Convolutional Neural Network for Detection and Classification of Dementia

CNNOne is a deep convolutional neural network built using PyTorch for multi-class image classification into 4 categories. The model is designed with a deep feature extraction pipeline using multiple convolutional layers, followed by fully connected layers and dropout for regularization.

Model Architecture

The architecture consists of:

• Input: RGB image with 3 channels
• Conv Block 1:
  • Conv2d(3, 16, kernel_size=3, padding=1)
  • ReLU
  • Conv2d(16, 16, kernel_size=3)
  • ReLU
  • MaxPool2d(2, 2)
• Conv Block 2:
  • Conv2d(16, 64, kernel_size=3, padding=1)
  • ReLU
  • Conv2d(64, 64, kernel_size=3, padding=1)
  • MaxPool2d(2, 2)
• Conv Block 3:
  • Conv2d(64, 256, kernel_size=3, padding=1)
  • ReLU
  • MaxPool2d(2, 2)
• Batch Normalization on 256 channels
• Flatten Layer
• Fully Connected Layer 1: Linear(186624, 64) + ReLU + Dropout(0.2)
• Fully Connected Layer 2: Linear(64, 4)
• Output Activation: LogSoftmax(dim=1)

The final output is a log-probability distribution over 4 classes.

---

🔧 Requirements

• Python 3.x
• PyTorch
• torchvision (optional, for dataset utils)
• numpy

Install dependencies:

pip install torch torchvision numpy