Introduction

Cycling is an active mode of mobility with significant health and environmental benefits. In England, there has been a significant upward trend in cycling since 2002 [1]. Despite many health, environmental, and independent travel benefits, there can be a risk of trauma in bicycle falls and collisions. In Great Britain, cyclists had a reduction in fatalities in 2021 (down 21%) compared with a significant peak during the 2020 COVID-19 pandemic. Despite this reduction, cyclist fatalities in 2021 remained higher than the 2017 to 2019 average (increase of 17%) [2]. Head injuries are a key cause of fatal and life-changing injuries in cyclists [3]. Some cyclists choose to wear a helmet as a key line of defense against head injuries if they are involved in a collision or fall. Several previous studies, including a large meta-analysis of data relating to 64,000 cyclists, have shown that helmets have a protective effect to the head against head injury (including serious and fatal injury) and facial injury in cycle incidents, such as collisions and falls [3, 4]. In particular, helmet use has been found to reduce the risk of skull fractures, hemorrhages (extradural, subdural, subarachnoid, intraparenchymal, and intraventricular) and facial fractures when a cyclist is involved in a collision or fall event [5,6,7,8,9,10,11,12,13].

All helmets which come to market must pass a minimum safety threshold, set out by standards [14,15,16,17,18,19,20]. At the time of publication, current standards use metrics based on linear motion of the head to assess the safety of helmets. These metrics can assess the protection of helmets against head injuries caused by linear mechanisms, such as skull fractures and associated focal injuries, including extradural haematoma [21]. Rotational motion is also a known head injury mechanism, leading to diffuse brain injuries and hemorrhage [22,23,24]. New initiatives in the injury biomechanics field have led to a better understanding of rotational effects on brain injury outcomes, particularly diffuse brain injuries which are seen in cyclists [3]. Injured cyclists are commonly reported to lose consciousness [25, 26]. Loss of consciousness (LOC) is associated with rotational motion [23, 27]. LOC can occur across a broad range of head injury severities, with a range of long-term outcomes. In more severe instances, cyclists can sustain another rotationally driven injury, diffuse axonal injury (DAI), which often results in unfavorable long-term outcomes [28,29,30]. Cyclists also sustain subdural haematoma, which is strongly associated with rotational motion [11, 13, 22, 31]. Despite the importance of rotational effects and the broad range of head injuries sustained by cyclists, which can be attributed to them, rotational motion is not yet assessed in current standards.

A few recent studies have assessed the performance of bicycle helmets under oblique impacts, which better represent real-world incidents and produces larger head rotation [32,33,34,35]. These studies show a wide range of performance measured with metrics based on head translation and rotation, including peak linear acceleration (PLA), peak rotational acceleration (PRA), peak rotational velocity (PRV), and the brain injury criterion (BrIC) [36]. However, these studies use the HIII headform, which has biofidelity shortcomings. For example, the HIII headform has a large coefficient of friction, leading to overprediction of head rotational metrics [37]. A state-of-the-art, more biofidelic headform that can measure linear and rotational motion of the head more accurately has been developed recently. The key physical properties of this headform, including coefficient of friction, mass, and mass moments of inertia, better represent those of the medium human head than the HIII headform [37]. Hence, the new headform is better able to assess head protection performance of helmets. The development of the headform has been underpinned by international groups of experts at the CEN/TC158/WG11 working group to shape future helmet standards like cycle helmet standard, EN1078. This biofidelic headform provides a new opportunity for assessing the performance of bicycle helmets using more biofidelic lab test methods.

The aim of this study is to determine the performance of a range of bicycle helmets under oblique impacts using the new, biofidelic headform. We provide the ranges of linear and rotational head kinematics for 30 popular helmets used by cyclists in the UK. A widely adopted head rotation management technology is MIPS (Multi-direction Impact Protection System). Previous work with the Hybrid III headform has shown MIPS to provide protection against rotation [38, 39]. Here, we test whether this remains the case with the newly adopted headform, which has a lower coefficient of friction and more accurate moments of inertia than the HIII headform. In the absence of more objective ways to compare helmet performance, consumers may rely on helmet price as the indicator of safety [40, 41]. Hence, we also test whether there is a relationship between helmet price and its linear, rotational and overall performance. Mass is a factor considered to be important in the road cycling community. We therefore investigate whether mass influences linear, rotational or overall risk.

Methods

An overview of the test method is shown in Fig. 1. The following sections explain each component of the test method.

Fig. 1
figure 1

The test method summary. a The Cellbond-CEN 2022 headform. b Each helmet was tested at four locations. c For each test, three components of head linear acceleration and rotational velocity time-history were recorded. The rotational velocities were differentiated to obtain the rotational accelerations. d Data were analyzed to derive kinematics-based injury metrics, two of which were used to calculate the linear risk and rotational risk, with mean of these risks representing the overall risk

Full size image

The New Headform

For testing helmets, we used a new headform manufactured by Cellbond, a division of Encocam Limited, under the instruction of the European Committee for Standardization Working Group 11 (CEN/TC158/WG11). The Cellbond-CEN headform is made of nylon through an injection molding process. The significant advantage of this headform over previous headforms, such as the Hybrid III and EN960 headform, is its improved biofidelity in terms of the moments of inertia (MoIs) and coefficient of friction (CoF) of the headform surface (for more details please see [37]). MoIs and CoF of the headform are key factors in determining its rotational motion during helmeted head impacts [37, 42,43,44]. In a recent study, we compared the kinematics of an earlier version of the Cellbond-CEN headform with the Hybrid III headform in various oblique impact scenarios while wearing helmets [37]. The Cellbond-CEN headform produced much lower peak rotational kinematics than the Hybrid III headform. These results emphasize the importance of incorporating realistic MoIs and CoF to accurately assess helmet performance.

In this study, we used a version of the Cellbond-CEN headform, manufactured in 2022, which includes two further improvements (Fig. 1a): (i) the headform has a realistic head and face geometry derived from data of a large human population [37] and (ii) the headform has a small portion of the neck, in contrast to the HIII headform, ensuring a more realistic interaction between the chin strap and the neck. The physical properties of this version of the headform are provided in Table 1.

