Abstract:Motor function Evaluation is an indispensable part of rehabilitation treatment for stroke patients. As a rule, this process is manually performed by the experience of therapists which is subjective. Recently, research on rehabilitation assessment based on data provided by wearable sensors has provided convenient and practical solutions. In order to objectively evaluate the function of hemiplegic upper-extremities (HUEs) patients, we propose 2-dimension convolution neural network (2DCNN) for the Brunnstrom Stage (BS) classification. As a contrast, an attention Bidirectional Long Short-term Memory model (Attention-BiLSTM) and traditional machine learning approaches Support Vector Machine (SVM), Random Forest (RF) and multi-layer perceptron (MLP) were also adopted, which uses the surface electromyography (sEMG) and inertial measurement unit (IMU) data provided by a single wearable sensor on the upper extremity of stroke patients. The 2DCNN can achieve classification accuracy of 93.46% in segments for 3 levels (BS IV, BS V and healthy status), which is better than Attention-BiLSTM and traditional machine learning approaches (p < 0.05). The proposed model can fastly and efficiently assess stroke patients-functional level at a short time interval (300 ms), which can assist therapists with stroke rehabilitation.

Brunnstrom Stage Evaluation Using a Single Sensor for Hemiplegic Upper-Extremities Based on Deep Features