Vous êtes sur la page 1sur 11

Annals of Biomedical Engineering, Vol. 43, No. 8, August 2015 ( 2014) pp.

18961906
DOI: 10.1007/s10439-014-1197-z

Measurement of Hybrid III Head Impact Kinematics Using


an Accelerometer and Gyroscope System in Ice Hockey Helmets
MARI A. ALLISON,1,2 YUN SEOK KANG,3 MATTHEW R. MALTESE,4,5 JOHN H. BOLTE IV,3
and KRISTY B. ARBOGAST1,5
1
The Center for Injury Research and Prevention, The Childrens Hospital of Philadelphia, 3535 Market St. Suite 1150,
Philadelphia, PA 19104, USA; 2Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, USA; 3Injury
Biomechanics Research Center, The Ohio State University, Columbus, OH, USA; 4Department of Anesthesiology and Critical
Care Medicine, The Childrens Hospital of Philadelphia, Philadelphia, PA, USA; and 5Department of Pediatrics, University of
Pennsylvania, Philadelphia, PA, USA
(Received 27 June 2014; accepted 19 November 2014; published online 5 December 2014)

Associate Editor Stefan M Duma oversaw the review of this article.

AbstractHelmet-based instrumentation is used to study the 1015%. It is important to dene system accuracy for a
biomechanics of concussion. The most extensively used particular helmet brand, sensor location, and impact direc-
systems estimate rotational acceleration from linear acceler- tion in order to interpret real-world data.
ation, but new instrumentation measures rotational velocity
using gyroscopes, potentially reducing error. This study KeywordsmTBI, Concussion, Head injury, Head accelera-
compared kinematics from an accelerometer and gyroscope-
tion, Head impact biomechanics, Impact monitoring, Helmet
containing system to reference measures. A Hybrid III (HIII)
adult male anthropometric test device head and neck was t sensors.
with two helmet brands, each instrumented with gForce
Tracker (GFT) sensor systems in four locations. Helmets
were impacted at various speeds and directions. Regression INTRODUCTION
relationships between GFT-measured and reference peak
kinematics were quantied, and inuence of impact direc- With millions of mild traumatic brain injuries (mTBIs)
tion, sensor location, and helmet brand was evaluated. The
occurring each year,4,6 and recent data suggesting that
relationship between the sensor output and the reference
acceleration/velocity experienced by the head was strong. they can cause long term neurological consequences,13,25
Coefcients of determination for data stratied by individual prevention of these injuries has been declared by the
impact directions ranged from 0.77 to 0.99 for peak linear Centers for Disease Control and Prevention as a research
acceleration and from 0.78 to 1.0 for peak rotational velocity. priority.20 The thresholds that lead to mTBI are still a
For the data from all impact directions combined, coef-
matter of debate;10 however, understanding the biome-
cients of determination ranged from 0.60 to 0.80 for peak
resultant linear acceleration and 0.83 to 0.91 for peak chanical inputs that lead to these injuries is essential in
resultant rotational velocity. As expected, raw peak resultant preventing them from occurring.
linear acceleration measures exhibited large percent differ- In recent years, helmet-based instrumentation con-
ences from reference measures. Adjustment using regressions sisting of accelerometers has been used to monitor the
resulted in average absolute errors of 1015% if regression
biomechanics of impacts sustained by athletes in high-
adjustments were done by impact direction or 2540% if
regressions incorporating data from all impact directions impact sports. These sports provide a unique oppor-
were used. Average absolute percent differences in raw peak tunity to study head injury in a real-world scenario, as
resultant rotational velocity were much lower, around impacts occur frequently in these settings and the
athletes are at an increased risk for mTBI during
normal play. Furthermore, neurocognitive baseline
measures can be taken on the athletes playing these
sports, so that biomechanical inputs can be correlated
to neurocognitive changes.
Address correspondence to Kristy B. Arbogast, The Center for
Injury Research and Prevention, The Childrens Hospital of Phila-
The most widely-used helmet-based systems to date
delphia, 3535 Market St. Suite 1150, Philadelphia, PA 19104, USA. have been the head impact telemetry (HIT) systems for
Electronic mail: arbogast@email.chop.edu football and ice hockey (Simbex LLC; Lebanon, NH).
1896
0090-6964/15/0800-1896/0 2014 Biomedical Engineering Society
Helmet-Based Measurement of Head Kinematics 1897

