There are many methods to establish a gait database with experiments done when the sampled person walks for a few minutes. Thats means the person walked repetitively for some gait cycles (one complete cycle is 100% gait cycle) as it only takes between 1-2 seconds to complete a gait cycle (either from a heel strike to next heel strike of the same leg or from toe off to next toe off of the same leg).
As there are many gait cycles involved in the experiment duration, how is the gait data obtained is pre and processed to finally come out as data between 0 - 100% gait cycle only? I mean, the data is only plotted over 100% gait cycle. Is the whole cycles data averaged to only one cycle? If so, how? For examples joint angle vs 100% gait cycle, joint torque vs 100% gait cycle.
Great question. You’ve actually outlined most of it in your post.
First, you are right, there is essentially an averaging of all the different gait cycles. This is done to try and get a more robust estimate of the joint angle or moment over the gait cycle.
To do it, we identify all of the individual gait cycles of interest from the data collection. So, if you have lots of data, you could identify all of the heel strikes and take the data between each of the heel strikes as Individual gait cycles. Let’s assume you do this for 10 gait cycles. You now have 10 curves showing your outcome of interest (joint angle, moment, etc.) over the gait cycle. One potential problem with this is that each cycle has a different duration. To overcome this, people will normalize the curve to span 0-100 (representing 0-100%) instead of the original time. If all of the curves are normalized in this same way, it means that every curve is now the same number of points across it (101). Now that all of the curves are the same length, we can calculate the mean and standard deviation for every point on the curve. Finally, we can now plot the mean values from each point along the curve to get the mean curve. We can also using the standard deviation at each point to calculate a confidence interval for each point along the curve.
The described method is for a single person, but the nice thing about this approach is that you can normalize each person to the same length and can do the same mean type analysis across all people.
Thanks Anthony for the great answer.
Yes, each curve may have different time duration or period or stride time. That means for each cycle, the number of data points may be different. So it is hard to find the mean of data points from different curve at any instant. I found that we cannot plot the data for each cycle over time because time is increasing over the experiment duration. The only way is to plot each cycle over gait cycle as cycle repeats (independant of time).
From your answer, does it mean that we have to interpolate each curve with the same time interval i.e. for example we take interpolate 100 points on each curve than find the mean of the points at an interval?
Another way, I tried to plot the whole experiment data with scatter plot (using Excel - see plot) over gait cycle as the cycle repeats after 100%. The pattern is there but I cannot find a way to find the average and standard deviation at each interval. Its like normalizing the scatter plot.
Regarding the method you described for a single person, the normalize plot is only applicable to that sampled person, so the normalized plot cannot be applied to population?
Yes, interpolate to 101 points (you need 101 points to get 0-100%).
For your population (sample) curve comment, first it's not a population unless you include every single person in the group of interest - i.e., for osteoarthritis it would be a population curve if we included every single person in the world with osteoarthritis - otherwise it's just a sample of people with osteoarthritis.
Now, to create the sample curve, you could first create the mean curves for each individual person (using the normalization/interpolation to 101 points method). Then, take the mean for each all 101 points for each individual person (the same as you averaged over trials within a person), and again you can calculate a standard deviation to create a confidence interval on this. Remember, when calculating your confidence interval, if you are want to have a confidence interval of the distribution of individual participants curves used the STD * zscore of interest, and if you want a confidence interval on the mean value over all of the people you'll have to convert your STD to standard error and do the same calculation (SE * zscore of interest).
Hi Anthony A Gatti ,
I have this Lokomat data from its recording for one therapy session. It consists of 75k rows of data of many parameters for 5 minutes recording (0 - 299.996s). The frequency of sampling is 250Hz (thats why in 5 minutes producing 75k rows).
Lokomat uses gait index (0-1000) to identify a complete cycle but the start of cycle is different from standard normalized gait graph which starts from heel strike. The gait index 0-1000 can be normalized to gait cycle 0%-100%.
See attached graphs :
1. hip position vs time and gait index vs time for 30s
2. whole data from 300s plotted with normal plot (left) and scatter plot (right)
3. standard single cycle plot
Approximately in 30s there are 10 cycles, so in 300s there should be about 100 cycles. I tried to plot them with normal plot and scatter plot (similar to graph before); the pattern is there. The question is, how can I produce a normalized graph (mean +/- SD) for the whole 5 minutes (~100 cycles) in one cycle? Next how to reduce the data from 0-1000 to 0-100 points? Thanks.
- Badges
- Science topic
- Similar topics
- Correction