Automated Identification of Postural Control for Individuals with Parkinson’s Disease using a Machine Learning Approach

Abstract

Machine learning is a branch of artificial intelligence that enables computer systems to learn from data and analyze data without being explicitly programmed. Interest in machine learning has grown rapidly in clinical settings because the diagnosis of diseases or disorders can be automated by computer systems with high accuracy and minimum human intervention. However, the use of machine learning to identify postural control patterns for people with Parkinson’s disease (PD) is not well established. PURPOSE: The purpose of the study was to develop and validate an automated identification of PD postural control patterns using a machine learning approach. METHODS: 12 participants with PD (age = 75.3 ± 6.6 yr, height = 1.71 ± 0.12 m, mass = 83.1 ± 12.4 kg) and 18 healthy controls (age = 83.3 ± 5.5 yr, height = 1.62 ± 0.08 m, mass = 73.1 ± 16.2 kg) were recruited. Participants were instructed to stand on a force plate and maintain still for 2 minutes during eyes-open and eyes-closed conditions. The center of pressure (COP) data were collected at 50 Hz; sway area, linear displacements, total distances, standard deviations of COP positions and average velocities were calculated. 3 supervised machine learning algorithms (i.e., logistic regression (LR), k-nearest neighbors (KNN) and naïve Bayes (NB)) were used to identify PD postural control patterns. All participants were divided into two datasets: 70% for training and 30% for testing. RESULTS: KNN achieved the highest overall accuracy rate (0.90) to identify PD postural control. LR and NB also exhibited satisfactory performance. The overall accuracy of LR ranged was 0.86; and the overall accuracy of NB was 0.81. Though all three models are capable of analyzing small-sample data, model performance to identify PD postural control could be potentially improved by recruiting a larger sample size and exploring other machine learning models in future research. CONCLUSIONS: Computer-aided machine learning models successfully identified postural control patterns of PD patients with high accuracy. The use of machine learning may provide a valid and efficient approach to better understand PD postural control features and thus, could be beneficial for the early diagnosis and early intervention in individuals with PD.