Using Flexible-Printed Piezoelectric Sensor Arrays to Measure Plantar Pressure during Walking for Sarcopenia Screening

Shulang Han, Qing Xiao, Ying Liang, Yu Chen, Fei Yan, Hui Chen, Jirong Yue, Xiaobao Tian, Yan Xiong2024 · DOI

TL;DR

This paper introduces a novel wearable system using flexible-printed piezoelectric sensor arrays to measure dynamic plantar pressure during walking. This system, combined with machine learning, achieves high accuracy in screening for sarcopenia, offering a more convenient and cost-effective alternative to traditional methods.

What Problem Does This Paper Address?

Sarcopenia, a progressive loss of muscle mass and function, is a significant age-related syndrome impacting independence and quality of life. Early detection is crucial for intervention, but current screening methods like DXA, CT scans, and grip strength tests are often costly, complex, and require specialized equipment or expertise, making them impractical for widespread community screening. Plantar pressure, reflecting gait and balance, has emerged as a potential indicator for sarcopenia, but existing measurement tools are often bulky, not designed for dynamic, real-world use, or require controlled lab environments. This study aims to bridge this gap by developing a wearable, flexible, and convenient system for dynamic plantar pressure monitoring to facilitate early sarcopenia screening.

Methods

The core of this work is a wearable flexible-printed piezoelectric sensor array fabricated using barium titanate (BaTiO3) thin films. These films, known for their piezoelectric properties, are integrated onto a flexible PET substrate using a flexible printer. This printing technique enhances the array's compressive strength and measurement range, making it suitable for the dynamic pressures experienced during walking.

The sensor array is integrated into an insole and connected to a signal processing circuit that converts the piezoelectric sensor's charge signals into voltage signals. This processed data is then transmitted wirelessly via Bluetooth to a mobile phone for real-time monitoring and analysis.

For sarcopenia screening, the study extracts various dynamic plantar pressure features from the walking data, including peak pressures (Pmax), average pressures (Pmean), and pressure standard deviations (Pstd) for five sensor locations on the forefoot. Crucially, they also consider weight-calibrated versions of these features (CPmax, CPmean, CPstd) to account for individual weight differences. These features, along with basic physiological data (age, height, weight, BMI), are then fed into five machine learning models (Decision Tree, Logistic Regression, Naive Bayes, Support Vector Machine, and K-Nearest Neighbors) for classification. To address the slight class imbalance in the dataset, the Synthetic Minority Oversampling Technique (SMOTE) is employed. Model performance is evaluated using standard metrics like accuracy, precision, recall, specificity, F1 score, and AUC. Feature selection using the ReliefF algorithm is also performed to identify the most informative features.

Datasets

The study collected data from 51 participants aged over 55, comprising 32 control participants and 19 sarcopenia patients. Participants were recruited based on specific foot size criteria (38–43 yards) and exclusion criteria (e.g., knee osteoarthritis, cognitive impairment, foot deformities). The data consists of dynamic plantar pressure measurements collected during a 6-meter walk. No new benchmark datasets were introduced; rather, existing diagnostic criteria for sarcopenia (AWGS2019) were used to classify participants.

Key Insights and Ideas

  • Weight Calibration Matters: The study highlights that for some plantar pressure features, weight calibration is crucial for revealing significant differences between sarcopenia patients and control groups. This suggests that simply looking at raw pressure values might be misleading due to variations in body weight.
  • Dynamic Pressure is Informative: The dynamic nature of plantar pressure during walking provides richer information than static measurements, capturing subtle changes related to gait abnormalities and musculoskeletal function that can be indicative of sarcopenia.
  • Forefoot Pressure is Key: The sensor array focuses on the forefoot (metatarsal bones), where peak dynamic plantar pressure typically occurs during walking. Alterations in pressure distribution in this region appear to be a strong indicator.
  • Machine Learning for Early Detection: The integration of wearable sensors with machine learning models offers a powerful and practical approach for early sarcopenia screening, moving beyond traditional, less accessible methods.

Key Results

  • The developed flexible-printed piezoelectric sensor array achieved an average voltage sensitivity of 4.844 mV/kPa.
  • Statistical analysis revealed significant differences in several plantar pressure features (both calibrated and uncalibrated) between sarcopenia patients and control groups. Notably, weight-calibrated features showed more significant differences in some cases.
  • The Support Vector Machine (SVM) model, after feature selection (using the top 21 features), achieved the highest accuracy of 93.65% for sarcopenia screening.
  • Other models also demonstrated good performance, with SVM generally outperforming others, especially when using selected features. For instance, with 21 selected features, SVM achieved high accuracy (93.65%), high recall (96.77%), and a high F1 score (93.75%).

Limitations and Future Work

The authors acknowledge several limitations:

  • Validation Population Size: The machine learning screening model needs validation in a larger, more diverse population to ensure better generalization.
  • Regional Differences: Plantar pressure characteristics might vary across different regions and populations, requiring further investigation to identify universally effective features.
  • Limited Sole Coverage: The current study focuses on the forefoot. Expanding the sensor coverage to the entire sole could provide more comprehensive data but would increase the complexity of data analysis and models.
  • Early Screening Only: This study focused on early screening and did not involve classifying the severity of sarcopenia.

Future work should focus on addressing these limitations by conducting larger-scale validation studies, exploring broader sole coverage, and potentially developing models that can assess sarcopenia severity.

Summary updated

BibTeX
@misc{han2024using,
  author = {Shulang Han and Qing Xiao and Ying Liang and Yu Chen and Fei Yan and Hui Chen and Jirong Yue and Xiaobao Tian and Yan Xiong},
  title = {{Using Flexible-Printed Piezoelectric Sensor Arrays to Measure Plantar Pressure during Walking for Sarcopenia Screening}},
  year = {2024},
  doi = {10.3390/s24165189},
}
Figure