Using A Soft Conformable Foot Sensor to Measure Changes in Foot Strike Angle During Running

The potential association between running foot strike analysis and performance and injury metrics has created the need for reliable methods to quantify foot strike pattern outside the laboratory. Small, wireless inertial measurement units (IMUs) allow for unrestricted movement of the participants. Current IMU methods to measure foot strike pattern places small, rigid accelerometers and/or gyroscopes on the heel cap or on the instep of the shoe. The purpose of this study was to validate a thin, conformable IMU sensor placed directly on the dorsal foot surface to determine foot strike angles and pattern. Participants (n=12) ran on a treadmill with di erent foot strike patterns while videography and sensor data were captured. Sensor measures were compared against traditional 2D video analysis techniques and the results showed that the sensor was able to accurately (92.2% success) distinguish between rearfoot and non-rearfoot foot strikes using an angular velocity cut-o value of 0 /s. There was also a strong and signi cant correlation between sensor determined foot strike angle and foot strike angle determined from videography analysis (r=0.868,p<0.001), although linear regression analysis showed that the sensor underestimated the foot strike angle. Conformable sensors with the ability to attach directly to the human skin could improve the tracking of human dynamics and should be further explored. Keywords:foot strike pattern; running; sensor; inertial measurement unit 1. Introduction Interest in running foot strike pattern (FSP) analysis has increased due to the potential association with reduced injury risk, better running economy, and improved performance [1–3]. Traditionally, FSP has been quanti ed using a variety of methods including two-dimensional (2D) video analysis, three-dimensional (3D) video analysis, center of pressure, and force plate readings [4–7].

sensor; inertial measurement unit 1. Introduction Interest in running foot strike pattern (FSP) analysis has increased due to the potential association with reduced injury risk, better running economy, and improved performance [1–3]. Traditionally, FSP has been quanti ed using a variety of methods including two-dimensional (2D) video analysis, three-dimensional (3D) video analysis, center of pressure, and force plate readings [4–7]. Although these methods have been shown to be reliable tools to classify FSP, the ability to evaluate FSP in more ecologically valid settings is of interest. For example, it is known that treadmill running is not the same as over ground running [8]. Furthermore, even though videography allows for data collection outside the traditional laboratory setting, videography still requires a speci c area of focus to accurately detect FSPs. Using these traditional methods, it is not possible, for example, to detect FSP during an entire 800 m middle distance race on a track, or during an entire road marathon event. Therefore, it is important to nd practical and e ective ways to evaluate FSP in any location where runners typically run. The use of inertial measurement units (IMUs) has become a popular way to collect data in the eld. The use of these systems in determining gait and FSP allows for unrestricted movement of the participants due to the IMUs' small size and wireless capabilities. Various researchers have used accelerometers for determining changes in gait during walking and running [9–15]. Boutaayamou et al. (2015) [10] validated the use of two foot mounted accelerometers, one on the heel and one above the proximal end of the big toe on the shoe, in detecting heel-strike, toe-strike, heel-o , and toe-o during Sports2019,7, 184; doi:10.3390 /sports7080184 /journal/sports

Sports2019,7, 184 2 of 12 running gait against the conventional 3D analysis system. The authors found that using wireless accelerometers applied to the right and left foot could accurately and precisely detect these events in the gait cycle. Similarly, Giandolini et al. (2014) [11] compared the use of accelerometers placed on the heel and metatarsal external to the shoe with 2D video analysis to determine foot strike patterns. The authors compared the time between heel and metatarsal accelerations and foot strike angle obtained from the video analysis. The method was determined to be reliable for a wide range of speeds, slopes, and foot strikes. In another application, two IMUs (one accelerometer and one gyroscope) on the top of the shoe were used to determine foot strike angle during running. Researchers found a signi cant correlation between their determinants of strike angle and sagittal plane angles from a 3D motion camera system [16]. The above-mentioned techniques made use of two separate IMUs, either accelerometers, or a combination of an accelerometer and a gyroscope to determine the characteristics of foot strike during running. Furthermore, all of the IMUs used in these studies were mounted on the surface of the shoe. Potential sensor movement with respect to the shoe reduces the ability of the sensor to accurately track the actual foot movement. It has also been shown that the addition of weight on the exterior of the shoe a ects the metabolic cost of running [17]. Although current sensors are lightweight, placement at a distance from the center of the foot could potentially a ect running economy. Recently, a wireless, skin-mounted, and conformal inertial sensor (BioStampRC, mC10 Inc., Lexington, MA, USA) has been developed with the ability to collect 3D accelerometer and 3D gyroscope data as well as sense electrical activity. The fact that this sensor is soft, thin, and conformable allows it to be placed directly on the foot to give a more accurate representation of foot dynamics. This allows the sensor to be worn inside the shoe during shod running, but also allows for the analysis of foot dynamics during barefoot

