Symptom-based, Dual-channel LSTM Network for The Estimation of Unified Parkinson's Disease Rating Scale III

Unified Parkinson Disease Rating Scale (UPDRS) is assessed by neurologists to detect patients with Parkinson's disease (PD) and rate the disease severity. However, the assessment has shown to be subjective. In this work, we developed a novel symptom-based, dual-channel LSTM network to estimate UPDRS part III (severity of motor complications) from motion data recorded using two wearable sensors. Unlike the existing sensor-based methods that require the subjects to perform specific activities, our new approach is entirely passive and based on subjects' free-body movements. The network was trained and evaluated using a subject-based, leave-one-out cross-validation, on motion data recorded from 24 subjects with idiopathic PD as they performed a variety of daily routine activities. The estimated UPDRS-III scores had a moderate correlation with the scores from the clinical examinations. Our analysis demonstrates the potential of our approach in estimating PD severity scores during subjects' routine daily-life, free-body movements. Such an algorithm could provide objective and comprehensive information to detect subtle changes during the disease progression and help the treating physician to provide effective management of the disease.