Australasian Physical & Engineering Sciences in Medicine Volume 26 Number 3, 2003 TECHNICA REPORT Quadrices muscles vastus medialis obliues, rectus femoris and vastus lateralis comared via electromyogram bicoherence analysis R. J. Simeoni and P. M. Mills School of Physiotheray and Exercise Science, Griffith University, Gold Coast, Australia Abstract Bicoherence analysis is alied to electromyogram (EMG) data for vastus medialis obliues (VM), rectus femoris (RF) and vastus lateralis (V) uadrices muscles of 18 adult male subjects for isometric knee extension exercise. Mean average bicoherence for VM, RF and V is 30.9 5.8, 26.0 1.2 and 25.4 1.4% resectively and reeated measures ANOVA differentiates the muscles on the basis of average bicoherence (F = 16.2 (1, 17), = 0.0009, VM cf. V and F = 15.4 (1, 17), = 0.0011, VM cf. RF). Prominent regions reresentative of strong second-order hase couling between constituent EMG freuencies are identified within VM and RF bicoherence sectra. No such rominent regions are identified for V which is thought to be less activated than VM during the secified task. Hence, the degree of second-order hase couling may increase as the level of muscle activation increases. The subject grou consists of young (24.0 0.9 years) and elderly (68.9 0.9 years) subgrous that cannot be differentiated by standard indices (median and sectral edge freuency) to within < 0.05 using the Mann-Whitney test. Average bicoherence differentiates the subgrous for RF (T = 9 (8,10), < 0.005) but not for VM or V. The alication of a bicoherence threshold that takes harmonic amlitude into account grahically differentiates the subgrous for all muscle tyes. The findings suggest that nonlinear rocesses lay a role within the EMG generation rocess and suort a mechanomyogram bicoherence analysis study that shows nonlinear rocesses occur within active muscle fibre twitch summation atterns. A otential exists for bicoherence analysis to comlement standard EMG freuency analysis techniues. Key words bicoherence, bisectrum, electromyogram, EMG, uadrices Introduction Bicoherence analysis rovides information on the degree of second-order hase couling that exists between constituent freuencies of electrocardiogram (ECG), electroencehalogram (EEG) and mechanomyogram (MMG) biosignals 1-8. Such hase couling characterises nonlinear biological systems such as the brain and central nervous system and can reflect the clinical state of these biological systems (e.g., the state of consciousness) 1, 3. Desite the clinical usefulness of bicoherence analysis when alied to the above biosignals and the fact that MMG bicoherence sectra have shown nonlinear rocesses to be significant within active muscle fibre twitch summation atters, a recent study 9 reorts a negative finding for bicoherence analysis alied to vastus lateralis (V) uadrices electromyogram (EMG) data during isometric knee extension exercise. Other EMG bicoherence Corresonding author: R. Simeoni, School of Physiotheray and Exercise Science, Griffith University, PMB50 Gold Coast Mail Centre, Gold Coast QD 9726, Australia, Tel: +61 7 5552 8451, Fax: +61 7 5552 8674, Email: r.simeoni@griffith.edu.au Received: 20 November 2002; Acceted: 4 August 2003 analysis studies are not aarent in the literature. The aim of the resent study is to further investigate whether a otential exists for bicoherence analysis to be used as an assessment tool in the area of EMG analysis. The study is a follow-u to the above-mentioned EMG bicoherence analysis study 9 and comares the resonses of vastus medialis obliues (VM), rectus femoris (RF) and V uadrices muscles for 18 adult male subjects during isometric knee extension exercise. During this task, there is a likelihood that the levels of activation of VM, RF and V will differ, and hence the resonses of the three uadrices muscles to bicoherence analysis may also differ. The rimary reason why the levels of activation could be exected to differ is that a number of studies 10-14 cited in a comrehensive review 15, reort a significantly greater level of activation