Table 1 The physical properties of the Cellbond-CEN 2022 headform
Full size table

Helmet Impact Conditions

The impact test conditions involved three factors: impact speed, impact angle, and impact location [46]. The values of these three factors were aligned with previous studies and supported by the findings in our recent, extensive literature review investigating cyclist head injury and impact characteristics [3, 32, 37, 45, 47]. Our literature review revealed that the head impact speed resulted from cyclist-ground impacts is concentrated around 6.5 m/s. In addition, the reviewed literature showed that the head impact angle (the angle between the impact velocity and the ground) is often between 30° and 60°. Hence, we chose a 6.5 m/s impact speed and a 45° impact angle.

Supported by previous studies and our literature review, we conducted helmet testing at four specific locations [3, 48,49,50]: left (pXR configuration), front (pYR configuration), rear (nYR configuration), and front-left (pZR configuration), as illustrated in Fig. 1b. Each impact location induced rotational motions predominantly about the specific axis of the headform which the impact location is named after.

Helmet Impact Test Method

The oblique impact tests were conducted with the drop tower helmet test rig at the Human Experience, Analysis and Design (HEAD) Lab, Imperial College London. The test rig has been purposely built and previously used for testing helmets under various conditions [37, 46, 51]. The bicycle helmet (with visor attached if applicable) was securely fitted onto the headform following the manufacturer’s instructions. To ensure consistency, a distance of 27.5 mm was maintained between the edge of the helmet and the upper edge of the eye orbit marked on the headform (Fig. 1). The chin strap was then tightened in accordance with normal usage. Prior to tightening, a rigid cylinder with a diameter of 10 mm was placed between the chin and the strap and subsequently removed once the strap was tightened.

Next, the helmeted headform was positioned on a free-falling U-shape testing platform. To ensure proper alignment, we used an inclinometer to adjust the orientation of the headform (Fig. 1a). The goal across all test configurations was to achieve a nearly horizontal position with an inclination of 0° ± 1° for pXR/pYR/nYR or an inclination of 65° ± 1° for the pZR test. We then used a gripper to hold the helmet onto the platform, maintaining the helmet’s position and orientation during the free fall (Fig. 1a). The gripper was released by a mechanical trigger just prior to the helmet-anvil impact.

Helmets were dropped onto a metal anvil, with a 130 mm diameter and 45° inclined surface. The anvil was covered with an 80-grit abrasive paper to simulate the road surface, as suggested in previous studies and helmet test standards [46, 52,53,54]. The impact speed was recorded using a photoelectric sensor trigger system, which also triggered the video capture. A high-speed video camera was placed behind the anvil to record the impacts at a rate of 3500 frames per second. Following each test, the high-speed video was examined to verify that the helmet maintained its intended orientation and position on the platform until impact with the anvil.

For each helmet model, we conducted the four impact tests using two helmet samples. pXR and pYR were performed on helmet sample 1, and nYR and pZR were performed on helmet sample 2. The closest impact points (i.e., nYR and pZR) are at least 135° apart from each other. This is to ensure that the impact locations were separated from each other to minimize the influence of accumulated damage on the subsequent tests. To enhance the reliability of our results, each test was repeated three times using three different samples. Therefore, each helmet required 12 tests performed on 6 samples.

Kinematic Data Capture and Processing

The headform was equipped with a DTS 6DX PRO sensor package along with a wireless datalogger system. This sensor package enabled the measurement of linear accelerations and rotational velocities along the three axes (Fig. 1c). Linear accelerations and rotational velocities were measured for 0.5 s either side of the time of impact at a sampling frequency of 20 kHz. All linear accelerations were filtered at CFC600 and rotational velocities were filtered at CFC180 according to ISO 6487 [55]. All signals were filtered before they were combined or used to derive other kinematic metrics, including calculating the resultant values and obtaining rotational acceleration via differentiating rotational velocity (Fig. 1d). We investigated N-point moving average to differentiate the filtered rotational velocity to obtain rotational acceleration. There was minimal difference between the peak rotational acceleration values obtained via the 1-point (no smoothing), 3-point, and 5-point moving averages, and therefore 1-point was adopted. Python was used to filter and differentiate the test data and perform subsequent analysis. The filtering was conducted using a fourth order Butterworth phaseless digital filter function (written according to SAE J211-1) [56]. The differentiation was done without any smoothing using the numpy library “gradient” function [57]. We extracted peak values of linear and rotational acceleration (PLA and PRA), rotational velocity (PRV), and BrIC. The mean value, averaged across all repeats for a given helmet and test configuration, were calculated as well as the standard deviation and coefficient of variation (CV).

Injury Risk Calculation

Helmeted cyclists sustain both focal and diffuse brain injuries [3, 4]. Hence, we incorporated methods for evaluating the risk of these injuries from the measured head kinematics (Fig. 1d). We used the peak linear acceleration (PLA) to predict the risk of skull fractures and associated focal injuries. The focal injury risk function was based on a recent work where 30 elderly vulnerable road user collisions were reconstructed and a risk function for PLA at AIS4+ severity was produced [58]. Out of the 20 cases with AIS4+, 19 suffered skull fracture, SAH or contusions, which were predicted by PLA. Since the risk function was established using data from older casualties (60+), we adjusted it by multiplying PLA by the ratio of PLA of older population to the general population at 50% risk of AIS4+. We used the threshold adopted by helmet standards for the latter, i.e., 250 g [59] and 200 g for the former [60], leading to the following risk function:

$$P\left(\text{linear}\right)=\frac{1}{\left[1+{e}^{\left(3.3202-0.01312*\text{PLA}\right)}\right]}$$
(1)

where

$$\text{PLA}=\text{max}\left(\sqrt{{{a}_{x}\left(t\right)}^{2}+{{a}_{y}\left(t\right)}^{2}+{{a}_{z}\left(t\right)}^{2}}\right).$$
(2)

\({a}_{x}(t)\), \({a}_{y}(t)\), and \({a}_{z}\left(t\right)\) represent the components of the head linear acceleration measured at the CoG.

