Related Experiment Video
Updated: Dec 14, 2025
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
16.0K
Evaluation of Vertical Ground Reaction Forces Pattern Visualization in Neurodegenerative Diseases Identification
Che-Wei Lin1,2, Tzu-Chien Wen1, Febryan Setiawan1
1Department of Biomedical Engineering, College of Engineering, National Cheng Kung University, Tainan 701, Taiwan.
Sensors (Basel, Switzerland)
|July 16, 2020
Summary
This study introduces a deep learning algorithm for classifying neurodegenerative diseases (NDDs) using gait analysis. The novel method achieves high accuracy with significantly shorter gait data, improving early diagnosis potential.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Neurology
Background:
- Clinical diagnosis of neurodegenerative diseases (NDDs) relies on symptoms, which are often unreliable, especially in early stages.
- Gait analysis, specifically vertical ground reaction force (vGRF) data, reveals irregular patterns indicative of NDDs compared to healthy controls (HC).
Purpose of the Study:
- To develop and validate a novel deep learning classification algorithm for accurate NDD diagnosis.
- To utilize recurrence plots of vGRF data for enhanced feature representation in NDD classification.
- To improve diagnostic efficiency by reducing the required gait signal length.
Main Methods:
- A deep learning approach was employed, involving preprocessing of 10-second vGRF data windows.
- Time-domain vGRF data were transformed into recurrence plots (images) and enhanced using Principal Component Analysis (PCA).
- A Convolutional Neural Network (CNN) classifier was trained and validated using leave-one-out cross-validation (LOOCV) on data from HC, ALS, HD, and PD subjects.
Main Results:
- High classification accuracies were achieved in two-class (e.g., HC vs. ALS: 100%, NDDs vs. HC: 98.91%) and multiclass (HC: 98.99%, ALS: 98.32%, PD: 97.41%, HD: 96.74%) classifications.
- The proposed method demonstrated superior performance compared to existing methods, requiring only 10-second gait signals versus the typical 5-minute signals.
- Specific pairwise accuracies included (HC vs. HD): 98.41%, (HC vs. PD): 100%, (ALS vs. PD): 95.95%, (ALS vs. HD): 100%, (PD vs. HD): 97.25%.
Conclusions:
- The proposed deep learning algorithm effectively classifies NDDs using vGRF-based recurrence plots.
- This method offers a highly accurate and efficient approach for NDD diagnosis, significantly reducing data requirements.
- The findings suggest a promising tool for earlier and more reliable detection of neurodegenerative diseases through gait analysis.