for VM than for V during some tyes of knee extension exercise. However, it should be noted that whether the level of activation of VM differs significantly from that of V during various knee extension exercises remains a toic of controversy in biomechanics and hysiotheray 15, 16. The resent study therefore contributes to the body of literature that addresses this controversy. From a hysiological ersective, sustained muscle activity is known to roduce grouings of motor unit action otentials (MUAPs) into distinct bursts of activity. These grouings have been described as a synchronisation, though technically the term synchronisation does not correctly 125
describe the effect 17. However, the onset of synchronouslike behaviour (harmonisation) suggests that hase couling between constituent EMG freuencies may also increase with sustained muscle activity. Hence, a ossible hysiological basis exists as to why VM, RF and V might become differentiable through bicoherence analysis if their levels of activation differ. The subject grou consists of young and elderly subgrous and the resent study also aims to investigate the otential for EMG bicoherence analysis to differentiate these subgrous. Standard EMG freuency analysis differentiates the young and elderly 18. Such age differentiation results from an increase in MUAP amlitude with age due to a loss of motor neurons and a subseuent eriheral reinnervation of muscle fibres by surviving motor neurons, as reorted for V 17, 18. Also, motor control differs between the young and elderly (e.g., during walking 19 ) and VM and V activations aear to differ in resonse to a change in motor control 16. EMG hase couling differences between the young and elderly might therefore occur for our isometric situation and be detectable through bicoherence analysis. If the uadrices muscles or subgrous for the same muscle are differentiable through bicoherence analysis, then a otential would exist for bicoherence analysis to comlement standard EMG freuency analysis techniues. The comlementary information rovided, i.e., the degree of second-order hase couling between constituent EMG freuencies, would give insight into MUAP harmonisation which is known to be affected by muscle fatigue and changes in neuromuscular function. Theory Bicoherence and bisectral analysis investigate the degree of second-order hase couling between the constituent harmonics, X n ( n ) cn sin( nt n ), of a biosignal, where n is angular freuency, c n is amlitude, n is hase and t is time. The bisectrum, B,, is a measure of the degree of second-order hase couling between X, X and X, while also taking into account the amlitudes of these harmonics. B, is calculated by breaking the biosignal into successive eochs and alying the following euation: B, X X X m1 m1 c i cc e where the index, m, is the current eoch and X is the comlex conjugate of X. Bicoherence, bic, m m (1), is often described as the Simeoni et al Quadrices muscles comared via bicoherence normalised bisectrum 2, 3, 5, since it is found by dividing B, by the suare root of a real (not comlex) trile roduct, harmonic: cc c, that involves the ower, c n 2, of each i c cc e m1 m bic, 100% (2) m1 c cc m bic, = 0% indicates that the hases of X and X bic X, vary indeendently, while, = 100% indicates that a constant hase relationshi is maintained between the harmonics. For more detail on and an introduction to bicoherence analysis theory, the reader is referred to references 3, 9. For the resent study, bicoherence and bisectrum information is also combined in a new manner and the theory behind this new analysis method is now exlained. A bicoherence sectrum is theoretically indeendent of harmonic amlitude. That is, bic, may eual 100% for harmonics that only make a small contribution to the EMG freuency sectrum. To give more weighting to harmonics with large amlitudes, we introduce a threshold whereby bic, is set to zero if the c cc sum in E.2 is less than an arbitrary threshold that relies on the identification of the maximum c cc for each eoch. The threshold is defined