Abnormal foot posture and mechanics have long been associated with lower limb injuries [
]. For instance, clinical measures of a pronated foot posture, such as a low arch [
] and excessive navicular drop [
], have been retrospectively identified with knee injuries and anterior cruciate ligament injury respectively. Similarly, measures indicating a more supinated foot posture, such as a high arch [
] and rearfoot varus [
] have been retrospectively associated with stress fractures and patellofemoral pain respectively. Consequently, the measurement and classification of foot posture in a clinical setting has become a central focus of lower extremity medicine, and now is widely used to evaluate injury risk and monitor treatment efficacy.
Despite the existence of many different techniques to evaluate foot posture in the clinical setting, there is still disagreement as to which method is the most clinically useful [
]. Indeed, some studies have found contrasting results regarding the association between abnormal foot type and injury depending on the clinical technique employed [
], with some researchers arguing that these conflicting findings may be at least partly due to the lack of reliability and validity of many of these measures [
]. Moreover, the inability of many of the static measures of foot posture to predict dynamic function also calls in to question their clinical utility [
To address these concerns, Redmond, Crosbie and Ouvrier [
] developed a subjective measure of static foot posture termed the Foot Posture Index (FPI). The FPI is comprised of one palpatory and five visual criteria used to determine whether the foot is in a supinated, neutral or pronated position [
]. Research has reported that the FPI possesses acceptable intra-rater reliability [
], and the tool has been validated against both static and dynamic three-dimensional (3D) lower limb models [
]. However, despite these advantages, the subjective nature and limited five-point Likert-type scoring scale of the FPI may limit the tool’s precision, with some researchers suggesting that the results need to be interpreted with caution and may actually have limited value, especially in a research setting [
]. Consequently, there remains a need for an inexpensive, portable and accurate assessment tool that can quantifiably assess static foot posture, which could be implemented in a clinical setting for everyday patient assessment.
The Microsoft Kinect™ is an inexpensive and portable video game accessory that combines a video and infrared-sensing camera to create a 3D model of the body. Recent research has shown that the Kinect system is capable of creating this 3D human model with similar accuracy to expensive and complex 3D body scanning systems [
]. Similarly, early work has also shown promising results for the Kinect to evaluate gait velocity [
], hand and elbow movements [
] and anatomical landmark displacement and trunk angle [
] when compared to 3D motion analysis systems. Combined, these studies demonstrate that the Kinect is able to obtain some kinematic and anatomical mapping data with a similar degree of accuracy to more expensive 3D motion analysis and scanning systems [
]. Consequently, the Kinect may have the potential to objectively evaluate static foot posture in a clinical setting with more accuracy than the subjectively based FPI. This in turn may permit better injury prediction accuracy, increased measurement reliability and improved clinical utility. Therefore, this study aimed to evaluate whether the Kinect is able to reliably and validly evaluate static foot posture, as measured using the FPI. A secondary aim was to validate the Kinect-derived data with assessment of static foot posture using a 3D motion analysis system. Lastly, a third aim was to examine whether Kinect measures of foot posture were able to predict the variance in the total visual FPI score.
This study was the first to examine the reliability and validity of the Microsoft Kinect to evaluate static foot posture. The Kinect demonstrated moderate to good reliability for four out of six items of the modified FPI (lateral malleolar curvature, talo-navicular joint bulging and medial arch height and peak). Additionally, the Kinect also displayed moderate to good concurrent validity for four items of the FPI when compared to the Vicon 3D analysis system (lateral malleolar curvature, talo-navicular joint bulging, medial arch height and forefoot abduction/adduction). However the relationship between Kinect and visual FPI items was found to be poor. Regression analysis revealed that the Kinect FPI items with moderate to good reliability were able to predict 61% of the variance in the total visual FPI score, with the only significant variable (talo-navicular joint bulging) also demonstrating a good to excellent relationship with the total visual FPI score. The advantage of the Kinect in comparison with the individual items of the visual FPI is that it provides quantified data on a continuous scale rather than on a limited ordinal scale. Although somewhat mixed, these results support the future potential use of the Kinect as a tool to assess static foot posture in a clinical setting.