We also used the injury metric based on head rotational velocities, BrIC, to predict the risk of diffuse injuries [36]. It has been shown that BrIC has a strong correlation with the maximum principal strain within the brain, produced in cycle helmet oblique impacts [61]. In addition, brain strain is shown to predict diffuse brain injuries, including white matter damage [62,63,64,65]. Risk functions have been developed for BrIC based on different definitions of brain strain (maximum principal strain, or MPS, and cumulative strain damage metric). These risk curves have been produced based on large animal experiments scaled to the human. The original risk curves were developed for severe diffuse brain injuries observed in animals, which were assumed to be of AIS4+ severity. These curves were scaled to obtain risk curves for other severities. We used the MPS-based BrIC risk function for AI2+ severity to evaluate the performance of helmets in preventing diffuse brain injuries [36]. This function is as follows:

$$P\left(\text{rotational}\right)=1-{e}^{-{\left(\frac{\text{BrIC}}{0.602}\right)}^{2.84}},$$
(3)

where

$$\text{BrIC}=\sqrt{{\left(\frac{\text{max}(|{\omega }_{x}\left(t\right)|)}{{\omega }_{xC}}\right)}^{2}+{\left(\frac{\text{max}(|{\omega }_{y}\left(t\right)|)}{{\omega }_{yC}}\right)}^{2}+{\left(\frac{\text{max}(|{\omega }_{z}\left(t\right)|)}{{\omega }_{zC}}\right)}^{2}.}$$
(4)

\({\omega }_{x}\left(t\right)\), \({\omega }_{y}\left(t\right)\), and \({\omega }_{z}\left(t\right)\) are the components of the head rotational velocity. \({\omega }_{xC}\) = 66.25 rad/s, \({\omega }_{yC}\) = 56.45 rad/s, and \({\omega }_{zC}\) = 42.87 rad/s are the critical values of rotational velocity. These critical values were calculated such that they corresponded to 50% risk of severe diffuse axonal injuries in large animals [36].

After evaluating the risk of focal (linear) and diffuse (rotational) injuries for each impact location, we combined them to obtain an overall risk for each impact location. As there is currently insufficient data available to provide accurate weighting of these two types of injuries [3], we allocated them a 0.5 weighting to determine an overall injury risk for each impact location:

$$P\left(\text{injury }@ \text{location}\right)=0.5P\left(\text{linear }@\text{ location}\right)+0.5P\left(\text{rotational }@\text{ location}\right)$$
(5)

Finally, we used the weighting of each impact location determined from a meta-analysis of real-world cycle incident data to determine the overall injury risk for the helmet:

$$P\left(\text{injury }@\text{ helmet}\right)={\sum }P\left(\text{injury }@\text{ location}\right)\times P\left(\text{location}\right).$$
(6)

The meta-analysis included three studies that followed the same convention for defining the impact location, with a total number of 815 impact locations recorded [3]. The analysis provided the impact frequency for different helmet regions, namely front, side, rear, and crown. It showed that the crown is the least frequently impacted area. We used the weighting for the four impact configurations, pXR, pYR, nYR, and pZR. The pYR and pZR are both within the front region of the helmet as defined in the meta-data analysis. Hence, we equally split the weighting of front impacts between them. The weightings are provided in Table 2.

Table 2 The probability of impact location as determined from our previously published meta-analysis [3]
Full size table

Selecting Popular Helmets

Two approaches were combined to select a cohort of 30 helmets that included the most popular helmets purchased by UK consumers. Firstly, we used best-selling lists of major retailers, which yielded 10 helmets. 3 additional helmets with an additional protective technology, MIPS, were selected from the best-selling list to provide comparison. We additionally surveyed a broad cyclist population to yield a further 17 helmets. Importantly, all these helmets meet the European standard EN 1078.

Helmets Selected from Popular Retailers’ Best-Selling Lists

Halfords and Decathlon are two of the largest in-store and online helmet providers in the UK. On each respective retailer’s website, adult cycling helmets were sorted using the “Best-Selling” function and the popularity of each helmet by website was recorded accordingly as a “Selling Rank” (as accessed in May 2022). Mountain biking-only, full-face, and time-trial helmets were removed from the longlist, resulting in a list of 76 helmets. The top 10 best-selling helmets were identified from the combined best-selling list (Table 3). In addition, 3 helmets from the top 10 had counterpart models which use MIPS, namely Lazer Compact DLX MIPS, Giro Angon MIPS, and Lazer Tonic MIPS. These helmets were additionally included to allow direct comparison between MIPS and non-MIPS versions of the same helmet.

Table 3 The top 10 best-selling cycling helmets were determined from a combined best-selling list from popular, major UK helmet retailors, Halfords, and Decathlon with three additional counterpart models which were fitted with MIPS also selected to give the 13 helmets listed
Full size table

Cyclist Helmet Use Survey Design and Distribution for Helmet Selection

We surveyed UK cyclists to determine more information about their cycling habits, including helmet use and preferences. In line with Imperial College London’s ethics requirements, a survey was approved by the Head of Department for the Dyson School of Design Engineering and Research Governance Integrity Team (RGIT). The survey was circulated widely across various social media platforms to a large range (> 50) of online groups designed for cyclists. A significant effort was made to distribute the survey to groups attracting and consisting of a range of geographic locations, ages, genders, races, and other demographics. The survey invited any cyclists who owned or wore a helmet any proportion of the time to list the helmet make and model they use (if known). The following questions were used to obtain information about UK cyclist helmet-related preferences:

  1. (1)

    Do you wear a helmet when you cycle?

  2. (2)

    Do you own a cycling helmet?

  3. (3)

    What is the retail price of your current helmet?

  4. (4)

    What is the model name and brand of your current cycle helmet?

Cyclist Helmet Survey Responses Informing Helmet Selection: Helmets and Their Prices

Of 1132 respondents, 1060 (93.6%) wore a helmet at least some of the time. Interestingly, a greater proportion owned a helmet (1083, 95.7%) all of whom provided information about the current retail price of the helmet. Around half (532/1083, 49.1%) of respondents who owned helmets provided information about the make or model of helmet. A frequency distribution was created, resulting in a list of popular helmets among survey respondents. Seven of the helmets already selected via the best-selling lists were present. Secondly, the survey results were used to better understand the price of helmets commonly worn by respondents. For each unique listed helmet, a corresponding retail price at the time of purchase selection (May 2022) was identified. The continuous distribution was subsequently classified into price bands, allowing for a distribution of cycling helmet prices to be obtained. This price distribution was then populated with the most popular road cycling helmets highlighted in the survey. Further details including figures can be found in the Appendix. The Btwin 100 City (£9.99) ranked eighth was unable to be purchased. This was replaced with another popular helmet, the Specialized Tactic MIPS (shown at the bottom of Table 4).

