This repository contains mini projects in machine learning in deep learing with notebook files

End-to-end brain tumor segmentation on BraTS2020 with a modified DUCKNet (U-Net + DenseNet). Includes data preprocessing/augmentation, Keras training loops, and rigorous eval (Dice/IoU), achieving 88.7% validation Dice with 0.010 validation loss. Reproducible notebook and comparisons vs baseline U-Net; trained on A100.