Frequency drift in MR spectroscopy at 3T

Steve C. N. Hui, Mark Mikkelsen, Helge J. Zöllner, Vishwadeep Ahluwalia, Sarael Alcauter, Laima Baltusis, Deborah A. Barany, Laura R. Barlow, Robert Becker, Jeffrey I. Berman, Adam Berrington, Pallab K. Bhattacharyya, Jakob Udby Blicher, Wolfgang Bogner, Mark S. Brown, Vince D. Calhoun, Ryan Castillo, Kim M. Cecil, Richard A. E. Edden, Yeo Bi Choi, Winnie C. W. Chu, William T. Clarke, Alexander R. Craven, Koen Cuypers, Michael Dacko, Camilo de la Fuente-Sandoval, Patricia Desmond, Aleksandra Domagalik, Julien Dumont, Niall W. Duncan, Ulrike Dydak, Katherine Dyke, David A. Edmondson, Gabriele Ende, Lars Ersland, C. John Evans, Alan S. R. Fermin, Antonio Ferretti, Ariane Fillmer, Tao Gong, Ian Greenhouse, James T. Grist, Meng Gu, Ashley D. Harris, Katarzyna Hat, Stefanie Heba, Eva Heckova, John P. Hegarty, Kirstin-Friederike Heise, Shiori Honda, Aaron Jacobson, Jacobus F. A. Jansen, Christopher W. Jenkins, Stephen J. Johnston, Christoph Juchem, Alayar Kangarlu, Adam B. Kerr, Karl Landheer, Thomas Lange, Phil Lee, Swati Rane Levendovszky, Catherine Limperopoulos, Feng Liu, William Lloyd, David J. Lythgoe, Maro G. Machizawa, Erin L. MacMillan, Richard J. Maddock, Andrei V. Manzhurtsev, María L. Martinez-Gudino, Jack J. Miller, Heline Mirzakhanian, Marta Moreno-Ortega, Paul G. Mullins, Shinichiro Nakajima, Jamie Near, Ralph Noeske, Wibeke Nordhøy, Georg Oeltzschner, Raul Osorio-Duran, María Concepción García Otaduy, Erick H. Pasaye, Ronald Peeters, Scott J. Peltier, Ulrich Pilatus, Nenad Polomac, Eric C. Porges, Subechhya Pradhan, James Joseph Prisciandaro, Nicolaas A. Puts, Caroline D. Rae, Francisco Reyes-Madrigal, Timothy P. L. Roberts, Caroline E. Robertson, Jens T. Rosenberg, Diana-Georgiana Rotaru, Ruth L. O'Gorman Tuura, Muhammad G. Saleh, Kristian Sandberg, Ryan Sangill, Keith Schembri, Anouk Schrantee, Natalia A. Semenova, Debra Singel, Rouslan Sitnikov, Jolinda Smith, Yulu Song, Craig E. L. Stark, Diederick Stoffers, Stephan P. Swinnen, Rongwen Tain, Costin Tanase, Sofie Tapper, Martin Tegenthoff, Thomas Thiel, Marc Thioux, Peter Truong, Pim van Dijk, Nolan Vella, Rishma Vidyasagar, Andrej Vovk, Guangbin Wang, Lars T. Westlye, Timothy K. Wilbur, William R. Willoughby, Martin Wilson, Hans-Jörg Wittsack, Adam J. Woods, Yen-Chien Wu, Junqian Xu, Maria Yanez Lopez, David Ka Wai Yeung, Qun Zhao, Xiaopeng Zhou, Gasper ZupanNeuroImage 2021 · DOI

TL;DR

This study benchmarks frequency drift in 3T MRI scanners across multiple vendors and sites, revealing that while typical drift is low, some scanners exhibit significant instability after fMRI scans. This drift can impact MR spectroscopy (MRS) data quality, highlighting the need for real-time field stabilization, especially in multi-site collaborations.

What Problem Does This Paper Address?

