Transfer learning based quantitative assessment model of upper limb movement ability for stroke survivors

Abstract

Stroke survivors often suffer from movement disability. The accurate assessment of their movement function is an important part of rehabilitation therapy and is the premise of making individualized movement prescriptions. Many previous studies have shown that inertial measurement unit (IMU), which contains accelerometer, gyroscope, and magnetometer, etc., can be used to quantitatively assess the movement function of stroke survivors. However, the assessment results can be influenced by sensor placement. To solve this problem, this paper proposed a novel method which combines random forest and transfer learning algorithm. The experimental results showed that by using the proposed method, the traditional quantitative assessment models established at one sensor placement can be easily transferred to adapt to other sensor placements. In other words, a quantitative assessment model that is free of sensor placement can be achieved.