부산대학교 도서관

Cerebellar Ataxia is a neurological disorder without an approved treatment. Patients will have impaired and uncoordinated motor functionality making them unable to complete their day-to-day activities. Ataxia clinics are established around the world to facilitate research and rehabilitate patients. However, the patients are generally evaluated by human – observation. Therefore, machine learning based data analysis is popular on motion captured via sensors. There are many neurological tests designed to analyse the motor impairments in different domains (such as upper limb, lower limb, gait, balance and speech). Clinicians follow scoring protocols to record the severity of patients for each domain test. This paper delivers a clinical assessment platform combining 12 neurological tests in 5 domains. It captures motion (from BioKin sensors), haptic and audio data (from the tablet or laptop screen). A data analysis system is hosted in a remote server which evaluates data to produce a severity score via different models built for each neurological test. The assessment platform clients and server communicate via a cloud buffer system. The scores input by the clinicians and predicted by the machine learning models are logged in the cloud database. This enables clinicians and doctors to view and compare the history of patient diagnosis. The server system is structured for automated score model upgrades via prompted approval. Thus, the most viable scoring model could be accommodated for each test based on longitudinal studies.