Alzheimer Image Classification using Convolutional Neural Network (CNN)

📌 About Alzheimer’s Disease

Alzheimer's disease is a progressive neurological disorder that leads to memory loss and cognitive decline. It is one of the most common causes of dementia, affecting millions of people worldwide. Early detection through MRI imaging can assist in diagnosis and treatment.

🧠 About the Dataset

This dataset contains MRI images categorized into four classes:

• Mild Demented: 8,960 images
• Moderate Demented: 6,464 images
• Very Mild Demented: 8,960 images
• Non Demented: 9,600 images

📌 Dataset Link: Alzheimer Dataset

🏗 Project Structure

This project classifies Alzheimer’s disease using CNN. Below is the structured workflow:

1️⃣ Importing Required Libraries

2️⃣ Data Preparation

3️⃣ Dataset Splitting

The dataset is divided into:

• 70% Training
• 20% Validation
• 10% Testing

4️⃣ Data Loading

5️⃣ Preprocessing Data

• Resize Images to 224x224 pixels for uniform input size.
• Normalize Pixel Values to improve model stability.
• Shuffle Data to prevent order bias during training.
• Batching & Prefetching to optimize training speed.

6️⃣ Model Architecture

The CNN model consists of:

• Feature Extraction
  • Conv2D + BatchNormalization + ReLU Activation (Extracts features from MRI images)
  • MaxPooling2D (Reduces feature dimension and prevents overfitting)
  • 5 Convolution Layers to capture complex patterns
• Classification Layers
  • Dense(128) + BatchNormalization + ReLU (Fully connected layer for feature representation)
  • Dropout(0.5) (Reduces overfitting)
  • Dense(4, activation='softmax') (Outputs probability for 4 classes)
• Training Setup
  • Adam Optimizer (Adaptive learning for stable convergence)
  • Sparse Categorical Crossentropy (Handles integer class labels efficiently)
  • Callbacks:
    • EarlyStopping: Stops training if no improvement after 5 epoch.
    • ReduceLROnPlateau: Reduces learning rate if validation loss stagnates.
    • Custom Callback: Stops training when validation accuracy hits 95%.

7️⃣ Evaluation and Visualization

After training, the model achieves:

• Test Accuracy: 96.59%
• Test Loss: 0.0991

Training & Validation Accuracy Analysis

📈 Accuracy Trends:

• Sharp accuracy increase at the beginning, stabilizing around 95% for validation accuracy.
• Training accuracy almost reaches 100%, with validation accuracy around 95%.
• Minimal gap between training and validation accuracy, indicating good generalization.

Training & Validation Loss Analysis

📉 Loss Trends:

• Training loss consistently decreases, showing the model learns well.
• Validation loss also decreases but fluctuates slightly mid-training.
• The small gap between training and validation loss confirms no significant overfitting.

📢 Conclusion

This CNN-based Alzheimer image classification model demonstrates high accuracy and stability in detecting different stages of Alzheimer's disease using MRI images. The balanced preprocessing, proper augmentations, and deep network architecture contribute to its efficiency and generalization.

🚀 Next Steps:

• Improve model robustness with more data augmentation.
• Experiment with transfer learning (e.g., ResNet, EfficientNet).
• Optimize hyperparameters for further accuracy gains.

📌 Developed with TensorFlow & Keras.