AI‑Based Alzheimer’s Detection Using MRI Brain Scans

Loss of only 0.0982 in 15 epochs
Accuracy of 96.56% in 15 epochs
Data of MRI Scans

Inspiration

Alzheimer’s disease is often diagnosed late, when symptoms are already severe. We were inspired to explore how AI and medical imaging can help detect early patterns in brain MRI scans, making diagnosis faster, more accessible, and more objective.

What it does

Our project uses a Convolutional Neural Network (CNN) to analyze MRI brain scans and classify them into different stages of Alzheimer’s disease. Given an MRI image, the model predicts the most likely cognitive condition, helping demonstrate how AI can assist in early screening.

How we built it

We trained a CNN-based deep learning model on an MRI dataset stored in parquet format. The pipeline includes image decoding, preprocessing, normalization, data splitting, and model training using TensorFlow/Keras. We evaluated performance using validation accuracy and visualized predictions and sample MRI scans for interpretability.

Challenges we ran into

Understanding and decoding MRI images stored inside parquet files
Handling data preprocessing correctly to avoid corrupted inputs
Debugging low accuracy caused by incorrect labels, scaling, and loss functions
Managing model stability and training performance in Google Colab

Accomplishments that we're proud of

Successfully visualized real MRI brain scans from the dataset
Built a working CNN that classifies Alzheimer’s stages from MRI images
Achieved 96.56% accuracy on validation data
Created clear visual outputs (sample scans and performance charts) for judges

What we learned

We learned how critical correct data preprocessing is for medical AI, how CNNs extract meaningful spatial features from images, and how to debug deep learning models systematically. We also gained experience working with non-traditional image formats like parquet files.

What's next for AI‑Based Alzheimer’s Detection Using MRI Brain Scans

Next, we aim to improve model generalization with larger datasets, add explainability methods like Grad-CAM, test on real-world clinical data, and explore deployment as a lightweight screening tool to support medical professionals.