MORSE-PI -- Flexible and artefact-free image reconstruction for structural and functional QSM and other phase-critical imaging applications
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: We're looking at the paper: "MORSE-PI -- Flexible and artefact-free image reconstruction for structural and functional QSM and other phase-critical imaging applications".
Mira: Here's the gist of the paper. Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo acquisitions.
Kai: We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR, artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.
Mira: Purpose of the Method MORSE-PI is introduced to address the sensitivity of phase imaging applications like QSM to noise amplifications, phase singularities, and other artefacts in challenging scenarios.
Kai: The method aims to produce high-SNR, artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.
Paper summary: Kai: So we're looking at this paper called "MORSE-PI -- Flexible and artefact-free image reconstruction for structural and functional QSM and other phase-critical imaging applications".
Mira: This paper is addressing how sensitive phase imaging techniques are to noise amplification, phase singularities, and other artifacts, especially in tough spots like ultra-high field seven tesla or when you only get a single echo.
Lev: So the main goal of this work is to create a way to get high signal-to-noise ratio images that are free from those kinds of problems for both structural structural imaging like SWI and QSM, and functional ones called fQSM.
Kai: Exactly, they introduce this new method called MORSE-PI to tackle those sensitivity issues head on.
Mira: The core idea is that it produces phase images that are high signal-to-noise ratio, free of artifacts, and without those pesky singularities.
Lev: They claim this works for both structural and functional phase based brain imaging, which is a big deal because those applications really need clean phase information to be useful.
Kai: The method extends a previous approach called MORSE by adding something they call a Virtual Reference Coil or VRC.
Mira: This VRC isn't just some random thing; it's built as a linear combination of the coil sensitivity maps, and they enhance the correlations between those coil elements using the noise covariance matrix.
Lev: They say this Virtual Reference Coil ensures robust signal support across the entire brain and is used to correct phase offsets in those MORSE derived coil sensitivity estimates, which results in artefact-free, high signal-to-noise ratio phase.
Kai: What's interesting is they build this VRC from smooth MORSE derived coil sensitivities instead of using individual coil images for construction.
Mira: This smoothing step is what helps them enhance the intercoil correlation between the coils, which strengthens that Virtual Reference Coil idea.
Lev: So, if you were trying to run this on real hardware, you’d need those smooth sensitivities calculated first before you can build that VRC structure.
Paper summary: Kai: Compared to other methods they looked at, like GRAPPA with ASPIRE phase correction or ESPIRiT and GRAPPA combined methods, MORSE-PI shows better robustness against things like noise amplification and aliasing.
Mira: They also see improved reproducibility in structural imaging when they test it at both three tesla and seven tesla.
Lev: That improved reproducibility is important because when you're trying to build real hardware, you need results that are consistent across different field strengths, not just one perfect setup.
Kai: They specifically point out that unlike ESPIRiT or GRAPPA combined with adaptive coil combination methods, MORSE-PI yields singularity free phase maps.
Mira: That means they are tackling a specific problem in phase imaging where you can get those weird sharp spikes in the data, and MORSE-PI seems to handle that better than what those other combinations do.
Lev: From an engineering standpoint, getting singularity free phase maps is a big win because it simplifies downstream processing; you don't have to spend so much time trying to remove these specific artifacts before you can even start the main reconstruction.
Kai: So, this method is designed to make the whole pipeline smoother from a computational perspective.
Mira: And they show high signal-to-noise ratio reconstructions even in really tough scenarios, like single echo EPI acquisitions at seven tesla.
Lev: That single echo scenario is particularly hard because you lose a lot of information, so achieving high SNR there is quite impressive.
Kai: It seems like the paper shows that MORSE-PI provides high signal-to-noise ratio, fold over free and singularity free phase images across various acquisitions.
Mira: This flexibility means it works for single echo and multi echo structural gradient echo scans as well as functional EPI scans.
Lev: And the method doesn't require you to add an extra reference body coil or multi echo reference scans for this to work, which is a nice thing.
Kai: The method also works with a wide range of B zero map estimation techniques and background field removal methods, and even dipole inversion algorithms and denoising pipelines.
Paper summary: Mira: It's quite versatile when you put it into a larger framework of image processing tools.