Data collected using these accelerometer systems have helmet construction. The helmets were t to the HIII
been used to compare head impact magnitudes based head following USA hockey guidelines, with the helmet
on a number of characteristics, including playing po- tight enough to prevent longitudinal rotation and cen-
sition, gender, awareness of impending impact, cervical tered on the head with the rim one nger width above
muscle strength, and sport specic scenar- the eyebrows.11 Differences in helmet construction and
ios.2,5,15,1719,21,23,26 Concussion risk curves for foot- sizing recommendations based on head circumference
ball players based on peak resultant linear head led to the use of a large-sized Easton helmet and a
acceleration have also been developed using data col- medium-sized Bauer helmet. The alignment of the hel-
lected via the HIT System for football.7,8,24 met on the HIII head was checked before each impact to
Accurate measurement of rotational kinematics is conrm repeatability of the testing conditions. Each
particularly important when studying thresholds of helmet had its corresponding facemask attached, as
traumatic brain injury. The rotational movement is hockey players under the age of 18 are required by USA
correlated to measures of strain within the brain, which Hockey rules to wear a facemask during play.
are predictive of diuse axonal injury across the The GFT consists of a triaxial accelerometer and a
spectrum of severity.3,9,14,27 Validation studies have triaxial gyroscope housed in a casing that is attached to
shown that previously-used helmet instrumentation a helmet via adhesive or Dual Lock Reclosable Fas-
systems have average absolute errors in the range of teners (3M, St. Paul, MN). This allows integration into
1560% in estimating rotational kinematics based on a helmet of choice, and implementation across a range
measures from linear accelerometers.1,12 However, it is of helmeted sports. Impact data is stored on-board and
likely more accurate to measure than calculate rota- uploaded via USB connection. The data obtained by the
tional kinematics and new helmet-based instrumenta- end-user is the raw acceleration and velocity data, and
tion systems have been developed that use gyroscopes has not been processed apart from use of a simple rst-
in combination with accelerometers. One such system, order hardware low pass lter on the accelerometers
the gForce Tracker (GFT) (Markham, ON), employs a with a cut-off frequency of 300 Hz, so it has not been
triaxial accelerometer and a triaxial gyroscope to transformed to an approximate center of gravity of the
measure linear acceleration and rotational velocity. head or adjusted to account for the helmets dissipation
Therefore, the objective of this study was to compare of energy. Impact data is recorded over a 40 ms time-
linear and rotational kinematic measures from the span at 3000 Hz for linear acceleration data and 760 Hz
GFT to reference measures during a series of impacts for rotational velocity data. However, if the acceleration
to a helmeted HIII headform to evaluate the systems remains above the user-set threshold beyond this time-
accuracy to monitor on-ice impacts in ice hockey. point, the system continues recording 40 ms timespans
of data until the acceleration falls below the threshold.
For this study, that threshold was set to 8 g.
MATERIALS AND METHODS A 23.9 kg pneumatic linear impactor, with a 0.6 kg
ultra-high molecular weight polyethylene impacting
A HIII 50th percentile male ATD head and neck with surface, was used to directly hit the helmet at 1.5, 2.5,
the 3-2-2-2 accelerometer array22 and three angular rate 3.75, and 5 m/s in front, side, back, and oblique back
sensors was rigidly mounted at T1. Rigid mounting of the directions (Fig. 2). The oblique back direction was
ATD head and neck simulates a large effective mass of measured 45 degrees from the sagittal plane. Direct
the torso. In order to best-simulate a real-world scenario, impact to the sensor changes the mechanics of the
a human hair wig was used as it was previously found impact and creates the possibility of altering the data
that the interface between the ATD head and the helmet collected at the other three sensor locations. Therefore,
can signicantly affect the relationship between helmet- due to impact interference, the back sensor was
based system and reference measures.1 The wig was ad- removed from the Bauer helmet during the back im-
hered to the HIII head using industrial-strength double- pacts, and impact data is not available for this helmet-
sided tape, which kept the wig in place, and it was sensor location-impact direction combination.
sprayed with water to simulate perspiration. The impacting surface was cylindrical (4 in. in
An Easton S9 hockey helmet (Easton-Bell Sports diameter) with rounded edges, with the at end of the
Inc., Van Nuys, CA) and a Bauer RE-AKT hockey cylinder contacting the helmet. The impacting speeds
helmet (Bauer Hockey, Inc., Exeter, NH) were each were chosen to produce accelerations across the range
instrumented with GFT sensors in four dierent loca- of measures observed during on-ice play,16 and the test
tions: outside top, outside back, outside right, and in- matrix included three or four impacts per speed-
side top (Fig. 1). The same set of sensors was used on direction combination (Fig. 3). The methodology used
both the Easton helmet and the Bauer helmet to com- in this study is similar to that of a previously published
pare differences in measures that may occur due to validation study on the HIT System for ice hockey.1
1898 ALLISON et al.

FIGURE 1. GFT sensor locations.

FIGURE 2. Depicts front (top left), side (top right), oblique (bottom left) and back (bottom right) impact directions. Front and back
impacts were done along the saggital plane, side impacts were done along the coronal plane, and oblique impacts were measured
at 45 from back.
Helmet-Based Measurement of Head Kinematics 1899

FIGURE 3. Test matrix showing the number of impacts per helmet-speed-impact direction combination.

