Quantitative and Qualitative Running Gait Analysis through an Innovative Video-Based Approach

uantitative and qualitative running gait analysis allows the early identi cation and the longitudinal monitoring of gait abnormalities linked to running-related injuries. A promising calibration- and marker-less video sensor-based technology (i.e.,Graal), recently validated for walking gait, may also offer a time- and cost-ef cient alternative to the gold-standard methods for running. This study aim was to ascertain the validity of an improved version of Graal for quantitative and qualitative analysis of running. In 33 healthy recreational runners (mean age 41 years), treadmill running at self-selected submaximal speed was simultaneously evaluated by a validated photosensor system (i.e.,Optogait—the reference methodology) and by the video analysis of a posterior 30-fps video of the runner through the optimized version of Graal. Graal is video analysis software that provides a spectral analysis of the brightness over time for each pixel of the video, in order to identify its frequency contents. The two main frequencies of variation of the pixel's brightness (i.e., F1 and F2) correspond to the two most important frequencies of gait (i.e., stride frequency and cadence). The Optogait system recorded step length,

video analysis software that provides a spectral analysis of the brightness over time for each pixel of the video, in order to identify its frequency contents. The two main frequencies of variation of the pixel's brightness (i.e., F1 and F2) correspond to the two most important frequencies of gait (i.e., stride frequency and cadence). The Optogait system recorded step length, cadence, and its variability (vCAD, a traditional index of gait quality). Graal provided a direct measurement of F2 (re ecting cadence), an indirect measure of step length, and two indexes of global gait quality (harmony and synchrony index). The correspondence between quantitative indexes (Cadence vs. F2 and step length vs. Graal step length) was tested via paired t-test, correlations, and Bland–Altman plots. The relationship between qualitative indexes (vCAD vs. Harmony and Synchrony Index) was investigated by correlation analysis. Cadence and step length were, respectively, not signi cantly different from and highly correlated with F2 (1.41 Hz 0.09 Hz vs. 1.42 Hz 0.08 Hz,p= 0.25, r 2 = 0.81) and Graal step length (104.70 cm 013.27 cm vs. 107.56 cm 13.67 cm,p= 0.55,r 2 = 0.98). Bland–Altman tests con rmed a non-signi cant bias and small imprecision between methods for both parameters. The vCAD was 1.84% 0.66%, and it was signi cantly correlated with neither the Harmony nor the Synchrony Index (0.21 0.03,p= 0.92, r 2 = 0.00038; 0.21 0.96,p= 0.87, r 2 = 0.00122). These ndings con rm the validity of the optimized version of Graal for the measurement of quantitative indexes of gait. Hence, Graal constitutes an extremely time- and cost-ef cient tool suitable for quantitative analysis of running. However, its validity for qualitative running gait analysis remains inconclusive and will require further evaluation in a wider range of absolute and relative running intensities in different individuals. Keywords: gait analysis; treadmill running; video-based systems; harmony; fast Fourier transform; video sensors 1. Introduction Running is one of the most popular recreational physical activities in the world, as it provides substantial health bene ts at minimal expense [1,2]. However, together with Sensors2021,21, 2977.

absolute and relative running intensities in different individuals. Keywords: gait analysis; treadmill running; video-based systems; harmony; fast Fourier transform; video sensors 1. Introduction Running is one of the most popular recreational physical activities in the world, as it provides substantial health bene ts at minimal expense [1,2]. However, together with Sensors2021,21, 2977.

