Crown Tools - Head Trauma Measurement Discussion
This blog explains the design philosophy associated with the Crown Tools Project which is an entry in the Sudden Impact Design Challenge.
I hope this discussion helps others tackle this difficult design challenge.
The Crown Tools are going to monitor helmet impacts that athletes might encounter and transmit results via Bluetooth to an android smart phone. Analog Devices is supplying phenomenal sensors and a great MCU, so the major problems for this challenge will be understanding the kinematics of impact, packaging and programming. Packaging accelerometers in a helmet such that the readings will accurately indicate what trauma the head is experiencing is a very difficult physics and biomedical engineering problem. It is also a difficult mechanical engineering job and a difficult electrical design job requiring a custom printed circuit board. Of course, since it is mounted on the head, it needs to be small and light weight.
Since I play hockey every week all year, I expect to mount the Crown Tools on my hockey helmet and view results on my smart phone. I certainly would not be looking to bang my head, but I can capture some video of the device in a hockey game and separately show video of the helmet and sensor sustaining impacts without a real head involved and of course show the resulting impact information.
Some Background on Measurement of Head Trauma
I have to apologise to the audience for the following long winded dissertation relating acceleration of a helmet to head trauma, but this challenge is being judged on performance and it is necessary to explain what the system does and why. Actually, as long as this explanation looks, it is hard to coherently condense years of research into a few paragraphs.
The ultimate objective of a user carrying instrumentation to measure impacts is to provide information that is useful in providing optimal treatment for users who suffer high acceleration of their heads.
The mechanisms of how a head gets injured from high acceleration are exceedingly complex and multifaceted. For example the brain tissue may be bruised by contact with the skull, or there may be micro lesions, or broken blood vessels, or cell damage, or relative displacements caused by extreme shear stresses, compressive stresses, tension stresses or local pressures as shockwaves propagate and reverberate through the brain tissue. The precise distribution of forces on the head and their waveforms all affect which mechanisms cause damage and where. The resulting injuries can also exhibit a wide range of adverse symptoms, such as loss of consciousness, disorientation, headaches, confusion, nausea, concentration problems, mood swings, dizziness, memory loss, insomnia, fatigue, anxiety, drowsiness, irritability, depression, blurred vision, sensitivity to light or sound, loss of balance, ringing in the ears, altered smell or taste, and even changes in personality. Since the brain controls many other bodily functions, they can also be affected. And these are just the range of mild TBI symptoms.
The head has many built-in methods of protection, such as a strong round skull and cerebrospinal fluid surrounding the brain, but every brain and its protection are different and injuries due to similar events on different heads can be quite different.
Measuring injuries at the cellular level in real time is likely a long ways off and would likely require extensive invasive instrumentation. However what we can do is measure the magnitude of impacts and try to correlate this to subsequent medical diagnoses. Once a large enough database is accumulated, and statistically analyzed, the impact data may be useful in making medical decisions regarding treatment.
Much work has been done to develop a model of head injury, particularly in vehicle crash scenarios and has resulted in a head injury criterion (HIC). This method uses accelerations of the skull to calculate a HIC "score", which can be used to statistically predict general levels of injury.
Obtaining measurements on a human for calculation of an accurate HIC score are problematic because placing a series of accelerometers on the skull requires a fairly invasive amount of instrumentation. It would be better to embed the accelerometers in the bone, but even if that is ruled out it at least probably starts with shaving the head and having accelerometers in direct contact with the skin, which is already too invasive for the average person to tolerate. If the impact force happens to be applied directly where a sensor is, it will not have the same readings as if it were mounted nearby, as the surface of the skin will have a faster acceleration (as it compresses) than the underlying skull. and point load readings will be very different from distributed load readings. You could mount sensors all around the head and try to deduce what the skull is doing by correlation with extensive experimental data (which may not exist yet), however this now becomes too expensive for a project like this, not to mention a lot of wires and apparatus on the head. And finally, any apparatus inside the helmet has the potential to cause injury or increase injury severity.
To turn this into a manageable problem, we can take a less invasive, more global approach and use helmet accelerations measured on the outside of the helmet to determine the magnitude of the impact which can then be correlated to medical outcomes. We could use a "HIC" calculation on helmet accelerations to compute an impact magnitude, but this number would not be directly useful in terms of normal HIC injury prediction because a "HIC" calculated using helmet acceleration will be dramatically different from a HIC calculated from skull acceleration in the same event.
The helmet surface is usually a relatively hard material and when it encounters another hard material, the local acceleration or deceleration can be enormous - orders of magnitude greater than the acceleration of the head in the helmet. This fact makes it difficult to determine head acceleration from helmet acceleration. Additionally, the hard shell of a helmet will vibrate and resonate when struck and these vibrations can have high accelerations at the helmet surface without being transmitted through the pads to the wearer.
One other issue is that angular acceleration can also cause traumatic brain injury and depending on the geometry of mounting, a linear accelerometer may not see such angular accelerations. It is exceedingly difficult to accurately determine angular acceleration of the head by measuring angular acceleration of the helmet because the helmet will rotate on the head an unpredictable amount under angular acceleration. Additionally angular accelerometers are very expensive and relatively large. Small angular rate sensors exist, but few have sufficient bandwidth to measure the high rates of change involved and differentiating angular rate to obtain angular acceleration amplifies noise enough to make this an unattractive technique. I have discovered some very useful methods of obtaining angular acceleration and velocity, but they are not in the public domain and I likely won't get into them in this project. Linear accelerations are still very useful in quantifying head trauma and plenty hard enough for this project.
