Parkinson's Disease Severity Estimation using Deep Learning and Cloud Technology

The management of motor complications in Parkinson's disease (PD) is an unmet need. This paper proposes an eHealth platform for Parkinson's disease (PD) severity estimation using a cloud-based and deep learning (DL) approach. The system quantifies the hallmark symptoms of PD using motor signals of patients with PD (PwPD). In this study, the dataset named "The Michael J. Fox Foundation-funded Levodopa Response Study" is used for the development and evaluation of computational methods focusing on severity estimation of motor function in response to the levodopa treatment. The data is derived from a wearable inertial device, named Shimmer 3, to collect motion data from a patient's upper limb which is more affected by the disease during the performance of some standard activities selected by MDS-UPDRS III and at home while performing daily life activities (DLAs). Seventeen PwPD were enrolled from two clinical sites, who have varying degrees of motor impairment. An incorporated cloud-based framework is proposed where patients' motion data is saved in MS Azure cloud where an automatic evaluation of patients' motor activities in response to the levodopa dose is performed using continuous wavelet transform and CNN-based transfer learning approach. Experimental results show that the efficiency and the robustness of the proposed procedure are proven by 90.0% accuracy for tremor estimation and 86.4% for bradykinesia, with good performance in terms of sensitivity and specificity in each class.