Forty-four (20 females, 24 males) healthy recreational runners (age 25.2 ± 6.5 years; height 167.2 ± 7.0 cm; body mass 62.8 ± 8.1 kg) were recruited. The inclusion criteria for the study were: (1) individuals who were recreational runners, with more than 1 year of regular running months (Honert et al., 2020), (2) with a minimum of one weekly run for the past three, (3) between the ages of 18–45, (4) comfortable to run on a treadmill without the need for handrails, and (5) capable of running continuously at a speed of 8 km/h for at least 10 min (Gazendam & Hof, 2007). Participants were excluded from the study if they (1) were pregnant during the time of the study (for female participants), (2) answered positively to any questions on the Physical Activity Readiness Questionnaire (PARQ+ 2020) indicating the presence of serious health conditions, (3) had undergone surgery on the legs within the past year, or (4) had lower limbs injuries requiring more than 7 days of rest in the past 6 months. All participants provided written informed consent prior to their participation and ethics approval was granted by the Nanyang Technological University Institutional Review Board (IRB-2021-124).

All participants were invited to run at their comfortable paces on a treadmill (h/p/cosmos saturn®, h/p/cosmos® sports & medical gmbh, Nusseldorf-Traunstein, Germany) using their own running shoes. To facilitate video analysis, cloth tapes were placed on the anterior and posterior talus, calcaneus, and mid-gastrocnemius on both limbs in each participant (Figure 2). The study protocol commenced with a 5-min warm-up session (Riley et al., 2018) where participants were allowed to transition from walking to running. Immediately after, the participants were directed to incrementally increase their running speed to a self-selected comfortable pace (9.4 ± 1.1 km/h) which they had to maintain for approximately 10 min. Verbal confirmation of their comfortable running speed was obtained prior to the start of the video recording. Participants were also instructed to maintain their running position at a consistent area of the treadmill. For recording the participants’ running, a single video camera (Casio Exilim EX-100) was utilized, and the camera was set at different spots for video recording from the front and back view. The video recordings were captured sequentially, with the front view being recorded first, followed by the back view.

The video camera was placed 1.5 m away, perpendicular to the participants’ coronal plane (Figure 3). The camera height was set at the hip (front view) and ankle (back view) levels respectively for each participant. Each camera view was recorded for approximately 30 s at 120 Hz while participants were running at their self-selected speed during the 10-min window. Participants continued running while researchers move the camera position. Multiple recordings were taken in each camera position, and the video in which participants displayed minimal movement (moving sideways, etc.) was selected by the researchers for analysis.

Souza (2016) provided guidelines that were followed by researchers to standardize the identification of the initial foot strike frame. After identifying the frame of initial foot strike, the researchers used the following procedures to derive angle measurements for each camera view. In the back camera view, the inversion angle was calculated between the line extending from the posterior talus to the midpoint of the gastrocnemius and calcaneus (Figure 4). In the front camera view, the inversion angle was calculated by measuring the angle formed between the line from the base of the foot to the anterior talus and the perpendicular angle from the base of the foot (Figure 5). Positive values indicate inversion, while negative values indicated eversion in both measurement methods. Figures 4, 5 outline the steps taken in detail for angle measurements in the front and back camera views, respectively. Subsequently, 18 consecutive gait cycles of each foot from each camera view were analyzed in Kinovea (Version 0.8.27). The mean values across the 18 gait cycles were used for analysis. The reason for selecting 18 gait cycles was based on previous research indicating that a minimum number of steps, ranging between 7 and 25 gait cycles, were necessary to attain stable kinematic mean values (Dingenen et al., 2018a; Oliveira & Pirscoveanu, 2021).

To address the aims of the study, two types of analyses were conducted. Preliminary testing was first conducted to screen for normality and homogeneity of variance. As the assumption of normality was violated in the back view data according to the results of Shapiro-Wilk test, all data were log-transformed prior to statistical analysis. A 2 × 2 (Camera × Foot) Analysis of Variance (ANOVA) with repeated measures was performed to compare the measured angles from 2 camera views for each foot. As 2 × 2 ANOVA revealed no significant differences between the left and right feet, data from both feet were combined when performing the Pearson’s correlation test. Thereafter, linear regression was performed to derive a single equation for the conversion of the front-view angle to the typical back-view angle. The Wilcoxon signed-rank test was conducted to compare the equation-derived measurements and the actual measurement obtained from the back-view perspective for both the left and right foot. Effect size (r) was interpreted as small (0.1 ≤ r < 0.3), medium (0.3 ≤ r < 0.5), or large (0.5 ≤ r ≤ 1.0) (Cohen, 1988). Statistical significance was set at p < 0.05 level and the statistical analyses were performed using SPSS (Version 28; SPSS Inc., Chicago, IL, United States) and R (Version 4.1.1). As this study reported discrete values instead of continuous time-series kinematic data, no filtering was performed.