Modelling the Impact Energy
What we are measuring is helmet acceleration. What we want to know is how much energy reached the head, how long was the accelerating force applied and what was the peak acceleration.
This analysis will assume that initially on impact, the energy transfer is all to the helmet so the acceleration of the helmet mass accounts for the entire initial impact energy. As the helmet moves, it starts to compress the helmet pads which apply force to accelerate the head and the impact energy starts to be shared between accelerating the helmet and accelerating the head. Once the helmet pads are fully compressed, the helmet and head see virtually the same acceleration and they can be considered as one mass. Full compression of helmet pads is likely less than .25 inches, depending on the how distributed the impact load is and how stiff the helmet shell is. This can be interpreted as a linear increase in head acceleration for the first .25 inches of helmet travel and thereafter the head acceleration matches the helmet acceleration. Of course for lighter impacts the pads will not be fully compressed, so accumulation of energy should be terminated when high acceleration ceases. When acceleration ceases, velocity is at its maximum because velocity cannot increase without acceleration. Likewise the force causing acceleration will have stopped when velocity hits its maximum. The accelerating force on the head doesn't actually stop until the pads have returned to their original shape. It is a bit tricky to compute the time at which the pads are no longer exerting accelerating forces because the head has moved and both the motion of the helmet and the head must be taken into consideration. If the helmet motion reaches maximum velocity before the pads are fully compressed, then the peak helmet velocity will likely exceed the maximum head velocity. Maximum head velocity could be calculated from the maximum helmet velocity with some derating for short impulses. I will simply use a linear derating so whatever fraction of full compression is achieved at maximum helmet velocity will be the ratio of head velocity over helmet velocity.
Maximum head velocity will be assumed to have occurred when helmet acceleration becomes small, because the pads will exert an acceleration force until they reach equilibrium.
Peak acceleration of the head is also not going to be the same as the helmet, but I will assume the peak head acceleration is less than the helmet by the ratio of their masses. Since a helmet weighs about .5 Kg and a head weights about 5 Kg the peak acceleration of the head is about 1/10 of the peak helmet acceleration. This factor is reduced by the fact that the acceleration measured on the helmet is also attenuated due to mechanical cushioning and flexing between the impact point and the accelerometer location, and additionally there is some electrical low-pass "filtering". In fact I am intentionally adding mechanical filtering specifically to try and make this factor close to 1. As far as I am aware this is a unique approach which not only makes the measurements easier, but the filtering also attenuates helmet shell vibrations and natural resonance frequencies that can manifest as large high-frequency accelerations on the helmet surface, but are never even felt by the head. The real reduction in acceleration is essentially the main protective mechanism of the helmet. Another way to think of it is the pads distribute the load over a larger area, minimizing the peak load at any location and at the same time, while the pads are compressing, they spread out the duration of the impact, also minimizing the peak acceleration. If we can make the accelerometer measure the equivalent of what is happening on the inside surface of the pads (at the head), then the head/helmet factor is roughly 1.
To a first approximation the acceleration waveform of the head will be assumed to have the same shape as the helmet waveform, even though the helmet pads distort the coupling, and the amplitudes will be different. Since the main interest in the waveform is to integrate it, hopefully the distortion has minimal effect on the integral.
Note that the way we calculate velocity is by integrating acceleration but integration is also used to sum up the energy of the impact, so maximum velocity is proportional to the energy of impact.
Also note that we deduce displacement or movement of the helmet by integrating velocity, which is a double integral of acceleration. The big issue with all of these integrals is that a small residual offset in the acceleration circuit results in the velocity never quite getting to zero, which in turn results in the calculated displacement and movement never stopping.
To eliminate this problem, we can trigger an event when the acceleration exceeds a threshold. If we can collect pre-trigger samples, we can start velocity and position integrals when acceleration first starts or we can simply extrapolate acceleration back to zero from the trigger threshold. When velocity has reached a peak and subsided to less than 10% of the peak we can ignore further data and extrapolate to zero velocity with very little error.
Each event will be characterized by 4 numbers:
- peak acceleration
- peak velocity
- duration
- total displacement
These quantities are relatively simple to calculate on the fly and avoid the need to store large numbers of samples taken at high sample rates. Even if the samples are all stored in RAM until these calculations can be completed, it avoids having to store complete waveforms from multiple events and avoids having a large non-volatile mass storage device in the system. It also means the data download times will be quick. I may include the direction of insult as a fifth characteristic, time permitting. The direction of insult would be computed from the vector addition of the 3 orthogonal velocities at the time of peak velocity.
At this time it is unlikely that storing entire waveforms would result in more meaningful analysis since helmet acceleration waveforms do not match head acceleration waveforms and the differences due to different heads and different helmets will be larger than any inaccuracy due to this method of on-the-fly data reduction.
Another reason for only presenting 4 numbers to characterize each event is that virtually nobody, including doctors, will be able to use waveform data from a helmet to directly diagnose any injury relationship.
It will be an on-going task to collect a database of these impact characterization numbers along with helmet and head characteristics and relate them to injury diagnoses so that eventually useful statistical relationships can be determined.
The most immediate requirement is to help diagnose when to stop playing a sport and give the brain time to heal, since subsequent accelerations to an injured brain cause far worse injuries than they would to a healthy brain. Hopefully with good instrumentation and analysis by physicians, some guidelines can be developed in the not too distant future.
This blog is background for the Crown Tools development project.