From a Stalled Implementation to a 90-Mark Dissertation: Brain Tumor Segmentation with 3D U-Net
The Challenge
He had chosen a strong but technically demanding topic — brain tumor segmentation using a 3D U-Net architecture — but was stuck on turning the idea into a working implementation. Between setting up the right environment, preparing medical imaging data correctly, and getting a deep learning model to actually train and validate properly, this is exactly the kind of technical gap that stalls a lot of AI/CS dissertations at the implementation stage.
What the Mentoring Covered
- Setting up a GPU-enabled Google Colab environment so training was actually feasible on his system
- Data preprocessing for the medical imaging dataset
- Data validation, to make sure the pipeline was sound before investing time in training
- Guidance through implementing the 3D U-Net model itself
- Working through to final results and how to interpret and present them
The implementation and dissertation were his own work — the mentoring was focused on unblocking each stage he was stuck at, not producing the work for him.
OutcomeScored 90 marks