Table 4 List of 30 popular UK helmets selected for testing (based on manufacturer best-selling lists and survey responses) showing purchase price at the time of purchase (May 2022)
Full size table

Statistical Analysis

The mean value, averaged across all repeats for a given helmet, were calculated as well as the standard deviation and coefficient of variation (CV). The CV was used to assess helmet test variability across all kinematic metrics. To directly compare a cohort of helmets composed of the same helmets with and without MIPS, the Mann-Whitney U test was used due to its ability to account for non-parametric data.

Ordinary least square (OLS) linear models from the Python package ‘statsmodels’ were used to investigate the influence of different factors on injury risk [66]. Three separate OLS models were considered. The first OLS model investigates the influence of impact location, mass, price, presence of MIPS, and distinct helmet model on the location-specific overall risk of injury associated with each test repeat, referred to previously as \(P\left(\text{injury }@ \text{location}\right)\). This enables the influence of impact location to be understood in conjunction with other listed factors. The input data included all test repeats collected from 360 tests (all helmets, impact locations and repeats). The mean overall risk at a specific location was used as a baseline for the inter-helmet comparison. In order to compare the influence of impact location, the non-exposure-weighted overall risk was used. The pXR impact location was selected as the baseline impact location as it occurs most frequently in real-world impacts [3]. The second OLS model investigates the influence of almost the same parameters (only impact location is necessarily excluded) on the helmet-specific risk, referred to earlier as \(P\left(\text{injury }@\text{ helmet}\right)\). The third OLS model investigates how individual parameters (e.g., purchase price and mass) are affected by linear and/or rotational motion. The helmet mass was taken from online vendors of the helmets. In all instances, we ensure that the assumptions of OLS models (linearity, no multicollinearity, no autocorrelation, homoscedasticity, and a normal distribution of errors) are upheld via the Omnibus and Jarque-Bera tests [67,68,69,70].

Results

Overview of the Head Kinematics

Figure 2 shows a snapshot from high-speed footage at 20 ms following the helmet/anvil contact initiation. The headform rotation about the x-axis at this time point is noticeable. This figure shows the difference in headform rotation across different helmets, e.g., a large rotation of the headform fitted with helmet 16 compared with helmet 27.

Fig. 2
figure 2

High-speed footage showing helmet response to impact at 20 ms following the helmet/anvil contact initiation. The pXR configuration is shown

Full size image

The head kinematic distributions of the whole tested helmet population for all impact locations and repeats are shown in Fig. 3. Each low transparency line represents one helmet drop. The variation in the headform kinematics can be seen by the faded distributions. This figure shows that the duration of impacts is 10–15 ms. It also shows some trends across impact locations. For instance, larger variation in linear acceleration can be seen in nYR impacts than other impact locations. In addition, when looking at the rotational velocity curves, the results are more homogenous for the pZR impact than the other three impact locations. This figure shows that some helmets produce distinctly lower rotational velocities than the rest of the helmets in pYR, pXR, and nYR impacts.

Fig. 3
figure 3

Time-history of head kinematics, showing resultant linear acceleration, resultant rotational acceleration and resultant rotational velocity for all helmets and impact locations

Full size image

Good Repeatability of the Tests

The coefficient of variation (CV) was below 10% for the majority of helmets, impact locations, and kinematic metrics (Fig. 4). 86.9% (417/480) of CVs across all metrics were below 10%, which includes 93.3% of PLA CVs, 91.7% of PRV and BrIC CVs, and 70.8% of PRA CVs. 98.5% (473/480) of CVs across all metrics were below 20%. The CV is generally higher for PRA than other metrics [CV PLA < CV PRA: U = 3483.0, p < 0.000; CV PRV < CV PRA: U = 3511.0, p < 0.0001; CV BrIC < CV PRA: U = 3328.0, p < 0.0001]. Except for a few helmets and impact locations, the CV for the other metrics was low and does not differ significantly between any other kinematics metrics, showing the good test repeatability.

Fig. 4
figure 4

The coefficient of variation of head kinematics metrics for all helmets and impact locations

Full size image

Different test locations had different CV distributions. In general, nYR and pYR tests had lower CV across all metrics, while pXR and pZR were higher. For all metrics, pYR had the lowest variability [pYR < pXR: U = 4905.0, p < 0.0001; pYR < pZR: U = 4414.0, p < 0.0001; pYR < nYR: U = 6213.0, p = 0.0333]. In addition, nYR variability was lower when compared to pXR and pZR [nYR < pXR: U = 5710.0, p = 0.0028; nYR < pZR U = 5040.0, p < 0.0001]. When using a linear model to investigate the interaction between impact location and kinematic metric on CV, the pXR and pZR tests in addition to PRA increased the CV [location = pXR, t = 2.629, p < 0.001; location = pZR, t = 2.897, p = 0.004; metric = PRA, t = 7.448, p < 0.001]. A full summary can be found in “Appendix 2”.

Head Kinematics Metrics and Injury Risks for Different Helmets

The head kinematics metrics and injury risks vary across the different helmets and impact locations, as shown in Figs. 5 and 6. The ranges of head kinematics and injury risks are provided in Table 5 for each impact location and across all locations, showing large differences between lowest and highest values recorded across all kinematics metrics and injury risks.

Fig. 5
figure 5

The head kinematics metrics for all helmets and impact locations; the mean value and standard deviation are shown

Full size image
Fig. 6
figure 6

The head injury risks for all helmets and impact locations. The linear risk is based on PLA, the rotational risk is based on BrIC, and the overall risk is the average of these risks

Full size image
Table 5 Ranges of kinematic values and three risk metrics associated with different impact locations for the 30 tested helmets
Full size table

Distinct Effect of Impact Location on Head Kinematic Metrics and Injury Risks

The mean PLA across all helmets and repeats was highest in the pXR impact location (159.9 g) than other impact locations (pYR: 147.7 g, pZR: 134.3 g, nYR: 131.1 g) (Fig. 7). The PLA across all tests in the pXR impact location was significantly higher than the tests in the other three impact locations [pXR > pYR: U = 5255.0, p = 0.0003; pXR > pZR: U = 6299.0, p < 0.0001; pXR > nYR: U = 6184.0, p < 0.0001]. Additionally, the PLA in the pYR impact location was significantly higher than in the nYR and pZR locations [nYR < pYR: U = 2857.0, p = 0.0003; pZR < pYR: U = 2761.0, p = 0.0001].

