Accelerating 4D image reconstruction for magnetic resonance-guided radiotherapy

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Reconstruction

Background:

• Physiological motion complicates radiation dose delivery in external beam and particle therapy.
• Four-dimensional magnetic resonance imaging (4D-MRI) offers excellent soft-tissue contrast for tracking intra-fractional motion.
• Current 4D-MRI reconstruction times are too long for online treatment adaptation in MR-guided radiotherapy.

Purpose of the Study:

• To accelerate 4D-MRI reconstruction times using high-performance computing (HPC).
• To enable the use of 4D-MRI for online adaptive workflows in MR-guided radiotherapy.
• To explore the application of accelerated 4D-MRI in particle therapy.

Main Methods:

• Developed fast, parallelized, open-source implementations of the extra-dimensional golden-angle radial sparse parallel algorithm for CPU and GPU architectures.
• Scanned four patients with pancreatic adenocarcinoma using a radial stack-of-stars gradient echo sequence on a 1.5T MR-Linac.
• Assessed reconstruction time, image quality (SSIM), and scalability across different architectures and binning strategies.

Main Results:

• Reconstructed 4D-MRI images were highly consistent with the reference implementation (SSIM > 0.99).
• CPU + GPU implementation achieved reconstruction times of 60 ± 1 seconds, over 17 times faster than the reference.
• Hyper-scaling with multiple GPUs further reduced reconstruction times, demonstrating efficient parallelization.

Conclusions:

• High-performance computing methods can significantly reduce respiratory-resolved 4D-MRI reconstruction times.
• Accelerated 4D-MRI is feasible for online workflows in MR-guided radiotherapy.
• This technology holds potential for improving precision in particle therapy by accounting for motion.