Related Experiment Video
Updated: May 28, 2025
10:28
Dynamic Digital Biomarkers of Motor and Cognitive Function in Parkinson's Disease
Published on: July 24, 2019
15.1K
A novel method for assessing cycling movement status: an exploratory study integrating deep learning and signal
Yingchun He1,2, Yi-Haw Jan3,4, Fan Yang1,5
1Department of Rehabilitation Medicine, School of Health, Fujian Medical University, Fuzhou, 350122, China.
BMC Medical Informatics and Decision Making
|February 11, 2025
Summary
This study introduces a deep learning motion assessment method using Keypoint RCNN (KR) and signal processing. The KR algorithm reliably evaluates motor function, showing excellent consistency and accuracy for individualized assessments.
Area of Science:
- Biomechanics
- Computer Vision
- Signal Processing
Background:
- Accurate motion assessment is crucial for evaluating motor function.
- Traditional methods may require specialized equipment or environments.
- Developing accessible and reliable motion analysis tools is an ongoing need.
Purpose of the Study:
- To propose and validate a deep learning-based motion assessment method.
- To integrate pose estimation (Keypoint RCNN) with signal processing for motion analysis.
- To assess the reliability and effectiveness of this integrated method for motor function evaluation.
Main Methods:
- Utilized the Keypoint RCNN algorithm to extract 2D skeletal keypoint coordinates from video.
- Employed inertial sensors and a smartphone for simultaneous motion capture.
- Applied signal processing techniques, including multiscale entropy analysis, for data interpretation.
- Performed statistical analyses: Spearman's rank correlation, ICC, error analysis, and t-tests.
Main Results:
- Keypoint RCNN demonstrated excellent consistency (ICC=0.988) with inertial sensors for peak acceleration frequency.
- Strong correlations (r > 0.70) and good agreement (ICC > 0.750) were found for peak acceleration frequencies and complexity index average (CIA).
- The method showed low error values (MAE = 0.001-0.040) and reliably distinguished movement statuses (p < 0.05).
Conclusions:
- The proposed deep learning and signal processing method is reliable and effective for motion assessment.
- Keypoint RCNN accurately captures movement dynamics and enables individualized motor function evaluation.
- This approach offers a promising solution for accessible, home-based motor function monitoring.
