Keras 3d U‑net For Multiclass Brain Tumor Segmentation (4 Classes)

3D Multiclass Brain Tumor Segmentation

This project implements a deep learning pipeline for 3D brain tumor segmentation using multi-modal MRI scans. It is designed to detect and classify different tumor subregions in volumetric brain images.

Key Features

Input: Four MRI modalities per patient (T1, T1ce, T2, FLAIR).

Output: Voxel-wise segmentation map with 4 classes:

0 – Background

1 – Tumor Region A (shown in red)

2 – Tumor Region B (shown in green)

3 – Tumor Region C (shown in blue)

Model: 3D U-Net with encoder–decoder architecture.

Loss Function: Combined Dice + Cross-Entropy for robust multiclass segmentation.

Metrics: Per-class Dice and mean Dice score.

Visualization: 2D slice viewer and interactive 3D tumor rendering (via VTK.js in the Flask frontend).

Deployment: Flask web application where users can upload MRI volumes, run segmentation, and visualize results in both 2D and 3D.

Workflow

Preprocess MRI scans (resampling, normalization, augmentation).

Train 3D U-Net on labeled MRI datasets.

Predict tumor regions and export segmentation masks.

Visualize results interactively in the web app (slice-by-slice and 3D tumor view).

Applications

Medical image analysis & tumor monitoring.

Clinical decision support for neurosurgeons and radiologists.

Research in brain tumor progression and treatment planning.