Lev: That versatility means it could potentially be integrated into many existing MRI analysis software packages without needing a complete overhaul of the reconstruction pipeline.
Kai: So, the MORSE-PI paper is about developing this flexible reconstruction method that delivers high quality phase images for various acquisitions at different field strengths.
Mira: It’s specifically targeting those hard cases where noise and artifacts usually kill the image quality in QSM or fQSM.
Lev: The main implication for real hardware is that if you can implement this VRC construction smoothly, you get more reliable results from your expensive high-field scanners.
Kai: So it boils down to using a Virtual Reference Coil built cleverly with noise covariance information to correct phase errors and keep the signal strong across the brain.
Mira: It’s about making the reconstruction process less dependent on having perfect coil data or ideal acquisition settings for those tricky phase sensitive measurements.
Lev: The limitation they mention is that you still need that initial noise calibration scan to get all this going, which is something you have to do anyway in many protocols.
Kai: So, it's a method that offers high quality phase images without needing extra scans, as long as you do the initial calibration scan.
Mira: This means it’s a practical tool for researchers wanting better results from QSM and fQSM studies on high field systems.
Lev: It’s about getting more consistent and artifact free data from those challenging acquisitions.
Kai: So, to wrap up this MORSE-PI paper, it's about using a Virtual Reference Coil enhanced by noise covariance to produce singularity free phase images for structural and functional QSM.
Mira: It shows better robustness at both three tesla and seven tesla compared to GRAPPA with ASPIRE phase correction.
Lev: The paper points out that you still need that initial noise calibration scan for this reconstruction method to function correctly.
Kai: That’s the essence of MORSE-PI, focusing on creating a flexible and artefact free way to get phase images for QSM and fQSM.
Conclusion: Kai: So this paper is about MORSE-PI, which stands for something that lets you reconstruct images without those nasty phase artifacts in QSM and fQSM scans.
Mira: Yeah, the authors are Dymerska and others, looking at how to make those phase images really clean across different field strengths.
Lev: They’re aiming for high signal-to-noise ratio results that actually hold up when you take things to seven tesla or when you only get a single echo scan.
Kai: It seems like the title itself says it's about being flexible and artifact free, which is exactly what they claim this method achieves.
Mira: The authors are showing how their approach tackles those noise amplifications and phase singularities that usually ruin structural and functional scans.
Lev: What matters for us is that they demonstrate improved reproducibility in structural imaging at both three tesla and seven tesla, which is a tough comparison to make.
Kai: So what this means for us on the hardware side is that if you’re running an expensive scanner, this method could give you more consistent results without needing extra complex setups.
Mira: It suggests that by using a virtual reference coil built with noise information, we can get high quality phase images without having to rely on perfect coil data or extra reference scans.
Lev: The catch is they still need that initial noise calibration scan, which is something you usually have to do anyway in many protocols.
Kai: And that's the main thing they flag—it’s not a magic fix; it depends on having that calibration measurement first.
Mira: It lays out a solid framework for how we can handle phase-critical imaging more reliably, even in challenging scenarios like single echo EPI scans.
Lev: So it’s less about inventing a whole new physical principle and more about clever mathematical correction using the noise covariance matrix.
Kai: It’s basically an advanced way to correct the phase offsets by building this smart virtual reference coil that understands the noise correlations.
Mira: It really shows how important those underlying assumptions are, because if your initial assumptions about the noise model are off, you lose that high SNR.
Lev: This makes me think about how we need to ensure our error correction models account for those non-ideal noise environments in real hardware testing.
Kai: So it's a solid step forward in making these complex imaging techniques more usable for actual clinical and research settings at higher fields.
