ALZHEIbyAI

• 

  AI Minds for a Clearer Tomorrow

Inspiration

• Early detection in Alzheimer's disease can really help in aiding patients to recover as swiftly as possible.
• Brain MRI scans contain subtle structural patterns that can indicate early stages of dementia, enabling timely diagnosis through AI based analysis.

What it does

• Our model takes input in the form of MRI scans and classifies the patient into ne of the four categories:
• Non demented
• Very mild demented
• Mild demented
• Moderate demented
• Uses Grad-CAM visualizations to highlight affected brain regions, improving interpretability.
• Since dementia is a strong indicator of Alzheimer’s disease, the model helps in early risk detection and classification.

How we built it

• Trained a machine learning based image classification system using labeled MRI scans.
• Implemented a ResNet-18 CNN architecture with transfer learning.
• We used a pretrained model to extract low-level and high-level features such as edges, textures, and structural brain patterns.
• Integrated Grad-CAM to generate heatmaps showing regions influencing predictions.

Challenges we ran into

• Achieving consistent results as beginners in machine learning.
• Handling computational constraints while training deep neural networks.
• Ensuring the model predictions aligned with the clinical goal of early Alzheimer’s detection.

Accomplishments that we're proud of

• Built a complete MRI-based classification pipeline.
• Achieved >80% validation accuracy.
• Delivered a healthcare-focused ML solution within a hackathon timeline.

What we learned

• The importance of selecting an appropriate loss function for multi-class classification.
• How CNNs and transfer learning improve performance on limited medical datasets.
• Ethical and practical considerations in deploying AI for healthcare.
• Combining visualization tools like Grad-CAM improves model transparency.
• Deep learning models require careful tuning when applied to medical data.

What's next for ALZHEIbyAI

• Introduce patient interaction features, such as: 1.Cognitive questionnaires 2.Memory and reaction-based tasks 3.Speech and response-time analysis
• Combine MRI predictions with interaction-based signals for multi-modal dementia detection.
• Enhance the web interface to support patient engagement and clinician insights.

Built With

• claude
• gardio
• googlecolab
• gpu
• gradcam
• kaggle
• numpy
• python
• pytorch
• torchvison