HIII acceleration time histories, measured at the were applied to peak values for that impact direction
center of gravity of the head, were processed using a CFC and then absolute error was calculated (Fig. 4b).
1000 lter. Similarly, rotational velocities measured via
angular rate sensors were processed using a CFC 60 lter. Percent error of power relationship adjusted data

a  GFTb
Maximum values of resultant linear acceleration and max  HIIImax
resultant rotational velocity were compared between the  100 1
HIIImax
HIII and GFT measurement systems for the same im-
pact. The correlation between these measures was dened where a and b are coefcients of the power regression
using power t regression and assessed via coefcient of equation
determination (R2). Power regressions were found for All statistical analyses were performed using the mixed
data from all impact directions combined and then for models procedure in SAS 9.3 (SAS Institute Inc.; Cary,
data stratied by impact direction. All regressions were NC). The relationship between HIII peak resultant kine-
helmet brand-specic. matic values (e.g., linear acceleration and rotational
The data was separated by helmet brand and the velocity) and the raw GFT peak resultant kinematic values
average absolute percent dierences between the raw was assessed by using the ratio of HIII peak value to
data and ATD reference measures for all impacts to corresponding peak raw GFT value as the outcome
each helmet were calculated. The versions of GFT measure. Impact direction (front, side, back, or oblique
hardware and software used in this study provide raw back), helmet brand (Easton or Bauer), and sensor loca-
data and do not incorporate an algorithm to transform tion (outside top, outside right, outside back, or inside top)
the raw data to the CG of the head. However, from an were included in the model as categorical variables (Model
injury biomechanics and injury risk standpoint, it is 1). The eect of these categorical variables on the rela-
important to understand how accurately such sensors tionship between HIII and GFT peak resultant measures
describe the kinematics of the head CG. Therefore, was assessed. In cases of statistically signicant interac-
adjustment factors were applied to attempt to estimate tions, data was stratied to test for specic eects, using
CG measures. Since the sensors are attached to the Bonferroni corrections to adjust for multiple tests (Model
helmet, this cannot be done using a direct rigid body 2, Model 3). Statistical analyses were performed for both
transformation as the helmet is not rigid relative to the peak linear accelerations and peak rotational velocities.
head. For these reasons, an empirical method was used HIIIpeak
to adjust the data, applying the two types of power b0 b1  V1 b2  V2 b3  V3
GFTrawpeak
regressions mentioned above to the peak resultant
b4  V1  V2 b5  V1  V3 b6
measures of linear acceleration and rotational velocity
(Eq. 1). First, the regression equation developed using  V2  V3 b7  V1  V2  V3 e
data from all impact directions combined was applied to
the peak values before calculating absolute error Model 1: Statistical model including all vari-
(Fig. 4a). As a second approach, the regression equa- ables. V1 = helmet brand, V2 = impact direc-
tions developed from data stratied by impact direction tion, and V3 = sensor location.
1900 ALLISON et al.

FIGURE 4. Peak resultant linear acceleration data for GFT measures from the inside top sensor location. Both graphs depict the
same data, but the graph to the left shows the power regression for all impact directions combined while the graph to the right
shows impact direction-specific power regressions.

Comparison between the raw GFT measures and


HIIIpeak
b0 b1  V2 b2  V3 b3  V2 the reference acceleration/velocity at the ATD CG
GFTrawpeak showed expected large dierences (100150%,
 V3 e Table 2a). These differences can be attributed to sev-
eral reasons: no in-built algorithm to transform the
Model 2: Statistical model for data stratied by data to the CG of the ATD, the role of the helmet in
helmet brand. V2 = impact direction and dissipating forces imparted to the head, and sensor
V3 = sensor location. error. After applying the regression equations devel-
HIIIpeak oped using data stratied by specic impact directions,
b0 b1  V2 e these differences were substantially reduced. Average
GFTrawpeak
absolute errors for peak linear acceleration ranged
Model 3a: Statistical model for data stratied by from 4.8 to 22.0% for the Bauer helmet and from 3.5
helmet brand and sensor location. V2 = impact to 23.0% for the Easton helmet (Table 2a). The aver-
direction. age absolute errors for data adjusted using direction-
HIIIpeak specic regressions were generally larger for the Easton
b0 b1  V3 e helmet than for the Bauer helmet, particularly for back
GFTrawpeak
and front impacts, and were generally smaller for side
Model 3b: Statistical model for data stratied by impacts than other impact directions for both helmet
helmet brand and impact direction. V3 = sensor brands (Table 2a). After applying the regression
location. equations developed using data from all impact
directions combined, average absolute errors for peak
linear acceleration ranged from 24.9 to 35.4% for the
RESULTS Bauer Re-Akt helmet and from 24.8 to 39.6% for the
Easton S9 helmet (Table 2a). The outside back sensor
The total test matrix was 118 impacts and analyz- location generally had the largest error.
able data was obtained for all impacts. Power regres- The dierences between GFT-measured peak
sion analysis of peak GFT kinematics compared to resultant rotational velocities and the corresponding
peak reference HIII kinematics showed strong rela- ATD measures were lower than those for linear
tionships between these measures. Coecients of acceleration (Table 2b). Raw data average percent
determination for data stratied by individual impact differences ranged from 9.2 to 17.1%. After applying
directions ranged from 0.77 to 0.99 for peak linear the regression equations developed using data stratied
acceleration (Fig. 4 right, Fig. 5; Table 1a) and from by specic impact directions, average absolute errors in
0.78 to 1.0 for peak rotational velocity (Fig. 5; peak rotational velocity ranged from 2.8 to 10.8% for
Table 1b). For the data from all impact directions the Bauer helmet and from 2.1 to 18.8% for the Easton
combined, coefcients of determination ranged from helmet. After applying the regression equations
0.60 to 0.80 for peak resultant linear acceleration and developed using data from all impact directions com-
0.830.91 for peak resultant rotational velocity (Fig. 4 bined, average absolute errors in peak rotational
left; Table 1). velocity ranged from 13.2 to 14.2% for the Bauer
Helmet-Based Measurement of Head Kinematics 1901

FIGURE 5. Exemplar data collected from impacts to the helmeted HIII head. The top two graphs depict peak linear and rotational
measures from the outside right sensor during oblique impacts. The bottom two graphs depict peak linear and rotational measures
from the outside right sensor during back impacts.

