Deep Learning Approach to Medical Diagnostics
This project focuses on automated epilepsy detection using EEG (electroencephalogram) data analysis. Using advanced signal processing and deep learning techniques, I developed a Convolutional Neural Network (CNN) model capable of identifying epileptic patterns in brain activity.
The work was conducted under the mentorship of a UC Berkeley professor and involved processing large-scale EEG datasets from the TUH EEG Corpus using Python's MNE library for signal processing and wavelet coefficient extraction.
TUH EEG Corpus and CHB-MIT datasets - World's largest publicly available EEG datasets, ensured that all demographics are represented in the dataset
Wavelet transform for feature extraction + CNN for pattern classification
Potential to assist clinicians in faster, more accurate epilepsy diagnosis
Explore the code, methodology, and results on GitHub. The repository includes Jupyter notebooks with detailed explanations, model architecture, and visualization of results.
Signal Processing Pipeline: Implemented a comprehensive preprocessing pipeline to clean and prepare EEG signals, including artifact removal, filtering, and normalization.
Feature Extraction: Utilized wavelet transforms to extract time-frequency features from EEG signals, capturing both temporal and spectral characteristics of epileptic activity.
Deep Learning Model: Designed and trained a CNN architecture optimized for EEG signal classification, achieving strong performance in distinguishing between normal and epileptic brain activity patterns.
Clinical Relevance: This work demonstrates the potential of AI-assisted diagnostic tools in neurology, potentially reducing the time and expertise required for epilepsy screening.