Spectral and Spatial Analysis of Plantar Force Distributions Across Foot-Strike Patterns During Treadmill Running

1. Introduction

Runners, whether barefoot or shod, have distinctive footstrike patterns categorised as heel, midfoot, or forefoot [1]. Typically, the best runners are forefoot strikers [2], yet over 90% of recreational runners are expected to be heel strikers [2]. Visually, it requires high-speed video [3] to capture the specific region of the foot making initial contact with the ground [4]. However, this method does not reveal the spatial application of plantar forces. Alternatively, it can be inferred from foot centre of pressure trajectories or vertical ground reaction force profiles [5] and more recently tri-axial accelerometer’s [6].

In respect to vertical ground reaction force traces, the presence of two force peaks indicates a heel or mid-foot strike (Figure 1A,B). A large 1^st peak signifies a heel strike, a spur is a midfoot strike, and no peak within 13% of ground contact time from point of initial contact signifies a forefoot strike [7,8] (Figure 1C).

These landmarks are important as they reflect the impact absorption phase of stance, which occurs before skeletal muscles can actively engage [9], thereby capturing the unattenuated collision between the foot and ground [10]. While such patterns are typically derived from overground running, the average velocity of the centre of mass is zero in treadmill running [11], and joint kinematics differ during the stance phase [12]. However, since the advent of instrumented treadmills [13], ground reaction force profiles at matched speeds suggest modalities are very similar [14]. Though some variability may occur between treadmill brands or when compared to in-shoe pressure insoles [15]. Furthermore, peak plantar pressures are generally lower during treadmill running, though directionally similar [16]. These modality differences may influence strike-pattern classification [17] and impact-related metrics derived from them.

The magnitude of this collision force along with its loading rate have been thoroughly researched in relation to the very high injury prevalence within runners [18,19,20] and the impact of shoe-midsole cushioning interventions to attenuate the impact force and associated loading rate [10,21,22,23,24]. However, in both cases the links are tenuous, where higher impact forces and loading rates are observed within heel strikes, compared to midfoot or forefoot strikers [25]. Likewise, research into the cushioning properties of a shoe’s ability is somewhat inconclusive. Questions have been raised [26] regarding the efficacy of traditional kinetic metrics used as indicators of foot-strike mechanics [5] and impact severity. The authors highlighted how total force measurements may be influenced by spatially non-specific components, masking the differences related to footwear cushioning interventions [26] and, by extension, foot-strike patterns. Additionally, the reliance on total peak values may overlook the spectral composition of the force signal, where low-frequency (<10 Hz), more related to the active processes rather than impact [27], can hide [26] the high-frequency (>10 Hz) component more directly related to impact and caused by the rapid deceleration of the limb [28].

Therefore, the aim of this study is to characterise foot-strike patterns during treadmill running based on the spatial distribution of in-shoe plantar forces, and to identify differences in impact kinetics using spectral analysis, with a specific focus on low- and high-frequency power components.

2. Materials and Methods

2.1. Participants

Eleven recreational to nationally competitive endurance runners, free of injury, participated in the study after providing written consent in accordance with the University Human Ethics Committee approval. This sample size was based on priori statistics (G*power V 3.1.9.7, Heinrich-Heine University, Dusseldorf, Germany) for analysis of variance (ANOVA) repeated measures within factors, using an alpha value of 0.05, power of 0.95, and effect sizes based on previously reported differences in foot strike patterns and kinetic variables: peak impact force, 2.43 [23]; heel pressure, 1.08 [26]; loading rate, 1.62 – 2.44 [23,29,30]; active peak force, 0.56 – 1.41 [5,23,29].

Participant characteristics (mean ± SD) where: age (yrs.) 29.6 ± 9.8, height (cm) 174.0 ± 8.7, body mass (kg) 65.4 ± 8.9 and Body Mass Index 21.5 ± 1.3) and weekly training volume (km∙wk^-1) 77.3 ± 23.6.

Prior to the study, participants had recently displayed individual peak impact force spatial distribution values (Figure 2) during treadmill running at 12 km∙h^-1 (mean ± SD: n-steps 138 ± 10; heel component (%) 47.7 ± 29.4; midfoot component (%) 29.1 ± 10.4, forefoot component (%) 23.2 ±15.6).

