北京生物医学工程

基于Hill肌肉模型的人体关节力矩智能预测

Human joint moment prediction based on artificial neural network

作者：熊保平史武翔林昱黄美兰杜民

单位：福建工程学院数理学院（福州 350116）福州大学物理与信息工程学院（福州 350116）福建中医药大学附属人民医院 (福州350004) 福建省医疗器械和医药技术重点实验室（福州 350116）

关键词：关节力矩预测；人工神经网络；极限学习机； Hill肌肉模型；输入变量

分类号：R318.01

出版年·卷·期（页码）：2021·40·1（11-23）

摘要：

目的人体关节力矩是康复评估和人机交互中非常关键的因素之一。它可以通过人工神经网络（artificial neural network ，ANN)模型以肌电等信号作为输入进行预测，但是由于人体结构的复杂性导致缺乏有效方法确定人工神经网络模型的输入变量。为此本文提出了一种基于Hill肌肉模型获取关节力矩神经网络预测最优输入变量的方法。方法利用Hill肌肉模型结合人体几何学知识建立关节力矩智能预测输入输出关系的数学模型，把Hill肌肉模型中肌肉纤维长度和收缩速度以及以关节自由度为支点的肌肉力臂等不可活体测量输入变量转换成关节自由度所关联肌肉的肌电信号以及这些肌肉所驱动关节自由度角度和角速度等可在线测量变量。结果实验中以本文获取变量作为极限学习机的输入，对一位在跑步机上以0.4, 0.5, 0.6, 0.7, 和0.8 m/s等5个不同速度行走的右下肢偏瘫患者所获数据进行测试。为了评估本文所提方法智能预测的泛化能力，实验在两个不同的泛化水平下进行，它们分别是只把前面三个低速数据（0.4, 0.5, 和0.6m/s）和全部五个速度的数据用于神经网络的训练并预测所有速度下的关节力矩值。实验是通过预测值与反向动力学计算值之间的归一化绝对误差和互相关系数评估。结果表明，本文所提输入变量与其他关节角和角速度作为输入方法相比预测关节力矩值更精准，除右踝关节内外翻展外其他关节力矩预测结果的最大归一化绝对误差为12.93%，最小平均互相关系数为0.89。结论该方法比目前通用多体反向动力学的输入变量少且可实现关节力矩值的在线预测，可为运动康复中实时步态分析和外骨骼机器人控制提供技术支持。

Objective Human joint moment is one of the most important factors in rehabilitation assessment and human-robot interaction, which can be predicted using an artificial neural network model. However, challenge remains as lack of effective methodologies to determine the input variables for the artificial neural network (ANN) model. Methods This study develops a novel method to determine the optimal input variables to the ANN based on the Hill muscle model for estimating lower extremity joint moments. In this method, we translate the muscle-tendon moment arm, velocity and length in the Hill muscle model to the online measurable variables, i.e. muscle span joints’ angles and angular velocities signals. We use moment-associated muscles’ electromyography (EMG) signals together with these variables of the muscle as the inputs to the ANN for joint moment prediction. Results The method is tested on the experimental data collected from a subject with a highly functional hemiplegic stroke，who is walking on a treadmill with different speeds, i.e. 0.4, 0.5, 0.6, 0.7, and 0.8 m/s. We first use the data collected in all speeds as the training data to predict the joint moment. To evaluate the generalization ability of our method, we then repeat the same test using only low speed (0.4, 0.5 and 0.6 m/s) data to train the ANN model for the joint moment prediction at all speeds. The accuracy of prediction is evaluated by using the normalized root-mean-square error and cross correlation coefficient between the predicted joint moment and multi-body dynamics moment. Our results suggest that our method can predict joint moments with a higher accuracy than those obtained by using other joint angles and angular velocities as inputs, except the right ankle inversion-eversion, the maximum normalized root mean square error of each joint moment of the lower limb was 12.93%, and the minimum average cross-correlation coefficient was 0.89. Conclusion The proposed method provides us with a useful tool to predict joint moment using online measurable variables, which uses less variables than the commonly used multi-body dynamics approach with a comparable accuracy. This method may facilitate the research on real-time gait analysis and exoskeleton robot control in motor rehabilitation.

