Related Experiment Video
Updated: Jan 15, 2026
07:52
Evaluating Postural Control and Lower-extremity Muscle Activation in Individuals with Chronic Ankle Instability
Published on: September 18, 2020
9.0K
A Preliminary Data-Driven Approach for Classifying Knee Instability During Subject-Specific Exercise-Based Game with
Priyanka Ramasamy1, Poongavanam Palani2, Gunarajulu Renganathan3
1Graduate School of Advanced Science and Engineering, Hiroshima University, Hiroshima 739-8527, Japan.
Sensors (Basel, Switzerland)
|October 16, 2025
Summary
This study developed a gaming system to track knee instability during squats, achieving 96% accuracy in detecting issues. The system uses multiple sensors to enhance lower limb training safety and effectiveness.
Area of Science:
- Biomechanics
- Rehabilitation Engineering
- Machine Learning
Background:
- Lower limb functional degeneration is increasing, impacting core strength and motor control.
- Improper squat techniques can cause dynamic knee instability, reducing motivation for training.
- Exergame systems are needed to improve user experience and prevent injuries during lower limb training.
Purpose of the Study:
- To develop and validate a gaming-based exercise tracking system for real-time detection of knee instability during squats.
- To assess the effectiveness of multimodal sensor fusion for improving the accuracy of knee instability classification.
- To investigate the performance of Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) models in identifying knee instability events.
Main Methods:
- 28 healthy subjects participated in exergame-based squat training.
- Dynamic kinematic features were collected using a depth camera-based inertial measurement unit (IMU) and an Anima forceplate sensor.
- Spearman correlation was used for feature selection, and LSTM and SVM models were trained for binary classification of knee instability.
Main Results:
- Knee instability events were successfully classified with high accuracy (96%) using both LSTM and SVM models.
- Feature selection identified key indicators of knee instability, including knee shakiness, knee distance, squat depth, sway velocity, and sway area.
- The multimodal sensor approach significantly improved classifier performance compared to single-modality methods.
Conclusions:
- The proposed system effectively tracks knee instability in real-time using multimodal sensor data and machine learning.
- This approach enhances the safety and efficacy of lower limb training, particularly in gamified rehabilitation settings.
- The findings support the use of integrated sensor systems for personalized and adaptive physical therapy.
