Automated plantar contact area estimation in a dynamic state using K-Means clustering

BACKGROUND

Estimation of plantar contact area (PCA) can be used for a variety of purposes such as classification of foot types and diagnosis of foot abnormalities. While some techniques have been developed for assessing static PCA, understanding dynamic PCA may improve understanding of gait biomechanics. This study aims (1) to develop an approach to estimate PCA from video images of footprints during walking and (2) to assess the accuracy and generalizability of this method.

METHODS

A sample of 41 ambulatory, young adults (age = 24.3 ± 3.2 years, mass = 67.2 ± 16.9 kg, height = 1.63 ± 0.08 m) completed 10 trials walking on a raised transparent plexiglass platform. Foot contact during walking was recorded using a video camera placed under the platform. An image processing algorithm, Clustering Segmentation, was developed based on identifying color intensity between the PCA and the rest of the foot and plantar contact morphology.

RESULTS

The proposed approach was compared to manual hand tracing, which is widely accepted as the Gold Standard, as well as with an earlier automated approach (Lidstone et al., 2019). Results showed that Clustering Segmentation followed the Gold Standard closely in all phases of gait. The maximum PCA and the maximum PCA length and width generally increased with foot size, indicating that the algorithm could successfully estimate the PCA across a wide range of foot sizes. Results also showed that the proposed approach for obtaining the PCA may be used to characterize various foot types in a dynamic state.

CONCLUSION

Clustering Segmentation algorithm eliminates the need for subjective interpretation of the PCA. The results showed that the algorithm was considerably faster and more accurate than the earlier automated method. The proposed algorithm will be appropriate for assessment of foot abnormalities and provides complementary information to gait analysis.

Electronic measurement of plantar contact area during walking using an adaptive thresholding method for Medilogic® pressure-measuring insoles

BACKGROUND

Pressure-measuring insoles have the potential to measure plantar contact area (PA) during walking. However, they are not widely used for this purpose because of the need for a reliable process that can convert the insole output into PA. The purposes of this study were to: (1) develop an adaptive-threshold method for pressure-measuring insoles that can improve the accuracy of the PA measurements during walking, and (2) experimentally assess the accuracy and generalizability of this method.

METHODS

A sample of 42 healthy, ambulatory, young adults (age=24.3±3.2years, mass=67.2±16.9kg, height=1.63±0.08m) completed 10 trials walking on an elevated walkway while wearing Medilogic^® pressure-measuring insoles (sizes 35-45). A total of six insole sizes were considered. Insole data were converted to PA using three unique adaptive-thresholds that were based on percentages of the maximum sum of digital values (MSDV) during an analyzed step. Three values were considered: 0.1%, 0.2%, and 0.3% of the MSDV. Additionally, a fixed-threshold, which is typically used to estimate PA, was assessed. These two techniques, applied to the insole worn on the left foot, were compared with PA obtained from high-resolution reference footprints obtained from optical pedography of the right foot and processed using digital image processing algorithms. An assumption of PA symmetry between the left (insole) and right (barefoot) feet was made and comparisons were conducted over the entire stance phase of walking. The generalizability of the algorithm was assessed by comparing PA errors from insoles with respect to the optical pedography results based on insole size criteria.

RESULTS

The adaptive-thresholds of 0.1%, 0.2%, and 0.3% of MSDV produced mean errors of 7.31±17.44%, -8.62±15.01%, and -20.45±14.18%, respectively. Using the 2-digital value fixed-threshold produced a mean error of 20.88±22.44%. The best performing adaptive-threshold varied among insole sizes.

CONCLUSION

It was observed that the fixed-threshold technique produced large magnitudes of errors. The proposed adaptive-thresholds of 0.1% and 0.2% of the MSDV reduced PA error to ±10% during walking. The adaptive-threshold method consistently reduced PA error vs. the fixed-threshold for each insole size.