<!doctype html public "-//w3c//dtd html 4.0 transitional//en"> <html> <head> <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"> <meta name="GENERATOR" content="Mozilla/4.75 [en]C-CCK-MCD BDP81800 (Windows NT 5.0; U) [Netscape]"> <title>toe-off.html</title> </head> <body> <h2> Determining Initial-contact &amp; Toe-off from Kinematic Data alone</h2> Very often in gait analysis we need to normalise data to the gait cycle, which means we need to know when initial-contact (IC) and toe-off (TO) occur. Of course, usually we can determine these times by looking at the ground reaction force (GRF) data. But what do we do if we don't have a force-plate? <br>&nbsp; <h3> Vertical kinematics</h3> <p><br>The first thing we could do is look at the <b>height</b> of the toe marker. <p><img SRC="ipsiheight.JPG" BORDER=0 height=183 width=182> <p>At toe-off, the toe obviously lifts off the floor. However, as you can see, it is difficult to decide exactly when it comes off the floor. The peak toe height also occurs (obviously) quite a lot later than than toe-off. The reverse is true for initial-contact. <p>How about the derivative of the toe height (<b>vertical velocity</b>)? <p><img SRC="ipsivel.JPG" BORDER=0 height=183 width=236> <p>It's clear that the positive (up-going) spikes of vertical velocity approximate to the toe-off events, although there's also a second wave immediately afterwards. Heel-strike is not at all defined. <p>What about if we low-pass filter the data at the natural frequency, or cadence, of the gait (about 1 Hz)? <p><img SRC="filtvert.JPG" height=181 width=243> <p>Now both the positive and negative peaks of the velocity signal nicely define both toe-off and initial-contact. <br>&nbsp; <h3> Antero-posterior kinematics</h3> <p><br>What about the <b>AP velocity</b>? <p><img SRC="ipsiapvel.JPG" BORDER=0 height=182 width=170> <p>Clearly, this signal is of no use. However, the <b>acceleration</b> signal (smoothed at 6 Hz) again shows clear peaks at toe-off: <p><img SRC="ipsiAPacc.JPG" BORDER=0 height=182 width=200> <p>In fact, if you look at the <a href="/faq/acc.mov">acceleration of the foot</a> recorded by an accelerometer attached to the heel, you will see that the acceration has large spikes at these two times. <br>&nbsp; <h3> Medio-lateral kinematics</h3> <p><br>Finally, for completeness, let's look at the <b>medio-lateral </b>direction: <p><img SRC="contrafvel.JPG" BORDER=0 height=183 width=210> <p><img SRC="contrafacc.JPG" BORDER=0 height=181 width=216> <p>The peaks of the acceleration signal have a reasonable correlation with contralateral toe-off, but also with foot contact. <br>&nbsp; <h2> Conclusion</h2> <p><br>The best kinematic indicators of TO are <b>peak</b> <b>positive vertical toe velocity (especially if it is low-pass filtered at the gait cadence), positive vertical toe acceleration</b> and <b>positive AP toe acceleration</b>. To improve reliability, it might be best to combine these three measures. The only indicator of IC appears to be <b>peak negative filtered vertical velocity</b>.&nbsp; <hr WIDTH="100%">Chris, <p>&nbsp; Predicting toe off from kinematic force-time curves has one or two subtle <br>problems. The phrase, "toe off," refers to the instant of final contact <br>between the shoe and the floor. The point of final contact between shoe and <br>floor is generally the very front, bottom edge of the shoe. A marker placed <br>at this position could not reach peak vertical velocity until sometime after <br>toe off, when it was actually moving up. <p>&nbsp; Many biomechanists place a marker on the lateral side of the 5th <br>metatarsal head and not the front edge of the shoe. I examined our data it <br>agrees with your statement: the instant of toe off occurs at the time of the <br>peak vertical velocity of the metatarsal head. So, one subtlety is that we <br>can predict the instant the toe or front edge of the shoe leaves the floor <br>by examining the kinematics of another body point. <p>&nbsp; Our data do not produce a good prediction of heel contact from the <br>metatarsal head kinematics. The metatarsal head is still moving forward and <br>downward after heel contact as the ankle joint plantarflexes. The peak <br>downward velocity of the met head occurs about 20 ms after heel contact in <br>our data. The second subtlety in this process is that methead kinematics can <br>be used as you suggested to estimate heel contact but a fudge factor (i.e. a <br>prediction equation) would be needed to adjust the predicted time of heel <br>contact to an earlier value. <p>Thanks for your time, <br>Paul&nbsp; <hr WIDTH="100%"> <br>Hi Chris, <p>Looked at your page regarding the use of toe marker kinematics as <br>indicators of heel contact and toe off and was surprised by your comment in <br>relation to toe a/p velocity that "Clearly, this signal is of no use." I <br>encountered the same problem of determining heel contact and toe off in <br>amputees ambulating on a treadmill. I had some footswitch data initially <br>but the tests required about an hour of walking in total and the <br>footswitches (FSR type, from a Motion Labs EMG Footswitch system) caused <br>problems inside the sound shoe, often stayed in the closed position in the <br>prosthetic shoe and soon wore out if I placed them on the outsole of the <br>shoe. After examining the footswitch and toe marker data that I had, I <br>opted for an algorithm which involved using the MPJ horizontal velocity. <br>The threshold for heel contact and toe off was <br>Vmin + 0.30(Vmax-Vmin) <br>and it seemed to work very well for the data I had. Just looking at the <br>data on your toe-off.html page, it looks to me like it would work very well <br>for your data as well. <p>I'm attaching a figure and caption as a word document. Hope you can read it. <p>Cheers, <br>Tim Bach. <br><img SRC="image772.JPG" height=543 width=549> <br><b>Figure 5.3. Relationship between MPJ horizontal velocity and footswitch data</b> <p>In this trial, horizontal velocity of the MPJ marker varied between approximately 1.5 m&middot;s-1 and -1.0 m&middot;s-1, the speed of the treadmill. Footswitch data is shown below with larger signals indicating times of heel switch closure and smaller signals indicating times of toe switch closure. Note the variability in the footswitch signals particularly for the toe which is absent altogether for the first step. The 30% velocity threshold used to detect foot contact data is indicated by a horizontal line in the top graph. The relationship between this threshold and the timing of heel contact and toe off for a single step is indicated by the two vertical lines. <h3> <hr WIDTH="100%">From BIOMCH-L March 2002</h3> I am seeking a  robust' method to accurately identify a user-defined <br>cyclical  event' in a sequence of kinematic data.&nbsp; Specifically, given <br>the 3D trajectory of a marker placed on the human body, measured with <br>noise, at equal space time intervals, with some missing data, and given <br>a user- defined  event' or a (small) series of  events', automatically <br>locate the remainder of the  events' within the sequence.&nbsp; For example, <br>given the 3D trajectory of a right heel marker during gait, measured at <br>180 Hz for 480 s (86400 samples), and given the first 6  heel-floor <br>contacts' visually identified by an operator, automatically locate the <br>remainder of the  heel-floor contacts' within the sequence. <p>Option one would be to reduce the frequency content of my input signal <br>to only include frequencies below 5 Hz (digital filter), or 4-5 <br>harmonics (Fourier), or some optimally restricted cut-off frequency <br>(GCVSPL available at http://isb.ri.ccf.org/software/sigproc.html). <br>Differentiation of the  smoothed' data will then allow me to reliably <br>identify zero-crossings or inflection points in the neighborhood of the <br> event'.&nbsp;&nbsp; But I'm reluctant to reduce the frequency content of my <br>signal to the point where I can reliably find  events' since I also <br>require high-accuracy estimates of the  event' times to calculate <br>high-accuracy estimates of the cycle-times (I will fit a curve using <br>GCVSPL at the located  event' frame, then interpolate frames to find the <br>exact time at which a zero-crossing or inflection point occurs). <p>Option two could use a user-defined  worm', centered at an event, to <br>find similar later occurring events by minimizing RMS difference. <br>Stanhope et al. (Stanhope SJ, Kepple TM, McGuire DA, Roman NL. <br>Kinematic-based technique for event time determination during gait. Med <br>Biol Eng Comput 1990 Jul;28(4):355-60) published a paper using this <br>method and they recommend the use of a sagittal plane, 5-7 frame  worm' <br>to identify gait events.