Manual Motor Function as a Classifier for Alzheimer’s Disease

Abstract

AbstractBackgroundIn contrast to numerous reports of gross motor problems in Mild Cognitive Impairment (MCI) and Alzheimer’s disease (AD), fine motor function has been relatively understudied. We examined if performance on manual motor tasks can distinguish between cognitively normal (CN) subjects, and individuals with MCI or AD, and which fine motor features are most informative.Method47 CN, 25 MCI, and 16 AD participants completed computerized tests of unimanual (dominant and non‐dominant hand) tapping, synchronous bimanual tapping, and alternate bimanual tapping. Outcome measures included initial reaction time, tapping speed, and variance. We used imbalanced random forest learning for classification modeling. The model was trained on 70% of the data and tested on the remaining 30% using 5‐fold cross‐validation. To investigate the contribution of individual variables to the classification model, we conducted permutation feature importance analysis.ResultThe overall classification accuracy of the training data was 56%, while the overall classification accuracy of the test data was 70%. The group‐specific accuracy, defined as the proportion of subjects predicted to belong to a certain group who are truly a member of that group, was 76% for CN, 33% of amnestic MCI, and 86% for AD. In 50% of the cases amnestic MCI subjects were classified as control subjects, and in 17% of the cases as AD subjects. The group‐specific precision, defined as the proportion of subjects who received a certain group label who truly belonged to that group (i.e., ‘true positives’) was 76% for controls, 100% for amnestic MCI, and 45% for AD. The three most important features that were used for classification of the test hold‐out data were 1) variance in tapping speed during dominant hand finger tapping; 2) the number of dual tapping trials without gaps or onset delays; and 3) initial reaction time for non‐dominant hand finger tapping.ConclusionSupervised machine learning can discriminate CN and AD subjects based on finger tapping performance features, but is less effective for distinguishing MCI subjects from CN and AD subjects. Finger tapping performance could be a cost‐effective tool for augmenting existing AD biomarkers.