MRI-Assisted PET Motion Correction for Neurologic Studies in an Integrated MR-PET Scanner

Ciprian Catana, Thomas Benner, André van der Kouwe, L. Byars, Michael Hamm, Daniel B. Chonde, Christian Michel, Georges El Fakhri, M. Schmand, A. Gregory SorensenJournal of Nuclear Medicine 2010 · DOI

Abstract

An MRI-based motion correction (MC) algorithm was implemented for an integrated MR-PET scanner. High-temporal-resolution MRI-derived motion estimates (obtained while simultaneously acquiring anatomic or functional MRI data) can be used for PET MC. An MRI-based MC method has the potential to improve PET image quality, increasing its reliability, reproducibility, and quantitative accuracy, and to benefit many neurologic applications.

Introduction

Positron Emission Tomography (PET) is a cornerstone of functional brain imaging, yet its long acquisition times—often ranging from minutes to an hour—make it highly susceptible to patient head motion. Even minor movements during a scan result in significant image blurring, which degrades spatial resolution and compromises the quantitative accuracy of radiotracer concentration measurements. While external tracking systems (e.g., optical cameras) or physical restraints have been employed, they often introduce logistical complexities or are obstructed by the scanner's bore. This study explores a novel approach: using the MRI component of an integrated MR-PET system as an intrinsic, high-precision motion tracker to correct PET data in real-time.

Methods

The researchers utilized a prototype integrated BrainPET scanner, consisting of a PET insert designed to operate within a 3T MRI magnet. The core of the proposed solution is a Line-of-Response (LOR) rebinning algorithm. Unlike post-processing methods that attempt to deblur a reconstructed image, LOR rebinning corrects the raw coincidence data. Each detected photon pair is mathematically shifted back to its original coordinate system based on the subject's head position at the microsecond of detection.

Integrated MR-PET scanner design and detector array geometry showing transaxial and axial gaps.

Two MRI-based tracking techniques were evaluated:

  1. EPI-derived tracking: Motion estimates are extracted from standard functional MRI sequences every 3 seconds.
  2. Cloverleaf Navigators (CLN): A high-speed navigator sequence providing motion updates every 20 milliseconds, capable of capturing rapid, involuntary movements.

Temporal synchronization was achieved by injecting MRI triggers directly into the PET list-mode data stream, ensuring a synchronization precision within 0.2 milliseconds. The system was validated using a Derenzo phantom for spatial alignment, a Hoffman brain phantom for quantitative verification, and human volunteers performing intentional head movements.

Results

The implementation of LOR-based motion correction yielded significant improvements in both visual and quantitative metrics. In phantom studies, the algorithm successfully recovered fine details that were otherwise lost to motion blur.

Comparison of PET phantom images demonstrating the effectiveness of the motion correction algorithm.

Key findings include:

  • Visual Clarity: Corrected images showed sharp delineation of brain structures, whereas uncorrected images were severely blurred.
  • Quantitative Integrity: The LOR-based method proved as accurate as gold-standard image-based correction, preserving the true tracer concentration levels.
  • Synchronization and Speed: The sub-millisecond synchronization allowed for the correction of jerky movements. While the total reconstruction time for human studies was approximately 4 hours, the individual motion estimates were processed in roughly 16 seconds, demonstrating the feasibility of the workflow.
  • Biological Signal Recovery: Time-activity curves, which track tracer kinetics, only reflected expected biological patterns after motion correction; uncorrected data was dominated by motion-induced noise.

Discussion and Limitations

This research demonstrates that MRI data, often collected simultaneously during PET scans in integrated systems, can be effectively repurposed for motion correction without additional hardware. By correcting in LOR space, the system accounts for the physical gaps between detector blocks, preventing artifacts that occur when a subject moves into a "blind spot."

However, several challenges remain. The computational demand of LOR rebinning is high, currently requiring hours for full reconstruction. Furthermore, the ultra-fast CLN tracking loses accuracy for large displacements (>10mm or >10°). Future developments will focus on optimizing processing speed via parallel computing and integrating motion tracking as a universal feature across all MRI sequences.

Conclusion

MRI-assisted motion correction significantly enhances the quality and reliability of PET imaging in integrated scanners. By leveraging the high temporal resolution of MRI, this method provides a robust solution for neurologic studies, ensuring that patient motion does not compromise the diagnostic or research value of the PET data.

Reference: Catana, C., et al. (2011). MRI-Assisted PET Motion Correction for Neurologic Studies in an Integrated MR-PET Scanner. The Journal of Nuclear Medicine, 52(1), 154–161.

Summary updated

Abstract

An MRI-based MC algorithm was implemented for an integrated MR-PET scanner. High-temporal-resolution MRI-derived motion estimates (obtained while simultaneously acquiring anatomic or functional MRI data) can be used for PET MC. An MRI-based MC method has the potential to improve PET image quality, increasing its reliability, reproducibility, and quantitative accuracy, and to benefit many neurologic applications.
BibTeX
@article{catana2010mri,
  author = {Ciprian Catana and Thomas Benner and André van der Kouwe and L. Byars and Michael Hamm and Daniel B. Chonde and Christian Michel and Georges El Fakhri and M. Schmand and A. Gregory Sorensen},
  title = {{MRI-Assisted PET Motion Correction for Neurologic Studies in an Integrated MR-PET Scanner}},
  journal = {Journal of Nuclear Medicine},
  year = {2010},
  doi = {10.2967/jnumed.110.079343},
}
Figure