most frequently selected features that achieved highest classification performance were selected. 10-fold cross-validation was applied via a linear SVM classification on the entire dataset. Multivariate analysis was run to test the between population differences while controlling for the covariates. Classification accuracies of ≥ 80% and 78% were achieved for the One-versus-All classification approach and AD versus AD-CVD separation, respectively. The results also held true after controlling for the effect of covariates. AD/AD-CVD participants showed smaller/larger EVestG averaged field potential signals compared to healthy controls and AD-CVD/AD participants. These characteristics are in line with our previous study results.