Skip to content

Deep Learning Medical Imaging Research

Brain Tumor Segmentation & Classification: CNN/U-Net

When
Jan – Feb 2025
Length
2 months
Status
Completed (research phase)

Trained a brain-tumor classifier on 7,023 MRI scans with two teammates and benchmarked it against a 3D U-Net segmentation model.

Where it sits

EarlierAI-BITSDec 2024LaterIn2neFeb 2025

The project

Implemented and benchmarked multiple Convolutional Neural Network (CNN) architectures, including transfer learning with Xception for classification, a custom 2D U-Net for tumor segmentation, and an advanced 3D U-Net using NIfTI volumes for comprehensive volumetric analysis. The pipeline integrates custom loss functions (Dice/Jaccard), handles image preprocessing for diverse MRI modalities (FLAIR, T1, T1CE, T2), and uses TensorFlow/Keras for GPU-accelerated training. Project adheres to the Cookiecutter Data Science template for organization.

Built with