Sensors2021,21, 2977 2 of 12 numerous health bene ts, there is a relatively high incidence of running-related injuries (6.8 to 59 injuries per 1000 h of running) [3,4]. The etiology of running-related injuries is multifactorial and still not fully understood [5]. Quantitative (e.g., stride length, cadence) and qualitative (e.g., gait variability, gait harmony) biomechanical features are modi able risk factors that play a relevant role in the occurrence of injuries, particularly in recreational and novice runners [6,7]. Quantitative gait analysis measures the spatiotemporal parame- ters, kinematics, and kinetics of running. Spatiotemporal parameters (e.g., cadence, stride length, and contact and ight times) describe the basic features of gait pattern and are useful to easily de ning the ability of the individual to ful l the general requirements of running (e.g., symmetry, coordination, and gait economy) [8–10]. Alteration in these parameters, such as the presence of overstride (i.e., an excessive stride length associated with a decreased cadence) or an excessive cadence variability, contribute to increased risk of injuries and to decreased running economy. Moreover, the longitudinal monitoring of these parameters allows the follow-up of functional outcomes after rehabilitation treat- ments [11–13]. Qualitative gait analysis identi es speci c running features (e.g., foot strike pattern, presence of cross-over, and low gait harmony) that are associated with increased risk of overuse injuries [13] or global movement scores (e.g., Volodalen Scale) indicative of running economy and inversely related to the risk of overuse injuries [14]. The quantitative and qualitative analysis of running biomechanics can be performed in either a laboratory or an outdoor setting, depending on the aim and on the available equipment. Running gait analysis in the laboratory setting has the main advantages of fully controlled environmental conditions and the use of gold-standard methods (i.e., op- tical motion capture systems associated with force platforms). Limitations of the use of gold-standard methods are a not fully ecological walking or running style, due to the use of a treadmill, and the high costs and complexity in terms of experimental setup[15–18] . In recent years, more low-cost, easy-to-use, and marker-less alternatives to gold-standard methods have been developed

(i.e., op- tical motion capture systems associated with force platforms). Limitations of the use of gold-standard methods are a not fully ecological walking or running style, due to the use of a treadmill, and the high costs and complexity in terms of experimental setup[15–18] . In recent years, more low-cost, easy-to-use, and marker-less alternatives to gold-standard methods have been developed for large-scale gait evaluation [6,8,13,14]. The most fre- quently used alternative technologies for quantitative gait analysis are optical timing systems and inertial sensors. Despite the several advantages compared to gold-standard methods, these systems require setup/subject preparation, calibration, and data post- processing procedures that reduce the ease of use. In recent years, no-calibration and marker-less video analysis methods have been developed, but their accuracy and precision remain to be assessed by validation studies [19–21]. Qualitative analysis of running is traditionally performed by clinicians or expert coaches using evaluation scales during direct observation of gait or of ine video analy- sis [13,14,22]. Objective methods have been proposed to overcome the limitations of the above subjective approach, the most frequently used alternative being inertial sensors. Through the study of the biomechanics of separate body segments that represent the entire body and/or through the evaluation of the interaction, coordination, and symmetry of mul- tiple body segments, these techniques permit a simple evaluation of gait quality [6,23–27]. Traditionally, running gait harmony is evaluated by considering the variability of spa- tiotemporal gait parameters, left vs. right asymmetries, and the step-by-step rhythmicity of acceleration patterns of the center of mass (i.e., the harmonic ratio) [6,28–31]. The above cited traditional and alternative methods for the quantitative and qualitative evaluation of gait all have in common a focus on speci c body segments (e.g., center of mass, lower limbs, or lower and upper trunk) that are considered representative of the motion of the entire body [5,28,32–34]. However, taking into account that all forms of gait are characterized by a simultaneous movement of all the lower and upper body segments, recent studies have shown that a global approach to movement characterization can provide a synthetic view of the harmony and

or lower and upper trunk) that are considered representative of the motion of the entire body [5,28,32–34]. However, taking into account that all forms of gait are characterized by a simultaneous movement of all the lower and upper body segments, recent studies have shown that a global approach to movement characterization can provide a synthetic view of the harmony and quality of gait [19,32,35] while considerably reducing the cost and time of gait analysis. Low-cost and calibration-/marker-less technologies would allow gait evaluation on a large scale in both clinical and sport sciences towards early identi cation of injury risk, injury prevention, and monitoring of recovery for a swift and safe return to sport. In the context of sport coaching, the same technologies

Sensors2021,21, 2977 3 of 12 would provide indexes that are objective, accurate, and repeatable to describe and monitor running technique for performance enhancement. However, a methodology providing quantitative and qualitative indexes of running gait with a global approach is still missing. With the aim to provide a tool for large-scale gait analysis, we recently proposed an innovative, video-based method that relies on the analysis of the frequency content of the variations of pixel brightness in digital video of a cyclic movement [35]. This method, validated for the evaluation of physiological walking at self-selected speed, is able to extract quantitative parameters such as the main frequency contents of rhythmic movement (i.e., cadence and step frequency), along with a qualitative index of the harmony of gait [35]. While the above ndings are very promising, some technical optimizations and additional con rmatory studies are needed before the technique can be extensively applied to the study of human movement. One of the main limitations of the method was the complexity of the algorithm used to extract the main frequencies of interest and, thus, the time needed for the computation. Therefore, in this study we optimized the calculations by reducing the signal noise and by averaging the frequency content of the video into a single spectrum. Furthermore, in search of a possibly more informative index of gait quality, we explored the performance of an alternative index of dispersion/variability of the average spectrum. This study tested the hypothesis that the optimized version of this innovative video- based method for gait analysis is able to accurately extract quantitative parameters and indexes of gait quality. The method's performance in terms of the quantitative and qualita- tive evaluation of gait was tested on running, a form of locomotion that is characterized by a higher frequency and complexity compared to walking [32]. 2. Materials and Methods In the present work, running data from 33 healthy individuals (21 males, 12 females) were acquired and studied. Participants signed a written informed consent, and the study was approved by the ethical committee of the IRCCS Fondazione Don Carlo Gnocchi ONLUS (number 13663_oss). Runners