The Kinect showed improved intra-rater reliability for individual items of the FPI when compared with the visual FPI observations. Specifically, individual FPI items recorded by the Kinect showed moderate to good reliability whereas visual FPI items demonstrated poor to moderate intra-rater reliability, which is consistent with previous research [
]. The total visual FPI score demonstrated good to excellent intra-rater reliability, which has also been reported previously [
]. The improved item reliability found with the Kinect could be attributed to the continuous data of the Kinect which, compared with the limited ordinal scale of the FPI, potentially provides improved accuracy in the evaluation of foot posture.
Similar to the finding of improved reliability of FPI items recorded using the Kinect, Vicon FPI items also demonstrated superior reliability when compared to the visual assessment of FPI. However, there was some variation in reliability levels between the systems. For instance, arch peak, which is part of the congruence of the medial longitudinal arch item of the FPI, demonstrated moderate to good intra-rater reliability for the Kinect system whilst poor reliability was shown for the Vicon system. In contrast, forefoot abduction/adduction demonstrated moderate to good reliability for the Vicon system whereas poor reliability was shown for the Kinect system. Furthermore, the item calcaneal inversion/eversion demonstrated poor intra-rater reliability for both systems.
The variable reliability results for the Kinect and Vicon systems may be partly due to the different data collection methods employed, with the Kinect system using depth data and the Vicon system using a wand to locate landmarks and trace over regions of the foot in 3D space. For instance, the precision along the longitudinal axis of the foot of the medial mesh techniques used in the Vicon analysis of the FPI, which involved performing sweeping movements over the medial surface of the foot using the wand tip, was poor relative to the depth data from the Kinect for measurement of the arch peak along the longitudinal axis of the foot. The distance between each sweep may have been too large (approximately 10 mm at the top of the arch), whilst the Kinect was able to measure the 3D position of the arch in each pixel with a longitudinal axis precision of approximately 3 mm. Furthermore, the wand tip may depress the soft tissue of the foot as the sweeps were performed along the medial arch. This may limit the ability to identify the position of the arch peak due to the varying compressions in the medial-lateral plane of the soft tissue, which would not affect the measure of arch height in the vertical plane. Instrumenting the wand and controlling force application during the assessment may have reduced this error, however during our pilot testing this proved difficult to control via feedback given that the forces applied through the wand tip were quite low (< 5 Newtons).
Additionally, the Kinect demonstrated poor reliability for the FPI item of forefoot abduction/adduction, which may be attributed to errors in visual anatomical landmark identification. The reliability for this item may be improved by identifying the furthest point of the rearfoot to the Kinect compared to the closest point of the forefoot to the Kinect. Finally, calcaneal inversion/eversion also demonstrated poor reliability for both the Kinect and Vicon systems. As discussed for the Vicon assessment of arch peak, the precision of the mesh technique for evaluation of calcaneal inversion/eversion may have been affected by the distance between each sweep and the varying compression of the soft tissue. Similarly, the poor reliability found for the Kinect assessment of calcaneal inversion/eversion may have been due to errors in visual estimation of calcaneal position, as suggested previously for forefoot abduction/adduction. Indeed, difficulties in visually estimating rearfoot position have been highlighted previously [
], and given the present study visually assessed a depth image rather than the actual foot, it is likely that this technique may have led to further errors in the evaluation of this item.
Validity analysis revealed that all individual items of the FPI derived from visual observations were poorly correlated with items from both the Kinect and 3D motion analysis system. This inability to correlate the visual FPI with other measures of foot posture is consistent with Scharfbillig et al. [
], who reported that four items of the FPI were poorly correlated with radiographic measures, which they partly attributed to a lack of agreement between bony architecture and the overlying skin. Given each item of the visual FPI has only five possible scores, this limited spread of data will reduce the likelihood of finding strong relationships when compared with a continuous set of data such as depth or radiological-based measures. Further, the reduced reliability of the individual visual FPI items when compared to both the Kinect and Vicon FPI items may limit the appropriateness of the visual inspection of foot posture as a concurrent validity measure. In contrast, mostly moderate to good correlations were found between the Kinect and Vicon, which is likely due to the two analysis systems using continuous data and having the same outcome measures. Bland-Altman plots revealed poor absolute agreement between the devices, although this may be explained by the different scales used by the Kinect and Vicon.