Barbara Dymerska, Oliver Josephs, Benjamin James, Vahid Malekian, Nadine N. Graedel, Martina F. Callaghan
University College London
physics.med-ph
Submitted: 2026-06-19
Updated: 2026-06-19
Comments: 33 pages including supplementary material
Journal ref: Dymerska, B., Josephs, O., James, B. et al. MORSE-PI: flexible and artefact-free image reconstruction for structural and functional QSM and other phase-critical imaging applications. Magn Reson Mater Phy (2026)
DOI: 10.1007/s10334-026-01434-8
Code: https://github.com/fil-physics/gadgetron-matlab
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 89/100
The gist: Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and other artefacts…
Key concepts
- Virtual Reference Coil (VRC)
- The VRC is a mathematical construct created by combining coil sensitivity maps. Crucially, its elements are correlated using the noise covariance matrix to ensure robust signal support across the entire brain. This VRC helps correct phase offsets in coil sensitivity estimates, leading to cleaner, high-SNR phase images.
- Noise Covariance Matrix
- This matrix quantifies how noise varies across different MRI coils and locations. By using this information, MORSE-PI enhances the spatial correlation between neighboring coil sensitivities before calculating the VRC. This enhancement ensures the VRC has consistent signal support throughout the field of view.
- Phase Singularities
- Phase singularities are specific artifacts that appear in phase images, often caused by noise or reconstruction errors. MORSE-PI is specifically designed to eliminate these singularities, resulting in singularity-free phase maps. This is vital for accurate quantitative measurements like QSM.
- Structural and Functional QSM (SWI/fQSM)
- Quantitative Susceptibility Mapping (QSM) measures magnetic susceptibility in the brain tissue. MORSE-PI is applied to both structural (SWI, QSM) and functional (fQSM) phase-based brain imaging. The method provides the high-quality phase images necessary for these techniques to accurately map tissue properties.
Terminology
Summary
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo acquisitions. p. 1 Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging. Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.Where the paper says it
Where the paper says it
Purpose of the Method
MORSE-PI is introduced to address the sensitivity of phase imaging applications like QSM to noise amplifications, phase singularities, and other artefacts in challenging scenarios. The method aims to produce high-SNR, artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging. p. 1 Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging. Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.Where the paper says it
Where the paper says it
Core Reconstruction Mechanism
MORSE-PI extends the previous approach, MORSE, by introducing a Virtual Reference Coil (VRC). The VRC is constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements using the noise covariance matrix. p. 1 Methods:
MORSE-PI extends our previous approach, MORSE, by introducing a Virtual Reference Coil (VRC). The VRC is
constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements
using the noise covariance matrix. Methods:
MORSE-PI extends our previous approach, MORSE, by introducing a Virtual Reference Coil (VRC). The VRC is
constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements
using the noise covariance matrix. The VRC is
constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements
using the noise covariance matrix. Such a VRC ensures robust signal support across the entire brain and is used
to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR
phase. The VRC is
constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements
using the noise covariance matrix. Such a VRC ensures robust signal support across the entire brain and is used
to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR
phase. 11
, the original VRC approach often leads to phase singularities, which can move around
depending on the selection of the seed voxel or the weights used for the linear combination of the signal from
the separate coils. To improve the robustness of the VRC approach in MORSE-PI, we construct the VRC from smooth MORSEderived coil sensitivities, rather than from individual coil images, having enhanced the inter-coil correlation,Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Key Improvements Over Previous Methods
MORSE-PI offers several advantages over existing methods in phase imaging. Specifically, it demonstrates greater robustness to artefacts such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T and 7T. p. 1 Such a VRC ensures robust signal support across the entire brain and is used