Magnetic Resonance Imaging (MRI) relies on a stable magnetic field (B₀) to generate accurate signals. However, rapid switching of gradient coils during imaging sequences, particularly functional MRI (fMRI) and diffusion tensor imaging (DTI), generates heat. This heat can alter the magnetic susceptibility of scanner components, leading to B₀ field instability, known as frequency drift. This drift is a significant concern for Magnetic Resonance Spectroscopy (MRS), which requires precise frequency alignment for accurate quantification, especially for techniques like J-difference editing. While retrospective correction methods exist, their efficacy is not fully characterized, and prospective real-time correction is often preferred. This paper aims to establish a benchmark for typical frequency drift across a large, multi-vendor, multi-site cohort to assess the necessity and impact of real-time field-frequency locking.

Methods

The researchers developed a standardized protocol to measure frequency drift on 99 3T MRI scanners from three major vendors (GE, Philips, Siemens) across 80 participating sites. The protocol involved acquiring unsuppressed water spectra using the PRESS sequence before and after a 10-minute BOLD-weighted fMRI acquisition. Real-time frequency stabilization was intentionally switched off to observe the natural drift.

Key methodological aspects include:

  • Standardized Protocol: A consistent protocol across vendors ensured comparability.
  • Phantom-based Acquisition: Using a water phantom minimizes biological variability and allows for precise measurement of field stability.
  • Pre- and Post-fMRI Measurements: This design directly assesses the impact of fMRI on B₀ field stability.
  • Spectral Analysis: Water peak frequency was extracted from each transient using a Voigt lineshape function, allowing for precise tracking of frequency shifts over time.
  • Simulation: Simulated MRS spectra were convolved with observed frequency drift traces to visualize the impact on metabolite signals (NAA and GABA).

Datasets

The study utilized data from 99 3T MRI scanners across 80 participating sites. These scanners were from three major vendors: GE (22 scanners), Philips (30 scanners), and Siemens (47 scanners). Sixty of these sites also provided a second dataset acquired on a different day to assess reproducibility. No new benchmark datasets were introduced; rather, existing scanners were used to establish a real-world performance benchmark.

Key Insights and Ideas

  • fMRI is a significant driver of drift: The study clearly demonstrates that even a short 10-minute fMRI scan can induce a measurable and significant increase in frequency drift compared to pre-fMRI baseline.
  • Vendor and scanner variability: While typical drift is low, there's substantial variation between scanners, with some exhibiting much higher drift rates than others. This suggests that scanner hardware and cooling systems play a crucial role.
  • Impact on MRS is quantifiable: Simulations show that frequency drift can lead to noticeable signal intensity reductions in key metabolites like NAA and GABA, underscoring the practical implications for MRS data quality.
  • Reproducibility is good: Despite drift, the pattern of drift is largely reproducible within individual scanners across different days, suggesting that consistent scan ordering within a protocol can help mitigate variability.

Key Results

  • Typical Drift: Before fMRI, the median absolute frequency offset was 0.44 Hz. After a 10-minute fMRI scan, this increased to 0.83 Hz for the first 5:20 minutes and 3.15 Hz for the full 30-minute post-fMRI acquisition.
  • Drift Rates: Average drift rates were 0.29 Hz/min before fMRI and 0.43 Hz/min after fMRI.
  • Extreme Cases: The most extreme cases showed drifts up to 26 Hz within 30 minutes after fMRI.
  • Metabolite Impact: Simulations indicated that NAA singlet intensity could be reduced by up to 44% after fMRI due to drift.
  • Reproducibility: Repeat acquisitions on different days showed good agreement, with ICC values around 0.85 for pre-fMRI and 0.95 for post-fMRI data.
  • Long-term Stability: In one extended acquisition, drift rate decreased significantly after about three hours, suggesting a return to equilibrium.

Limitations and Future Work

  • Outlier Emphasis: The visualizations tend to over-emphasize sites with large drift, potentially obscuring the performance of more stable scanners.
  • Limited Sensor Data: The study lacked consistent gradient temperature monitoring data across all sites, which would have provided deeper insights into the physical mechanisms of drift.
  • Passive Shim Information: Detailed information on passive shim elements, which influence heat dissipation, was often inaccessible from participating sites.
  • RF Heating Neglected: The study focused on gradient-induced heating and did not account for RF energy absorption by the phantom, which could also contribute to temperature changes.
  • Future Work: The authors suggest that real-time field-frequency locking is crucial for MRS, especially after multiple long EPI acquisitions or in multi-center collaborations. Further investigation into cooling system variations and the impact of passive shim configurations could also be beneficial. The collected data is made available for further research.

