Analysis of Walking Trajectories and Physical Fitness Using Machine Learning To Identify Mild Cognitive Impairment

Abstract

With the aging of the global population, the proportion of patients with dementia has increased. The current method for diagnosing and evaluating mild cognitive impairment (MCI) involves an assessment and blood examinations. Clinically, a question-and-answer test between psychologists and patients is used as the basis for diagnosing MCI. However, this method is subjective. Research results have indicated that patients’ gait or posture balance can be used as an index for the early diagnosis of MCI. Therefore, we tested the gait behavior and physical fitness of 66 people aged over 65 years. The average age of the participants in this study was 73.2 ± 5.66 years. Of these participants, 33 had MCI, and 33 did not. The characteristic parameters of the participants’ walking trajectories were captured using a self-developed system by using image analysis to track and record the coordinates of a patient’s trajectory. The trajectory parameters of the patient were calculated, then. A total of 28 trajectory parameters were obtained from seven physical fitness tests, and these parameters were screened using the Montreal Cognitive Assessment scale to select the parameters with high discrimination ability for MCI. Moreover, the classification results obtained using multiple machine learning models were analyzed and compared to determine the optimal model for evaluating MCI. The results indicated that the classification accuracy was higher when physical fitness parameters and walking trajectory parameters were used in the adopted models than when only physical fitness parameters were used. In the classification with physical fitness parameters, Extreme Gradient Boosting (XGBoost) and decision tree showed the highest accuracy (84.76%) among the adopted models. The duration of stay of an individual in a straight region and physical fitness parameters had high discrimination ability.