TABLE 1. Power regression equations and their associated coefficients of determination derived from regression analyses for
both helmet brands and each sensor location/impact direction combination for peak resultant linear acceleration (a) and rotational
velocity (b).

Outside top sensor Outside right sensor Outside back sensor


Inside top sensor location location location location

Helmet Impact direction Regression equation R2 Regression equation R2 Regression equation R2 Regression equation R2

(a)
Bauer All data y = 0.37x1.08 0.74 y = 0.37x1.05 0.77 y = 0.37x1.07 0.72 y = 0.50x1.00 0.69
Side impacts y = 0.33x1.20 0.96 y = 0.57x1.05 0.99 y = 0.50x1.05 0.99 y = 1.10x0.92 0.99
Oblique impacts y = 0.14x1.28 0.93 y = 0.02x1.68 0.95 y = 0.11x1.40 0.95 y = 0.04x1.48 0.92
Back impacts y = 0.52x1.01 0.92 y = 0.57x0.95 0.94 y = 1.06x0.85 0.94 N/A N/A
Front impacts y = 1.20x0.71 0.91 y = 0.70x0.85 0.88 y = 0.12x1.19 0.90 y = 0.28x1.07 0.98
Easton All data y = 0.41x1.02 0.80 y = 0.16x1.21 0.80 y = 0.25x1.12 0.66 y = 1.23x0.77 0.60
Side impacts y = 1.14x0.87 0.99 y = 0.75x0.94 0.92 y = 0.46x1.09 0.99 y = 1.24x0.87 0.99
Oblique impacts y = 0.03x1.58 0.93 y = 0.05x1.48 0.98 y = 0.05x1.46 0.90 y = 0.004x1.87 0.93
Back impacts y = 0.42x0.98 0.81 y = 0.14x1.19 0.89 y = 0.29x1.07 0.79 y = 0.16x1.13 0.77
Front impacts y = 0.41x0.99 0.89 y = 0.07x1.34 0.87 y = 0.17x1.11 0.84 y = 0.25x1.21 0.96
(b)
Bauer All data y = 2.86x0.86 0.84 y = 2.28x0.88 0.85 y = 1.39x0.95 0.83 y = 1.23x0.96 0.86
Side impacts y = 0.72x1.04 0.96 y = 0.18x1.22 0.98 y = 0.55x1.07 0.99 y = 0.15x1.25 0.97
Oblique impacts y = 1.15x0.99 0.96 y = 0.84x1.04 0.96 y = 0.35x1.16 0.98 y = 0.85x1.03 0.96
Back impacts y = 1.94x0.94 0.94 y = 2.78x0.87 0.95 y = 2.14x0.92 0.92 N/A N/A
Front impacts y = 14.0x0.62 0.92 y = 12.7x0.63 0.90 y = 2.14x0.92 0.91 y = 5.09x0.75 0.90
Easton All data y = 1.67x0.93 0.88 y = 2.84x0.86 0.91 y = 0.95x1.00 0.85 y = 2.96x0.86 0.91
Side impacts y = 1.28x0.94 0.97 y = 1.51x0.93 0.99 y = 1.09x0.96 0.98 y = 0.55x1.09 0.99
Oblique impacts y = 0.60x1.09 1.00 y = 0.39x1.14 1.00 y = 0.25x1.17 0.78 y = 0.59x1.09 0.99
Back impacts y = 0.93x1.02 0.86 y = 2.34x0.89 0.98 y = 0.39x1.14 0.91 y = 2.83x0.85 0.98
Front impacts y = 5.46x0.76 0.97 y = 12.6x0.65 0.97 y = 1.56x0.94 0.97 y = 14.13x0.64 0.98
1902

TABLE 2. Average absolute percent differences (6SD) in peak resultant linear acceleration (a) and peak resultant rotational velocity (b) for raw GFT measures, after
adjustment using power regression equations developed from the data for all impact directions combined, and after adjustment using impact direction-specific regression
equations.