Fig. 7
figure 7

The distributions of head kinematics and risk metrics separated by impact location. The risks shown are not exposure-weighted and follow the definition of \(\text{P}\left(\text{injury }@\text{ location}\right)\) set out in the method section

Full size image

The mean PRA across all helmets and repeats was highest in the pZR (6.31 krad/s2) impact location, followed by the pYR (5.94 krad/s2), nYR (4.69 krad/s2), and finally lowest in the pXR impact location (4.15 krad/s2). Statistically, the PRA across all tests at pXR was lower when compared to test conducted at the other three impact locations [pXR < pYR: U = 1630.0, p < 0.0001; pXR < pZR: U = 1054.0, p < 0.0001; pXR < nYR: U = 2975.0, p = 0.0010]. The PRA of the nYR tests was lower than for the pZR and the pYR tests [nYR < pZR: U = 6131.0, p < 0.0001; nYR < pYR: U = 2407.0, p < 0.0001].

The mean PRV for the nYR location was highest (24.4 rad/s), followed by pYR (24.1 rad/s), pZR (22.8 rad/s) and finally pXR (17.2 rad/s). Statistically, the PRV across all repeats and helmets was significantly lower for pXR than the other three impact locations [pXR < pYR: U = 686.0, p < 0.0001; pXR < pZR: U = 1263.0, p < 0.0001; pXR < nYR: U = 832.0, p < 0.0001]. The PRV was lower across all tests at pZR than the nYR and pYR locations [pZR < nYR: U = 2761.0, p = 0.0001; pZR < pYR: U = 2958.0, p = 0.0008].

The mean BrIC across all helmets and repeats was highest in the pZR (0.523) impact location, followed by the pYR (0.450), nYR (0.436), and finally lowest in the pXR impact location (0.272). When comparing all tests at a given impact location statistically, pZR tests produced higher BrIC across all tests [pZR > nYR: U = 6254.0, p < 0.0001; pZR > pYR: U = 6434.0, p < 0.0001; pXR < pZR: U = 91.0, p < 0.0001].

Since the linear and rotational injury risks are monotonic functions of PLA and BrIC, respectively, the impact location has the same effect on these risks as it has on PLA and BrIC. Most notably, the largest linear risk was seen in pXR impacts [pXR > pYR: U = 5255.0, p = 0.0003; pXR > pZR: U = 6299.0, p < 0.0001; pXR > nYR: U = 6184.0, p < 0.0001] (Fig. 7). The largest rotational risk was seen in pZR impacts [pZR > nYR: U = 6254.0, p < 0.0001; pZR > pYR: U = 6434.0, p < 0.0001; pZR > pXR: U = 91.0, p < 0.0001]. The largest overall risk was seen in pZR impacts [pZR > nYR: U = 6215.0, p < 0.0001; pZR > pYR: U = 6079.0, p < 0.0001; pZR > pXR: U = 563.0, p < 0.0001].

The OLS model built to investigate the influence of impact location, helmet type, mass, price, and presence of MIPS on non-exposure-adjusted overall risk, \(P\left(\text{injury}@\text{location}\right)\), confirmed that pYR, nYR, and pZR all differed significantly from pXR. Significantly higher overall risk was seen in pYR [coefficient: 0.0960, t = 16.7, p < 0.001], nYR [coefficient: 0.0883, t = 15.3, p < 0.001] and pZR [coefficient: 0.1635, t = 16.7, p < 0.001] when compared to the pXR baseline. The coefficients show that the overall risk is approximately 0.1 higher for pYR and nYR compared to pXR, and 0.17 higher for pZR (the location with the highest associated risk). The findings related to other OLS model parameters are detailed within the relevant sections.

Exposure Weighted Linear, Rotational, and Overall Injury Risk: Ranking of Helmets

Finally, we used the impact location weighting to calculate one value for the linear, rotational, and overall risk for each helmet type, \(P(\text{injury}@\text{helmet})\) and rank them based on the overall risk (Table 6). We observed a larger variation in rotational risk than the linear risk, as shown in Fig. 8. The worst performing helmet of the 30 cohort (rank #30) had a 2.62 times higher overall injury risk compared to the best performing helmet (rank #1). When considering the linear and rotational components of the overall worst and best performing helmets, this ratio was 1.76 for the linear risk and 4.21 for the rotational risk.

Table 6 The overall, linear, and rotational injury risk (mean value of the three repeats) and overall rank of the helmets are shown with the purchase price (2022–23) in GBP and mass in grams, where overall risk was calculated as an average of the linear and the rotational risk
Full size table
Fig. 8
figure 8

A visualization of the influence of linear and rotational risk on the overall risk rank and values. The first subplot shows (from left to right) linear, overall, and rotational risk ranks while the second subplot shows (from left to right) linear, overall, and rotational risk values including their distributions. The colored table at the base acts as a color-coded reference to the helmets on the plot

Full size image

Of the 9 helmets which had rotational risk lower than the mean rotational risk of 0.256, 6 (66.7%) were also below the mean linear risk of 0.175, including the 6 helmets with the lowest rotational and overall risks. However, of the 15 helmets that had linear risk lower than the mean linear risk of 0.175, only 6 (40%) were also below the mean rotational risk of 0.257. A single variate OLS model predicting rotational risk using linear risk suggests there is a trend toward lower rotational risk being associated with lower linear risk, significant at p = 0.10 level but not p = 0.05 level [coefficien t = 0.173, t = 2.0, p = 0.055].

MIPS Reduces Rotational Kinematics and Risk

Ten of the top eleven most protective helmets in terms of overall risk had the MIPS add-on technology (Table 6). MIPS was a factor included in the non-exposure-weighted OLS model that also included helmet model, price, and mass. The presence of MIPS reduced overall non-location weighted risk by 0.0729 [t = − 21.3, p < 0.001].

