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.
python data machine-learning image supervised-learning imageprocessing neuralnetworks datacollection unet-image-segmentation datapreparation braintumorsegmentation modelevaluation brats2020 featureselection ducknet-tensorflow ducknet-keras datatraining ducknet
-
Updated
Dec 6, 2024 - Jupyter Notebook