as 0.1 the minimum of the identified maximum c cc values. It should be noted that the resulting bicoherence sectrum differs from the bisectrum. That is, although the bisectrum reflects harmonic amlitude, a relatively large B, value may be obtained if the amlitudes of three harmonics are small but the hase couling strong. Our roosed sectrum would set bic, to zero in this case. Also, a relatively large B, value may be calculated if the magnitudes of the harmonics are large but the hase couling weak, if random vectorial cancellation in E.1 is non-zero (ractically, ideal vectorial cancellation to zero is unlikely). Our roosed sectrum, being bicoherence based, would give a relatively small bic, value that would be set to zero in this second case. Thresholds have been alied to the bicoherence sectrum 3 but not in the manner described above. Method Subjects 10 young (24.0 0.9 years) and 8 elderly (68.9 0.9 years) male subjects articiated in the study. All subjects were free from known musculoskeletal and neurological abnormalities. The subjects were considered as 18 adult 126
males for comarisons between uadrices muscles and as subgrous for young versus elderly comarisons for the same uadrices muscle. EMG exerimental rocedure Biolar configurations of Sentry Ag-AgCl 10 mm regelled, disosable disk electrodes were laced on the skin suerficial to VM, RF and V, with an inter-electrode distance of 20 mm. A reference electrode was laced on the skin suerficial to the left anterior suerior iliac sine. Electrode imedance did not exceed 2 k for any activereference electrode air. A total of 54 EMG data segments were collected for each uadrices muscle as each subject erformed three 6 s isometric knee extension actions with the right leg which was attached to a strain gauge cell. The task goal was to maintain a knee extension torue of 25 Nm -1. Subjects received visual feedback on their generated torue in real time. The knee remained in 45 degrees of flexion and the hi in 90 degrees of flexion. An isometric task was chosen to minimise non-stationarity effects 20. EMG data were samled at 1000 Hz using a Bioac Data Acuisition system in conjunction with Acknowledge software and raw EMG data were amlified by a factor of 1000. To test for reeatability between EMG channels, measurements were reeated for two subjects with leads and amlifiers swaed between VM, RF and V. Bicoherence analysis rocedure The three 6 s segments of EMG data collected from each subject for each uadrices muscle were adjusted to a mean of zero and filtered with a second-order, zero hase Butterworth filter (12 db/octave rolloff) with low- and high-ass filter cut-offs set to 450 and 10 Hz resectively. The 10 Hz cut-off and filter order were chosen to allow the detection of nonlinearities associated with motor unit firing rates below 20 Hz. The EMG data segments were divided into 32 eochs, each consisting of 500 oints and of 0.5 s duration. Eoch overla was 75%. A Blackman window 21 was alied to, and a Fast Fourier Transform (FFT) calculated for, each eoch. FFT resolution was 2 Hz. A bicoherence sectrum was then generated for each data segment by calculating bic, for all linear freuencies, fn n / 2, above 10 Hz such that f f 300 Hz, using a rogram develoed in the Matlab rogramming language. The 300 Hz maximum was chosen in accordance with the average sectral edge freuency (SEF) for the V EMG freuency sectra (see Results section), where SEF is defined as the freuency below which 95% of the ower exists. Bicoherence sectra were also calculated with the 10% of maximum real trile roduct threshold described in the Theory section for the same eoch regime described above. Statistical analysis rocedure All mean values are reorted as mean SD. Median freuency (MF) and SEF comarisons between the young and elderly subgrous for each uadrices muscle were made using the Mann-Whitney test as skew and kurtosis Simeoni et al Quadrices muscles comared via bicoherence distribution differences between the subgrous were observed in some cases. bic(, ) is the average bicoherence of a given