ABSTRACT
A detection approach for Parkinson’s disease on-off period based on wrist posture is proposed to address the problem of high-precision drug on-off period detection for Parkinson’s disease patients in a universal medical scenario. To categorize the condition of Parkinson’s disease on-off phase, the motion IoT-based sensor data on the wrist is utilized to compute the attitude, get the information aspects of the wrist posture, and use it as the input of the convolutional neural network. Comparative experiments on clinical patient test data in the hospital show that using attitude information improves detection accuracy by 20.3% compared to the optimal results using motion sensor raw data; when compared to the current optimal network structure, the convolutional neural network used in this method maintains a similar detection accuracy (88.7%) but is reduced by 90.4%. Furthermore, investigations on free movement data of clinical patients in the hospital reveal that this system can accurately forecast the patient’s on-off state under unconstrained actions, with a 91.5% on-phase and 94.4% off-phase accuracy rate.
