TremorSense: Tremor Detection for Parkinson's Disease Using Convolutional Neural Network

Parkinson's Disease (PD) hand tremors are common symptoms in all stages of PD. PD tremors have a severe influence on patients' daily quality of life. Wearable technology can be used to help detect, quantify, and mitigate these PD tremors. Among the wearable technology, PD tremor detection is the primary step for further analysis and treatment using wearable devices. Some researchers have explored PD rest tremor detection. However, less research has been done concerning postural tremor and action tremor detection, which are difficult to classify only using frequency-domain features. In this paper, we propose TremorSense, a PD tremor detection system to classify Parkinson's Disease hand tremors. TremorSense utilizes accelerometers and gyroscopes as wearable sensors on patients' wrists to collect data from 30 PD patients. We develop the TremorSense Android application that connects the sensors via Bluetooth to save the data. Furthermore, we design an 8-Layer Convolutional Neural Network (CNN) to classify PD rest, postural, and action tremors. We evaluate the CNN model with self-evaluation, cross-evaluation and leave-one-out evaluation, and the accuracies for all three evaluations are greater than 94%.