bicoherence sectrum and subgrou comarison for each uadrices muscle, based on this index, was also made using the Mann-Whitney test. The test statistic uoted for the Mann-Whitney test is T = S n e (n e +1)/2, where S is the sum of the ranks for the elderly subgrou and n e is the number of elderly subjects. Reeated measures ANOVA was erformed in SAS software using the Proc Mixed rocedure to detect for differences in bic(, ) between uadrices muscles for all subjects. This statistical method was emloyed because second-order hase couling indeendency between muscles for the same subject could not be assumed. The effects of muscle and trial on bic(, ) were assessed using a two factor within-subject design. As no significant effect of trial (F = 0.056 (2,17), = 0.538) or interaction between muscle and trial (F = 2.88 (2,17), = 0.0548) were identified, the bic(, ) values were then averaged across trials for each muscle and subject and the effect of muscle on bic(, ) assessed using a single factor within-subject design. The selection of the aroriate covariance matrix was based on evaluation of Akaike s Information Criterion and Schwarz s Bayesian Criterion. Where an effect of muscle was identified, contrasts were used to determine which levels of muscle were significantly different. was set at 0.05 for all analyses. Reeated measures ANOVA was also used to detect for bic(, ) differences between the subgrous for all muscles combined. Results MF and SEF of EMG freuency sectra for all subjects and trials are 63 6 and 172 30 Hz (VM), 64 10 and 194 39 Hz (RF) and 91 22 and 282 39 Hz (V). An examle V biosignal and its corresonding freuency sectrum (MF = 102 Hz, SEF = 288 Hz) are shown in Figs.1 and resectively. Figure 1. Examle vastus lateralis EMG biosignal and its corresonding freuency sectrum. Median and sectral edge freuencies are 102 and 288 Hz resectively. 127
Examle bicoherence sectra for VM, RF and V for simultaneously collected EMG data for the same subject are shown in Figs.2, and resectively. bic, ) ( for each sectrum is 38.8, 28.0 and 25.5% resectively. Mean bic(, ) for all subjects and trials is 30.9 5.8% (VM), 26.0 1.2% (RF) and 25.4 1.4% (V), and reeated measures ANOVA differentiates VM from RF and V based on bic(, ) (F = 16.2 (1, 17), = 0.0009, VM cf. V and F = 15.4 (1, 17), = 0.0011, VM cf. RF). RF is not differentiated from V (F = 2.96 (1, 17), = 0.1033, RF cf. V). Figs.3, and show for VM, RF and V resectively, bicoherence sectra suerimosed to give an average sectrum for all subjects and trials. MF and SEF Mann-Whitney test comarisons between the young and elderly subgrous do not reveal a significant (T = 21 (8,10), < 0.05) difference between the subgrous for any of the uadrices muscles, with T > 21 for all comarisons. Subgrou MFs and SEFs for each muscle are omitted for conciseness. Mean bic(, ) for the young and elderly subgrous are resectively 33.1 6.4 and 28.2 4.0% (VM), 26.7 0.9 and 25.2 0.9% (RF), and 25.7 1.6 and 25.0 0.9% (V). Differentiation of the subgrous based on bic(, ) occurs for RF (T = 9 (8,10), < 0.005) but not for VM or V. Reeated measures ANOVA gives a significant age effect for all muscles combined (F = 5.23 (1, 16), = 0.0362). Figs.4 and show for the young and elderly subgrous resectively, VM bicoherence sectra with the 10% of maximum real trile roduct threshold alied, suerimosed for all subjects and trials. Similar sectra are shown for RF in Figs.4 and (d) and for V in Figs.4(e) and (f), while Figs.4(g) and (h) show such sectra for V without the threshold alied. bic(, ) calculated with threshold does not differentiate the subgrous for any of the uadrices muscles. Discussion and conclusions The examle V biosignal in Fig.1 shows the maintenance of consistent muscle activity and is tyical for the subjects. The corresonding freuency sectrum in Fig.1 is tyical of other standard EMG freuency sectra reorted in the literature 17 with a eak freuency near 40 Hz and a smaller initial eak at lower freuencies due to the grouing of MUAPs. Quadrices muscle comarison VM bicoherence sectra exhibit