2.2. Procedures and Measurements

The experimental protocol consisted of three trials (heel, midfoot, and forefoot strike patterns) while treadmill running at 10 km∙h^-1. Each trial was performed in one session on the same day in a counter balanced order of strike pattern, with participants all wearing the same brand-model of shoe that they were familiar with.

Upon arrival at the laboratory, participants were weighed, measured and presented with the required shoe size. The prototype cushioned shoe (Figure 3A) has a 38 mm heel (12% from heel) and a 32 mm forefoot stack height (75% from heel) midsole, featuring a single layer of ATPU (Aliphatic Thermoplastic Polyether Polyurethane) super critical bead foam cushioning. The foam had a factory pre-processing density of 0.08 ± 0.01 g∙cm^-2. Shoe cushioning property assessment using a modified industry standard test (ISO 20344:2021 (5.17)) [31] - Instron 4467 (Instron, Norwood, MA, United States) - compressed the midsole vertically at a rate of 10 mm per minute with a maximum applied force of 2.2 kN. Results of this test are provided in Figure 3B.

Following a self-selected warm-up period on the treadmill (Life Fitness, Hamilton, New Zealand) at 10 km∙h^-1, participants were asked to use one of three strike pattern techniques (Heel, Midfoot, or Forefoot). Prior to and during each trial, where necessary, instructions were given to land (make initial contact) on the heel, flat footed, or on the balls of their feet. Once they were comfortable that they were using the specified technique, 10 s of data were logged for each condition [31,32]. Time-force data from Loadsol® pressure sensitive in-soles, segmented into posterior, central, and anterior zones (Figure 2 legend) was logged at 200 Hz. The data was then uploaded to MATLAB (R2022b, (MathWorks, Inc., Natick, MA, United States), re-sampled to 1,000 Hz, and processed using a 10-20 N force threshold to detect initial foot contact and toe-off in the total force trace. Based on these identified points, the following variables were identified or calculated [31] including spatial zone data where applicable:

• Stride duration was the time from one initial impact to the next initial impact for the same foot.
• Ground contact time (s) was the time the foot remained in contact with the treadmill surface i.e. toe-off – initial contact times.
• Swing time (s) was the time the foot has no contact with the treadmill surface.
• Active peak vertical force (N∙BW^-1) was the second peak for heel strike runners, or the highest force reading of each step for midfoot and forefoot strikers.
• Peak vertical impact force (N∙BW^-1) identified as the first peak in the total force trace between initial contact and the active peak. If the first peak was not present in running (mid-forefoot strike) it was defined as the force at 13% of the ground contact time.
• Average loading rate (N∙BW^-1∙s^-1) is the difference between forces at 20% and 80% of the peak impact force divided by the corresponding time interval (s).
• Impulse (N∙s∙BW^-1), the area under the force-time curve.
• Power Spectral Density (PSD) analysis: During the stance phase, while the right foot was in contact with the ground, force data (total and spatial zones) were analysed in the frequency domain. PSD was computed using Welch’s [33] method (MATLAB, default windowing), and applied to the first 20% of each step’s stance phase to isolate impact related portion of ground contact. A frequency resolution of 2 Hz was used from 0-100 Hz. PSD values were retained in their raw form (N²∙Hz^-1), preserving magnitude to reflect actual spectral energy differences across the zones in relation to strike pattern. Spectral power was partitioned into low-frequency (<10 Hz) and high-frequency (10-100 Hz) bands to quantify relative contributions and frequency dependent loading characteristics.

All dependent variable data were calculated per step and expressed as mean±SD for each independent variable. Differences in spatiotemporal and kinetic parameters determined from total force were compared between strike patterns using a one-way ANOVA, with Tukey post-hoc multiple comparisons if significance was found. Differences between strike pattern and spatial zones and or low-frequency and high-frequency bands were analysed using a two-way repeated-measures ANOVA, with 2 within-subject variables (strike-pattern*Spatial-zone) or (strike-pattern*Frequency-band). Where significance was found, Sidak’s post-hoc multiple comparisons testing was performed. All statistics were performed using GraphPad Prism (V 8.4, (GraphPad Software, San Diego, CA, United States) with significance set at p<0.05.