The Kinect items with moderate to excellent reliability were shown to predict 61% of the variance in the total visual FPI score. Similarly, the Vicon items with moderate to excellent reliability predicted 58% of the variance in the total FPI score derived from visual observations. In both cases, talo-navicular joint bulging was the only item entered into the regression model to independently demonstrate significance (
< 0.05). Although no previous study has investigated the ability of objective 3D measures of foot posture or an individual analysis of each of the FPI items to predict total FPI score, Redmond et al. [
] reported that total FPI score was able to predict 64% of the variance in a measure of ankle joint position in 3D space. The validation of the Kinect in explaining a high proportion of the variance in total visual FPI scores suggests the potential feasibility of the Kinect and custom analysis software to be further refined to classify overall foot posture. Indeed, given the greater reliability of the individual items of the FPI derived from the Kinect, and the greater similarity in Kinect FPI items to FPI items from the 3D analysis system, future studies may attempt to develop a total Kinect FPI-type score that could be used to classify foot posture with greater accuracy and reliability compared to the FPI derived from visual observations.
Interestingly, one item of the FPI, talo-navicular joint bulging, demonstrated good to excellent correlations with the total visual FPI score when measured by the Kinect. This may indicate the potential of this particular item of the FPI derived from the Kinect to be used as a stand-alone measure to classify foot posture. Similarly, other research has reported that clinical measures of the midfoot strongly correlated with radiographic measures of foot posture [
]. Future research should further examine the agreement between the measure of talo-navicular joint bulging derived from the Kinect and total FPI score.
A limitation of the current study is the lack of a total score for the Kinect FPI items. Given the different scales used within each FPI item for the Kinect, this made the generation of a total FPI-type score and foot posture classification problematic. Future research may wish to first implement techniques, such as multiple regressions, to derive a total FPI-type score and foot classification from the Kinect items and secondly to collect a comprehensive data set of the full range of foot posture using the Kinect, from highly supinated to highly pronated. Additionally, future research should examine the inter-rater reliability of the Kinect to evaluate foot posture given previous research has shown poor inter-rater reliability of the FPI based on visual observation [
]. If inter-rater reliability was found to be superior, this may further the potential use of the Kinect as a tool to assess static foot posture in a clinical setting. Although mostly moderate to excellent correlations were found between the Kinect and the 3D motion analysis system, another potential limitation may have been the use of the Vicon system as a benchmark reference. This is supported by the poor test re-test reliability of the Vicon data in evaluating two FPI items, and may be partially explained by the limited accuracy from the distance between sweeps along the longitudinal axis of the foot and from soft tissue deformation as explained previously. In contrast, previous research has shown that a 3D foot scanner is able to reliably and accurately provide a 3D digital representation of the foot [
]. Although such a tool may provide a more appropriate benchmark reference with which to compare the Kinect, the Vicon system was used in this study due to the need of further research to develop standardised 3D foot scanning protocols for evaluating foot posture and morphology, and the limited availability of such systems. Future studies may wish to examine the concurrent validity of the Kinect to evaluate static foot posture when compared to 3D foot scanners. Furthermore, the generalizability of the study may be comprised due to the male only participants and the use of a novice rater for assessment. To attain generalised results, future research is required using many raters with larger participant numbers.
This study was funded in part by ASICS Oceania. All authors have received funding from ASICS Oceania either directly or indirectly via research grants or employment. Author RC designed the software and may at some stage release it either for free or at a cost.
Authors BM, KP and RC were involved in all aspects of the study. Author AM was involved in the pilot testing, data collection and drafting stage of the study. Authors SB and AB were involved in the preliminary design and drafting stages of the study. All authors read and approved the final manuscript.