Gadgetron Parallel Reconstruction Framework

A practical framework for translating offline MR reconstructions to inline deployment, built on the open-source Gadgetron platform.

---

📝 Citation

If you use this framework in your research, please cite:

Ning Z, et al. From offline to inline without pain: A practical framework for translating offline MR reconstructions to inline deployment using the Gadgetron platform. Magn Reson Med. 2026. https://doi.org/10.1002/mrm.70304

---

✨ Key Features

✅ Easy Prototyping of Offline Reconstructions

• Converts ISMRMRD raw data into a Twix-like structure for minimal modification of existing offline code
• Provides a general-purpose MATLAB-based input converter and customizable ParameterMap covering commonly used headers

⚙️ Inline Execution Without Scanner Disruption

• Enables long custom reconstructions without interrupting the MR exam flow
• Inline reconstructions are executed asynchronously on an external server
• Supports retrieval scans and retro-reconstruction
• Scanner-native reconstructions are preserved for early review

🔁 Multi-input & Joint Reconstruction Support

• Allows inline reconstructions to access shared or external input scans (e.g., reference coils)

🧠 Resource-aware Parallel Scheduling

• Includes built-in GPU monitoring and queueing
• Schedules jobs intelligently across multi-GPU servers to avoid overloads

🧬 Integrated Scanner Post-processing

• Enables scanner-based post-processing (e.g., bias field and distortion correction) on custom reconstructions
• Ensures visual consistency with Siemens-reconstructed images

---

📁 Folder Structure

Gadgetron_Parallel_Framework/
│
├── ParameterMap/             # Metadata mapping (Twix → ISMRMRD)
├── config/                   # Gadget chain configuration files (*.xml)
├── matlab_handle_archive/    # MATLAB handlers for "Read & Save" and background recon
├── Gadgetron_tools/          # Input converters, Twix-like wrappers, utilities
├── Bash_script/              # Headless / queued job launchers
├── Useful_tools/             # Misc tools: ISMRMRD readers, resource monitors, etc.
├── SENSE_Recon_Demo/         # Demo: SENSE (offline + inline)
├── AlignSENSE_Recon_Demo/    # Demo: AlignSENSE (offline + inline)
├── radial_NUFFT_Demo/        # Demo: NUFFT recon for radial sequence (offline + inline)

---

🚀 Getting Started

Requirements

• Gadgetron
• MATLAB ≥ R2021a with Gadgetron Toolbox
• CUDA-enabled Conda environment (for launching recon scripts)
• Linux server with NVIDIA GPUs
• Siemens scanner with ICEGadgetron integration

📖 Full Documentation

All setup instructions and tutorials are available at Wiki of this repo

Or see: 👉 User Manual on Notion

---

🧪 Demo Cases

Inline implementations included:

• SENSE: SENSE_Recon_Demo/

  • Reference: Pruessmann KP et al., SENSE: Sensitivity Encoding for Fast MRI. MRM 1999. DOI:10.1002/mrm.1910420526

  

  Figure 1. SENSE reconstruction results of a SWI sequence using the inline demo script

• AlignSENSE: AlignSENSE_Recon_Demo/

  • Reference:
    • L. Cordero-Grande, et al., Sensitivity Encoding for Aligned Multishot Magnetic Resonance Reconstruction. IEEE Transactions on Computational Imaging, 2016. DOI: 10.1109/TCI.2016.2557069
    • L. Cordero-Grande, et al., Motion-corrected MRI with DISORDER: Distributed and incoherent sample orders for reconstruction deblurring using encoding redundancy. MRM. 2020. DOI: 10.1002/mrm.28157

Figure 2. AlignSENSE reconstruction results of a MPRAGE sequence using the inline demo script. (A) before MoCo; (B) after MoCo via AlignSENSE

Figure 3. AlignSENSE reconstruction results of a SWI sequence using the inline demo script. (A) before MoCo & using auto-calibration lines for coil sensitivity estimation; (B) after MoCo via AlignSENSE & using auto-calibration lines; (C) before MoCo & using an external reference for coil sensitivity estimation and reconstruct to a lager FOV (wrapping removed compared to A); (D) after MoCo via AlignSENSE & using an external reference

• radial NUFFT: radial_NUFFT_Demo/
  • Reference: Blunck Y, et al., 3D-multi-echo radial imaging of 23 Na (3D-MERINA) for time-efficient multi-parameter tissue compartment mapping. MRM, 2018. DOI: 10.1002/mrm.26848

Figure 4. NUFFT reconstruction results of a multi-echo GRE Sodium imaging sequence with radial trajectory using the inline demo script for three views (A,D, axial; B,E, sagittal; C,F, coronal; A-C, first echo; D-F, second echo)

Each demo contains:

• Offline version
• Inline version (inline/ subfolder)
• diff/ subfolder showing minimal modifications for inline adaptation

The demo datasets (raws) can be downloaded at Zenodo: 👉 demo datasets at Zenodo

---

✅ Validation Summary

Validated on:

• 3T MAGNETOM Vida (XA60) — tested with MPRAGE, FLAIR, SWI

• 7T MAGNETOM Terra X (XA60) — tested with multi-echo radial GRE

• TwinsUK cohort study on 3T MAGNETOM Vida - applied on MPRAGE

  • N = 480 subjects
  • 99% successful inline retrieval rate (via retrieval scans or retro-recon)

---

📄 License

This project is released under the MIT License.