is characterized by a higher frequency and complexity compared to walking [32]. 2. Materials and Methods In the present work, running data from 33 healthy individuals (21 males, 12 females) were acquired and studied. Participants signed a written informed consent, and the study was approved by the ethical committee of the IRCCS Fondazione Don Carlo Gnocchi ONLUS (number 13663_oss). Runners had to be engaged in a running program (at least two sessions per week with a minimum continuous running time of 20 min per session) for a minimum of six months, be free from injuries in the last two months, be free from chronic musculoskeletal diseases, and be adapted in the use of a treadmill [36]. Subjects were requested to run on a at motorized treadmill (MTC-Climb, Runner, Italy) following a previously developed protocol [37]. In brief, subjects started with ve minutes of a warm-up and familiarization phase: the treadmill velocity was initially set at 7 km/h and was self-modulated until a subjectively comfortable speed was reached [38,39]. The subjectively comfortable speed was chosen to minimize the fatigue factor, which has been shown to affect running mechanics [36]. All participants wore their habitual running shoes. Following warm-up/familiarization, after a 1-minute recovery, subjects ran at the predetermined self-selected speed for 2 min. The analysis of running parameters was performed on the nal 60” of the running test at self-selected speed, in parallel, through a validated photosensor system (Optogait) [40] that was used as a reference method and through custom-designed, Web-based video analysis software (Graal) [35]. Both Optogait and Graal are gait analysis systems developed by the company Microgate (Bolzano, Italy). Among the parameters measured directly by the Optogait system, we extracted step length (cm) and step length normalized for stature, contact time (s), ight time (s), cadence (in Hz), and within-subject cadence variability across the entire test (vCAD, %) [7]. The latter was taken as an index of gait harmony [27,30,41,42]. Concurrently, using a Logitech Brio 4 K (2.5 m behind the subject at 95 cm height), we recorded videos at 30 Hz and with a resolution of 640 480 pixels

time (s), ight time (s), cadence (in Hz), and within-subject cadence variability across the entire test (vCAD, %) [7]. The latter was taken as an index of gait harmony [27,30,41,42]. Concurrently, using a Logitech Brio 4 K (2.5 m behind the subject at 95 cm height), we recorded videos at 30 Hz and with a resolution of 640 480 pixels that were successively analyzed by Graal software. The original method of analysis, optimized in the present work, was previously reported in detail [35]. In brief, the following steps were performed via MATLAB-based software (The MathWorks, USA, version R2018) in order to identify the main frequency content of gait (i.e., F2, in Hz, representative of cadence), to derive the step length (Graal step length, in cm), and to calculate an index of global gait quality

Sensors2021,21, 2977 4 of 12 (i.e., Harmony Index). In the present study an additional index (i.e.,Synchrony Index) was developed and proposed to describe the global gait quality. Both the Harmony and Synchrony Indexes are in arbitrary units [35]. The analysis starts with the subdivision of the video le into frames, allowing us to obtain a number (n) of individual JPEG images. For each pixel of each image, the software computes the brightness via a brightness edit algorithm [35,43], followed by the application of a speci c correction method called the Hanning window and a high-pass Butterworth lter with order 2 at 0.5 Hz to attenuate the component due to direct current [44]. After this preprocessing, the fast Fourier transform (FFT) algorithm used in Simoni et al. (2020) [35] is applied to the vector of brightness values over time for each pixel to obtain its magnitude in the frequency domain. The result is a power spectrum for each pixel that displays the magnitude as a function of frequency and permits us to identify the peaks of the dominant frequencies [35]. To ameliorate the process of peak frequency detection presented in Simoni et al. (2020), power spectra for each pixel are then averaged in order to obtain the Averaged Power Spectrum of the video, and the signal is then further processed using the detrend MATLAB function in order to attenuate the effect of ickering or pink noise [44] (Figure).Sensors 2021, 21, 2977 FOR PEER REVIEW 4 of 13 Concurrently, using a Logitech Brio 4 K (2.5 m behind the subject at 95 cm height), we recorded videos at 30 Hz and with a resolution of 640 × 480 pixels that were succes- sively analyzed by Graal software. The original method of analysis, optimized in the pre- sent work, was previously reported in detail [35]. In brief, the following steps were per- formed via MATLAB-based software (The MathWorks, USA, version R2018) in order to identify the main frequency content of gait (i.e., F2, in Hz, representative of cadence), to derive the step length (Graal step length, in cm), and to calculate an index

