大腿截肢患者的残肢肌电运动识别

目的研究利用大腿残肢肌电信号进行下肢运动模式识别的方法,探讨肌电信号控制下肢假肢的可能性。方法采集15名大腿截肢者残肢侧股直肌、股外侧肌、阔筋膜张肌、股二头肌、半腱肌、臀大肌6块肌肉的表面肌电信号,提取肌电信号的6种时域、频域特征,利用支持向量机对平地行走、上楼梯、下楼梯、坐下、起立5种下肢运动模式进行识别。结果利用残肢肌电信号可以实现5种下肢运动模式的在线识别,对同一受试者同次测试数据识别率为94%,同一受试者的多次混合数据识别率为85%,对不同受试者混合数据识别率为74%。通过特征优化,仅利用3块肌肉的2个特征,对同一受试者的同次测试数据识别率仍可达92%。对平地行走、上楼梯、下楼梯3种动作的识别,同一受试者同次测试数据识别率为100%,同一受试者的多次混合数据识别率为98.33%,对不同受试者混合数据识别率为93.33%。结论仅仅利用残肢肌电信号能够实现运动意图的在线识别,通过对同一患者使用前的多次数据训练,有望达到较高的识别率。研究结果为肌电运动识别用于下肢假肢控制奠定了基础。

Motion recognition based on EMG signals of residual limb in transfemoral amputee

Objective To study the method of lower limb movement pattern recognition using the electromyographic (EMG) signals of the residual thigh muscles, and explore the possibility of lower limb prosthesis control based on the EMG signals. Methods Fifteen transfemoral amputees were selected as subjects, and the subjects were required to complete 5 kinds of motion, including level walking, stair ascent, stair descent, standing up and sitting down. The surface EMG signals from 6 muscles of the thigh stump were collected from each subject, including rectus femoris, vastus lateralis, tensor fascia lata, biceps femoris, semitendinosus and gluteus maximus. Six kinds of time-domain and frequency domain features of the EMG signals were extracted, and 5 kinds of motion patterns were recognized by the support vector machine. Results Five kinds of motion patterns could be recognized online by EMG signals of the residual thigh muscles. By single experimental data from one subject, the recognition rate was 94%; for the same subject, by the data mixed from two experiments, the recognition rate was 85%; for different subjects, the recognition rate was 74%. By feature optimization, using only two EMG features of 3 muscles, the recognition rate could reach 92% by single experimental data from one subject. For 3 kinds of motion patterns (level walking, stair ascent, stair descent), the recognition rate respectively was 100% using single experimental data from one subject, 98.33% using the data mixed from two experiments for the same subject, and 93.33% using the data from different subjects. Conclusions Simply using the thigh stump EMG signals to recognize movement intention is proved to be feasible. For each patient, by several times of training before using the EMG signals, the recognition rate is expected to reach an ideal state. The present work will lay a foundation for lower limb prosthesis control based on the EMG signals.