参考文献：

[1] Ardestani MM, Zhang X, Wang L, 等. Human lower extremity joint moment prediction: A wavelet neural network approach[J]. Expert Systems with Applications, 2014, 41(9): 4422–4433.

[2] Park H S, Peng Q, Zhang L Q. A Portable Telerehabilitation System for Remote Evaluations of Impaired Elbows in Neurological Disorders[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2008, 16(3): 245–254.

[3] Zhang S, Guo S, Gao B, 等. Design of a Novel Telerehabilitation System with a Force-Sensing Mechanism[J]. Sensors, 2015, 15(5): 11511–11527.

[4] Schmidt A. Movement pattern recognition in basketball free-throw shooting[J]. Human Movement Science, 2012, 31(2): 360–382.

[5] Pfeiffer M, Hohmann A. Applications of neural networks in training science[J]. Human Movement Science, 2012, 31(2): 344–359.

[6] Joshi D, Mishra A, Anand S. ANFIS based knee angle prediction: An approach to design speed adaptive contra lateral controlled AK prosthesis[J]. Applied Soft Computing, 2011, 11(8): 4757–4765.

[7] Rupérez M J, Martín-guerrero J D, Monserrat C, 等. Artificial neural networks for predicting dorsal pressures on the foot surface while walking[J]. Expert Systems with Applications, 2012, 39(5): 5349–5357.

[8] Burnie J, Brodie D A. Isokinetics in the assessment of rehabilitation: a case report[J]. Clinical Biomechanics, 1986, 1(3): 140–146.

[9] Callaghan M J, Mccarthy C J, Al-omar A, 等. The reproducibility of multi-joint isokinetic and isometric assessments in a healthy and patient population[J]. Clinical Biomechanics, 2000, 15(9): 678–683.

[10] R, Happee, and, et al. The control of shoulder muscles during goal directed movements, an inverse dynamic analysis[J]. Journal of Biomechanics, 1995.

[11] Kim Y-H, Phuong B T T. Estimation of Joint Moment and Muscle Force in Lower Extremity During Sit-to-Stand Movement by Inverse Dynamics Analysis and by Electromyography[J]. Transactions of the Korean Society of Mechanical Engineers A, 2010, 34(10): 1345–1350.

[12] Lloyd D G, Besier T F. An EMG-driven musculoskeletal model to estimate muscle forces and knee joint moments in vivo[J]. Journal of Biomechanics, 2003, 36(6): 765–776.

[13] Meyer A J, Patten C, Fregly B J. Lower extremity EMG-driven modeling of walking with automated adjustment of musculoskeletal geometry[J]. PLOS ONE, 2017, 12(7): e0179698.

[14] Sartori M, Reggiani M, Farina D, 等. EMG-Driven Forward-Dynamic Estimation of Muscle Force and Joint Moment about Multiple Degrees of Freedom in the Human Lower Extremity[J]. PLOS ONE, 2012, 7(12): e52618.

[15] Uchiyama T, Bessho T, Akazawa K. Static torque–angle relation of human elbow joint estimated with artificial neural network technique[J]. Journal of Biomechanics, 1998, 31(6): 545–554.

[16] Luh J-J, Chang G-C, Cheng C-K, 等. Isokinetic elbow joint torques estimation from surface EMG and joint kinematic data: using an artificial neural network model[J]. Journal of Electromyography and Kinesiology, 1999, 9(3): 173–183.

[17] Song R, Tong K Y. Using recurrent artificial neural network model to estimate voluntary elbow torque in dynamic situations[J]. Medical and Biological Engineering and Computing, 2005, 43(4): 473–480.

[18] Hahn M E. Feasibility of estimating isokinetic knee torque using a neural network model[J]. Journal of Biomechanics, 2007, 40(5): 1107–1114.