optimized in the pre- sent work, was previously reported in detail [35]. In brief, the following steps were per- formed via MATLAB-based software (The MathWorks, USA, version R2018) in order to identify the main frequency content of gait (i.e., F2, in Hz, representative of cadence), to derive the step length (Graal step length, in cm), and to calculate an index of global gait quality (i.e., Harmony Index). In the present study an additional index (i.e., Synchrony Index) was developed and proposed to describe the global gait quality. Both the Harmony and Synchrony Indexes are in arbitrary units [35]. The analysis starts with the subdivision of the video file into frames, allowing us to obtain a number (n) of individual JPEG images. For each pixel of each image, the software computes the brightness via a brightness edit algorithm [35,43], followed by the application of a specific correction method called the Hanning window and a high-pass Butterworth filter with order 2 at 0.5 Hz to attenuate the component due to direct current [44]. After this preprocessing, the fast Fourier trans- form (FFT) algorithm used in Simoni et al. (2020) [35] is applied to the vector of brightness values over time for each pixel to obtain its magnitude in the frequency domain. The result is a power spectrum for each pixel that displays the magnitude as a function of frequency and permits us to identify the peaks of the dominant frequencies [35]. To ameliorate the process of peak frequency detection presented in Simoni et al. (2020), power spectra for each pixel are then averaged in order to obtain the Averaged Power Spectrum of the video, and the signal is then further processed using the detrend MATLAB function in order to attenuate the effect of flickering or pink noise [44] (Figure 1). Figure 1. The Averaged Power Spectrum of a representative subject (black curve). The blue dashed line shows the signal after the application of the detrend Matlab function to remove the effect of the flickering noise. The two largest peaks correspond to stride frequency (F1) and cadence (F2). A simple peak detection is

of flickering or pink noise [44] (Figure 1). Figure 1. The Averaged Power Spectrum of a representative subject (black curve). The blue dashed line shows the signal after the application of the detrend Matlab function to remove the effect of the flickering noise. The two largest peaks correspond to stride frequency (F1) and cadence (F2). A simple peak detection is thus performed to extract the peaks with the highest power between 0 and 15 Hz, as indicated in Simoni et al. (2020) [35]. The frequencies of the first two largest peaks are named F1 and F2 and have been demonstrated to corre- spond, respectively, to stride frequency and cadence, expressed in Hz units, for walking gait [35]. Assuming that F2 corresponds to step frequency or cadence, the Graal system indi- rectly estimates the spatial parameter step length as follows [35,45,46]: Figure 1. The Averaged Power Spectrum of a representative subject (black curve). The blue dashed line shows the signal after the application of the detrend Matlab function to remove the effect of the ickering noise. The two largest peaks correspond to stride frequency (F1) and cadence (F2). A simple peak detection is thus performed to extract the peaks with the highest power between 0 and 15 Hz, as indicated in Simoni et al. (2020) [35]. The frequencies of the rst two largest peaks are named F1 and F2 and have been demonstrated to correspond, respectively, to stride frequency and cadence, expressed in Hz units, for walking gait [35]. Assuming that F2 corresponds to step frequency or cadence, the Graal system indi- rectly estimates the spatial parameter step length as follows [35,45,46]: Graal Step Length[cm]=speed h m s 1 i F2 1 h s 1 i 100 (1) F2 is expressed in Hz, but in Equation (1),Hzhave been represented ass 1to facilitate readability. Moreover, in search of a more informative descriptor of global gait harmony, we computed the Synchrony Index, intended to restitute the degree of synchrony in the change

This study evaluates an innovative video-based method for analyzing running gait.