Identifying cognitive impairment in older adults using machine learning on combined fNIRS and motion data during an upper extremity dual task function.

It is critical that dementia clinical interventions begin early in the disease progression to be effective. Similar clinical manifestations may be observed in both cognitively healthy older adults and older adults with early-stage cognitive impairment, posing a challenge for early disease identification. This study explored classification models using a combination of motor-(gyroscope) and brain- (functional near infrared spectroscopy (fNIRS)) based features for potential use in screening cognitive impairment in older adults. Cognitively normal older adults (CNOA, n = 43; age = 75.47 ± 7.15) and cognitively impaired older adults (CIOA, n = 32; age = 75.90 ± 7.48) completed a 3-minute resting period followed by 3-minute upper extremity dual task function (UEF) involving simultaneous serial subtraction and elbow flexion. The selected features included motor variability and fNIRS anterior prefrontal cortex connectivity outcomes. Logistic regression, support vector machine (SVM), and bootstrap aggregated decision trees predicted the cognitive classification of participants. Cross-validation results suggest SVM models had superior performance with an average accuracy of 76%, Receiver Operating Characteristic - Area Under Curve (ROC-AUC) of 0.86, and F1 score of 69. When used with classification algorithms, the UEF dual task may offer an objective technique for early dementia identification.