P2‐131: Applying Random Forest Machine Learning to Diagnose Alzheimer's Disease and Dementia with Lewy Bodies: A Combination of Electroencephalography (EEG), Clinical Parameters and Biomarkers

Discrimination between Alzheimer's disease (AD) and dementia with Lewy bodies (DLB) can be difficult, yet is important for patient care. Recently there has been increased interest in automated classification algorithms to provide support for clinical decision-making. Our aim was to develop a machine learning based classifier to improve diagnostic accuracy for AD and DLB. We aimed to quantify the importance of multimodal diagnostic measures (i.e. features), with a focus on electro-encephalography (EEG). 66 AD patients, 66 DLB patients and 66 controls, matched for age and gender, were selected from the Amsterdam dementia cohort. Quantitative EEG (qEEG) measures (spectral analysis, theta/alpha ratio, connectivity and network measures) were derived from resting state EEGs. Clinical, neuropsychological, visual EEG, visual MRI and cerebrospinal fluid (CSF) data were added. A random forest classification algorithm was applied, calculating variable importance (VIMP) scores per diagnostic feature. For discrimination between AD and controls, an accuracy of 91% was achieved. The most important features were cognitive tests and CSF tau/abeta-42 ratio. For discrimination between DLB and controls, the accuracy was 94% and the most important feature was EEG theta/alpha ratio. For discrimination between AD and DLB, the accuracy was 86% and relative beta power was the most important feature. This study is the first to use a machine learning based approach to combine visual and qEEG data with other diagnostic features. In this dataset, the random forest classifier could discriminate between AD, DLB and controls with high accuracies. To discriminate AD from controls the clinical features were the most important. However, to discriminate DLB from both controls and AD, qEEG measures were the most valuable features. Possibly these could play a role in clinical diagnosis.