Related Experiment Video
Updated: Jan 17, 2026
07:51
Video Movement Analysis Using Smartphones ViMAS: A Pilot Study
Published on: March 14, 2017
17.2K
Classifying simulated gait impairments using privacy-preserving explainable artificial intelligence and mobile phone
Lauhitya Reddy1, Ketan Anand2, Shoibolina Kaushik3
1Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, Georgia, United States of America.
PLOS Digital Health
|September 16, 2025
Summary
A new mobile phone artificial intelligence (AI) system accurately classifies gait impairments using video analysis. This privacy-preserving tool offers an accessible, objective method for gait assessment, moving beyond costly or subjective traditional approaches.
Area of Science:
- Biomedical Engineering
- Computer Science
- Rehabilitation Medicine
Background:
- Current gait impairment diagnosis relies on subjective clinical observation or expensive multi-camera systems.
- There is a need for accessible, objective, and privacy-preserving gait assessment tools.
Purpose of the Study:
- To develop and evaluate a mobile phone-based artificial intelligence (AI) system for classifying gait impairments.
- To assess the system's accuracy and identify key features for gait classification.
Main Methods:
- A novel dataset of 743 videos of simulated normal and six pathological gait types (circumduction, Trendelenburg, antalgic, crouch, Parkinsonian, vaulting) was created using mobile phone cameras.
- An AI system was developed to classify gait impairments from frontal and sagittal video views.
- Feature importance analysis was conducted to identify critical classification features.
Main Results:
- The AI system achieved 86.5% accuracy in classifying gait impairments using combined frontal and sagittal views.
- Sagittal views generally outperformed frontal views, except for specific gait types like circumduction.
- Lower limb keypoints, frequency-domain features, and entropy measures were identified as critical for classification.
Conclusions:
- Mobile phone-based AI systems can effectively classify diverse gait types while preserving privacy through on-device processing.
- The system shows potential for rapid prototyping of gait analysis tools.
- Clinical validation with patient data is necessary to confirm efficacy in real-world settings.