3. Results

The data analysis period resulted in the analysis of 13 ± 1, 13 ± 1, 14 ± 1 right steps for the heel, midfoot, and forefoot strike patterns, respectively. No first peaks were identified in any foot-strike method for any participant despite there being clear spatial differences, as demonstrated in Figure 4.

Mean±SD data for total force derived variables stride duration, ground contact time, swing time, peak impact force, average loading rate, active peak force, and impulse are presented in Table 1. One-way ANOVA found no-significant differences for stride duration (F_{(1.76, 17.65)} =1.6, p=0.230), ground contact time (F_{(1.85, 18.47)}=1.05, p=0.365), or swing time (F_{(1.85, 18.47)} =0.98, p=0.389). There were overall significant differences for: peak impact force (F_{(1.63, 16.25)} =13.58, p=0.0006, Table 1) with forefoot impact being less than heel (P=0.024) and midfoot (p=0.002) strike patterns; average loading rate (F_{(1.75, 17.54)} =14.61, p=0.0003), with forefoot being less than heel (p=0.009) and midfoot (p=0.001) strike patterns; active peak force (F_{(1.36, 13.55)} =20.32, p=0.0002, Table 1) with the heel being less than both midfoot (p=0.023) or forefoot (p=0.001), and the midfoot less than the forefoot (p=0.006); impulse (F_{(1.11, 11.05)} =23.9, p=0.0004, Table 1) with the heel being less than the midfoot (p<0.0001), and the forefoot (p=0.0008), while the midfoot was also less than the forefoot (p=0.012) strike patterns.

Two-way ANOVA for peak impact identified a strike-pattern*Spatial-zone interaction (F_(4,40)=91.12, p<0.0001, Figure 5) with a main effect for strike pattern (F_(2,20)=14.14, p=0.0001) and spatial zone (F_(2,20)=13.37, p=0.0002). Within each strike pattern multiple post-hoc comparisons (Figure 5A, Table 2) presented increased contribution of the posterior zone compared to either the central or anterior zone (p<0.0001) during heel striking; increased contribution of the posterior compared to the central or anterior zones (p<0.01) during midfoot strikes; while the anterior zone was greater (p<0.0001) than either the central or posterior zones during forefoot strikes. Between strike pattern comparisons for the major zone of force contribution showed that the posterior of the heel strike was greater than the posterior in the midfoot or anterior zone in the forefoot strikes (p<0.0001). From a contribution perspective, the Mean±SD (%) values for posterior, central and anterior zones were: heel strikes - 81±13, 16±11, 3±3; midfoot strikes - 49±19, 29±11, 21±10; and forefoot strikes - 9±15, 28±9, 63±17, respectively (Figure 5C). As such, the overall spatial zone contribution was significantly different for each foot strike (F_(2,20)=10.27, p=0.0009,Figure 5C) where post-hoc identified significance (p<0.05) in every pairing accept between the central and anterior zones in the heel and midfoot strikes. There was also no difference across strike patterns for the central zone contribution.

Two-way ANOVA for average loading rate identified a strike-pattern*Spatial-zone interaction (F_(4,40)=74.7, p<0.0001, Figure 5B, Table 2) with a main effect for strike pattern (F_(2,20)=11.24, p=0.0007) and spatial zone (F_(2,20)=22.9, p<0.0001). Within each strike pattern multiple post-hoc comparisons (Figure 5B, Table 2) presented increased contribution of the posterior zone compared to either the central or anterior zone (p<0.0001) but not the central and anterior zones (p=0.227) during heel striking; increased contribution of the posterior compared to the central (p=0.002) or anterior zones (p<0.0001), but no difference between central and anterior (p=0.587) during midfoot strikes; while the anterior zone was greater than the central (p<0.0001) and the posterior (p<0.0001) zone, and the central was greater than the posterior zone (p=0.06).

