An Exploration of Parallel Imaging System for Very-Low-Field (50 mT) MRI Scanner

Lei Yang, Wei He, Sheng Shen, Yucheng He, Jiamin Wu, Zheng XuIEEE Trans. Instrumentation & Measurement 2025 · DOI

TL;DR

This paper introduces a novel parallel imaging (PI) technique, KF-NL-GRAPPA, specifically designed for very-low-field (VLF) MRI scanners. By combining a linear-prediction based Kalman filter for denoising with a nonlinear GRAPPA reconstruction and a custom 8-channel phased-array coil, they achieve significantly improved image quality and SNR at half the scan time compared to existing methods, paving the way for more accessible and efficient VLF MRI.

What Problem Does This Paper Address?

Magnetic Resonance Imaging (MRI) is a powerful diagnostic tool, but high-field MRI scanners are expensive and complex. Very-low-field (VLF) MRI scanners offer a more affordable and portable alternative, but they suffer from significantly weaker MR signals. This low signal-to-noise ratio (SNR) necessitates longer scan times to achieve diagnostic image quality, which can be uncomfortable for patients and reduce scanner efficiency. Parallel imaging (PI) techniques, which use multiple coil elements to accelerate data acquisition, are a standard solution in high-field MRI. However, directly applying existing PI methods and coil designs to VLF MRI is challenging due to the fundamental differences in magnetic field orientation and the even greater impact of noise. This paper aims to bridge this gap by developing a tailored PI system for VLF MRI that overcomes these limitations.

Methods

The authors propose a two-pronged approach:

  • KF-NL-GRAPPA Algorithm: This novel PI algorithm addresses the noise and signal limitations of VLF MRI.
  • Linear-Prediction based Kalman Filter (LP-KF): This component is used to denoise the acquired undersampled k-space data. It leverages linear prediction to model the signal and a Kalman filter to estimate the true signal while suppressing noise.
  • Nonlinear GRAPPA (NL-GRAPPA): The denoised data is then fed into a modified GRAPPA algorithm. Unlike standard GRAPPA, which uses a linear model, this NL-GRAPPA employs a polynomial feature mapping to capture more complex relationships between acquired and missing k-space data, leading to more accurate reconstruction.
  • 8-Channel Phased-Array Coil Design: A custom 8-channel phased-array RF coil was designed specifically for VLF MRI.
  • Configuration: The coil's geometry was optimized for head imaging, balancing filling factor for good SNR with subject comfort. It incorporates two 8-shaped coil units to improve image information in the top and bottom regions.
  • Decoupling: Geometric overlap and LII preamplifier decoupling techniques were used to minimize crosstalk between coil elements, which is crucial for maintaining the independence of coil sensitivities and enabling higher acceleration factors.

Datasets

The study utilized data from healthy volunteers for in-vivo head scanning experiments. No new benchmark datasets were introduced; instead, the performance of the proposed method was compared against existing methods on data acquired from a 50mT VLF MRI scanner.

Key Insights and Ideas

  • VLF MRI requires specialized PI: The significantly lower SNR and different magnetic field orientation in VLF MRI necessitate a departure from direct application of high-field PI techniques and coil designs.
  • Noise suppression is paramount: The Kalman filter's ability to effectively suppress Gaussian white noise and Johnson noise is critical for improving the quality of VLF MRI images, even surpassing the SNR of fully sampled images.
  • Nonlinearity improves reconstruction: Moving beyond linear models in GRAPPA (using polynomial feature mapping) allows for a more accurate estimation of missing k-space data, especially in the presence of noise.
  • Coil design matters: A well-designed phased-array coil with effective decoupling is essential for maximizing the benefits of PI in VLF MRI. The inclusion of 8-shaped coils addressed a previously observed limitation in image coverage.
  • Half the scan time, better or equal quality: The proposed method achieves comparable or even superior image quality to fully sampled images while reducing scan time by half, a significant improvement for patient comfort and scanner throughput.

Key Results

  • Image Quality: The KF-NL-GRAPPA method produced reconstructed images with superior quality compared to standard GRAPPA and NL-GRAPPA. Qualitative analysis showed fewer artifacts and better detail preservation.
  • SNR Improvement: Notably, the reconstructed images using the proposed method achieved higher SNR than the fully sampled images, a significant advantage for VLF MRI.
  • Scan Time Reduction: The method achieved a reduction factor of 2.03, enabling scans to be completed in half the time of fully sampled acquisitions.
  • Quantitative Metrics:
    • MSE: The proposed method achieved the smallest MSE (12.73) compared to GRAPPA (18.09) and NL-GRAPPA (16.82).
  • PSNR: The highest PSNR was achieved by KF-NL-GRAPPA (37.39 dB) compared to GRAPPA (35.55 dB) and NL-GRAPPA (36.27 dB).
  • SSIM: KF-NL-GRAPPA also yielded the highest SSIM score (0.93) compared to GRAPPA (0.90) and NL-GRAPPA (0.91).
  • Coverage: The new coil design with 8-shaped elements resolved the issue of missing image information in the top and bottom regions observed in previous VLF MRI coil designs.

Limitations and Future Work

  • Reduction Factor: The authors acknowledge that the reduction factor achieved (2.03) is relatively low and that further improvements are possible.
  • Coil Optimization: Future work could focus on optimizing the phased-array coil further by:
  • Improving decoupling techniques, as LII preamplifiers may not be ideal for eliminating crosstalk between nonadjacent coils. Novel decoupling technologies like magnetic walls or clip-path conductors are suggested.
  • Exploring more receiver channels to increase the number of coil units along the parallel imaging direction, which theoretically increases the reduction factor.
  • Further optimizing coil unit structure for better geometric overlap and reduced g-factor.
  • Gradient Coil Linearity: The linearity of the gradient coils can impact reconstruction accuracy, and improvements in this area could allow for higher reduction factors.

Summary updated

BibTeX
@article{yang2025exploration,
  author = {Lei Yang and Wei He and Sheng Shen and Yucheng He and Jiamin Wu and Zheng Xu},
  title = {{An Exploration of Parallel Imaging System for Very-Low-Field (50 mT) MRI Scanner}},
  journal = {IEEE Trans. Instrumentation \& Measurement},
  year = {2025},
  doi = {10.1109/tim.2025.3546402},
}
Figure