Summary updated

BibTeX
@article{hui2021frequency,
  author = {Steve C. N. Hui and Mark Mikkelsen and Helge J. Zöllner and Vishwadeep Ahluwalia and Sarael Alcauter and Laima Baltusis and Deborah A. Barany and Laura R. Barlow and Robert Becker and Jeffrey I. Berman and Adam Berrington and Pallab K. Bhattacharyya and Jakob Udby Blicher and Wolfgang Bogner and Mark S. Brown and Vince D. Calhoun and Ryan Castillo and Kim M. Cecil and Richard A. E. Edden and Yeo Bi Choi and Winnie C. W. Chu and William T. Clarke and Alexander R. Craven and Koen Cuypers and Michael Dacko and Camilo de la Fuente-Sandoval and Patricia Desmond and Aleksandra Domagalik and Julien Dumont and Niall W. Duncan and Ulrike Dydak and Katherine Dyke and David A. Edmondson and Gabriele Ende and Lars Ersland and C. John Evans and Alan S. R. Fermin and Antonio Ferretti and Ariane Fillmer and Tao Gong and Ian Greenhouse and James T. Grist and Meng Gu and Ashley D. Harris and Katarzyna Hat and Stefanie Heba and Eva Heckova and John P. Hegarty and Kirstin-Friederike Heise and Shiori Honda and Aaron Jacobson and Jacobus F. A. Jansen and Christopher W. Jenkins and Stephen J. Johnston and Christoph Juchem and Alayar Kangarlu and Adam B. Kerr and Karl Landheer and Thomas Lange and Phil Lee and Swati Rane Levendovszky and Catherine Limperopoulos and Feng Liu and William Lloyd and David J. Lythgoe and Maro G. Machizawa and Erin L. MacMillan and Richard J. Maddock and Andrei V. Manzhurtsev and María L. Martinez-Gudino and Jack J. Miller and Heline Mirzakhanian and Marta Moreno-Ortega and Paul G. Mullins and Shinichiro Nakajima and Jamie Near and Ralph Noeske and Wibeke Nordhøy and Georg Oeltzschner and Raul Osorio-Duran and María Concepción García Otaduy and Erick H. Pasaye and Ronald Peeters and Scott J. Peltier and Ulrich Pilatus and Nenad Polomac and Eric C. Porges and Subechhya Pradhan and James Joseph Prisciandaro and Nicolaas A. Puts and Caroline D. Rae and Francisco Reyes-Madrigal and Timothy P. L. Roberts and Caroline E. Robertson and Jens T. Rosenberg and Diana-Georgiana Rotaru and Ruth L. O'Gorman Tuura and Muhammad G. Saleh and Kristian Sandberg and Ryan Sangill and Keith Schembri and Anouk Schrantee and Natalia A. Semenova and Debra Singel and Rouslan Sitnikov and Jolinda Smith and Yulu Song and Craig E. L. Stark and Diederick Stoffers and Stephan P. Swinnen and Rongwen Tain and Costin Tanase and Sofie Tapper and Martin Tegenthoff and Thomas Thiel and Marc Thioux and Peter Truong and Pim van Dijk and Nolan Vella and Rishma Vidyasagar and Andrej Vovk and Guangbin Wang and Lars T. Westlye and Timothy K. Wilbur and William R. Willoughby and Martin Wilson and Hans-Jörg Wittsack and Adam J. Woods and Yen-Chien Wu and Junqian Xu and Maria Yanez Lopez and David Ka Wai Yeung and Qun Zhao and Xiaopeng Zhou and Gasper Zupan},
  title = {{Frequency drift in MR spectroscopy at 3T}},
  journal = {NeuroImage},
  year = {2021},
  doi = {10.1016/j.neuroimage.2021.118430},
}

Loading related content...

Figure