sahas025/CNN-for-brain-tumor — reverse-engineered prompt

Build me a Jupyter Notebook project for brain tumor detection from medical scan images using a custom CNN. I want it to classify multiple tumor types, train the model, validate it on a separate dataset, and show clear results like accuracy, loss curves, and a confusion matrix.

Please use PyTorch and common Python data science tools. Keep it understandable for someone learning deep learning, with clean notebook sections for setup, loading images, preprocessing, model building, training, evaluation, and comparison with a DenseNet121 baseline. Include comments explaining what each part is doing.

The final notebook should be runnable after installing the requirements, and it should make it easy to swap in a folder of brain tumor images. If you need exact current syntax or docs, look them up online.