A Random Forest Approach for Quantifying Gait Ataxia With Truncal and Peripheral Measurements Using Multiple Wearable Sensors

Gait disturbance is one of the most pronounced and disabling symptoms of cerebellar disease (CD). Generally, gait studies quantify human gait characteristics under natural walking speeds while mainly considering upper body movements. Therefore, the primary goal of this study was to investigate the influence of different walking speeds on different gait parameters of both the upper and lower body, as a result of disabilities caused by Cerebellar Ataxia (CA). We employed wearable sensor technology to identify the kinematic characteristics which best identify the gait abnormalities seen in CA. Measurements were made at self-selected slow, preferred and fast walking speeds. Velocity irregularity and resonant frequency characteristics were identified as key features of truncal and lower limb movements respectively. Subsequently, the differentiating features for both trunk and lower limb movements were combined to produce an even greater separation between the patients and the normal subjects, as well as better correlation with the expert clinical assessment (ECA) (0.86) and the Scale for the Assessment and Rating of Ataxia (SARA) (0.62). The different speed of walking conditions resulted in varying degrees of the separation and the correlation. Moreover, the contribution of the extracted features was examined using the random forest algorithm. Clinically observable truncal medio-lateral movements express the disability at relatively slow gait speeds while the anterior-posterior movements captured by the sensory mechanisms characterises the disability across all walking speeds. The importance of selected dominant features from the trunk and lower limb suggest that overall clinical assessments are predominantly influenced by the lower body peripheral movements, particularly at higher cadences.