EEG Epilepsy Detection with Streamlit

This project detects epilepsy from EEG recordings by converting signals into time–frequency scalograms, enhancing them with Sobel filtering, and classifying them using a VGG16-based deep learning model. A Streamlit web app allows users to upload a scalogram image and receive an instant prediction: “Epilepsy Detected” or “No Epilepsy Detected”.

---

🧠 Overview

The full pipeline consists of:

1. EEG signal preprocessing

   • 19-channel EEG recordings from epilepsy and non-epilepsy subjects
   • Segmentation into 1-minute windows
   • Normalization (mean–std scaling per segment)

2. Time–frequency representation

   • Short-Time Fourier Transform (STFT) to generate scalograms from each EEG segment

3. Feature enhancement

   • Sobel filters applied to the scalograms to emphasize edges and sharp transitions relevant to epileptic activity

4. Deep learning model

   • VGG16 (pretrained on ImageNet) used as a feature extractor
   • Custom fully-connected layers for binary classification: epileptic vs non-epileptic
   • Trained with Adam optimizer and binary cross-entropy
   • Achieves ~83% validation accuracy on the scalogram dataset

5. Web deployment

   • A Streamlit app (app.py) loads a saved model (model.h5)
   • User uploads an EEG scalogram image (PNG/JPG)
   • The app outputs a clear prediction:
     • Epilepsy Detected
     • No Epilepsy Detected

---

📂 Project Structure (example)

project-root/
│
├── app.py                 # Streamlit app for epilepsy detection
├── model.h5               # Trained VGG16-based model (not included in repo)
├── logo.png               # Optional logo for the app header
├── Signal_Processing.ipynb  # EEG preprocessing & STFT scalogram generation
├── VGG16+ANN.ipynb        # Model training notebook
├── requirements.txt
└── README.md
"# EEG-Epilepsy-Detection" 
"# EEG-Epilepsy-Detection" 
"# EEG-Epilepsy-Detection" 
"# EEG-Epilepsy-Detection" 
"# EEG-Epilepsy-Detection"