Two-way ANOVA (strike-pattern*spatial-zone) for active peak force resulted in an interaction (F_(4,40)=59.17, p<0.0001) with a significant main effect for spatial zone (F_(2,20)=35.84, p<0.0001) but not strike pattern (F_(2,20)=0.94, p=0.406). Multiple post-hoc comparisons for spatial zone across strike patterns highlighted significance (p<0.001) for the posterior and anterior zones but not the central zone (P>0.8), Table 2. Impulse had an interaction (F_(4,40)=104.4, p<0.0001) with main effect differences for both strike pattern (F_(2,20)=20.88, p<0.0001) and spatial zone (F_(2,20)=76.8, p<0.0001) with p<0.0001 post-hoc differences for both posterior and anterior zones but not the central zone (p>0.8), Table 2.

Two-way ANOVA for power spectral density data (Figure 6) found a frequency*strike-pattern interaction (F_(2,20)=11.74, p=0.0005) with a main effect for frequency (F_(1,10)=136.4, p<0.0001) and strike pattern (F_(2,20)=12.37, p=0.0004). Post-hoc comparison between strike patterns were not significantly different in the low-frequency but there were differences in the high-frequency band for all pairings (Figure 6A). Specifically, total high-frequency was lower in the heel strike compared to midfoot strike (255351±79702, 302406±100656 N²∙Hz^-1, p=0.011), heel strike was greater than the forefoot strike (255351±79702, 185948±51238 N²∙Hz^-1, p=0.0004) and the midfoot was greater than the forefoot strike (302406±100656, 185948±51238 N²∙Hz^-1, p<0.0001). Further analysis of the high-frequency component with regards to strike-pattern*spatial-zone (Figure 6B) showed an interaction (F_(2,20)=65.6, p<0.0001), and main effects for both strike pattern (F_(2,20)=16.91, p<0.0001) and spatial zone (F_(1,10)=28.47, p<0.0001). Key post-hoc comparisons of interest highlighted greater PSD high-frequency in the heel strike posterior zone vs. midfoot strike central zone (166845±48401, 32613±18687 N²∙Hz^-1, p<0.0001), heel posterior vs. forefoot anterior (166845±48401, 88075±30093 N²∙Hz^-1, p<0.0001), and heel posterior vs. midfoot central (166845±48401, 77517±44168 N²∙Hz^-1, p<0.0001). There was no post-hoc difference between midfoot central vs. forefoot central (32613±18687, 18082±15512 N²∙Hz^-1, p<0.0001) or midfoot posterior vs. forefoot anterior (77517±44168, 88075±30093 N²∙Hz^-1, p=0.990).

4. Discussion

This study aimed to characterise foot-strike patterns during treadmill running using in-shoe plantar force distributions to quantify impact related kinetics together with spectral analysis. While no significant spatiotemporal variations occurred between foot strike patterns, significant spatial differences in peak impact and average loading rate were evident, despite the absence of distinct peaks in the midfoot or heel strike force traces. As such, strike classification appears more useful when based on spatial force distribution than on total plantar force during treadmill running. Further, the high-frequency component of the PSD was greatest in midfoot and lowest in the forefoot strikes, with the posterior zone of heel strikes exhibiting the highest zone specific high-frequency power out of all strike pattern spatial zones. Lastly, total peak active force and total impulse were greater in forefoot strikes than in midfoot and heel strikes.

All participants successfully adopted the assigned footstrike techniques, enabling development of data driven classification guidelines based on spatial distribution of impact forces (Figure 5C). Where, runners exhibiting >70% of impact in the posterior zone were classified heel strikers; >50% in the anterior zone as forefoot strikers; and those with <70% posterior, <50% anterior, but >25% in the central zone were classified as midfoot strikers. Based on these classification thresholds, our cohort included four natural heel strikers and seven natural midfoot strikers (Figure 2).