To investigate whether MIPS was causal in this protective effect, we compared the three helmets which had models with and without MIPS using Mann–Whitney U tests (Table 7). Across all 72 tests conducted at different impact locations for the three helmets with and without MIPS, MIPS significantly reduced PRV by 19% [U = 263.0, p < 0.0001], PRA by 27.6% [U = 302.0, p < 0.0001], and BrIC by 18.5% [U = 336.0, p = 0.0002] but not PLA [U = 545.0, p = 0.1242]. The significance holds when considering the helmet models with and without MIPS individually. The Giro Angon MIPS had significantly lower PRV [U = 42.0, p = 0.044] and PRA [U = 37.0, p = 0.023] but not BrIC [U = 58.0, p = 0.218] or PLA [U = 47.0, p = 0.079] compared to the Giro Angon (without MIPS). The Lazer Tonic MIPS had significantly lower PRV [U = 21.0, p = 0.002], PRA [U = 26.0, p = 0.004], and BrIC [U = 27.0, p = 0.005] but not PLA [U = 57.0, p = 0.201] compared to the Lazer Tonic (without MIPS). The Lazer Compact with MIPS demonstrated significantly lower PRV [U = 25.0, p = 0.004], PRA [U = 29.0, p = 0.007], and BrIC [U = 27.0, p = 0.005] but not PLA [U = 61.0, p = 0.272].

Table 7 A comparison of the effect of MIPS technology on three helmet models tested with and without MIPS is presented with mean kinematics (PLA, PRV, PRA, and BrIC) and risk (linear, rotational, overall) values and associated statistics across all test repeats and impact locations (n = 72), with the differences between the MIPS and no-MIPS values (indicated by a ‘−’ in the MIPS column)
Full size table

Linear, rotational and overall risk were also compared across the three helmets which had MIPS and no-MIPS versions. Helmets with MIPS significantly reduced the overall risk by 33.8% [U = 1.0, p = 0.0003] and rotational risk by 53.8% [U = 0.0, p = 0.0002] but not the linear risk [U = 29.0, p = 0.1657]. When comparing individual helmets, the overall and rotational risks were reduced for the Giro Angon, Lazer Compact, and Lazer Tonic models when MIPS was included [all U = 0.0, p = 0.0404]. The linear risk was reduced for the Lazer Compact and Giro Angon [both U = 0.0, p = 0.0404], but not the Lazer Tonic [U = 2.0, p = 0.1914].

No Influence of Price on Protection

The price of helmets in our cohort varied between £9.99 and £135.00 (GBP). A visual summary of price vs linear and rotational risk does not indicate any association between them (Fig. 9). This was confirmed using the OLS models. Predicting overall, linear and rotational risk from price showed that there was not a significant influence of price on protection [overall risk: p = 0.755; linear risk: p = 0.263; rotational risk: p = 0.799]. The same conclusions were reached when considering all three risk values in conjunction to predict price [overall risk: p = 0.755; linear risk: p = 0.790; rotational risk: p = 0.747].

Fig. 9
figure 9

A scatter plot showing helmet purchase price vs overall risk for all thirty tested helmets

Full size image

This differed slightly when considering non-exposure-weighted risk. The OLS model combining helmet model, impact location, presence of MIPS, price, and mass showed that price had a small influence on non-exposure-weighted location-specific overall risk. For every 1GBP increase in price, the overall non-location-weighted risk increased by 0.0007 [t = 7.6, p < 0.001]. The difference between the cheapest (£10) and most expensive helmet (£135) in the sample is £125, which based on the OLS output corresponds to a difference of 0.0875 in risk, with more expensive helmets associated with a higher risk. Note that it was not possible to repeat this analysis for overall risk (adjusted for exposure) due to the assumptions of OLS modeling not being upheld in that instance. Details of this model can be found in “Appendix 3”.

Mass of Helmet Significantly Affects Linear Protection

Helmet masses in our cohort varied between 230 and 560 g. A visual summary of mass vs overall risk is shown in Fig. 10. In single variate OLS models, we observed no significant effect of mass on rotational risk [t = − 1.135, p = 0.266] or on overall risk [t = − 0.060, p = 0.952], which can also be seen in Fig. 10. However, there was a relationship between linear risk and mass [t = 2.307, 0.029], inferring that heavier helmets were associated with higher linear risk. An OLS model which assessed the combined effect of linear and rotational risk in relation to mass showed that mass depended on both the rotational risk and linear risk, in different directions [rotational risk can predict mass: t = − 2.297, p = 0.030; linear risk can predict mass: t = 3.118, p = 0.004]. This infers that increasing rotational risk was associated with decreasing mass and conversely, that increasing linear risk is associated with increasing mass.

Fig. 10
figure 10

A scatter plot showing the mass of the helmet in grams (taken from vendor or manufacturer websites) vs the overall risk for all thirty tested helmets

Full size image

When considering non-exposure-weighted overall risk, the OLS model combining helmet model, impact location, presence of MIPS, price, and mass showed that mass had a small influence on non-weighted location-specific overall risk. For every gram increase in mass, the overall non-location weighted risk increased by 0.0004 [t = 6.9, p < 0.001]. The 330 g difference between the lightest and heaviest helmet corresponds to a difference of 0.132 non-location adjusted overall risk, with heavier helmets associated with higher risk. Note that it was not possible to repeat this analysis for overall risk (adjusted for exposure) due to the assumptions of OLS modeling not being upheld in that instance.

Discussion

We presented the protection performance of 30 most popular bicycle helmets against skull fractures associated with the head linear motion and diffuse brain injuries associated with the head rotational motion. We used the new biofidelic headform for testing the helmets, allowing us to predict head linear and rotational responses more accurately [37]. In addition, we used an evidence-based test protocol enabling to assess performance under representative impact conditions [3]. Head kinematics and overall, linear, and rotational injury risks varied substantially across helmets, demonstrating that there are large differences in their protection, although all helmets had passed the standard impact tests of the EN1078 standard. Interestingly, we observed a greater difference in rotational compared to linear risk protection. The ratio between the highest and lowest rotational risk was over two folds the ratio between the highest and lowest linear risk. This suggests that the helmets tested are better optimized for managing head linear motion than rotational motion. This is likely due to the absence of rotational testing in helmet standards to date, an area that needs to be addressed in future standards in order to improve the protection of helmets against injuries caused by head rotation, particularly diffuse brain injuries.

