The dataset used in this project is the Brain MRI Segmentation Dataset from Kaggle. The dataset contains 110 patients' images of brain MRI scans with their corresponding segmentation masks.


The model used in this project is a custom Unet with a Resnext50 as backbone. The model is trained on the dataset for 50 epochs with a batch size of 16


The model managed to achieve a Mean DICE on validation: 0.938 and a Mean IoU of the test images of - 92.28%. The following is an example of the model's predictions on the test set.

Brain MRI segmentation overview

Project repository


A Pytorch implementation of brain tumor segmentation, using a Resnext50 as backbone and a custom Unet

Image segmentation has various applications in industry, particularly in the medical industry the possiblities for automated diagnoses and cheaper medical imaging programs motivate researchers to invest in producing robust image segmentation models.