Rotation was identified as the dominant mechanism of brain injury eight decades ago. The instruments that monitor athletes today still mostly report a single number: peak magnitude. This is the story of that gap — and why closing it means asking what an impact was, not just how big it was.
Alan Holbourn published a short, mechanically argued paper proposing that rotational acceleration — not linear acceleration — was the primary cause of concussive brain injury. Working from two-dimensional gelatin models of the brain, he reasoned that brain tissue is nearly incompressible, so it resists direct compression but readily deforms in shear. A blow that mostly pushes the head in a straight line loads the tissue in a way it tolerates well; a blow that rotates the head loads it in shear, which it does not tolerate well. Holbourn went further than "rotation matters" — he argued that pure translation, on its own, was not injurious at all.
Around the same period, Wayne State University researchers built the Wayne State Tolerance Curve from cadaver studies of skull fracture under direct linear impact — a plot of acceleration magnitude against pulse duration. It was rigorous, reproducible, and, importantly, buildable: it needed only a linear accelerometer and a well-defined impact. Gurdjian, Lissner, and colleagues extended this line of work through the 1950s and 60s, focused on skull deformation and pressure waves — a translational picture of injury that coexisted uneasily with Holbourn's shear-strain argument, and won out in practice.
Ayub Ommaya and colleagues developed an experimental concussion model in rhesus monkeys, publishing in Science in 1966. Through the 1970s, Ommaya, Thomas Gennarelli, Lawrence Thibault and coworkers systematically compared rotational and translational loading in primates — reported at the 1972 Stapp Car Crash Conference and synthesized in a 1974 Brain paper correlating the experimental findings with clinical concussion. The line of work reached a landmark in 1982: Gennarelli and colleagues, writing in the Annals of Neurology, showed that rotational acceleration in their primate model produced diffuse axonal injury and coma, while equivalent translational loading did not produce the same injury. Holbourn's 1943 argument now had direct experimental support.
In 1971, John Versace formalized the Head Injury Criterion (HIC) — a time-weighted integral of linear acceleration, built directly on the Wayne State cadaver data. The U.S. National Highway Traffic Safety Administration adopted it into federal motor vehicle safety regulation the following year. HIC said nothing about rotation, angular velocity, or waveform shape — by design, it couldn't. It became the template that automotive safety, and soon sports-equipment standards, were built around, arriving in force at almost exactly the moment the primate evidence for rotation was accumulating.
Linear accelerometers were small, cheap, and easy to trust. Angular motion is harder to measure well: gyroscopes had narrower dynamic range, and angular acceleration — the quantity closest to what Holbourn and Gennarelli's work implicated — has to be derived by differentiating angular velocity, which amplifies noise. Most field devices still trigger on a linear-acceleration threshold crossing, which introduces a structural bias: a predominantly rotational event can produce only a modest local linear signal, so a linear-only trigger can miss or under-weight exactly the events the 1943–1982 evidence said mattered most. Regulatory and commercial standards built around what could be measured reliably — not necessarily around what the mechanism actually was.
Steve Rowson and Stefan Duma published a rotational injury risk function for football concussion in 2012. Koshiro Kimpara and Naoki Iwamoto proposed the Rotational Injury Criterion (RIC) the same year — HIC's integral form, with angular acceleration in place of linear. Erik Takhounts and colleagues at NHTSA went further, publishing the Brain Injury Criterion (BrIC) at the 2013 Stapp Car Crash Conference: a criterion built from peak angular velocity about all three anatomical axes, calibrated against finite-element brain models rather than cadaver skull fracture. Seventy years after Holbourn, rotation had a regulatory-grade number of its own.
BrIC was real progress — but look at what it actually is: three peak angular velocities, normalized and combined into a single value. It answers the same kind of question HIC always asked, just about a different axis of motion: how much? Timing had entered the picture earlier than that — HIC's own integral already weighted magnitude by duration, and BrIC's calibration depends on the duration-sensitive relationship between rotational velocity and brain strain. But even with timing folded in, every widely used criterion — HIC, RIC, BrIC — still collapses an entire waveform down to a scalar. None of them ask what shape the motion took.
Field research on wearable head-impact sensors — mouthguards, helmet units, skin and headband patches — has matured considerably as a measurement discipline. But its center of gravity has stayed on the same three questions HIC-era research asked: was there an event, how big was it, and how many were there. Processing choices — filter cutoffs, coupling quality, trigger thresholds — can change reported event counts and peak magnitudes by several-fold between otherwise similar studies, which says something important: the field has spent its energy getting the peak right, not asking what else the waveform contains.
Machine-learning approaches to impact classification have used dozens or hundreds of waveform-derived features as inputs — spectral density, temporal statistics, coupling ratios — and clustering approaches have used similar features to define impact subtypes. But in nearly all of this work, the multidimensional feature space is a means to an end: a substrate for a classification label or a regression target, not something examined in its own right. What its dimensionality is, how much of it is independent of peak magnitude, whether it separates real events from artefacts on its own terms — those questions were rarely, if ever, asked directly.
HIC asks how much linear force. BrIC asks how much rotational velocity. Both are still answers to "how much" — just pointed at different axes. The Impact Feature Space asks a different kind of question: not how big was this event on some chosen axis, but what shape did it take, how were rotation and translation related in time, and what does its frequency content say about how it was generated. That's not a philosophical distinction — it shows up directly in the numbers.
Peak kinematic metrics, PCA'd: effectively one-dimensional. A ranking by intensity, however many axes you throw in.
IFS structural features, PCA'd: several independent mechanical dimensions, 55.3% of it orthogonal to every peak metric combined.