[19] Ardestani M M, Zhang X, Wang L, 等. Human lower extremity joint moment prediction: A wavelet neural network approach[J]. Expert Systems with Applications, 2014, 41(9): 4422–4433.

[20] Kipp K , Giordanelli M , Geiser C . Predicting net joint moments during a weightlifting exercise with a neural network model[J]. Journal of Biomechanics, 2018:S002192901830294X.

[21] May R , Dandy G , Maier H . Review of Input Variable Selection Methods for Artificial Neural Networks[M]// Artificial Neural Networks - Methodological Advances and Biomedical Applications. InTech, 2011.

[22] Hill Archibald Vivian. The heat of shortening and the dynamic constants of muscle[J]. Proceedings of the Royal Society of London. Series B - Biological Sciences, 1938, 126(843): 136–195.

[23] Kumar D, Rudolph K S, Manal K T. EMG-driven modeling approach to muscle force and joint load estimations: Case study in knee osteoarthritis[J]. Journal of Orthopaedic Research, , 30(3): 377–383.

[24] Feng G, Huang G B, Lin Q, 等. Error minimized extreme learning machine with growth of hidden nodes and incremental learning.[J]. IEEE Transactions on Neural Networks, 2009, 20(8): 1352–1357.

[25] Huang G B , Zhu Q Y , Siew C K . Extreme Learning Machine: Theory and Applications[J]. Neurocomputing, 2006, 70(1-3).

[26] Huang G-B, Chen L. Enhanced random search based incremental extreme learning machine[J]. Neurocomputing, 2008, 71(16): 3460–3468.

[27] Huang G B, Chen L. Letters: Convex incremental extreme learning machine[J]. Neurocomputing, 2007, 70(16): 3056–3062.

[28] Reinbolt J A, Schutte J F, Fregly B J, 等. Determination of patient-specific multi-joint kinematic models through two-level optimization[J]. Journal of Biomechanics, 2005, 38(3): 621–626.

[29] Hug F. Can muscle coordination be precisely studied by surface electromyography?[J]. Journal of Electromyography and Kinesiology, 2011, 21(1): 1–12.

[30] Menegaldo L L, Fleury A de T, Weber H I. Moment arms and musculotendon lengths estimation for a three-dimensional lower-limb model[J]. Journal of Biomechanics, 2004, 37(9): 1447–1453.

[31] Sartori M, Reggiani M, Bogert A J van den, 等. Estimation of musculotendon kinematics in large musculoskeletal models using multidimensional B-splines[J]. Journal of Biomechanics, 2012, 45(3): 595–601.

[32] An K N, Takahashi K, Harrigan T P, 等. Determination of muscle orientations and moment arms[J]. Journal of Biomechanical Engineering, 1984, 106(3): 280–282.

[33] Tanzil S M S, Hoiles W, Krishnamurthy V. Adaptive Scheme for Caching YouTube Content in a Cellular Network: Machine Learning Approach[J]. IEEE Access, 2017, 5(99): 5870–5881.

[34] Yang Y, Wu Q M J, Wang Y. Autoencoder With Invertible Functions for Dimension Reduction and Image Reconstruction[J]. IEEE Transactions on Systems Man & Cybernetics Systems, 2016, PP(99): 1–15.

[35] Liu Y, Shih S-M, Tian S-L, 等. Lower extremity joint torque predicted by using artificial neural network during vertical jump[J]. Journal of Biomechanics, 2009, 42(7): 906–911.

[36] Zhang F, Li P, Hou Z-G, 等. sEMG-based continuous estimation of joint angles of human legs by using BP neural network[J]. Neurocomputing, 2012, 78(1): 139–148.

[37] Henriksen M, Aaboe J, Bliddal H, 等. Biomechanical characteristics of the eccentric Achilles tendon exercise[J]. Journal of Biomechanics, 2009, 42(16): 2702–2707.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录