to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR
phase. Results:
Compared to GRAPPA with ASPIRE phase correction, MORSE-PI demonstrates greater robustness to artefacts
such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T
and 7T. Results:
Compared to GRAPPA with ASPIRE phase correction, MORSE-PI demonstrates greater robustness to artefacts
such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T
and 7T. Unlike ESPIRiT and GRAPPA combined with adaptive coil combination methods, MORSE-PI yields
singularity-free phase maps. Unlike ESPIRiT and GRAPPA combined with adaptive coil combination methods, MORSE-PI yields
singularity-free phase maps. MORSE-PI enables high-SNR reconstructions even for the most challenging
scenarios, such as single-echo EPI at 7T.Where the paper says it
Where the paper says it
Where the paper says it
Performance and Validation
The method was evaluated across various field strengths and acquisition types. The comparison with alternative image reconstruction schemes showed that MORSE-PI produced singularity-free phase images and high-quality CLEAR-SWI and QSM reconstructions. p. 9 2 Structural imaging at 7T and 3T - comparison with alternative image reconstruction schemes
MORSE-PI was compared with the original MORSE implementation, as well as methods commonly used for
phase imaging: vendor-provided GRAPPA26 + Adaptive Combine27 (with and without SVD) and state-of-the-art
phase imaging with GRAPPA26 + ASPIRE14
, and ESPIRiT
18
. The GRAPPA + Adaptive Combine (with and without
SVD) and GRAPPA + ASPIRE images were reconstructed using their default image reconstruction chains (via
the Siemens Image Calculation Environment) on the native scanner hardware via the “retrospective
reconstruction” facility. 2 Structural imaging at 7T and 3T - comparison with alternative image reconstruction schemes
MORSE-PI was compared with the original MORSE implementation, as well as methods commonly used for
phase imaging: vendor-provided GRAPPA26 + Adaptive Combine27 (with and without SVD) and state-of-the-art
phase imaging with GRAPPA26 + ASPIRE14
, and ESPIRiT
18
. The GRAPPA + Adaptive Combine (with and without
SVD) and GRAPPA + ASPIRE images were reconstructed using their default image reconstruction chains (via
the Siemens Image Calculation Environment) on the native scanner hardware via the “retrospective
reconstruction” facility. Such a VRC ensures robust signal support across the entire brain and is used
to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR
phase. Results:
Compared to GRAPPA with ASPIRE phase correction, MORSE-PI demonstrates greater robustness to artefacts
such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T
and 7T. Such a VRC ensures robust signal support across the entire brain and is used
to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR
phase. Results:
Compared to GRAPPA with ASPIRE phase correction, MORSE-PI demonstrates greater robustness to artefacts
such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T
and 7T.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Conclusion and Applicability
MORSE-PI flexibly provides high SNR, fold-over-free and singularity-free phase images, as demonstrated for single-echo and multi-echo structural GRE and functional EPI scans. It is well-suited for phasebased imaging techniques such as structural and functional QSM. p. 1 Conclusion:
MORSE-PI offers a flexible and computationally efficient solution for generating high-SNR, artefact- and
singularity-free phase images in both single- and multi-echo GRE and EPI acquisitions. This makes it particularly
well-suited for structural and functional QSM, as well as other phase-based MRI applications. Using the noise covariance matrix to ensure good correlation between coil
sensitivities, prior to VRC formation, provides good VRC magnitude and phase support, typically over the entire
brain without phase singularities, as explained below. In many MRI applications, the noise covariance matrix is calculated from short noise measurements (without
any excitation of MR signal) prior to running the imaging sequence, and used to decorrelate, or whiten, the
complex signals from separate coils.Where the paper says it
Where the paper says it
Code and Data Availability Statement
The open-source code, as well as scripts necessary to compile a Gadgetron image and create a Docker container encapsulating the MORSE-PI reconstruction, are available on https://github.com/fil-physics/gadgetron-matlab. p. 6 Methods
The MORSE-PI image reconstruction scheme has been deployed for a diverse range of 3T and 7T functional
and quantitative structural imaging studies at The Functional Imaging Laboratory (The FIL), University College