comaratively high bic(, ) values and reeated measures ANOVA differentiates VM from V ( = 0.0009) based on this index. Similarly, VM is differentiated from RF ( = 0.0011). VM bicoherence sectra also dislay regions of ronounced second order hase couling, as seen in Simeoni et al Quadrices muscles comared via bicoherence Fig.2, for a number of subjects. If these rominent regions are random in nature, then the suerosition of all VM sectra is exected to give a relatively featureless sectrum due to random cancellation. In actuality, when VM sectra are suerimosed for all subjects and trials (Fig.3), rominent regions of second-order hase couling remain, indicating hase couling commonality within the EMG generation rocess. These rominent regions in Fig.3 are broad with indistinct boundaries and bic, values >35%. The regions indicate that freuencies aroximately between 30 and 100 Hz dislay strong second-order hase couling with two narrow bands of freuencies near 20 and 40 Hz, while freuencies aroximately between 100 and 175 Hz dislay strong second-order hase couling with freuencies from 80 to 120 Hz. The differentiation between VM and V may be directly related to comaratively greater levels of VM activation found during some tyes of knee extension exercise 10-14. Hence, the findings suggest that nonlinear rocesses, in the form of second-order hase couling, may lay an increasingly imortant role within the EMG generation rocess as muscle activation level increases. Muscle force and fatigue levels might therefore also influence any nonlinear rocesses that occur. A rominent region of second-order hase couling is also identified for all RF bicoherence sectra suerimosed (Fig. 3) and this region generally involves freuencies below 100 Hz. Prominent regions of second-order hase couling are not aarent for V in Fig.3. Since the secified task was modest and easily maintained by most subjects, it is ossible that V bicoherence sectra would exhibit a ositive finding for a more strenuous task. The effect of increasing muscle fatigue on the bicoherence sectrum is the intended toic of a follow-u research roject. It is of interest that bic(, ) dislays a relatively small deviation between individuals for RF and V (coefficients of variation = 4.6 and 5.5% resectively), comared to the indices of MF and SEF which dislay resective coefficients of variation of 16 and 20% (RF) and 24 and 14% (V). This small deviation is of interest because standard indices such as MF, SEF and average rectified value of an EMG biosignal, can vary markedly between individuals erforming the same task or for muscles activated to similar levels, and thus these indices do not rovide an absolute measure of muscle activity that allows for direct individual comarison (unless maximum voluntary isometric contraction information is known in the latter case). Based on the small deviation between subjects dislayed by bic, ), it cannot be stated that ( bic(, ) will rovide the absolute measure described above. However, the ability to rovide such a measure should not be excluded and remains an interesting ossibility, articularly since bic, is inherently normalised. Tissue filtering is a common henomenon in the area of EMG analysis 22 and all EMG freuency sectra are affected 128
Simeoni et al Australas. Phys. Eng. Sci. Med. Vol. 26, No 3, 2003 x Quadrices muscles comared via bicoherence bic(z,z) bic(z,z) Figure 2. Bicoherence sectra for vastus medialis, Figure 3. Bicoherence sectra suerimosed for all subjects and trials for vastus medialis and rectus femoris, showing regions of rominent second-order hase couling between constituent EMG freuencies. The comarable sectrum for vastus lateralis is relatively featureless. bic(z,z ) = 38.8%, rectus femoris, bic(z,z ) = 28.0% and vastus lateralis, bic(z,z ) =25.5%, for simultaneously collected EMG data for the same subject. 129
