MRI Tumor CNN
A PyTorch convolutional neural network for binary classification of brain tumors from MRI scans, built as an applied deep-learning study after Google DevFest 2025.
MRI Tumor CNN is a convolutional neural network, written in PyTorch, that classifies brain MRI images as tumor or no-tumor. I built it as a hands-on study of applied computer vision, following an “AI for Good” prompt from Google DevFest 2025.
Why it exists
Medical imaging is a common entry point for applied deep learning because the task is concrete and the stakes make the modeling choices matter. I used it to work through a full image-classification pipeline end to end, from loading and augmenting data to evaluating a trained model, rather than treating any single step as a black box.
How it works
- Data pipeline: loads and preprocesses the MRI dataset, with augmentation to improve generalization on a limited number of samples.
- Model: a compact CNN with convolutional, pooling, and fully connected layers trained for binary classification.
- Training loop: standard supervised training with loss tracking and validation to watch for overfitting.
- Evaluation: accuracy and confusion-matrix analysis on a held-out test set to understand where the model fails, not just how often it succeeds.
The whole project lives in a readable Jupyter notebook so the reasoning behind each step is visible alongside the code.