EMG-Spectrogram-Empowered CNN Stroke-Classifier Model Development

Area of Science:

• Biomedical Engineering
• Neurorehabilitation
• Machine Learning in Healthcare

Background:

• Stroke is a primary cause of long-term disability, leading to motor dysfunction and reduced productivity.
• Limited access to rehabilitation services, especially in low- and middle-income countries, hinders post-stroke recovery.
• Current home-based rehabilitation (HBR) relies on subjective assessments, highlighting the need for objective evaluation methods like electromyography (EMG).

Purpose of the Study:

• To develop and validate a novel deep learning (DL) methodology for objective stroke assessment using EMG signals.
• To automate the classification of stroke patients versus healthy individuals based on EMG data.
• To explore the potential of this method for enhancing stroke rehabilitation procedures and monitoring in HBR settings.

Main Methods:

• EMG signals were transformed into time-frequency representation (TFR) spectrograms.
• A novel convolutional neural network (CNN) model, Tri-CCNN, was developed using these spectrograms as input.
• The Tri-CCNN model's performance was compared against Shallow CNN and LeNet-5 architectures.

Main Results:

• The proposed Tri-CCNN model achieved a classification accuracy of 93.33%, outperforming existing models.
• Analysis of spectrogram amplitude distributions revealed distinct patterns differentiating stroke patients from healthy individuals.
• The findings indicate the method's potential for objective stroke assessment and classification.

Conclusions:

• The developed DL approach using EMG spectrograms offers an effective tool for objective stroke classification.
• This methodology shows significant promise for automating rehabilitation monitoring in home-based rehabilitation (HBR) settings.
• The study paves the way for improved stroke rehabilitation strategies and accessibility.