London. The open-source code, along with scripts for creating a Docker container that encapsulates the
MORSE-PI reconstruction within the Gadgetron framework19, is available on GitHub23
. For all structural data, the MORSE and MORSE-PI reconstructions used:
𝑁𝑟𝑒𝑓 = 6, 𝑁𝑜𝑟𝑑𝑒𝑟 = 4, gaussian smoothing filter 𝑤 = 6 voxels, and global regularisation factor of 𝜆 = 3x10-4
,
as deployed in various in vivo studies at The FIL17
. Structural images were acquired at 3T and 7T using an in-house, but openly available, 3D multi-echo spoiled
gradient echo acquisition configured with either proton density (PD), T1 or magnetisation transfer (MT)
weighting. Table 1 Sequence parameters for 3T and 7T GRE and EPI scans on which MORSE-PI reconstruction was
evaluated. Protocol for Structural Imaging: 3D Gradient Echo
Field
[T]
Resolution
[mm3
]
Acceleration Reference data
Matrix
[HF X AP x RL]
TR
[ms]
TE, 𝚫TE
[ms]
TA
[min:s]
7
0. For all structural data, the MORSE and MORSE-PI reconstructions used:
𝑁𝑟𝑒𝑓 = 6, 𝑁𝑜𝑟𝑑𝑒𝑟 = 4, gaussian smoothing filter 𝑤 = 6 voxels, and global regularisation factor of 𝜆 = 3x10-4
,
as deployed in various in vivo studies at The FIL17
. Structural images were acquired at 3T and 7T using an in-house, but openly available, 3D multi-echo spoiled
gradient echo acquisition configured with either proton density (PD), T1 or magnetisation transfer (MT)
weighting. The 3T data were acquired on a Siemens Healthineers Prismafit
(software version VE11c) using a body coil for transmission and a 64-channel head and neck array for
reception. The 7T data were obtained on a Siemens Healthineers Terra (software version VE12u-SP01) with a
Nova Medical head coil featuring 8 transmit and 32 receive channels, operated in a quadrature-like transmit
mode. Using the noise covariance matrix to ensure good correlation between coil
sensitivities, prior to VRC formation, provides good VRC magnitude and phase support, typically over the entire
brain without phase singularities, as explained below. In many MRI applications, the noise covariance matrix is calculated from short noise measurements (without
any excitation of MR signal) prior to running the imaging sequence, and used to decorrelate, or whiten, the
complex signals from separate coils. Full details of all acquisitions are provided in Table 1. MR phase imaging is commonly used for SWI and QSM, we therefore prepared image processing pipelines for
these contrasts and share them on GitHub28
. Methods
The MORSE-PI image reconstruction scheme has been deployed for a diverse range of 3T and 7T functional
and quantitative structural imaging studies at The Functional Imaging Laboratory (The FIL), University College
London. The open-source code, along with scripts for creating a Docker container that encapsulates the
MORSE-PI reconstruction within the Gadgetron framework19, is available on GitHub23
.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
References
Reichenbach JR, Barth M, Haacke EM, Klarhöfer M, Kaiser WA, Moser E. High-Resolution MR Venography at 3. p. 2 This contrast also underpins blood oxygenation level-dependent (BOLD) functional
MRI, which detects changes in blood oxygenation via associated susceptibility variations4
. With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution,
phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5
, Quantitative Susceptibility Mapping (QSM)6
, and functional QSM (fQSM)7
. This contrast also underpins blood oxygenation level-dependent (BOLD) functional
MRI, which detects changes in blood oxygenation via associated susceptibility variations4
. With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution,
phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5
, Quantitative Susceptibility Mapping (QSM)6
, and functional QSM (fQSM)7
. For all structural data, the MORSE and MORSE-PI reconstructions used:
𝑁𝑟𝑒𝑓 = 6, 𝑁𝑜𝑟𝑑𝑒𝑟 = 4, gaussian smoothing filter 𝑤 = 6 voxels, and global regularisation factor of 𝜆 = 3x10-4
,
as deployed in various in vivo studies at The FIL17
. Structural images were acquired at 3T and 7T using an in-house, but openly available, 3D multi-echo spoiled
gradient echo acquisition configured with either proton density (PD), T1 or magnetisation transfer (MT)
weighting. Introduction
Magnetic Resonance Imaging (MRI) is highly sensitive to subtle variations in magnetic susceptibility across the
human brain, particularly when using T2*-weighted acquisitions with gradient-echo (GRE) or echo-planar
imaging (EPI) readouts, and at high magnetic field strengths. These techniques enable the study of cerebral
vasculature1 as well as regional variations in tissue myelin and iron — two major contributors to magnetic
susceptibility contrast2,3
