Essential tremor evaluation system with Eval-net method via accelerometer signal

Jianguo Pan,Shun Liu,Qin Ni

DOI: https://doi.org/10.1109/SOLI48380.2019.8955003

2019-01-01

Abstract:Essential tremor (ET), a common neurological movement disorder, is typically measured with rating scales by neurologists during a clinical visit. This method undervalues the tremor fluctuations during patients' daily activities. This paper proposes an ET severity evaluation system based on advanced deep learning algorithms. We combine the convolutional neural network (CNN) and long short-term memory (LSTM) to build the multi-layer ET evaluation model (Eval-net). Experiments based on acceleration data gathered from 20 ET patients and five healthy subjects are conducted to verify the feasibility of the proposed method. Experimental results show that the Eval-net model can classify six typical activities and four levels of ET severity with better performance than previous approaches in recognition accuracy. This study demonstrates that the proposed method makes improvements in recognizing human activities and evaluating tremor severity based on acceleration data.