MIPS was equipped to the top 9 and 11th overall most protective helmets, 8 of which were helmets with the lowest rotational risk (0.074–0.242). MIPS has been shown to be effective in reducing rotational motion when testing with the Hybrid III (HIII) headform across a range of headform surface conditions, including bare, with a stocking to reduce the coefficient of friction and with hair [71]. This study is the first to show that MIPS remains effective at mitigating rotational motion with the new headform that is more biofidelic than the HIII headform in terms of two key factors affecting head rotation, the coefficient of friction and moments of inertia [37, 72, 73]. Importantly, in the direct comparison between matched helmet models with and without MIPS, we found that linear acceleration was not significantly reduced by MIPS. While MIPS drives overall risk down by reducing rotational risk, it does not drive a reduction in linear risk. Interestingly, the helmet with the Wavecel technology, designed for rotational risk mitigation [74], produced the lowest linear risk (0.141) in our cohort of 30 helmets, although its overall rank was #13 due to the high rotational risk (0.301). These results show that it is vital to design helmets holistically to reduce both linear and rotational kinematics metrics to protect against a range of different head injury types caused by different mechanisms.

We observed a distinct effect of impact location on head kinematics and subsequently the linear and rotational risks calculated from PLA and BrIC, respectively. The largest PLA and linear risk were seen in pXR impacts. The largest BrIC and rotational risks were seen in pZR impacts. Interestingly although both BrIC and PRV are driven from head rotational velocity, the largest PRV was produced in nYR impacts, rather than pZR. This is because BrIC has a lower critical angular velocity for the z axis rotation than x and y, i.e., it exaggerates the effects of rotation about the z axis. A recent study of eight established brain FE models using kinematics data from cycle helmet oblique impacts has shown stronger correlation between BrIC and brain strain, than PRV [61]. This further supports adopting BrIC for assessing the effects of head rotational motion on the brain in our helmet impacts. Our findings show the particular importance of the pXR and pZR impact locations in producing highest linear and rotational risks, respectively, which should be a target for improving helmet design.

Although we found that there is no relationship between helmet price and the linear, rotational, or overall risk in isolation, our OLS model of non-exposure-weighted overall risk based on helmet model, mass, price, and presence of MIPS suggested an effect of increased price on increased risk. The lack of relationship in the isolated models is in keeping with previous work that has shown a weak negative correlation between cycle helmet price and risk of “concussion” measured under oblique impacts [33, 75]. The lack of relationship between price and safety performance is important because in the absence of objective information about helmet safety, consumers may rely on the price to indicate the safety performance [40, 41]. Previous work has shown that higher likelihood of helmet use is correlated with higher income and employment status [76]. Consumers should be able to access clear information about the protection performance of helmets, particularly relating to more affordable options.

Our survey demonstrated that helmet mass is a factor considered to be important in the road cycling community, likely due to lower mass equating to higher speed and comfort. Of 1083 helmet wearers, 890 (82.2%) found helmet mass to be either “very important,” “important,” or “somewhat important” compared to 193 (17.8%) who found mass to be “not important” or “not important at all.” The results of the multivariate OLS models suggested that increasing rotational risk is associated with decreasing mass and conversely, that increasing linear risk is associated with increasing mass (additionally supported by a single variate model). The fact that heavier helmets were associated with higher linear risk is in contrast to one of our incoming hypotheses that heavier helmets have more material and therefore offer better linear protection due to higher potential for the liner to remove energy in an impact. Our second hypothesis that lighter helmets are product of increased research and development, which is associated with better performance, is one possible reason that lighter helmets offered better linear protection. We found no studies on the relationship between bicycle helmet mass and protection, making this a novelty of our study. In one 1996 study of fatal motorcyclist incidents with axial load shift, heavier motorcycle helmets (> 1500 g among a range of 600–2000 g) were associated with higher basilar skull fracture risk [77]. The authors found no increased risk below 1500 g. Our helmets ranged from 230 to 560 g, making all bicycle helmets we tested lighter than the motorcycle helmet in the previously mentioned study. This previous work additionally assessed risk of a specific injury, whereas we use a combined overall risk metric assessing the risk of diffuse and focal injuries. Importantly, our OLS model shows that increasing mass is associated with increasing overall risk, a finding which may be of interest to the road cycling community who use mass as a purchase factor.

We repeated the tests three times using a new sample. This enabled us to quantify the variation across the tests, which can be attributed to both the test conditions and helmet manufacturing. Such information is missing in previous studies that have assessed helmet performance [32, 33]. The coefficient of variation for the peak kinematic values was found to be below 10% for 87% of tests across all metrics, with 99% below 20%. The variation was highest in the peak rotational acceleration, which is likely due to the fact that PRA was obtained by differentiating the filtered rotational velocity. We chose not to apply an additional filter to the rotational acceleration pulse, as this would act as a smoother when filtering for a second time. The CV for the other kinematic metrics was low and did not differ significantly, further demonstrating consistency of the test conditions and helmet samples. The low CV particularly provides more support for the adoption of the new headform in helmet standards and rating systems that use an isolated headform.

The Cellbond-CEN 2022 headform used in this study has further evolved since this study was completed. The focus of the improvements has been on lowering the CoF to fall within the 0.27–0.33 range, which overlaps with the range of CoF reported for the contact between human head and EPS foam (95% CI static CoF = 0.30–0.34, dynamic CoF = 0.26–0.28) and polyester liner (mean ± STD = 0.29 ± 0.07) [78, 79]. The mean value of the CoF that we measured for the headform was 0.4 ± 0.01, which is 21% higher than the upper value of the target CoF range. A study of cycle helmet oblique impacts using the HIII and NOCSAE headforms has shown that a 55–58% reduction in CoF to around 0.37–0.38 by adding a skull cap to the headforms had a small effect on PLA, PRA and PRV, decreasing them between 2 and 14% [44]. It is likely that using a more recent version of the CEN headform with lower CoF, our results will be affected within a range similar to that reported in this previous study.

