Machine learning algorithms for detection of visuomotor neural control differences in individuals with PASC and ME

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Millions worldwide experience long-term symptoms post-COVID-19 infection, known as post-acute sequelae of SARS-CoV-2 (PASC) or long COVID.
• Early detection and intervention are critical for managing PASC and Myalgic Encephalomyelitis (ME).
• Current diagnostic methods may not fully capture the neurological effects of these conditions.

Purpose of the Study:

• To develop and validate a novel method for detecting the likelihood of PASC or ME using electroencephalogram (EEG) data.
• To explore the efficacy of machine learning and deep learning models in analyzing EEG spectrograms for PASC/ME detection.
• To investigate the impact of data augmentation using Generative Adversarial Networks (WGANs) on model performance.

Main Methods:

• Collected four-channel EEG data using a wearable headband.
• Processed raw EEG signals into spectrogram-like matrices using Continuous Wavelet Transform (CWT).
• Trained and evaluated various machine learning models (CONVLSTM, CNN-LSTM, Bi-LSTM) and traditional models.
• Augmented the dataset with synthetic spectrograms generated by Wasserstein Generative Adversarial Networks (WGANs).

Main Results:

• The CNN-LSTM model achieved 83% accuracy using original EEG spectrogram data.
• Models trained on WGAN-generated synthetic data achieved an average accuracy of 93%.
• Synthetic data augmentation significantly improved detection accuracy and addressed data volume and privacy concerns.

Conclusions:

• The proposed EEG-based method is effective for detecting PASC and ME, enabling early identification.
• Machine learning models, particularly CNN-LSTM trained on augmented data, show high potential for clinical application.
• This approach can aid in evaluating patient conditions, monitoring recovery, and assessing intervention efficacy for PASC and ME.