-
-
Alzheimer Detection
Inspiration
Early detection of Alzheimer’s and other forms of dementia can significantly improve patient care, but traditional diagnostic methods are costly, time-consuming, and invasive. EEG data is widely available and non-invasive, making it ideal for rapid detection using AI.
What it does
This project classifies EEG signals into Alzheimer’s Disease (AD), Cognitive Normal (CN), and Frontotemporal Dementia (FTD). Users can upload EEG recordings via a web app, and the model predicts the likelihood of each condition, while also providing performance visualizations like confusion matrices and ROC curves.
How I built it
•Data Collection & Preprocessing: EEG signals processed and normalized to 19 channels × 1425 timepoints. •Model: EEGNet, a CNN optimized for EEG data, trained with cross-validation and hyperparameter tuning. •Backend: PyTorch for deep learning and model inference. •Frontend: Streamlit web app for uploading EEG files and displaying predictions. •Evaluation: Metrics include accuracy, precision, recall, F1 score, specificity, and ROC-AUC.
Challenges I ran into
•Limited EEG dataset and class imbalance among AD, CN, and FTD. •Handling different EEG file formats and ensuring correct input shapes for the model. •Deployment issues with PyTorch dependencies in Streamlit. •Preventing overfitting and optimizing the model for cross-subject generalization.
Accomplishments that I'm proud of
•Achieved over 92% cross-subject test accuracy after rigorous 5-fold cross-validation and Optuna hyperparameter tuning. •Built a user-friendly web interface that allows researchers and clinicians to test EEG data quickly. •Implemented detailed metrics visualization for model interpretability.
What I learned
•How to preprocess EEG data for deep learning. •Effective use of EEGNet for multi-class EEG classification. •Importance of cross-validation and hyperparameter optimization in medical AI projects. •Deploying PyTorch models with Streamlit and handling real-time inference challenges.
What's next for eeg-alzheimers-detection
•Add support for real-time EEG streaming for live diagnosis. •Expand dataset to include more diverse patient demographics. •Implement explainable AI techniques to highlight EEG features driving predictions. •Integrate with clinical workflows for practical use in hospitals and research labs.