Different headforms are currently used in standards and research studies, including EN960, NOCSAE, and HIII, but often their CoF or MoI are not reported or controlled, or they do not agree with human data [44]. Several studies however have investigated helmet response using headforms with more biofidelic MoI or CoF, showing that the headform kinematics is influenced by both [37, 42,43,44, 73, 80,81,82]. The new CEN headform is designed with MoI and CoF that better matches those of the human head, addressing the shortcomings of other headforms used for helmet testing. The adoption of this headform by researchers and test labs will enhance the reproducibility of helmet tests.

One novelty of this study is that injury risk functions developed for skull fractures and associated focal injuries and diffuse brain injuries were used, covering a large range of pathologies reported in helmeted cyclists’ incidents. Using a risk function allows for producing one overall risk for a helmet, in contrast to using kinematics values only [32, 83]. Previous studies have used risk functions focused on predicting the risk of “concussion” [33, 39]. Here we provided a more comprehensive picture of helmet protection against a range of injuries. A limitation of this approach is that the risks of linear and rotational injuries are averaged, assuming they have equal importance and presence in real-world casualties. With more data on the distribution of these injuries and their consequences, the weighting of these risks should be adjusted in future.

The rotational risk function is based on BrIC. This metric has had the best correlation with the strain predicted by a range of brain FE models [61]. The BrIC risk function used here was scaled by the developers of BrIC to predict mild diffuse brain injuries, which is a limitation [36]. A recent study has developed a mild traumatic brain injury risk function for BrIC based on data from professional American Football [84]. This function however provides zero risk for small values of BrIC. This study assumed that mild TBI is equivalent to “concussion,” an assumption that has been debated by neurologists, who suggest shifting the focus from symptoms to the likelihood of brain structural damage [85]. Supporting this suggestion, recent work has shown that the count of concussions is not associated with neurodegeneration in American Football players, while cumulative head linear and rotational accelerations can predict neurodegeneration [86]. Hence, in the absence of a better risk function for BrIC, here we used a more conservative risk function that associates a non-zero risk to small values of BrIC.

In this study, only four impact locations are tested, while real-world head impacts can occur in many different locations. We however ensured to select the locations impacted most frequently based on a meta-analysis of 815 cycle helmets [3]. We used the results of this analysis to weight the injury risk associated with each impact location for exposure in contrast to previous work [33, 39]. The exposure was highest in the pXR (side) impact, followed by the nYR (rear) impact and then the pYR and pZR (frontal) [3]. We however did not include a crown impact due to it being the lowest exposure impact location based a meta-analysis of 815 cycle helmets. One further recent study assessing head impact location by soft tissue damage also suggests that crown impacts are uncommon, with soft tissue injuries to the parietal region representing the lowest proportion of soft tissue injuries to the head region (excluding the face), at just 15% [87]. Additionally, all helmets which go to market in Europe and the UK must pass standards, which include impacting the crown region. Therefore, a minimum threshold of protection for impacts to the crown is obtained for all helmets in this cohort.

Consistent use of OLS models was limited due to some datasets violating the assumptions for modeling, in particular non-normal distribution of errors and residuals (identified using the Omnibus and Jarque-Bera tests). This led to an inability to use OLS models to assess the location exposure-adjusted overall risk, \(P(\text{injury}@\text{helmet})\).

Another limitation of this study is that we tested medium-sized helmets only, as dictated by the physical headform. Although future development will include a range of sizes for the new headform, at the time this study was conducted, the headform corresponds to an average adult (57 cm circumference based on a height of 175 cm) [88]. Our survey of cyclists showed that on average, males tended to wear medium or large helmets, while females tended to wear small or medium helmets, with a minority of males wearing small helmets and a minority of females wearing large helmets. This corresponds to previous work which demonstrates that head circumference is a function of height, with a slight difference between the sexes, whereby females have a 1.38 cm smaller head circumference at the same height [88]. Although a large proportion of the adult population do wear a medium helmet, future work should ensure that different helmet sizes are tested, promoting equitable research.

We selected the most popular cycle helmets used on the UK roads. This led to the inclusion of two distinct helmet technologies in the cohort of helmets studied here, MIPS and Wavecel. There are several other helmet technologies that are currently available in commercially available helmets [83]. Although these technologies were not within the helmets selected here, they warrant testing according to the protocol used in this study, allowing for a comparison between their performance and the performance of the most popular helmets included in this study. An aim of future work should be to provide the opportunity for the developers of new helmet technologies to submit their helmets for assessment by independent test labs, with the results being published for consumers’ information. This approach helps the designers to better understand the comparative performance of their helmets. It also helps consumers to learn about most protective helmet technologies that may not be popular yet.

In summary, we present the protective performance of 30 bicycle helmets which are popular in the UK under oblique impacts using a new, biofidelic headform and evidence-based test protocol enabling to assess performance under representative impact conditions. Below is a summary of the key findings:

  • The least protective helmet had a 2.62 times higher overall head injury risk than the most protective helmet. There was a lesser spread of linear risk (1.76 times) compared to rotational risk (4.21 times).

  • The pXR and pZR impact locations produced highest linear and rotational risks, respectively.

  • The nine helmets offering the best overall injury protection were all equipped with the anti-rotation technology MIPS, which in direct helmet comparisons between models with MIPS and no-MIPS versions, was shown to be effective in reducing rotational kinematics and risk under the impact conditions tested. However, not all helmets equipped with MIPS were the most protective.

  • Mass and price in isolation did not demonstrate a significant effect on exposure-weighted linear, rotational and overall risk, \(P(\text{injury}@\text{helmet})\). However, the OLS model predicting non-exposure-weighted overall risk, \(P(\text{injury}@\text{location})\), using price, mass, presence of MIPS and impact location showed that both price and mass had a small influence on the non-exposure-weighted overall risk, with increased mass and price shown to relate to increased non-exposure-weighted overall risk.

Our study highlights the need for distinct linear and rotational injuries with different mechanisms to be mitigated through continued improvement in helmet test methods and helmet designs. It also supports the need for providing consumers with objective information about helmet impact performance to help them with choosing most appropriate helmet.