Inside Top Sensor Outside Top Sensor Outside Right Sensor Outside Back Sensor

Corrected Corrected Corrected Corrected Corrected


Corrected with direction- with all with direction- Corrected with direction- Corrected with direction-
with all data specific Raw data specific Raw with all data specific Raw with all data specific
Raw data regression regression data regression regression data regression regression data regression regression

(a)
Bauer
Side Impacts 11.1% (7.9%) 4.8% (4.1%) 5.3% (3.0%) 6.2% (2.8%)
Oblique Impacts 96.8% 26.6% 12.3% (8.0%) 120.5% 24.8% 22.0% (12.3%) 110.6% 28.3% 11.0% (8.5%) 112.4% 35.4% 12.8% (10.2%)
Front Impacts (65.9%) (22.9%) 10.9% (7.1%) (65.6%) (17.0%) 12.0% (7.5%) (82.9%) (30.8%) 11.0% (7.2%) (75.8%) (13.9%) 5.1% (4.5%)
Back Impacts 9.8% (7.5%) 9.6% (7.4%) 8.1% (6.0%) N/A
Easton
Side Impacts 4.8% (3.3%) 13.4% (9.0%) 3.5% (3.3%) 4.3% (2.8%)
Oblique Impacts 134.1% 28.4% 17.9% (15.4%) 153.7% 24.9% 8.6% (7.5%) 152.3% 37.0% 19.6% (19.6%) 158.6% 39.6% 15.8% (15.7%)
ALLISON et al.

Front Impacts (72.0%) (15.8%) 17.2% (8.8%) (86.6%) (22.4%) 19.1% (15.8%) (110.1%) (28.6%) 18.1% (23.4%) (124.1%) (33.5%) 9.4% (8.3%)
Back Impacts 21.5% (13.7%) 16.3% (11.1%) 23.0% (13.8%) 18.2% (13.7%)
(b)
Bauer
Side Impacts 6.5% (4.1%) 6.6% (4.0%) 2.8% (2.6%) 5,5% (4.5%)
Oblique Impacts 14.5% 13.6% 4.1% (5.9%) 14.3% 13.8% 6.0% (5.5%) 14.6% 14.2% 4.5% (3.8%) 15.0% 13.2% 5.0% (5.5%)
Front Impacts (9.9%) (11.0%) 8.8% (4.6%) (11.7%) (10.1%) 10.1% (6.2%) (9.4%) (9.4%) 9.2% (6.0%) (12.8%) (8.6%) 10.8% (5.4%)
Back Impacts 9.6% (5.6%) 8.1% (5.4%) 9.2% (6.9%) N/A
Easton
Side Impacts 6.3% (3.8%) 3.7% (3.0%) 4.2% (3.4%) 3.9% (3.0%)
Oblique Impacts 13.0% 12.9% 2.1% (1.1%) 10.6% 11.2% 2.3% (1.4%) 17.1% 15.4% 18.8% (9.5%) 9.2% 11.3% 3.5% (2.7%)
Front Impacts (8.1%) (8.3%) 6.7% (4.7%) (8.5%) (7.3%) 7.6% (3.2%) (14.2%) (11.4%) 6.3% (4.6%) (10.3%) (8.0%) 6.0% (3.1%)
Back Impacts 13.7% (7.3%) 4.9% (3.5%) 9.6% (9.9%) 4.8% (2.3%)
Helmet-Based Measurement of Head Kinematics 1903

Re-Akt helmet and from 11.2 to 15.4% for the Easton these three contributions to the dierences observed.
S9 helmet. The average absolute errors in peak resul- Our study demonstrated that, combined, these factors
tant rotational velocity did not vary substantially by led to expectedly large average percent dierences in
sensor location, helmet brand, or impact direction. raw peak resultant linear acceleration (over 100%), but
The statistical model evaluating the relationship relatively small average percent dierences in raw peak
between GFT peak kinematic measures and reference resultant rotational velocity (1015%). Because of
kinematic measures for all of the impact directions these large dierences in the raw linear acceleration
combined showed a signicant three-way interaction data, in using this version of the system to estimate
between helmet brand, sensor location, and impact measures at the CG of the head it is important to
direction. Since helmet brand alone had a statistically develop regression equations that describe the rela-
signicant eect (p = 0.04) and this is a controllable tionships between peak GFT measures and peak linear
parameter in on-eld implementation, the data was acceleration ATD measures so that these can be used
stratied by helmet brand for all other analyses. For to empirically adjust the data and improve linear
both the Easton and Bauer peak linear acceleration acceleration accuracy. This would require impact
data, there were signicant two-way interactions testing on a given helmet brand-sensor location com-
between sensor location and impact direction (p < bination as these controllable parameters signicantly
0.0001). When the data was stratied by impact aect the relationship between peak head kinematics
direction to test for signicance of sensor location, and peak GFT measure. This particular helmet model
adjusted p values were signicant for all directions and sensor location should then be used for all players
(Table 3). Similarly, when the peak linear acceleration on a given team to collect real-world data that will be
data was stratied by sensor location to test for sig- adjusted using the experimentally derived regressions.
nicance of impact direction, adjusted p values were Given that average absolute errors in peak linear
signicant for all sensor locations (Table 3). Statistical acceleration were larger for certain sensor locations,
analysis of the peak rotational velocity data had sim- even after adjustment using the all-impact direction
ilar results for signicance of impact direction, but regression equations, such testing would ideally in-
some differences in signicance of sensor location. clude multiple sensor location options. The best loca-
While sensor location remained signicant in the tion could then be determined by comparing average
majority of impact directions, for front impacts to the absolute errors after data were corrected using the
Bauer helmet, and for front and back impacts to the regression method of choice (all impact directions
Easton helmet, sensor location was not statistically combined or direction-specic). An ideal choice for
signicant (Table 3). sensor location would also be one that is less likely to
be directly impacted, such as the top of the helmet.16,19
Similar to evaluation of another helmet-based
DISCUSSION accelerometer system,1 this study showed a statistically
signicant effect of impact direction on the relation-
This study evaluated a helmet-based system con- ship between peak sensor-measured and reference
sisting of both accelerometers and gyroscopes and its kinematics. One method to account for this nding in
expected on-eld accuracy under impact conditions real world implementation would be to determine the
similar to those used in the laboratory test setup. The impact direction on-ice and apply direction-specic
ndings highlight factors to consider when using this regressions to the data. This should result in the
system to collect real-world head impact kinematics, smallest possible error, but may be logistically difcult
including specic implications for both data collection for some users. A more feasible approach for many
and data analysis. users may be to combine data from all impact direc-
For the version of GFT hardware and software tions and apply a single regression equation to the
evaluated in this study, the raw data is not processed data, but this approach results in larger error
through an in-built algorithm to transform the data to (approximately 2040% for linear acceleration and 10
the center of gravity of the head. As mentioned above, 15% for rotational velocity). This combined direction
this leads to large dierences in raw sensor measures approach may be further improved by developing a
for linear acceleration compared to those measured at combined direction regression that weights the data
the CG of the ATD for three primary reasons: the data from the individual directions (front, back, side, top)
has not been transformed from the location on the corresponding to the percentage of impacts in each
helmet to the center of gravity of the head, the energy direction seen on ice.16,19
dissipation of the helmet has not been accounted for, It is important to note that average absolute errors
and there is some amount of sensor error. This study in rotational kinematics, both for the raw data and
was not designed to calculate the independent eects of after correcting the data using direction-specic
1904 ALLISON et al.

