aehrc/synthrad2025_docker

SynthRAD2025 – Task 1 & Task 2 Solutions

This repository contains our solutions for the MICCAI Grand Challenge – SynthRAD2025, focusing on Task 1 and Task 2.
Our team achieved 1st place in the post-challenge leaderboard for both tasks (with Task 1 also ranking 1st during the official test phase).

---

🏆 Challenge Overview

• Task 1: MRI-to-CT synthesis (MR → sCT)
• Task 2: CBCT-to-CT synthesis (CBCT → sCT)

Our methods emphasize robust image synthesis, reproducible pipelines, and multi-region generalization.

---

File Descriptions

• docker_task_1/
  Contains all necessary files to build and run the Docker image for Task 1 (MR → sCT).

  • process.py: Script that performs inference, converting MR images into synthetic CT (sCT).

• docker_task_2/
  Contains all necessary files to build and run the Docker image for Task 2 (CBCT → sCT).

  • process.py: Script that performs inference, converting CBCT images into synthetic CT (sCT).

• Normalization Config Files

  • 260_gt_nnUNetResEncUNetLPlans.json/540_gt_nnUNetResEncUNetLPlans.json: Normalization configuration for the Abdomen region.
  • 262_gt_nnUNetResEncUNetLPlans.json/542_gt_nnUNetResEncUNetLPlans.json: Normalization configuration for the Head & Neck region.
  • 264_gt_nnUNetResEncUNetLPlans.json/544_gt_nnUNetResEncUNetLPlans.json: Normalization configuration for the Thorax region.
    These files are essential for inverse normalization, ensuring that the synthesized CT intensities are mapped back to their correct clinical ranges.

• Dockerfile
  Defines all steps and dependencies needed to build the Docker image. It ensures reproducibility and consistency across environments.

• base_algorithm/
  Contains the baseline algorithm files provided by the official SynthRAD2025 template, serving as the foundation for our solution.

• build.sh
  Shell script for automating the Docker build process.

• export.sh
  Shell script for exporting the built Docker image into a compressed archive for submission or deployment.

• requirements.txt
  Lists all Python dependencies required to run the code.

• revert_normalisation.py
  Script to apply inverse normalization to synthesized CT outputs, restoring them to the correct intensity distributions for downstream evaluation.

🚀 Getting Started

• nnunet_results
  Before starting inference, you also need to create a folder called nnunet_results/ in docker_task_1/docker_task_2 and place your trained models under the nnunet_results/ directory so that inference can correctly locate and load them.

1. Build the Docker Image

cd docker_task_1
bash build.sh

To test the algorithm locally, you can run the Docker container with GPU support, memory limit, and a larger shared memory (`/dev/shm`) size (e.g., 8 GB).