Elsevier

Journal of Biomechanics

Volume 41, Issue 12, 28 August 2008, Pages 2772-2775
Journal of Biomechanics

Short communication
Regional peak plantar pressures are highly sensitive to region boundary definitions

https://doi.org/10.1016/j.jbiomech.2008.06.029Get rights and content

Abstract

Traditional pedobarographic analyses subsample pressure data over a number of discrete anatomical regions of interest (ROIs). To our knowledge, the sensitivity of these data to ROI boundary definitions has not been previously addressed. Eight subjects each performed 20 trials of self-paced walking; commercial software was used to define 10 ROIs for each of the 160 total peak pressure images, and regional peak pressures (RPPs) were extracted for each image (total: 1600 values). We then asked three specific questions regarding RPP sensitivity to ROI boundary definition: (1) Is the ROI centroid representative of the RPP location? (2) How frequently do RPPs lie on the ROI boundary? and (3) By how much do RPP values change if the ROI boundary is changed by one pixel (resolution: 5.08×7.62 mm)? We found that the RPP locations differed from the ROI centroid in 80% of the cases and that the RPPs lay on the ROI boundary with a probability of 65%. We also found that a single-pixel change in the ROI boundary caused a mean RPP change of 10.8%. The most sensitive region was the midfoot for which a single-pixel ROI change yielded a median 29.4% change in RPP. These results indicate that RPP data are biased by regionalization schemes, which delineate pressure fields based on anatomy rather than on the field's geometric properties, and ultimately that regionalization may constitute a poor method of quantifying complex pressure fields. RPP sensitivity should be considered when making statistical inferences regarding foot function.

Introduction

Pedobarographic (foot pressure) images produced by modern commercial systems are sampled at frequencies of approximately 500 Hz over thousands of sensors (Alexander et al., 1990). Data reduction typically involves regional peak pressure (RPP) extraction from on the order 10 anatomical regions of interest (ROIs) (Rosenbaum and Becker, 1997). RPP statistical results are implicitly assumed to be representative of the anatomical regions from which they are extracted.

But how representative are such data? How sensitive are they to ROI boundary definitions? To our knowledge, these issues have not been previously addressed. We believe that this is an important gap in the literature for four reasons: (i) RPP is a max function and is hence highly sensitive to outliers, (ii) independent sensors (Alexander et al., 1990) are each prone to noise, (iii) ROIs are specified by pixel masks for which adjacent pixels can arbitrarily represent different anatomical regions, and (iv) commercial software often allows manual ROI boundary definition/adjustment. These factors and others make the RPP measure less reliable than the mean (Martinez-Nova et al., 2007). Reduced reliability, if caused by small subjective or algorithmic ROI changes, could at best reduce statistical confidence and at worst alter statistical inferences regarding foot function.

The purpose of the current study was to assess RPP sensitivity to ROI boundary definitions. To this end we employed a commercially available ROI masking protocol and formulated three specific questions regarding the relation between RPP and the ROI geometry: (1) Is the ROI centroid representative of RPP location? (2) How frequently does the RPP lay on the ROI boundary? and (3) By how much does the RPP value change if the ROI boundary definition changes by a single pixel?

Section snippets

Design

Eight male subjects (age: 30.3±8.9 years; mean±SD) each performed 20 self-paced walking trials over a 10 m gait runway. Right foot pedobarographic records were collected at 500 Hz using a 0.5 m Footscan 3D system (RSscan, Olen, Belgium) and a Kistler force plate (model 8281B, Winterthur, Switzerland). Prior to participation subjects gave informed consent according to the policies of the Research Ethics Committee of the University of Liverpool.

RPP and ROI centroid locations

Ten ROIs were defined automatically for each trial

RPP and ROI centroid locations

The RPP data exhibited spatial clustering, often far away from the regional centroids and frequently on or near the regional boundaries (Fig. 1B and C). The mean RPP location was farthest from the ROI centroid (∼20 mm) in the T2-5, MF, and HL regions (Table 1A). Hotelling's T2 tests found that the mean RPP location was significantly different from the ROI centroid in 64 (80%) of the 80 comparisons (eight subjects, 10 regions).

RPP and ROI boundary coincidence

There was a significantly greater than zero chance that the RPP lay on

Discussion

The main findings were that (1) RPP locations were significantly different from ROI centroids, (2) the majority (65%) of all RPPs lay on the ROI boundary, and (3) RPP values were highly sensitive to single-pixel changes in ROI boundaries. The magnitude of these effects was surprisingly high.

Single-pixel ROI changes induced an average |ΔRPP| of 10.8%, a value greater than the effect magnitudes (<10%) reported in a variety of recent studies (Taylor et al., 2004; Richter et al., 2006; Vincenzo et

Conflict of interest

The authors report no conflicts of interest.

Acknowledgements

Financial supported was provided by the Leverhulme Trust (Grant F/0025/x) and NERC (Grants GR3/11202 and GR3/12004).

References (13)

There are more references available in the full text version of this article.

Cited by (0)

View full text