TABLE 3. Summary of p values for statistical significance of impact direction and sensor location on the ratio of peak GFT
measure to peak corresponding HIII measure.

Bauer peak linear Easton peak linear Bauer peak rotational Easton peak rotational

Significance of impact direction


Inside top sensor 0.0004* 0.0004* 0.0004* 0.0004*
Outside top sensor 0.0004* 0.0004* 0.0004* 0.0012*
Outside back sensor 0.0004* 0.0004* 0.0004* 0.0024*
Outside right sensor 0.0004* 0.0004* 0.0004* 0.0004*
Significance of sensor location
Side impact 0.0004* 0.0004* 0.0020* 0.0004*
Front impact 0.0016* 0.0004* 0.4604 0.0512
Back impact 0.0004* 0.0268* 0.0004* 0.8636
Oblique impact 0.0004* 0.0004* 0.0116* 0.0004*

*Statistically significant.

regressions, were considerably lower for data from the the hard surfaces that may be encountered, such as the
gyroscope-containing GFT (218%) than for similar boards or ice. However a player may also contact other
data (also after direction-specic calibration) from the players or their equipment and this was not evaluated
accelerometer-only HIT System for ice hockey in this study. Furthermore only a single helmet t was
(1250%) found in a previous study.1 This is not a used, which was chosen to most closely mimic the real-
direct comparison, as the GFT measures rotational world scenario and USA hockey recommendations.
velocity while the HIT System calculated rotational This is a factor that varies by individual, and a study
acceleration. Rotational acceleration data is inherently by Jadischke et al. examined how helmet t may
noisier than measuring rotational velocity data, and inuence helmet-based measurement of head kine-
studies show that rotational velocity accurately pre- matics during impact.12
dicts strains within the brain and thus is a good metric Before analyzing data collected using this system to
for quantifying traumatic brain injury risk across the study the biomechanics of head impacts that lead to
spectrum of severity.28 Therefore, rotational velocity mTBI, more work should be done to translate the
rather than acceleration is likely a better measurement ndings from this study into practical strategies for
choice for helmet-based systems. In terms of peak implementation. It would be benecial for future work
resultant linear acceleration, there were slightly higher to include direct impacts to the sensor, as well as a
average absolute errors in GFT measures after sensitivity analysis to determine how small changes in
adjustment using direction-specic regressions impact direction aect the relationship between peak
(540%) than were previously found for the HIT sensor and reference kinematic measures. Further-
System for ice hockey after adjustment using direction- more, we understand that enhancements are continu-
specic regressions (530%). Given the evidence that ally being made to the GFT sensor and software to
rotational measures are better predictors of diffuse improve accuracy, including in-built algorithms to
brain injury across the spectrum of severity,3,9,14,27 transform the data to the CG of the head; this study
rotational accuracy is likely a higher priority. evaluated sensor version GFT3.s.19 along with soft-
The GFT is designed to estimate the azimuth and ware version gManager 1.10.
elevation of the impact when it is calibrated to its spe-
cic location in the helmet. For the GFT software
version that was used in this analysis (gManager 1.10) CONCLUSIONS
the algorithm to estimate azimuth and elevation was not
optimized to identify exact locations and therefore did This study evaluated a helmet-based system consist-
not always correspond well to the actual impact direc- ing of accelerometers and gyroscopes designed to collect
tion. We understand, however, that the algorithm to real-world data on the biomechanics of head impacts in
quantify impact direction has since undergone sub- ice hockey. Under the conditions evaluated in this
stantial revision. Further research should be done to study, the raw peak resultant linear acceleration mea-
verify the accuracy of GFT-estimated impact direction. sures exhibited expectedly large dierences compared to
Limitations of this study center around the idea that reference measures as the data had not been trans-
it does not incorporate the full range of impact sce- formed to the CG, had not been adjusted to account for
narios that players may experience on-ice. A single helmet energy dissipation, and included some level of
impact condition was tested over a range of speeds and sensor error. Therefore, if attempting to estimate mea-
directions. One impacting surface was chosen to mimic sures at the CG of the head, it is necessary to adjust
Helmet-Based Measurement of Head Kinematics 1905

