Brain Tumor Classifier

A deep learning project that classifies brain MRI scans into four tumor categories at 99.3% test accuracy on the Brain Tumor MRI Dataset (Kaggle, Masoud Nickparvar). The model is a custom PyTorch CNN I designed from scratch rather than fine tuning a pretrained ResNet, because the goal was to actually understand what each layer was doing on medical imagery: four convolutional blocks with batch normalization and ReLU, three max pooling layers with mixed kernel sizes, a fully connected layer of 512 units with 50 percent dropout, and a softmax head.

Training ran in a Jupyter / Kaggle notebook with image resizing to 224x224, ImageNet normalization, and model checkpoints per epoch. The best model (model_38) is loaded by a Streamlit app that exposes the classifier as a live web demo at brain-tumor-classification.streamlit.app. Users upload an MRI image, it is transformed through the same pipeline, and the predicted tumor type renders on screen.

The repo is public (github.com/HalemoGPA/BrainMRI-Tumor-Classifier-Pytorch) and includes the full training notebook, the model architecture, the evaluation plots, and the Streamlit inference code.