(d) bic(, ) Simeoni et al Quadrices muscles comared via bicoherence Prominent regions within Figs.3 and involve freuencies much lower than those affected by tissue filtering. Unfiltered comonents of EMG freuency sectra reserve hase relationshis. Desite the fact that average MF and SEF are similar for VM and RF, bic(, ) for RF is significantly less than that for VM and corresonds more closely with the value for V. For a given subject bic(, ) for VM does not significantly change when the maximum value of f f within bicoherence sectra is reduced to 200 Hz, a value closer to the average SEF for VM. Reeatability measurements with leads and amlifiers swaed between VM, RF and V show that the findings are not due to differences between channel sensitivity or signal-to-noise ratio. (e) (g) Figure 4. Bicoherence sectra suerimosed with 10% of maximum real trile roduct threshold alied for all young (left) and elderly (right) subjects and trials. and are for vastus medialis, and (d) for rectus femoris, and (e) and (f) for vastus lateralis. (g) and (h) are for vastus lateralis without the threshold alied. by this henomenon to various degrees. Noting that the average SEFs for VM and RF are lower than that for V due to tissue filtering, it must be established that the above findings are not a conseuence of this filtering. This ossibility, and the ossibility of the findings being due to different sensitivities between sets of EMG channels, are excluded as follows: (f) (h) Subgrou comarison The young and elderly subgrous are not differentiated on the basis of MF or SEF using the Mann-Whitney test for any of the muscles. The findings that the Mann-Whitney test differentiates the subgrous on the basis of bic(, ) for RF ( < 0.005) and that reeated measures ANOVA of bic(, ) gives a significant age effect for all muscles combined ( = 0.0362), raise the ossibility that known motor neuron oulation or motor control-associated changes in the elderly 17-19 might indirectly affect secondorder hase couling between constituent EMG freuencies during certain tasks. Since the suerosition of all sectra yields no distinctive features for V, it is of interest that for V the alication of the 10% of maximum real trile roduct threshold grahically differentiates the young and elderly subgrous (see Figs.4(e) and (f) resectively). Here, differentiation is on the basis of a more intense region of rominence existing within the sectrum for the young. Grahical differentiation is not aarent without the threshold alied as shown by comarative Figs.4(g) and (h). Sectra, with threshold alied, for VM (Figs.4 and ) and RF (Figs.4 and (d)) grahically differentiate the subgrous in a similar manner to that of V. Differentiation, based on bic(, ), of the subgrous is not significant for Figs.4 to (f), indicating that a more secific index based on defined regions within bicoherence sectra 6 is reuired to match the grahical differentiation result. That is for examle, Fig.4 contains higher bic, values than Fig.4, while Fig.4 dislays a broader region of rominence, resulting in similar bic, ) values for the two sectra. Hence, the ( calculation of bic(, ) over a confined range of freuencies is reuired for differentiation. 130
In conclusion, bicoherence analysis clearly differentiates the VM uadrices muscle from the RF and V uadrices muscles and this differentiation may be directly related to greater VM activation levels comared to V reorted in the literature. Hence, second-order hase couling may increase with the level of muscle activation. A searate study is needed to further exlore the hysiological basis of this differentiation and thus establish the manner in which bicoherence analysis might be used as a uantitative EMG assessment tool. It cannot be concluded that bic(, ) rovides an absolute measurement that allows for direct comarison of muscle activity level between individuals. However, the ability to rovide such as absolute measure remains a ossibility based on the small deviation in bic(, ) between individuals erforming the same modest task for RF and V and the normalised nature of the bicoherence calculation. The grahical differentiation of the young and elderly subgrous by the alication of a real trile roduct threshold is an interesting finding erhas indirectly