The adaptation of different strike patterns resulted in no paired-differences to stride duration or its components, ground contact time or swing time. This agrees with comparisons between non-paired natural heel vs forefoot strikers [34,35] and paired natural heel vs unnatural forefoot strikers [30]. This likely reflects a preferred movement path [36], where individuals modulate impact attenuation or force production from strike specific spatial zones ( Figure 5A,B , Table 2). However, this contrasts with observations where midfoot strikers presented shorter duration ground contact times and superior running economy than heel strikers [37]. In the current study, participants wore a prototype shoe that, in vitro testing (Figure 3B), demonstrated approximately 11% greater cushioning than a maximally cushioned shoe, tested under the same conditions [31]. Given the association between footwear geometry and reduced ground contact time [38], as well as the influence of cushioning on leg stiffness and gait adaptations [39,40], such a design could influence strike pattern related spatiotemporal variables. However, no differences were observed in the current dataset, agreeing with comparisons between minimalistic and conventional cushioned shoes [8,41]. Further investigation exploring the interaction between shoe geometry, cushioning, and strike pattern is warranted.

The absorption phase of ground contact is characterised by impact kinetics, typically determined through force-trace peak identification [5] or the time component and inability to consciously dampen the impact [9]. While the first peak is always present in overland running, this criteria is somewhat equivocal during treadmill running [17], where both instrumented-treadmill brand, and kinetic measurement system, have produced presence and absence of such peaks [15]. Data from the current study suggests that the sole reliance on total vertical force traces may lead to misclassification of strike patterns. Specifically, in-shoe total plantar force data revealed no distinct impact peaks for midfoot or heel strike peaks (Figure 4), despite clear distinctions in spatial zone differences in peak impact and average loading rate (Figure 5). These spatial differences aligned consistently with the designated strike pattern (Figure 5A,B). The posterior zone exhibited the highest magnitude impact peak and average loading rate in the heel strike, the anterior zone dominated in the forefoot strike, and midfoot strike displayed a more even distribution with the posterior zone still being the dominant zone (Figure 5). These within-strike pattern distinctions highlight the value of using spatial force data to accurately identify foot strike pattern and more advanced analysis to assess impact kinetics during running interventions.

Spectral analysis has been used to separate the true (high-frequency) impact component of the total force signal from that of the active processes [26,27] that occur during limb deceleration. Figure 6A supports the presence of a low-frequency active process component but suggests it may not significantly influence the overall kinetic signal. Alternatively, it could be that within the same participants, strike pattern did not alter the active processes. Nonetheless, strike pattern had a significant effect on peak impact, average loading rate (Table 1), and the high-frequency PSD. Interestingly, the midfoot strike had significantly greater values than either the heel or forefoot strike, despite most participants being natural midfoot strikers (Figure 2) based on the classification criteria (Figure 5C). This contrasts with previous research that has focussed on natural strike pattern preferences or overland running, which often reported lower impact kinetics in midfoot and forefoot striker patterns [6,27,42]. It is possible that the instructions given to run flat footed produced a gait distinct from participants habitual midfoot style, potentially increasing vertical limb acceleration and thereby amplifying impact deceleration. This is supported where peak impact accelerations were greater in midfoot than rear or forefoot strike patterns [43]. Although not a primary focus, the small, non-significant decrease in swing time (Table 1) warrants further exploration from a kinematic perspective. Additionally, spatial analysis of the high-frequency component revealed similarity with peak impact (Figure 5A) and average loading rate (Figure 5B), but with notable differences in zone contributions. Specifically, in the heel and forefoot strikes where non-dominant zones contributed less (Figure 6B). This suggests that the low-frequency components may inflate non-specific zone data during impact.

Although this study primarily focused on comparing strike pattern with respect to impact kinetics and force signal characteristics, active peak force associated with propulsion was also measured. The results showed significantly lower active peak force in heel strikes compared to midfoot and forefoot strikes, in both total and spatial zone measures (Table 1 and Table 2). These findings contrast with previous work where there were no differences between total forces across strike patterns [42]. However, the data presented shows considerable non-anterior contributions to total force in both heel and midfoot strikes indicating differences in plantar surface mechanics.

While no spatiotemporal differences were observed between foot-strike patterns during treadmill running, and total force trace structure appeared similar when measured by in-shoe devices, distinct differences in kinetic variables between strike patterns were present, particularly when examined using frequency domain analysis. These distinct kinetic differences, combined with the ability to classify strike pattern based on spatial force metrics, could be used to discern intervention effects in treadmill running studies, where previous findings have been equivocal.