Contents

• Overview
• Repo Contents
• System Requirements
• Installation Guide
• Instructions for Use
• License
• Citation

1. Overview

This project provides a Nerf-based CT volume reconstruction method using a projection single view. Using the code requires users to have basic knowledge about python, PyTorch, and deep neural networks.

2. Repo Contents

• simulate_DRR.py: DRR X-ray images generation.
• generator/renderers.py: network definition of proposed method.
• loss/edge_loss.py: edge loss functions definition.
• dataset/datasets.py: data loader definition for model training.
• train.py: main code for the training stage.
• configs.py: main settings for the method.
• You can perform your CT reconstruction by changing the settings here.

3. System Requirements

Prerequisites

• Ubuntu 18.04
• NVIDIA GPU + CUDA (Geforce RTX 3090 with 24GB memory, CUDA 11.4)

Package Versions

• python 3.8
• pytorch 1.10.1
• torchvision 0.11.2
• opencv-python 4.6.0.66
• numpy 1.19.5

4. Instructions for Use

Training

• Run python simulate_DRR.py to simulate X-ray images using the DRR method. Then change the parameter in configs.py for the subsequent training process.
• Run python train.py to start the module training stage.

5. License

This project is covered under the BSD-3-Clause License.