Monday, March 7, 2011

Prediction of muscle fiber type from powermeter data, part 4

by Andrew R. Coggan, Ph.D.

In previous blog entries in this series:

http://www.trainingandracingwithapowermeter.com/2010/12/prediction-of-muscle-fiber-type-from.html

http://www.trainingandracingwithapowermeter.com/2010/12/prediction-of-muscle-fiber-type-from_07.html

http://www.trainingandracingwithapowermeter.com/2010/12/prediction-of-muscle-fiber-type-from_20.html

I described a method for predicting muscle fiber type distribution from force-velocity (or power-velocity) data collected using an SRM. Unfortunately, as I have mentioned several times before other powermeters appear to be incapable of providing data with sufficient accuracy and/or temporal resolution to enable such calculations. For example, the original PowerTap recorded data only once every 1.26 s, which is generally insufficient for this sort of testing. Furthermore, data from a PowerTap are inherently "noisier" due to aliasing effects - that is, since the calculations are time- instead of event-based, each data point represents the average over a non-integer number of pedal revolutions, leading to values that tend to alternately high and low relative to the true value. Although more recent versions of the PowerTap can now record data at 1 s intervals (i.e., the current ANT+ standard), this aliasing problem still exists.

Unlike Saris's PowerTap, the Quarq CinQo is event-based, i.e., like the SRM it averages data over complete revolutions before broadcasting it as ANT+ messages. Unfortunately, however, the CinQo (or at least the one I had use of) seems prone to grossly overestimating power/grossly underestimating cadence when power and cadence are changing rapidly, e.g., when resuming pedaling after a brief pause. This is evident in Fig. 1 below, which shows the sort of abnormal power "spike" than can occur under such conditions:

Figure 1. Abnormal "spike" in power generated by Quarq CinQo when resuming pedaling after a brief pause.

In this particular example, the power is not especially high in absolute terms, but it is impossibly high relative to the cadence at which it was reportedly produced. This is more evident when the data from this file are analyzed using Quadrant Analysis, as shown in Fig. 2 below:

Figure 2. Quadrant analysis of the data file containing the abnormal power "spike" shown in Figure 1.

As shown in this figure, a number of data points were recorded in which the average effective pedal force calculated from the power and cadence exceeded ~1500 N, with a maximum value of ~2500 N. Since the peak force on the pedals is usually about twice the average force, this would imply peak forces of ~3000 to ~5000 N, or approximately 4-7x my body mass. Given that I can only lift ~1x my body mass when performing two-legged squats, such values are clearly artifactual in nature.

When performing force-velocity (or power-velocity) testing, the result of this tendency of the Quarq to overestimate power/underestimate cadence when resuming pedaling is a non-linear AEPF-CPV relationship, as shown in Figure 3 below:

Figure 3. Non-linear AEPF-CPV relationship generated by Quarq CinQo during force-velocity testing.


In theory it might be possible to avoid this issue by sampling the "raw" ANT+ data stream (which is broadcast at 4 Hz) and/or by judicious editing of the collected data. However, without access to additional equipment (in the first case) or more trustworthy data collected using another device (in the second case), such solutions seems to be markedly less-than-ideal.

The above data were recorded using an iBike iAero as the ANT+ receiving device, so it is possible that they reflect limitations in how the data are handled by it rather than limitations in how the data are originally generated by the Quarq. It is difficult to envision, however, how this might be possible and still obtain close agreement between Quarq/iBike and, e.g., PowerTap data under less challenging conditions. Furthermore, I have recently been successful in using a Lemond Fitness Revolution trainer paired with their Power Pilot to generate linear force-velocity data that closely match that provided by an SRM, even though, like the iBike, the Power Pilot uses the ANT+ protocol to record data once per second. This suggests (although certainly does not prove) that the non-linear nature of the Quarq/iBike force-velocity curve reflects the behavior of the former, and not the latter.

Method #2: Estimation of muscle fiber type from fatigability

Given that powermeters other than the SRM appear incapable of generating robust force-velocity cuves, how can somebody who doesn't own one estimate their muscle fiber type? One possibility is to rely on some other fiber type-specific property of muscle, e.g., fatigue resistance.

Compared with fast-twitch (type II) muscle fibers, type I muscle fibers contain more mitochondria, are surrounded by more capillaries, etc., making them better adapted for sustained, aerobic energy production. Conversely, type I fibers have a lower maximal ATPase activity, have lower activities of glycolytic and glycogenolytic enzymes, etc., meaning that they rely tend to rely less upon production of ATP from anaerobic sources, i.e., ATP/PCr breakdown and lactate production. For these and other reasons, type I muscle fibers are more fatigue resistant, even during short-duration, high intensity activities such as a 30 s Wingate-test type effort as shown in Fig. 4:

Figure 4. Relationship between muscle fiber type and fatigue index during a 30 s maximal effort.

The data in the above figure are drawn from a study that I conducted at Ball State University under the direction of Dr. Dave Costill, which eventually came to serve as my master's thesis (1). Compared to the regression relating optimal CPV to fiber type provided in the 2nd article of this series, the correlation between the fatigue index (i.e., the percentage decline in power during the test) and muscle fiber type is not as high. Given, however, that it is based upon a secondary characteristic of muscle fibers (i.e., fatigue resistance vs. contractile properties), this is perhaps not unexpected. The strength of the correlation is also comparable to that reported in the scientific literature in similar experiments. In any case, these data can be used to predict muscle fiber type using the equation below:

% type I area = 139.3 - 1.931 * fatigue index

R^2 = 0.596

P<0.05

S.E.E. = 9.7%

Final thoughts

In some regards, the formula presented above - and indeed, this entire series of blog entries - can be considered to be "navel gazing" of limited practical value. After all, when racing a bicycle it is the actual power you can produce over relevant durations (along with, e.g., tactics) that determine the outcome of a race, not your muscle fiber type per se. Furthermore, I strongly believe that "if it walks (sprints) like a duck (slow-twitcher) and talks (resists fatigue) like a duck (slow-twitcher), then it is a duck (slow-twitcher)", regardless of what a muscle biopsy might reveal or what formal tests/predictions I have described might suggest. Nonetheless, I do think that being able to "pin a single number on things" may help at least some people understand their own physiology just a little bit better, with this deeper insight hopefully helping them prepare better for/perform better in competition.

References

1. Coggan AR, Costill DL. Biological and technological variability of three anaerobic ergometer tests. Int J Sports Med 1984; 5:142-145.