. MORSE-PI – Flexible and artefact-free image reconstruction for structural
and functional QSM and other phase-critical imaging applications. Barbara Dymerska1
, Oliver Josephs1
, Benjamin James1
, Vahid Malekian1
, Nadine N. Introduction
Magnetic Resonance Imaging (MRI) is highly sensitive to subtle variations in magnetic susceptibility across the
human brain, particularly when using T2*-weighted acquisitions with gradient-echo (GRE) or echo-planar
imaging (EPI) readouts, and at high magnetic field strengths. These techniques enable the study of cerebral
vasculature1 as well as regional variations in tissue myelin and iron — two major contributors to magnetic
susceptibility contrast2,3
. 3 MORSE-PI with Virtual Reference Coil
To remove ambiguous phase offsets and singularities we adopt the concept of a Virtual Reference Coil (VRC)20
with scalar phase matching at a seed voxel21,22
. In de Zwart et al. This contrast also underpins blood oxygenation level-dependent (BOLD) functional
MRI, which detects changes in blood oxygenation via associated susceptibility variations4
. With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution,
phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5
, Quantitative Susceptibility Mapping (QSM)6
, and functional QSM (fQSM)7
. This contrast also underpins blood oxygenation level-dependent (BOLD) functional
MRI, which detects changes in blood oxygenation via associated susceptibility variations4
. With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution,
phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5
, Quantitative Susceptibility Mapping (QSM)6
, and functional QSM (fQSM)7
. Protocol for Functional Imaging: 3D EPI
Field
[T]
Resolution
[mm3
]
Acceleration Reference data
Matrix
[HF x AP x RL]
TR
[ms]
TE
[ms]
TA
[s/vol]
7
0. 8
None, 8 inplane segments None
Whole brain
160 x 240 x240 100. The
broader adoption of these methods has been facilitated by advances in acquisition efficiency, image
reconstruction methods, and improved understanding of phase contrast mechanisms. It is now well
established that susceptibility contrast not only reflects tissue composition but also microstructural
orientation, as exploited by methods such as Susceptibility Tensor Imaging (STI)8
. It is now well
established that susceptibility contrast not only reflects tissue composition but also microstructural
orientation, as exploited by methods such as Susceptibility Tensor Imaging (STI)8
. In parallel, emerging
techniques like magnetic susceptibility source separation have provided increasing insight into sub-voxel
distributions of key biological substrates such as iron and myelin9,10
. Protocol for Functional Imaging: 3D EPI
Field
[T]
Resolution
[mm3
]
Acceleration Reference data
Matrix
[HF x AP x RL]
TR
[ms]
TE
[ms]
TA
[s/vol]
7
0. 8
None, 8 inplane segments None
Whole brain
160 x 240 x240 100. This contrast also underpins blood oxygenation level-dependent (BOLD) functional
MRI, which detects changes in blood oxygenation via associated susceptibility variations4
. With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution,
phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5
, Quantitative Susceptibility Mapping (QSM)6
, and functional QSM (fQSM)7
. Graedel1
, Martina F. Callaghan1
1The Functional Imaging Laboratory, Department of Imaging Neuroscience, UCL Queen Square Institute of
Neurology, University College London, UK
Running Title: MORSE-PI
Corresponding author:
Barbara Dymerska
Department of Imaging Neuroscience,
UCL Queen Square Institute of Neurology,
University College London,
London, WC1N 3AR, UK
Email: b. The QSM pipeline included phase unwrapping and field map calculation using ROMEO30
, hybrid masking using
ROMEO30 and SPM31 segmentation32
, background field removal via PDF33
, and dipole inversion using STARQSM34
. We used CLEAR-SWI and ROMEO from the MriResearchTools package (v4. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging. Methods:
MORSE-PI extends our previous approach, MORSE, by introducing a Virtual Reference Coil (VRC). 3 MORSE-PI with Virtual Reference Coil
To remove ambiguous phase offsets and singularities we adopt the concept of a Virtual Reference Coil (VRC)20
with scalar phase matching at a seed voxel21,22
. In de Zwart et al. Several solutions exist that perform well for GRE-based phase imaging, particularly at 3T and lower field
strengths, as reviewed in detail elsewhere11,13. However, the challenges increase at ultra-high fields (≥7T),
where the lack of a body coil (often used as a phase reference in SENSE-based reconstructions) and the steeper
coil sensitivity profiles complicate the generation of high-SNR, whole-brain phase images. In de Zwart et al. and Hammond et al. These artefacts can propagate into