To provide a gauge on the frequency of crossover gait problems potentially encountered by clinicians and practitioners, the number of crossover gait was reported in this study. Approximately 30% of the participants in the present study displayed a crossover gait, resulting in an obstructed view of their foot when recording from the dorsal plane. This prevents practitioners from accurately measuring the foot inversion angle at the point of the initial foot strike when using the conventional technique. The frequent occurrence of crossover gait reaffirms that the traditional back-view measurement is insufficient and hence there is a need to look for alternative measurements, such as the front-view measurement proposed in our study.

This study found that the mean front and back camera view measurements were 11.4° (10.4°, 12.3°) and 6.5° (5.0°, 7.9°), respectively. The traditional back camera view measurements were comparable with those obtained by McCrory et al. (1999), who also used a 2D video analysis approach and found that healthy runners displayed rearfoot eversion angles of −7.0° ± 0.7°. These results were similar to Grau et al. (2008), where a 3D motion analysis involving healthy runners revealed foot eversion angles of 1.9° ± 2.6°. In the studies mentioned above, the negative values denote an inversion angle and were measured from the back camera view. This emphasizes the caution that should be taken in using the values interchangeably. In addition, it also highlights the need for a standard conversion equation to be established, enabling front-view foot measurements to be translated and used in clinical diagnosis.

There are a few possible reasons for the differences in foot inversion angle measurements between the front and back camera views. For simplicity, this study treated the foot as a single rigid body. However, it is understood that the foot is a complex structure in which the forefoot and rearfoot may behave independently (Davis, 2004). This implies that the forefoot may have inverted without affecting the rearfoot. Therefore, the measurements could have differed significantly due to the independent foot structures.

No significant differences were revealed between the left and right foot inversion angles, suggesting that participants utilized comparable foot-landing techniques for both feet. This study did not examine individual foot strike patterns. However, previous research has shown that these techniques can impact rearfoot and forefoot kinematics. For example, heel-strikers are more likely to display greater inversion angles at touchdown compared with forefoot strikers (Pohl & Buckley, 2008). As such, this may have contributed to the discrepancies in measurements between the two camera views.

The present study performed a regression analysis to obtain an equation that allows one to easily convert the front-view foot inversion angles to the traditional back-view rearfoot inversion angles. This regression equation can assist practitioners and clinicians in the event of an obstruction of the back camera view. For example, if a runner’s foot inversion angle is measured to be 10° from the front view, clinicians can calculate the supposedly back view angle to be approximately 5.7° (0.42 + 0.53 × 10°) and thereafter perform the necessary diagnosis for the runner. Additionally, the comparison between equation-derived measurements and actual measurements revealed no statistically significant difference. This confirms the possibility of using the equation to convert front view measurements to back view measurements despite having a low R ² value. Future research could define reference norms and cut-off values based on front-view camera measurements of the foot inversion angle. This novel strategy might potentially be applied to all runners, regardless of the presence of crossover gait.

This study used only one camera in each trial, positioned at different heights for the front and back views. The lack of synchronization between camera recordings may have compromised the accuracy of the results. The use of a single camera was done intentionally to simulate clinical settings where one camera view is often used to measure multiple kinematic variables, possibly due to the increased efficiency and limitations on manpower and equipment. For instance, the front camera view at the hip level could also measure knee adduction and contralateral hip drop (Dingenen et al., 2018a). Although this single-camera approach resulted in different gait cycles being analyzed and compared, it reflects the practical constraints encountered in clinical settings.

In addition, this study only measured 18 steps for each leg with varied gait patterns (e.g., crossover steps). Human locomotion such as running can present considerable variability between strides due to changes in motor control function (Stergiou et al., 2006). Previous studies that have attempted to identify the number of strides required for kinematic variables to reach a stable mean by performing a sequential estimation technique showed varying results. Dingenen et al. (2018a) reported a minimum of 7 steps while Oliveira and Pirscoveanu (2021) recommended a minimum of 25 steps were necessary to attain a stable kinematic mean. Therefore, although we have chosen a value within the higher range, it remains possible that the foot inversion angle measurements in this study did not reach a stable mean.

Another limitation is the use of a visual-based approach to identify the initial foot strike of running. This approach may be less accurate when compared to using an instrumented treadmill that provides ground reaction forces for quantitative identification (e.g., threshold of 20 N). From video recordings sampled at 60 Hz, Ghoussayni et al. (2004) found that the error in detecting the initial foot strike was less than 2 frames (0.033 s) when comparing to using a force platform. This present study used a high-speed video camera with 120 Hz recording frequency and hence a higher accuracy in initial foot strike identification can be expected.