Inspiration
Alzheimer’s Disease is one of the most challenging neurological disorders, affecting millions of people worldwide. Early detection plays a crucial role in slowing disease progression and improving patient care. However, manual MRI analysis is time-consuming and requires expert radiologists.
This inspired us to build NeuroLens, an AI-powered system that can automatically analyze brain MRI scans and assist in detecting different stages of Alzheimer’s Disease with high accuracy and transparency.
What it does
NeuroLens uses deep learning to classify brain MRI images into four stages of Alzheimer’s Disease:
Non-Demented Very Mild Demented Mild Demented Moderate Demented
In addition to accurate classification, NeuroLens uses explainable AI (Grad-CAM) to highlight the regions of the brain that influence predictions, helping users understand why a decision was made.
How we built it
Loaded MRI images stored in Parquet files Converted image bytes into pixel arrays using PIL Applied preprocessing (resizing, normalization, channel conversion) Used ResNet-18 (Transfer Learning) with PyTorch Trained the model using CrossEntropy Loss and Adam optimizer Evaluated performance using accuracy and classification metrics Applied Grad-CAM for explainability and visualization
Challenges we ran into
Handling MRI images stored as byte objects inside Parquet files Resolving datatype errors while converting images to NumPy arrays Class imbalance across Alzheimer’s stages Training deep models efficiently within limited computational resources Ensuring model predictions were explainable and not a black box
Accomplishments that we're proud of
Achieved 95.86% classification accuracy Built a complete end-to-end pipeline from data loading to evaluation Successfully integrated Explainable AI (Grad-CAM) Created a reproducible and well-documented notebook Delivered a healthcare-focused AI solution with real-world impact
What we learned
How to process large medical imaging datasets stored in Parquet format Practical application of transfer learning in healthcare Importance of explainability in medical AI systems Debugging real-world data issues in deep learning workflows Ethical considerations in medical AI applications
What's next for NeuroLens
Integrating clinical data (age, cognitive scores) with MRI images Expanding to multi-modal inputs such as PET scans Improving class balance with data augmentation Deploying NeuroLens as a web-based clinical decision support system Collaborating with healthcare professionals for real-world validation
Built With
- google-colab
- matplotlib
- numpy
- pandas
- pillow
- python
- pytorch
- resnet-18
- scikit-learn
