ThomasCdnns/EEG-Seizure-Detection

SeizureDetectionCNN

Model Description

SeizureDetectionCNN is a convolutional neural network designed for binary classification of seizure events using EEG data converted to images. The model employs a simple yet effective architecture with two convolutional layers followed by batch normalization and three fully connected layers.

Model Architecture

SeizureDetectionCNN(
  (conv1): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (pool): MaxPool2d(kernel_size=2, stride=2, padding=0)
  (conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (bn1): BatchNorm2d(32)
  (bn2): BatchNorm2d(64)
  (dropout): Dropout(p=0.5)
  (fc1): Linear(in_features=4096, out_features=120)
  (fc2): Linear(in_features=120, out_features=32)
  (fc3): Linear(in_features=32, out_features=2)
)

Input Description

Input images are preprocessed to 32x32 grayscale Images are normalized with mean=[0.5] and std=[0.5] Input tensor shape: (batch_size, 1, 32, 32)

Preprocessing

from torchvision import transforms

transforms.Compose([
    transforms.Grayscale(),
    transforms.Resize((32, 32)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5], std=[0.5])
])

Training Procedure

Architectural Features

2 Convolutional layers with ReLU activation Batch Normalization after each convolutional layer MaxPooling with kernel size 2 Dropout (p=0.5) for regularization 3 Fully connected layers

Parameters

Total Parameters: ~500K Input Channels: 1 (grayscale) Output Classes: 2 (binary classification)

Intended Uses & Limitations

Intended Uses

Research and development in seizure detection Processing of EEG data converted to image format Binary classification of seizure/non-seizure events