Classification of amyotrophic lateral sclerosis disease based on convolutional neural network and reinforcement...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease affecting motor neurons.
• Electromyogram (EMG) signals offer valuable insights into neuromuscular function and disease states.
• Accurate classification of EMG signals is crucial for diagnosing and monitoring neuromuscular disorders like ALS.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for efficient and accurate classification of EMG signals in patients with ALS compared to healthy individuals.
• To explore the effectiveness of various time-frequency representations for characterizing EMG signals.
• To demonstrate the superiority of the proposed deep learning model over existing methods.

Main Methods:

• Time-frequency (T-F) representations including spectrogram, continuous wavelet transform (CWT), and smoothed pseudo Wigner-Ville distribution (SPWVD) were utilized.
• A convolutional neural network (CNN) architecture with two convolution layers, two pooling layers, and a fully connected layer was designed.
• The CNN model was trained using a reinforcement sample learning strategy on a publicly available EMG dataset.

Main Results:

• The proposed deep learning method achieved a classification accuracy of 96.80% on the EMG dataset.
• The dataset comprised 89 ALS and 133 normal EMG signals, sampled at 24 kHz.
• Comparative analysis demonstrated the superior performance of the developed method against other existing techniques.

Conclusions:

• Deep learning, particularly CNNs with appropriate T-F representations, offers a highly effective approach for classifying EMG signals in ALS.
• The proposed method shows significant potential for improving the diagnostic accuracy and efficiency of ALS detection.
• Further research could explore larger datasets and diverse T-F methods to enhance robustness and generalizability.