gyroscope data as well as sense electrical activity. The fact that this sensor is soft, thin, and conformable allows it to be placed directly on the foot to give a more accurate representation of foot dynamics. This allows the sensor to be worn inside the shoe during shod running, but also allows for the analysis of foot dynamics during barefoot running. With the BioStampRC's onboard memory, the potential for data collection in the eld is expanded, and the ability to collect data wirelessly allows for movement without restriction. Therefore, the purpose of this study was to validate the use of the BioStampRC in determining foot strike pattern against traditional video analysis techniques using a high-speed video camera. Speci cally we aimed to determine: (i) whether these sensors could accurately detect changes or di erences in foot strike angle during running, and (ii) whether the sensors could detect di erent FSPs (i.e., rearfoot (RF), midfoot (MF), forefoot (FF)) during running. Apart from considering FSP, our rst aim was to determine whether the sensor could accurately detect foot strike angle. Since FSP is often determined by cut-o values in foot strike angle [6,18,19], the foot strike angle itself might also be of interest. It is plausible that a runner can signi cantly change foot strike angle without necessarily changing to a di erent FSP. Altman and Davis (2012) have suggested that foot strike angle might be a better measure as it is an `objective, quanti able, continuous indication of FSP' [4]. It was hypothesized that there would be a positive correlation between the sensor determined foot strike angle and that determined from a more traditional method (2D videography). Furthermore, it was hypothesized that the sensor would be able to distinguish between di erent FSPs, similar to human raters. 2. Materials and Methods 2.1. Participants The data from 12 participants were collected and analyzed in this study. The participants were healthy men and women (no injuries in the past three months, no diabetes, cardiovascular, or renal/kidney disease) aged between 18–45 years old. Participants completed the 2015 American College of Sports Medicine Exercise Pre-participation

erent FSPs, similar to human raters. 2. Materials and Methods 2.1. Participants The data from 12 participants were collected and analyzed in this study. The participants were healthy men and women (no injuries in the past three months, no diabetes, cardiovascular, or renal/kidney disease) aged between 18–45 years old. Participants completed the 2015 American College of Sports Medicine Exercise Pre-participation Health Screening Form and provided informed consent approved by the Institutional Review Board (IRB) of the Appalachian State University. The participants were cleared for vigorous activity in order to participate in this study.

Sports2019,7, 184 3 of 12 2.2. Experimental Protocol Subjects reported to the Biomechanics Laboratory for one visit lasting about an hour. Height and mass (height: 1.76 0.12 m; mass: 75.2 23.2 kg) were collected and participants were given ve minutes to warm up at a self-selected resistance and cadence on a cycle ergometer (Monark Exercise AB, Vansbro, Sweden). Before beginning a trial, participants were asked to stand stationary on a treadmill (Bertec, Columbus, OH, USA) followed by a jump into the air. This jump was later used as an event to synchronize data collection. Participants then began running at a self-selected comfortable running pace on the treadmill. If the participant did not know a comfortable running pace, the speed of the treadmill was modi ed until a comfortable speed was found. Participants ran for one minute using either a FF strike pattern or a RF strike pattern (randomized). After a one-minute rest, the opposite strike pattern was performed. Finally, after both FF and RF trials were completed, participants were asked to run a third trial using a MF strike pattern. If a subject was unfamiliar or unaccustomed with a speci c FSP, the FSP was described similar to the de nition by Hasegawa et al. (2007) [6]. 2.3. Data Collection Prior to the participant starting the trials, a BioStampRC sensor (mc10, Lexington, MA, USA) was placed on the dorsal side of the right foot underneath the sock and shoe to measure gyroscope and accelerometer data (Figure). The sensor (6.6 3.4 0.45 cm) was lightweight (7 g), soft, and exible, and contained an inertial measurement unit with a 3-dimensional accelerometer ( 16 G) and 3-dimensional gyroscope ( 2000 /s). Data were stored in the on-board memory. The sensor was attached with a double-sided sticker pressed rmly onto the skin. Accelerometer and gyroscope data were sampled at 250 Hz. Misthcke et al. (2017) [20] showed that the sampling rates of 200 Hz or above are su cient for kinematic measures such as the angular velocities and angles employed in this study.Sports 2019, 7, x FOR PEER REVIEW 3 of 11 2.2. Experimental

