Feasibility of Using a Novel, Multimodal Motor Function Assessment Platform With Machine Learning to Identify Individuals With Mild Cognitive Impairment

Abstract

Introduction: Early identification of clinical conditions associated with Alzheimer disease and related dementias (ADRD) is vital for intervention. One promising early detection method is the use of instrumented assessment to identify subtle motor declines associated with ADRD. This pilot study sought to establish the feasibility of building a machine learning model to identify individuals with mild cognitive impairment (MCI) using motor function data obtained from an inexpensive, portable device. Methods: Our novel, multimodal motor function assessment platform integrates a depth camera, forceplate, and interface board. Healthy older adults (n=28) and older adults with MCI (n=19) were assessed during static balance, gait, and sit-to-stand activities in both single- and dual-task conditions. Three machine learning models (ie, support vector machine, decision trees, and logistic regression) were trained and tested with the goal of classification of MCI. Results: Our best model was decision trees, which demonstrated an accuracy of 83%, a sensitivity of 0.83, a specificity of 1.00, and an F1 score of 0.83. The top features were extracted and ranked on importance. Discussion: This study demonstrates the feasibility of building a machine learning model capable of identifying individuals with mild cognitive impairment using motor function data obtained with a portable, inexpensive, multimodal device.