linear acceleration data collected using this system. Sci. Sports Exerc. 46(1):115123, 2014. doi:10.1249/
Adjustment using regressions resulted in average abso- MSS.0b013e3182a32d0d.
2
Brainard, L. L., J. G. Beckwith, J. J. Chu, et al. Gender
lute errors of 1015% if they were done on an impact
differences in head impacts sustained by collegiate ice
direction-specic basis, or 2540% if a regression hockey players. Med. Sci. Sports Exerc. 44(2):297304,
incorporating data from all impact directions was used. 2012. doi:10.1249/MSS.0b013e31822b0ab4.
3
Average percent dierences in raw peak resultant rota- Browne, K. D., X.-H. Chen, D. F. Meaney, and D. H.
tional velocity were much lower, in the range of Smith. Mild traumatic brain injury and diffuse axonal in-
jury in swine. J. Neurotrauma 28(9):17471755, 2011. doi:
1015%, suggesting that this sensor may be a particu-
10.1089/neu.2011.1913.
larly valuable tool in measuring the important parameter 4
Centers for Disease Control and Prevention. Nonfatal
of rotational kinematics and represents advancement traumatic brain injuries from sports and recreation activi-
over previous systems for a rotational metric. In the case tiesUnited States, 20012005. MMWR Morb. Mortal.
of rotational velocity, given that the raw data was an Wkl. Rep. 56(29):733737, 2007.
5
Crisco, J. J., R. Fiore, J. G. Beckwith, et al. Frequency and
excellent predictor of actual head kinematics, the
location of head impact exposures in individual collegiate
adjustment procedure via regression only slightly re- football players. J. Athl. Train. 45(6):549559, 2010. doi:
duced the differences between the sensor measures and 10.4085/1062-6050-45.6.549.
6
reference ATD measures (to less than 10%). Faul, M., L. Xu, M. M. Wald, and V. Coronado. Trau-
The study identied factors that statistically inuence matic brain injury in the United States: emergency
department visits, hospitalizations and deaths 20022006.
the relationship between peak helmet sensor and refer-
Atlanta, GA: Centers for Disease Control and Prevention,
ence measures, including the model of helmet, the sen- p. 7, 2010.
sor location, and the direction of impact. A specic 7
Funk, J. R., S. M. Duma, S. J. Manoogian, and S. Row-
helmet model and sensor location should be chosen if son. Biomechanical risk estimates for mild traumatic brain
implementing this technology for a given team. On-ice injury. Annu. Proc. Assoc. Adv. Automot. Med. 51:343361,
2007.
use of this system should account for these factors, and 8
Funk, J. R., S. Rowson, R. W. Daniel, and S. M. Duma.
if interested in estimating the kinematics of the CG of Validation of concussion risk curves for collegiate football
the head, analysis of the data should incorporate the players derived from HITS data. Ann. Biomed. Eng.
types of dierences quantied in this study. More work 40(1):7989, 2012. doi:10.1007/s10439-011-0400-8.
9
should be done to characterize the necessary adjustment Gennarelli, T. A., L. E. Thibault, and A. K. Ommaya.
Pathophysiologic Responses to Rotational and Transla-
factors for peak resultant linear acceleration before
tional Accelerations of the Head, 1972. doi:10.4271/720970.
analyzing data collected using this version of the system. 10
Guskiewicz, K. M., and J. P. Mihalik. Biomechanics of
sport concussion: quest for the elusive injury threshold.
Exerc. Sport Sci. Rev. 39(1):411, 2011. doi:10.1097/JES.
0b013e318201f53e.
11
USA Hockey. Equipment Sizing Guide. Available at:
ACKNOWLEDGMENTS
http://www.usahockey.com/Sizing_Guide.aspx. Accessed
This study was supported by a National Science July 12, 2012.
12
Jadischke, R., D. C. Viano, N. Dau, A. I. King, and J.
Foundation Graduate Research Fellowship and Fun-
McCarthy. On the accuracy of the head impact telemetry
damental Research Supplement, the Center for Child (HIT) System used in football helmets. J. Biomech.
Injury Prevention Studies, particularly the National 46(13):23102315, 2013. doi:10.1016/j.jbiomech.2013.05.030.
13
Highway Trac Safety Administration and Toyota Konrad, C., A. J. Geburek, F. Rist, et al. Long-term cog-
