Real – Time Locomotion Classification using Transient EMG

Sarthak Pati*1, Deepak Joshi2, Ashutosh Mishra2and Sneh Anand2

1Department of Biomedical Engineering, Manipal University – Manipal

2Centre for Biomedical Engineering, Indian Institute of Technology – Delhi

______

ABSTRACT

This study analyses the possibility of a real – time locomotion classifier based on transient surface Electromyography signals. This classifier aims to provide an improved control mechanism for use in an above – knee powered prosthesis. The types of locomotion under consideration are Fast Walk (FW), Normal Walk (NW), Running (RUN) and Slow Walk (SW).

Keywords:EMG-based classification, LDA, locomotion classification, real time.

INTRODUCTION

The present – day medical scenario is in dire need of intelligent prosthesis. Many studies have strived to deliver the near – natural abilities to prosthetics and have also succeeded to some extent. The major problem faced while designing an intelligent prosthesis is the necessity of real – time processing of the signal under consideration. Current work in the field of lower limb prosthesis has showcased prosthesis the use kinematic variables to classify locomotion (Deepak Joshi, et. al.) and use of Electromyography (EMG) signal to classify ascending/descending of stairs (Parker, et. al.).

This works encompass the use of various feature – extraction tools like Fourier Analysis and time – domain feature analysis to obtain reliable features that can be used to distinguish between the various locomotion modes under consideration.

METHODOLOGY

A total of 7 subjects in the age group of 21 – 27 years without any neuro – muscular disorders have been considered for this study. The experimental protocol was properly explained to them and the written permission was obtained to use their data for the study. The EMG signals are extracted using Noraxon © series wireless EMG acquisition System. Standard disposable plug-in type electrodes have been used. This classifier system has been designed for above – knee amputees; hence EMG signals from only the major thigh muscles have been considered. These muscles are enumerated below (refer figure 1) :

  1. Rectus Femoris– M1
  2. Vactus Laterali– M2
  3. Semitendinous– M3

Figure 1 : Human Thigh Muscles (Image courtesy : Henry Grey' Anatomy of the Human Body)

Figure 2 : Electrode Placements on subjects (Orthotics and Prosthetics Lab., AIIMS)

Since this classifier is intended for use in a real – time system, only time – domain features (Phinyomark A., et. al.; Huang H., et. al.; Preece S.J., et. al.) are being considered. The various extracted are given below :

For all the features stated below, “N” is the total number of samples in the EMG signal “”.

1)Integrated EMG (IEMG) – It is defined as the summation of the absolute values of the sEMG signal.

This parameter is generally used as an onset index to detect muscle activity. Its can be expressed as shown below,

2)Root Mean Square (RMS) – It is modelled as amplitude modulated Gaussian random process whose RMS is related to the constant force and non-fatiguing contraction. It is expressed as,

3)Slope Sign Changes (SSC) – This represents the number of times the slope of the signal changes from positive to negative. This can be calculated by,

Where,

4)Variance (VAR) – It is the measure of dispersion of a set of data points around their mean value. It’s expressed as shown,

Where “x” is the entire signal and“µ” is its mean.

5)Waveform Length (WL) – It represents the cumulative length of the waveform over the entire time segment. Its expression is given as,

7)Zero Crossings (ZC) – It denotes the number of times the sEMG signal changes its sign.

In this study, LDA has been used because of its extensive used in real time applications (K. Englehard, et. al.; F.H.Y. Chan, et. al.). There also have been some other studies done which have used Artificial Neural Network for use as a classifier (L.J. Hargove, et. al.). These features are analysed using Linear Discriminant Analysis (LDA)in the MATLAB® environment and the resultant transformation matrix is used to design a 2–stage threshold–based classifier system. The signals from all the three muscles are studied and the time–window where maximum variance of the signal occurs is considered. After extensive correlation coefficient analysis, this window has been fixed in the 66.7ms to 100ms time–frame.

RESULTS AND DISCUSSION

In the first stage of the LDA classifier (refer figure 2), it is observed that only FW and SW are not getting perfectly classified. Thus, to resolve this, a second stage LDA classifier is implemented (refer figure 3). The transformation matrix are obtained using the standard LDA Eigen problem (Maaten L.J.P. van der). This classification has been obtained by extensively studying the EMG signal data from all the three muscles. It was found that the muscles M2 and M3 contribute maximum to the net variance of the feature matrix and hence features from these two muscles have been considered.

Figure 2 : LDA classification between all the four locomotion modes

Figure 3 : LDA classification between FW and SW

In the single stage classifier, RUN and NW are being classified with 100% accuracy and FW with 20% accuracy. And in the second stage classifier, there is 100% accuracy in classification of FW and SW.

CONCLUSION

Thus, we can conclude that by using a Linear Discriminant Analysis based classification system, we can construct a real – time classifier which gives more than 99.89% accuracy rate. With such a high amount of accuracy and relatively low algorithm complexity, this system is practical and efficient enough to be implemented in an intelligent prosthesis device.

REFERENCES

F.H.Y.Chan, Y.S.Yang, F.K.Lam, Y.T.Zhang, P.A. Parker. Fuzzy EMG Classification for Prosthesis Control; IEEE Transactions on Rehabilitation Engineering, Vol.8, No.3, September [2000]

Deepak Joshi, Sneh Anand. Study of circular cross correlation and phase lag to estimate knee angle: an application to prosthesis; Int. J. Biomechatronics and Biomedical Robotics [in press]

K.Englehart, B.Hudgins, P.Parker, S.Maryhelen. Time-Frequency Representation for Classification of The Transient Myoelectric Signal; 20th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Vol. 20, No 5 [1998]

K.Englehart,B. Hudgins. A Robust, Real-Time Control Scheme for Multifunction Myoelectric Control; IEEE Transactions on Biomedical Engineering, Vol.50, No.7, July [2003]

L. J.Hargrove, H.Huang, A. E.Schultz, B. A.Lock, R.Lipschutz, T. A. Kuiken. Toward the Development of a Neural Interface for Lower Limb Prosthesis Control; Delsys Prize Winner [2008]

H.Huang, T.A.Kuiken, R.D. Lipschutz. A Strategy for Identifying Locomotion Modes Using Surface Electromyography; IEEE Transactions on Biomedical Engineering, Vol.56, No.1, January [2009]

L.J.P. van der Maaten – An Introduction to Dimensionality Reduction Using MATLAB

P.Parker, K.Englehart, B. Hudgins. Myoelectric signal processing for control of powered limb prostheses; Journal of Electromyography and Kinesiology [2006]

S.J.Preece, J.Y. Goulermas, L.P.J. Kenney, D. Howard, K. Meijer, R. Crompton – Activity Identification Using Body-Mounted Sensors : A Review of classification Techniques; Physiological Measurements, April [2009]

A. Phinyomark, C. Limsakul, P. Phukpattaranont. A Novel Feature Extraction for Robust EMG Pattern Recognition; Journal of Computing, Vol 1, Issue 1, December [2009], ISSN: 2151-9617

______

* Corresponding Author

E-mail address: