yashwanthadventure/brain_tumor

This project is a brain tumor classifier built on top of MRI scans of the human brain. The README first sets up the medical motivation: a brain tumor is a mass of abnormal cells inside the skull, and because the skull is rigid, any extra growth can raise pressure and damage the brain. Tumors can be cancerous or non-cancerous, and either kind can become dangerous. Catching them early helps doctors pick the right treatment sooner, which is the reason the author frames this as a useful problem to work on. The classifier is trained on a public Kaggle dataset of 7,022 MRI images. Each image falls into one of four labels: glioma, meningioma, pituitary, or no tumor. About 22 percent of the set is held back for testing, and the rest is used for training. The images do not all share the same size, so the README notes that you should resize them after preprocessing. Before training, the images go through a cropping step. The code looks for the extreme top, bottom, left, and right edges of the brain inside the scan and trims away the surrounding black border, so the model sees only the brain region. The README links to a Preprocessing.py file from a related repo that performs this step using OpenCV. For the actual classification model, the author uses ResNet50, a deep network that was already trained on a much larger image dataset. Reusing a pre-trained model is common because training one from scratch is expensive. The README links the original ResNet paper for readers who want the background, and it lists the package versions the project depends on: Python 3.7.12, TensorFlow 2.6.0, Keras 2.6.0, matplotlib, OpenCV, and scikit-learn. The rest of the work, including training steps and testing results, is kept inside a Jupyter notebook called Brain_Tumor_Classification.ipynb. The README itself is short and points the reader at that notebook for the details.