Collaborative Safety Research Center, and the SAFER nitive and emotional consequences of mild traumatic brain
injury. Psychol. Med. 41(6):11971211, 2011. doi:
Vehicle and Trac Safety Centre. The sensors were
10.1017/S0033291710001728.
provided free of charge by GForceTracker Inc and 14
Meaney, D. F., D. H. Smith, D. T. Ross, and T. A. Gen-
there is no nancial relationship between the authors, narelli. Diffuse axonal injury in the miniature pig: biome-
Childrens Hospital of Philadelphia, and GForce- chanical development and injury threshold. ASME
Tracker Inc. Dr. Arbogast serves as a consultant for Crashworth. Occup. Prot. Transp. Syst. 29:169175, 1993.
15
Mihalik, J., D. Bell, S. Marshall, and K. Guskiewicz.
the National Football League Players Association
Measurement of head impacts in collegiate football players:
(NFLPA) on head injury biomechanics topics unre- an investigation of positional and event-type differences.
lated to this manuscript. Neurosurgery 61(6):12291235, 2007. doi:10.1227/01.
NEU.0000280147.37163.30.
16
Mihalik, J. P., K. M. Guskiewicz, J. A. Jeffries, R. M.
Greenwald, and S. W. Marshall. Characteristics of head
REFERENCES impacts sustained by youth ice hockey players. Proc Inst
Mech Eng Part P J Sport Eng Technol. 222(1):4552, 2008.
1
Allison, M. A., Y. S. Kang, J. H. Bolte, M. R. Maltese, and doi:10.1243/17543371JSET4.
17
K. B. Arbogast. Validation of a helmet-based system to Mihalik, J. P., R. M. Greenwald, J. T. Blackburn, R. C.
measure head impact biomechanics in ice hockey. Med. Cantu, S. W. Marshall, and K. M. Guskiewicz. Effect of
1906 ALLISON et al.
23
infraction type on head impact severity in youth ice hockey. Reed, N., T. Taha, M. Keightley, et al. Measurement of
Med. Sci. Sports Exerc. 42(8):14311438, 2010. doi: head impacts in youth ice hockey players. Int. J. Sports
10.1249/MSS.0b013e3181d2521a. Med. 31(11):826833, 2010. doi:10.1055/s-0030-1263103.
18 24
Mihalik, J. P., K. M. Guskiewicz, S. W. Marshall, R. M. Rowson, S., and S. M. Duma. Development of the STAR
Greenwald, J. T. Blackburn, and R. C. Cantu. Does cer- evaluation system for football helmets: integrating player
vical muscle strength in youth ice hockey players affect head impact exposure and risk of concussion. Ann. Biomed.
head impact biomechanics? Clin. J. Sport Med. 21(5):416 Eng. 39(8):21302140, 2011. doi:10.1007/s10439-011-0322-5.
25
421, 2011. doi:10.1097/JSM.0B013E31822C8A5C. Sterr, A., K. A. Herron, C. Hayward, and D. Montaldi.
19
Mihalik, J. P., K. M. Guskiewicz, S. W. Marshall, J. T. Are mild head injuries as mild as we think? Neurobehav-
Blackburn, R. C. Cantu, and R. M. Greenwald. Head impact ioral concomitants of chronic post-concussion syndrome.
biomechanics in youth hockey: comparisons across playing BMC Neurol. 6:7, 2006. doi:10.1186/1471-2377-6-7.
26
position, event types, and impact locations. Ann. Biomed. Stojsih, S., M. Boitano, M. Wilhelm, and C. Bir. A pro-
Eng. 40(1):141149, 2012. doi:10.1007/s10439-011-0405-3. spective study of punch biomechanics and cognitive func-
20
National Center for Injury Prevention and Control (U.S.). tion for amateur boxers. Br. J. Sports Med. 44(10):725730,
Report to Congress on Mild Traumatic Brain Injury in the 2010. doi:10.1136/bjsm.2008.052845.
27
United States: Steps to Prevent a Serious Public Health Takhounts, E. G., S. A. Ridella, V. Hasija, et al. Investi-
Problem. Atlanta, GA, 2003, p. 26. gation of traumatic brain injuries using the next generation
21
Ocwieja, K. E., J. P. Mihalik, S. W. Marshall, J. D. of simulated injury monitor (SIMon) nite element head
Schmidt, S. C. Trulock, and K. M. Guskiewicz. The effect model. Stapp Car Crash J. 52:131, 2008.
28
of play type and collision closing distance on head impact Takhounts, E. G., M. J. Craig, K. Moorhouse, J. McF-
biomechanics. Ann. Biomed. Eng. 40(1):9096, 2011. doi: adden, and V. Hasija. Development of brain injury criteria
10.1007/s10439-011-0401-7. (Br IC). Stapp Car Crash J. 57:243266, 2013.
22
Padgaonkar, A. J., K. W. Krieger, and A. I. King.
Measurement of angular acceleration of a rigid body using
linear accelerometers. J. Appl. Mech. 42(3):552, 1975.

Vous aimerez peut-être aussi