Clustering and classification of regional peak plantar pressures of diabetic feet
Introduction
It has been estimated that the number of people worldwide with diabetes will surpass 365 million by 2030 (Wild et al., 2004). Foot ulceration in people with diabetes will continue to be a major public-health concern, considering the 15–25% lifetime risk of developing a foot ulcer (Abbott et al., 2005, Singh et al., 2005). At least 15% of these ulcers will lead to some form of foot amputation (http://www.diabetes.org). The burden of diabetic foot problems arises not only from the medical and economic effects, but also from the impact on the patient's quality of life (Jaksa and Mahoney, 2010).
Plantar pressures, among other clinical measurements, have been used to assess foot conditions in patients with diabetes (Singh et al., 2005). Elevated pressures are believed to increase the risk of ulceration in the diabetic foot, particularly when combined with deformity and peripheral neuropathy (Lavery et al., 2003). Footwear design to relieve elevated plantar pressure is an ongoing research area (Mueller, 1999, Cavanagh and Owings, 2006) although the evidence base for the effectiveness of various offloading techniques remains limited (Cavanagh and Bus, 2010).
Given the wide variety of foot types (Ledoux et al., 2003) and foot biomechanics, a single footwear design cannot successfully decrease peak plantar pressures for all distributions. Ideally, a customized footwear design for each patient would be preferable, but such an approach is not always feasible. A practical solution that accommodates patient-specificity would be to identify groups of patients with similar peak pressure distributions and to establish footwear design guidelines that reduce high regional pressures for each group. New patients could then be classified into one of these groups and the group-specific footwear solution could be prescribed with the expectation that it would reduce corresponding elevated plantar pressures. Waldecker (2012) successfully used logistic regression with plantar pressure, force and pressure–time integral to predict the risk of ulceration although the validity of the pressure–time calculation in this and other studies has been challenged (Waaijman and Bus, 2012, Melai et al., 2011). Acharya et al., 2008, Acharya et al., 2011 and See et al. (2010) used principal component analysis and artificial neural networks to classify patients as normal or diabetic, with and without neuropathy. The k-means clustering algorithm was chosen in the current study for its relative simplicity in comparison to these classification methods and its ability to independently classify data for different numbers of clusters, unlike hierarchical clustering algorithms (Vardaxis et al., 1998). Our goal was to use k-means cluster analysis to objectively and systematically determine characteristic regional peak plantar pressure distributions for patient classification that may be useful for footwear prescription.
Section snippets
Methods
The Diabetic Foot Clinic at the Cleveland Clinic has an established database of barefoot plantar pressures collected from patients diagnosed with diabetes. Measurements were collected using an EMED X pressure platform (Novel Inc., St. Paul, MN) on 819 different feet at a sampling frequency of 100 Hz. The subject group included 223 males and 215 females (mean age 59.5±s.d. 12.6 years). Both feet were included in the analysis if they were available. Data collection and analysis protocols were
Results
The accuracy of clustering and classification between the two independent subsets of plantar pressure data decreased as the total number of clusters increased (Fig. 2). For two clusters (k=2), the success rate was approximately 93%, which decreased to approximately 63% for ten clusters (k=10). The variability of the success rate (due to initial random assignment of data points to clusters) also changed as indicated by an increase in the standard deviation from 2.8% to 7.5%.
K-means analysis
Discussion
K-means cluster analysis provided an objective differentiation of regional peak plantar pressure data, identifying not only major foot groupings, but also revealing several plantar pressure distributions that may require special footwear interventions. Although subjective classifications can and are made clinically, predominant plantar pressure distributions observed within a population can only be reliably identified using an objective and quantitative method, such as the k-means algorithm
Conflict of Interest Statement
Ahmet Erdemir owns and operates innodof, LLC, a computational modeling and simulation company. Peter R. Cavanagh holds equity in DIApedia, LLC, and is a consultant for Langer, UK.
Acknowledgments
This study was supported by the National Institutes of Health (Grant # 5R01 HD037433). The authors would like to acknowledge efforts by Chris Borish for masking foot regions and the members of the Diabetic Foot Care Program at the Cleveland Clinic who assisted with the data collection.
References (33)
- et al.
A comparison of the 1-step, 2-step, and 3-step protocols for obtaining barefoot plantar pressure data in the diabetic neuropathic foot
Clinical Biomechechanics (Bristol, Avon)
(2005) - et al.
Off-loading the diabetic foot for ulcer prevention and healing
Journal of Vascular Surgery: Official Publication, the Society for Vascular Surgery [and] International Society for Cardiovascular Surgery
(2010) - et al.
Nonsurgical strategies for healing and preventing recurrence of diabetic foot ulcers
Foot and Ankle Clinics
(2006) - et al.
A functional foot type classification with cluster analysis based on plantar pressure distribution during jogging
Gait & Posture
(2006) - et al.
Cluster analysis to classify gait alterations in rheumatoid arthritis using peak pressure curves
Gait & Posture
(2009) - et al.
Calculation of plantar pressure time integral, an alternative approach
Gait & Posture
(2011) - et al.
Regional peak plantar pressures are highly sensitive to region boundary definitions
Journal of Biomechanics
(2008) - et al.
Pedobarographic statistical parametric mapping (pSPM): a pixel-level approach to foot pressure image analysis
Journal of Biomechanics
(2008) - et al.
Normal pressure values and repeatability of the Emed ST4 system
Gait & Posture
(2008) - et al.
Pedography and radiographic imaging for the detection of foot deformities in rheumatoid arthritis
Clinical Biomechanics (Bristol, Avon)
(2008)
Classification of able-bodied gait using 3-D muscle powers
Human Movement Science
The interdependency of peak pressure and pressure-time integral in pressure studies on diabetic footwear: no need to report both parameters
Gait & Posture
Pedographic classification and ulcer detection in the diabetic foot
Foot and Ankle Surgery: Official Journal of the European Society of Foot and Ankle Surgeons
Gait pattern classification of healthy elderly men based on biomechanical data
Archives of Physical Medicine and Rehabilitation
Abdominal and erector spinae muscle activity during gait: the use of cluster analysis to identify patterns of activity
Clinical Biomechanics (Bristol, Avon)
Foot ulcer risk is lower in South-Asian and African-Caribbean compared with European diabetic patients in the U.K.: the North-West diabetes foot care study
Diabetes Care
Cited by (57)
Clustering classification of diabetic walking abnormalities: a new approach taking into account intralimb coordination patterns
2020, Gait and PostureCitation Excerpt :In the last 40 years, several studies have been conducted to assess gait biomechanics alterations in diabetic neuropathic subjects [1–4].
Gait changes in persons with diabetes: Early risk marker for diabetic foot ulcer
2020, Foot and Ankle SurgeryA state-of-the-art review of foot pressure
2020, Foot and Ankle SurgeryCitation Excerpt :In foot pressure analysis, data mining was used to characterize and group the patterns of recorded data. For example, Bennetts et al. [62] used k-means clustering analysis to identify potential risks of foot ulceration among diabetic patients. The k-means clustering was applied for determining the peak plantar pressure distribution in patient classification.
The diabacare cloud: predicting diabetes using machine learning
2024, Acta Scientiarum - Technology