Problem Description

Brain tumors are often difficult to detect manually from MRI scans, especially due to their varying shapes and sizes. In this project, you will build an AI system that can automatically detect and segment (outline) brain tumor regions from MRI images.

You will be provided with MRI scans and corresponding ground-truth masks that mark the exact tumor boundaries. Your task is to design a preprocessing pipeline, train a segmentation model, evaluate the predictions, and visualize the results.

In this project, students will build a deep learning model that can analyze brain MRI scans and automatically segment (highlight) tumor regions. The project introduces students to the intersection of AI and healthcare using real-world medical imaging data and teaches them how modern models like U-Net perform medical image segmentation.

Methodology

1. Dataset Collection

Students will use publicly available MRI datasets such as:

• BraTS 2021 Brain Tumor Segmentation Dataset (recommended)
• Contains:
  • Four MRI modalities per patient (T1, T1Gd, T2, FLAIR)
  • A corresponding segmentation mask marking tumor regions

Dataset source examples:

• Kaggle: BraTS 2021 – MRI Brain Tumor Segmentation

2. Data Preprocessing

Students must:

• Load MRI files in NIfTI (.nii.gz) format
• Convert 3D MRI volumes into 2D slices
• Normalize intensity values
• Resize slices to a fixed resolution (e.g., 128 × 128 or 256 × 256)
• Prepare:
  • Input images (any chosen modality or stacked modalities)
  • Ground-truth masks (tumor segmentation)

3. Model Development

Students will: