Artificial neural network to classify cognitive impairment using gait and clinical variables

Combining gait and clinical variables could increase the accuracy of identifying cognitive impairment (CI) in geriatric patients. We aimed to classify geriatric patients with and without CI based on clinical variables, gait, or a combination of clinical and gait variables, using two machine learning methods, Random Forest (RF) and Artificial Neural Network (ANN). The most accurate classification model examined how interactions between clinical and gait variables would improve classification accuracy and determine the contributions of key variables. Based on Minimal Mental State Examination (MMSE) scores, 131 geriatric patients were divided into a cognitive impaired and a cognitively healthy (CH) group. From 3D accelerometer data collected during 3 min of walking at a habitual speed, we computed 23 dynamic gait variables. In conclusion, an ANN model incorporating the interaction between clinical and gait variables classified geriatric patients with an accuracy of 96%, an area of the receiver operating characteristic curve of 0.95, and a model validation score of 0.97 (F1) based on their clinical status. Machine learning analyses of gait and clinical variables can inform geriatricians about the diagnosis of geriatric patients’ cognitive status.