a double-sided sticker pressed rmly onto the skin. Accelerometer and gyroscope data were sampled at 250 Hz. Misthcke et al. (2017) [20] showed that the sampling rates of 200 Hz or above are su cient for kinematic measures such as the angular velocities and angles employed in this study.Sports 2019, 7, x FOR PEER REVIEW 3 of 11 2.2. Experimental Protocol Subjects reported to the Biomechanics Laboratory for one visit lasting about an hour. Height and mass (height: 1.76 ± 0.12 m; mass: 75.2 ± 23.2 kg) were collected and participants were given five minutes to warm up at a self-selected resistance and cadence on a cycle ergometer (Monark Exercise AB, Vansbro, Sweden). Before beginning a trial, participants were asked to stand stationary on a treadmill (Bertec, Columbus, OH, USA) followed by a jump into the air. This jump was later used as an event to synchronize data collection. Participants then began running at a self-selected comfortable running pace on the treadmill. If the participant did not know a comfortable running pace, the speed of the treadmill was modified until a comfortable speed was found. Participants ran for one minute using either a FF strike pattern or a RF strike pattern (randomized). After a one-minute rest, the opposite strike pattern was performed. Finally, after both FF and RF trials were completed, participants were asked to run a third trial using a MF strike pattern. If a subject was unfamiliar or unaccustomed with a specific FSP, the FSP was described similar to the definition by Hasegawa et al. (2007) [6]. 2.3. Data Collection Prior to the participant starting the trials, a BioStampRC sensor (mc10, Lexington, MA, USA) was placed on the dorsal side of the right foot underneath the sock and shoe to measure gyroscope and accelerometer data (Figure 1). The sensor (6.6 × 3.4 × 0.45 cm) was lightweight (7 g), soft, and flexible, and contained an inertial measurement unit with a 3-dimensional accelerometer (±16 G) and 3-dimensional gyroscope (±2000°/s). Data were stored in the on-board memory. The sensor was attached with a double-sided sticker pressed firmly onto the skin.

shoe to measure gyroscope and accelerometer data (Figure 1). The sensor (6.6 × 3.4 × 0.45 cm) was lightweight (7 g), soft, and flexible, and contained an inertial measurement unit with a 3-dimensional accelerometer (±16 G) and 3-dimensional gyroscope (±2000°/s). Data were stored in the on-board memory. The sensor was attached with a double-sided sticker pressed firmly onto the skin. Accelerometer and gyroscope data were sampled at 250 Hz. Misthcke et al. (2017) [20] showed that the sampling rates of 200 Hz or above are sufficient for kinematic measures such as the angular velocities and angles employed in this study. Foot strike was additionally captured by means of a Sentech high-speed USB3 camera (STC- MBS241U3V, Sentech Technologies America, Inc, Carrollton, TX, USA) using MaxTRAQ 2D Software (Version 2.8.1.1, Innovision Systems Inc, Columbiaville, MI, USA). The camera was set up at the level of the treadmill to enable the whole stride and foot strike to be clearly visible in the frame (Figure 1). The frame rate was set at 120 frames per second with a resolution of 800 × 600. Figure 1. (A) BiostampRC sensor attached to the dorsal surface of the foot to determine sensor based foot strike angle (FSA SENSOR) and foot strike classification (FSCSENSOR). (B, C) View from camera showing the participant running on treadmill. This view was used by human raters to determine the foot strike angle from the video (FSA VIDEO) and the foot strike classification from video (FSCVIDEO). The image on the left (B) shows an example of FSA VIDEO = 19° and FSCVIDEO = rearfoot (RF). On the right (C) is an example of FSA VIDEO = −11° and FSC VIDEO = forefoot (FF). 2.4. Data Processing 2.4.1. Data Synchronization Figure 1. (A) BiostampRC sensor attached to the dorsal surface of the foot to determine sensor based foot strike angle (FSA SENSOR) and foot strike classi cation (FSC SENSOR). (B,C) View from camera showing the participant running on treadmill. This view was used by human raters to determine the foot strike angle from the video (FSA VIDEO) and the foot strike classi cation from

sensor attached to the dorsal surface of the foot to determine sensor based foot strike angle (FSA SENSOR) and foot strike classi cation (FSC SENSOR). (B,C) View from camera showing the participant running on treadmill. This view was used by human raters to determine the foot strike angle from the video (FSA VIDEO) and the foot strike classi cation from video (FSC VIDEO). The image on the left (B) shows an example of FSA VIDEO=19 and FSC VIDEO=rearfoot (RF). On the right (C) is an example of FSA VIDEO= 11 and FSC VIDEO=forefoot (FF). Foot strike was additionally captured by means of a Sentech high-speed USB3 camera (STC-MBS241U3V, Sentech Technologies America, Inc, Carrollton, TX, USA) using MaxTRAQ 2D Software (Version 2.8.1.1, Innovision Systems Inc, Columbiaville, MI, USA). The camera was set up at the level of the treadmill to enable the whole stride and foot strike to be clearly visible in the frame (Figure). The frame rate was set at 120 frames per second with a resolution of 800 600.