related to known motor neuron oulation or motor control-associated differences between the subgrous. Acknowledgement Thanks to Garry Dennis and Justin Keogh for giving of their time in the Biomechanics laboratory whilst carrying out this study, and to Kerrie Evans, Belinda Beck and isa Gilkes for helful discussions on uadrices activation. The authors also wish to thank Clare Weber for roviding statistical analysis advice. References Simeoni et al Quadrices muscles comared via bicoherence 1. Barnett, T., Johnson,.C., Naitoh, P., Hicks, N. and Nute, C., Bisectrum analysis of electroencehalogram signals during waking and sleeing, Science. 172: 401-402, 1971. 2. Ning, T. and Bronzino, J. D., Autoregressive and bisectral analysis techniues: EEG alications, IEEE Eng. Med. Biol. March: 47-50, 1990. 3. Sigl, J. C. and Chamoun, N. G., An introduction to bisectral analysis for the electroencehalogram, J. Clin. Monit. 10: 392-404, 1994. 4. iton, J. M., Dabke, K. P., Alison, J. F., Cheng, H., Yates,. and Brown, T. I. H., Use of the bisectrum to analyse roerties of the human electrocardiograh, Austral. Phys. Eng. Sci. Med. 21: 1-10, 1998. 5. Muthuswamy, J., Sherman, D.., and Thakor, N. V., Higherorder sectral analysis of burst atterns in EEG, IEEE Trans. Biomed. Eng. 46: 92-99, 1999. 6. Nahm, W., Stockmanns, G., Petersen, J., Gehring, H., Konecny, E., Kochs, H. D. and Kochs, E., Concet for an intelligent anaesthesia EEG monitor, Medical Informatics. 24: 1-9, 1999. 7. Schack, B., White, H., Helbig, M., Schelenz, Ch. and Secht, M., Time-variant non-linear hase-couling analysis of EEG burst atters in sedated atients during electroencehalic burst suression eriod, Clin. Neurohysiol. 112: 1388-1399, 2001. 8. Orizio, C., iberati, D., ocatelli, C., Grandis, D. De. and Veicsteinas, A., Surface mechanomyogram reflects muscle fibres twitches summation, J. Biomech. 29: 475-481, 1996. 9. Simeoni, R. J. and Mills, P. M., Bicoherence analysis of uadrices electromyogram during isometric knee extension, Austral. Phys. Eng. Sci. Med. 26: 12-17, 2003. 10. Hanten, W. P. and Schulties, S. S., Exercise effect on electromyograhic activity of the vastus lateralis muscle, Phys. Ther. 70: 561-565, 1990. 11. Hungerford, D. S. and Barry, M., Biomechanics of the atellofemoral joint, Clin. Ortho. 144: 9-15, 1979. 12. King, A.C., Ahiles, T. A., Martin, J. E. and White, R., EMG biofeedback-controlled exercise in chronic arthritic knee ain, Arch. Phys. Med. Rehabil. 65: 341-343, 1984. 13. eveau, B. F. and Rogers, C., Selective training of the vastus medialis muscle using EMG feedback, Phys. Ther. 60: 1410-1415, 1980. 14. Soderberg, G. and Cook, T., An electromyograhic analysis of uadrices femoris muscle setting and straight leg exercises, Clin. Ortho. 63: 1434-1438, 1983. 15. Grabiner, M. D., Koh, T. J. and Draganich,. F., Neuromechanics of the atellofemoral joint, Med. Sci. Sorts. Exerc. 26: 10-21, 1994. 16. Cowan, S.M., Bennell, K.B., Hodges, P.W., Crossley, K.M. and McConnell, J., Delayed onset of electromyograhic activity of vastus medialis obliues relative to vastus lateralis in subjects with atellofemoral ain syndrome, Arch. Phys. Med. Rehabil. 82: 183-189, 2001. 17. Enoka, R. M., Neuromechanical basis of kinesiology, 2 nd ed., Human Kinetics, Chamaign, Illinois, 168-170, 1994. 18. Stalberg, E. and Fawcett, P.R., Macro EMG in healthy subjects of different ages, J. Neurol. Neurosurg. Psychiatry. 45: 870-878, 1982. 19. Mills, P. and Barrett, R., Anterior-osterior contact velocity during normal walking is greater in elderly than young men, Proceedings of the 3 rd Australasian Biomechanics Conference. Griffith University. January/February: 61-62, 2000. 20. Poivanov, D. and Mineva, A., Testing rocedures for nonstationarity and non-linearity in hysiological signals, Mathematical Biosciences. 157: 303-320, 1999. 21. Harris, F. J., On the use of windows for harmonic analysis with the discrete Fourier transform, Proc. of the IEEE. 66: 51-83, 1978. 22. De uca, C. J., The use of surface electromyograhy in biomechanics, J. A. Biomech. 13: 135-163, 1997. 131