Machine Learning-Based Approaches for Tandem Romberg Test of Vestibular Hypofunction Assessment Using Wearable Sensors

The vestibular system is an important physiological sensory system that contributes to human sense of proprioception and equilibrium. The most common symptoms of vestibular hypofunction are dizziness or impaired balance control. Balance tests and the Dizziness Handicap Inventory (DHI) are typically used clinically to assess vestibular hypofunction. However, the above assessments have subjective problems because of different clinical evaluators or discrepancies in comprehension of questionnaire. Additionally, inertial measurement units (IMUs) have been widely used in motion analysis recently. This study aims to establish an objective assessment method using machine learning to measure the balance performance of patients with IMUs. The recruited ten outpatients and ten healthy subjects are asked to perform the tandem Romberg test, a balance test used to evaluate vestibular function. Five parameters including maximum, average, root mean square, average sway velocity and relative position feature are used to extract the features of acceleration signals. Machine learning algorithms are developed to classify vestibular hypofunction patients from healthy controls. The accuracy 94%, recall 94% and F-score 94% of the Random Forest (RF) are the highest among the algorithms, while precision 97% of Support Vector Machine (SVM) is the highest. The results indicate that the proposed method is feasible and able to provide clinical objective assessment.