ChodBhangra/knee-cartilage-segmentation

Knee Cartilage Segmentation

Automated multi-class knee cartilage segmentation on magnetic resonance imaging (MRI) scans from the Osteoarthritis Initiative (OAI / OAIZIB-CM).

Overview

This repository provides a deep learning testbed for benchmarking and evaluating 2D, 2.5D, and 3D segmentation architectures on femoral cartilage (FC), medial tibial cartilage (MTC), and lateral tibial cartilage (LTC).

Supported Models

• UNet (2D Slice baseline)
• Pseudo3D (2.5D Triplet channel-stacking)
• nnUNet (2D residual blocks with deep supervision)
• TransUNet (Hybrid CNN-Transformer encoder with ViT bottleneck)
• UNet3D (Volumetric 3D U-Net with attention gates)

Quick Start

1. Install dependencies:

   pip install -r requirements.txt
   

2. Generate an experiment notebook:

   python run_config.py
   

   Or run non-interactively:

   python run_config.py --model unet --gpu 0 --fraction 50 --yes