&nbsp; However, to improve  event' location accuracy <br>they digitally filtered their data thereby potentially altering <br>frequency content and subsequently true  event' time. [I should note <br>that I have tried this method and get about 30% false-positives and 25% <br>false-negative return rates]. <p>Option three could use fuzzy system identification to identify gait <br>events.&nbsp; This technique was used by Ng and Chizeck (Ng SK, Chizeck HJ. <br>Fuzzy model identification for classification of gait events in <br>paraplegics IEEE Trans on Fuzzy Systems 1997 5(4): 536-544).&nbsp; Although <br>this process could use raw data, they filtered at 5 Hz, altering <br>frequency content.&nbsp; Even so, it seems that the technique was able to <br>only correctly identify 80% of  events', a value unacceptable for my <br>application. [Although I must note that this method is implementation <br>attractive since MATLAB has built in callable-functions.&nbsp; Maybe someone <br>knows how to improve accuracy?]. <p>Option four could use cluster analysis (Kaufman L, Rousseeuw PJ. <br>Finding groups in data:&nbsp; An introduction to cluster analysis, John Wiley <br>&amp; Sons, Inc., 1990) but to my knowledge this has not been attempted to <br>locate gait events and I'm reluctant to follow a path that may be <br>fruitless. <p>So, in conclusion I seek help from the Biomechanics community.&nbsp; Ideally, <br>someone out there has the perfect source code (I still program in <br>Fortran!) that I could embed within my application.&nbsp; As usual practice, <br>I will post a summary of the responses to my query. <p>Thanks, <p><a href="mailto:pierryn@MCMASTER.CA">Michael Pierrynowski</a> <br>McMaster University <br>Canada <br> <hr WIDTH="100%"> <h4> from CGA March 2002</h4> Hello CGA <p>Usually, in identfying the gait cycle events using Vicon motion <br>analysis, I identify the 'toe off' event with the point at which the foot begins to <br>move forward i.e. the start of swing. An alternative would be to identify the <br>point at which the foot leaves the ground i.e. when there is clearance <br>(which is what 'toe off' literally means). We have recently analysed the <br>gait of a little girl with marked foot drag . In her case, the latter <br>definition results in a 30% shorter 'swing phase' compared with the <br>former definition. Which definition of 'toe off' event is correct? Or do labs <br>have specific protocols for different presentations? <p>Please advise. <p><a href="MAILTO:dominic.lloyd-lucas@sth.nhs.uk">Dominic Lloyd-Lucas</a> <br>Clinical Scientist <br>Sheffield Teaching Hospitals <br>UK <br>email: dominic.lloyd-lucas@sth.nhs.uk <hr WIDTH="100%"> <br>Dear Dominic and others, <p>My thoughts are that this dilemma is due to the terminology used. Foot- <br>off defines the point where the foot comes 'off' the ground, however in <br>this case foot-off does not correspond with the point where foot <br>progression begins. It may be necessary to describe the cycle in terms <br>of both stance and swing phases (defined by the foot-off and foot- <br>contact events) as well as foot-stationary and foot-progression phases <br>(defined by the points at which foot progression begins and ceases). <p>It will be interesting to hear others' views on this topic. <p>Regards, <br><a href="mailto:p.mills@mailbox.gu.edu.au">Pete Mills</a> <hr WIDTH="100%"> <br>Hello readers, <p>It is the function of the leg that matters. Let's assume that we accept that <br>the conventional tasks during the gait cycle are weight acceptance, single <br>support and double support for the stance phase and limb advancement for the <br>swing phase. <p>If someone drags the foot then its main function at that moment is not <br>propulsion any more (as it was at the end of the stance phase) but <br>progression. In that sense this should be named the swing phase. <p>Obviously it's more convenient to look for the point in time when the size <br>of the force vector drops to zero but it probably does not relate to the <br>function of the leg. <p>Gabor <p>-- <br><a href="mailto:G.J.Barton@livjm.ac.uk">Dr Gabor Barton (MD)</a> <br>Senior Lecturer in Biomechanics <br>Centre for Sport &amp; Exercise Sciences, Liverpool John Moores University <br>Room 2.51 Henry Cotton Campus, 15-21 Webster Street, Liverpool, L3 2ET <br>Tel: +44 (0)151 231 4333/4321&nbsp;&nbsp; Fax: +44 (0)151 231 4353 <br>E-mail: G.J.Barton@livjm.ac.uk <hr WIDTH="100%"> <p><a href="/faq.html">Back to FAQ</a><a href="/faq.html"><img SRC="signpost.gif" BORDER=0 height=32 width=33></a> </body> </html>