Wave2Vec: Vectorizing Electroencephalography Bio-Signal for Prediction of Brain Disease

Area of Science:

• Artificial Intelligence in Healthcare
• Bio-signal Analysis
• Deep Learning for Time Series Data

Background:

• Growing interest in AI for health-medical information analysis, particularly deep learning for disease prediction and diagnosis using EHR and literature data.
• Limited research applying advanced AI to continuous bio-signal data (EEG, ECG) due to challenges in preprocessing, feature selection, and interpretability.
• Existing deep learning methods for time series bio-signals face issues like black-box learning, feature identification difficulties, and high computational complexity.

Purpose of the Study:

• To address the challenges of applying deep learning to bio-signal time series data.
• To propose an encoding-based Wave2vec time series classifier model combining signal processing and deep learning-based natural language processing.
• To enhance data readability, feature selection intuition, and reduce computational complexity for real-time bio-signal analysis and disease diagnosis.

Main Methods:

• Developed an encoding-based Wave2vec time series classifier model.
• Converted real-valued time series bio-signals (EEG) into sequences of symbols or wavelet patterns.
• Utilized deep learning-based natural language processing to vectorize and learn from these symbol sequences for classification.

Main Results:

• Demonstrated the model's advantages through three experiments using benchmark EEG data.
• The proposed method enhanced data readability and the intuition of feature selection and learning processes.
• Achieved drastic reduction in computational complexity without performance deterioration, facilitating real-time analysis.

Conclusions:

• The encoding-based Wave2vec model effectively classifies bio-signal data by converting time series into symbolic sequences.
• The approach improves the interpretability of deep learning models for bio-signal analysis.
• Facilitates the development of real-time disease diagnosis systems by enabling efficient analysis of large-capacity data.