AI for Alzheimer's MRI

• GIF

  Working

Alzheimer’s MRI Classification Using Deep Learning

Inspiration

Alzheimer’s disease affects memory, independence, and families. We wanted to explore how AI + MRI imaging could support research into early understanding of cognitive decline — not to replace doctors, but to learn how deep learning can recognize meaningful brain-scan patterns.

What We Built

We trained a ResNet-based deep learning model to classify MRI scans into four stages:

• Non-Demented
  
• Very Mild Demented
  
• Mild Demented
  
• Moderate Demented
  

We used transfer learning, a reproducible notebook, and a clean evaluation pipeline so others can easily follow and extend our work.

How We Built It

1. Preprocessed MRI scans: resize → normalize → augment
   
2. Fine-tuned a pretrained model on the dataset
   
3. Evaluated using accuracy and weighted F1-score
   
4. Integrated the trained model into a simple prediction app
   

Inline math example: ( \mathcal{L} = - \sum_i y_i \log(\hat{y}_i) )

Displayed math example:
$$ \text{Accuracy} = \frac{\text{Correct Predictions}}{\text{Total Samples}} $$

What We Learned

We learned a lot about:

• Transfer learning techniques
  
• Handling dataset imbalance
  
• Evaluation metrics and debugging model collapse
  
• The importance of reproducibility and transparency in healthcare AI
  

Results

Our final model achieved:

• Accuracy: 92.89%
  
• Weighted F1-score: 92.83%
  

These scores showed strong, consistent performance across all four classes.

Challenges

We encountered real-world issues such as:

• Class-label inconsistencies
  
• Environment differences
  
• Early model collapse (fine-tuning pitfalls)
  
• Maintaining full notebook reproducibility
  

Each challenge helped us strengthen the project and improve our workflows.

Example Code Block

puts "Hello World!"

Built With

• code
• colab
• github
• jupyter
• matplotlib
• numpy
• python
• pytorch
• resnet
• scikit?learn
• seaborn
• torchvision
• vs