A Transfer Learning Approach for Toe Walking Recognition Using Surface Electromyography on Leg Muscles

Area of Science:

• Biomedical Engineering
• Neurology
• Sports Medicine

Background:

• Gait monitoring is essential for early detection of motor abnormalities like toe walking.
• Persistent toe walking can lead to significant musculoskeletal issues, balance problems, and reduced quality of life.
• Surface Electromyography (sEMG) offers potential for early detection by capturing pre-movement muscle electrical activity.

Purpose of the Study:

• To propose a novel method for detecting toe walking using lower limb sEMG signals.
• To address the complexity and noise inherent in sEMG data for robust classification.
• To leverage Transfer Learning (TL) to improve model accuracy and generalizability across different sEMG devices.

Main Methods:

• Utilized surface Electromyography (sEMG) sensors to record lower limb muscle electrical activity.
• Applied Continuous Wavelet Transform (CWT) to convert 1-second sEMG signal windows into 2D scalogram images.
• Employed Transfer Learning (TL) with pre-trained neural network architectures to classify toe walking patterns.

Main Results:

• Achieved approximately 95% classification accuracy on a public dataset using the InceptionResNetV2 architecture.
• Demonstrated the effectiveness of the proposed sEMG and TL approach in identifying toe walking.
• Highlighted the potential of CWT-generated scalograms for robust feature extraction from noisy sEMG data.

Conclusions:

• The developed sEMG-based method, enhanced by Transfer Learning and CWT, shows high accuracy in detecting toe walking.
• This approach offers a promising non-invasive tool for early diagnosis and monitoring of gait abnormalities.
• Further research can explore broader applications in clinical settings and diverse populations for gait analysis.