Sports2019,7, 184 4 of 12 2.4. Data Processing 2.4.1. Data Synchronization Each trial began with the participant performing a standing jump into the air where both feet left the treadmill. This movement was used to time synchronize the videography data with the IMU sensor data. Two-dimensional (2D) videography data were downloaded and analyzed using Kinovea software (Version 0.8.15,). Absolute foot angle was measured over the time of the initial jump (from just before the participant took o for the jump until just after the participant landed). IMU sensor data, speci cally gyroscope (angular velocity) data measured about the z-axis (corresponding to sagittal plane axis of the foot-placed sensor), were integrated to give estimated sensor foot angles. Synchronization occurred by means of a MATLAB (Mathworks, Natick, MA, USA) function, which determined which part of the signal (2D angles of jump from videography) best matched a longer data stream (IMU sensor based angles). After the videography and sensor angle data were resampled (using shape-preserving piecewise cubic interpolation) to a common frequency (1000 Hz), this synchronization method shifted the signal (2D videography angle during jump) to best match the data (IMU sensor based angles throughout the complete trial) by calculating the smallest squared Euclidean distance between the signal and data array. This allowed a time match of the di erent signals with su cient accuracy to analyze the same foot contacts using the sensor and the videography data. Once the two datasets (videography and IMU data) were synchronized, a time 30 s into the trial was selected as the starting point for foot strike analysis. The rst ten right foot strikes after this starting point were further analyzed. 2.4.2. Foot Strike Determination Human rater using 2D videography: Ten steps were analyzed by three independent raters using two sets of criteria: Video Foot Strike Angle (FSAVIDEO): Using Kinovea software, raters measured the foot angle of each foot strike relative to the treadmill surface. The angle was measured between the top surface of the treadmill belt and the bottom surface of the outsole of the shoe with the vertex at the initial shoe contact point

independent raters using two sets of criteria: Video Foot Strike Angle (FSAVIDEO): Using Kinovea software, raters measured the foot angle of each foot strike relative to the treadmill surface. The angle was measured between the top surface of the treadmill belt and the bottom surface of the outsole of the shoe with the vertex at the initial shoe contact point with the treadmill. A positive angle indicated rearfoot and a negative angle indicating forefoot (Figure). Final FSA VIDEOwas the average across raters. Video Foot Strike Classi cation (FSCVIDEO): Determining an ordinal classi cation of foot strike pattern based o of Hasegawa et al. (2007) [6], raters were given the following instructions related to whether a footfall was RF, MF, or FF (Figure): Rearfoot strike (RF)— rst foot-ground contact with the heel or rear third part of the sole only. Midfoot or forefoot portion had no contact at foot strike. Midfoot strike (MF)— rst foot-ground contact with not only the rear third of the sole, but the midfoot or entire part of the sole. Forefoot strike (FF)— rst foot-ground contact was the forefoot, or front half of the sole, and the heel did not have any contact at foot strike. For nal analysis, foot strikes that were given the same rating by at least two of three raters were used, which were all of the 360 foot strikes (i.e., no foot strike was given three di erent classi cations of RF, MF, FF) by the three di erent raters. However, after analyzing the results from the rater classi cation, we found that only 12% (44 out of 360) of all foot strikes were classi ed as MF, and that 5 out of 12 participants never performed a MF strike pattern as rated by the raters. The low number of samples reduced the potential accuracy of the classi cation of MF strike pattern and valid discrimination between RF, MF, and FF. We therefore decided to collapse all of the MF and FF foot strikes into one category: non-rearfoot strikes. The two nal classi ers were: rearfoot (RF) and non-rearfoot (NRF).

rated by the raters. The low number of samples reduced the potential accuracy of the classi cation of MF strike pattern and valid discrimination between RF, MF, and FF. We therefore decided to collapse all of the MF and FF foot strikes into one category: non-rearfoot strikes. The two nal classi ers were: rearfoot (RF) and non-rearfoot (NRF).

This research explores a new method for analyzing foot strike patterns in runners.