the final combined phase image. We previously introduced MORSE15: a robust and computationally efficient method for coil sensitivity
estimation and regularised SENSE16 image reconstruction. Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging. With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution,
phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5
, Quantitative Susceptibility Mapping (QSM)6
, and functional QSM (fQSM)7
. The
broader adoption of these methods has been facilitated by advances in acquisition efficiency, image
reconstruction methods, and improved understanding of phase contrast mechanisms. Following this scalar phase matching, a VRC can be
formed by a linear combination of the phase signals of all the receiver coils20
. As discussed in the review paper
by Robinson et al. Unlike ESPIRiT and GRAPPA combined with adaptive coil combination methods, MORSE-PI yields
singularity-free phase maps. MORSE-PI enables high-SNR reconstructions even for the most challenging
scenarios, such as single-echo EPI at 7T. Following this scalar phase matching, a VRC can be
formed by a linear combination of the phase signals of all the receiver coils20
. As discussed in the review paper
by Robinson et al. Below, we describe the
implementation and summarize the performance evaluation carried out on representative in vivo data prior
to deployment. 1 Gadgetron and MATLAB implementation
Figure 1 visualises the computational steps of the MORSE-PI phase correction applied to the first order coil
sensitivities calculated using MORSE. Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Supplementary material
Figure S1 Noise covariance matrices illustrating the effects of whitening or increasing cross-channel correlation in a 64-channel coil at 3T. Figure S2 Reproducibility assessment at 7T and 3T for GRAPPA+ASPIRE and MORSE-PI image reconstructions of PD-weighted and T1-weighted scans. p. 9 ROMEO uses weighted echo averaging for field map calculation36, which is deemed the optimal
solution for QSM if the echo times are chosen appropriately13,37
. This solution can, however, lead to
propagation of phase singularities present in MORSE, GRAPPA + Adaptive Combine, and ESPIRiT
reconstructions to the final result. Unlike ESPIRiT and GRAPPA combined with adaptive coil combination methods, MORSE-PI yields
singularity-free phase maps. MORSE-PI enables high-SNR reconstructions even for the most challenging
scenarios, such as single-echo EPI at 7T. Below, we describe the
implementation and summarize the performance evaluation carried out on representative in vivo data prior
to deployment. 1 Gadgetron and MATLAB implementation
Figure 1 visualises the computational steps of the MORSE-PI phase correction applied to the first order coil
sensitivities calculated using MORSE. Below, we describe the
implementation and summarize the performance evaluation carried out on representative in vivo data prior
to deployment. 1 Gadgetron and MATLAB implementation
Figure 1 visualises the computational steps of the MORSE-PI phase correction applied to the first order coil
sensitivities calculated using MORSE. For all structural data, the MORSE and MORSE-PI reconstructions used:
𝑁𝑟𝑒𝑓 = 6, 𝑁𝑜𝑟𝑑𝑒𝑟 = 4, gaussian smoothing filter 𝑤 = 6 voxels, and global regularisation factor of 𝜆 = 3x10-4
,
as deployed in various in vivo studies at The FIL17
. Structural images were acquired at 3T and 7T using an in-house, but openly available, 3D multi-echo spoiled
gradient echo acquisition configured with either proton density (PD), T1 or magnetisation transfer (MT)
weighting.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Authors Contributions
Dymerska: Study conception and design, Code contributions, Acquisition design, Acquisition of data, Analysis and interpretation, Drafting of manuscript, Critical revision. Josephs: Code contributions, Analysis and interpretation. p. 2 In addition to magnetic susceptibility-induced
phase perturbations (i. , ΔB0-related phase terms), each coil also encodes phase contributions from its own
sensitivity profile, as well as system-related effects including frequency drifts, timing errors and eddy currents. In addition to magnetic susceptibility-induced
phase perturbations (i. , ΔB0-related phase terms), each coil also encodes phase contributions from its own
sensitivity profile, as well as system-related effects including frequency drifts, timing errors and eddy currents. MORSE-PI – Flexible and artefact-free image reconstruction for structural
and functional QSM and other phase-critical imaging applications. Barbara Dymerska1
, Oliver Josephs1
, Benjamin James1
, Vahid Malekian1
, Nadine N. MORSE-PI – Flexible and artefact-free image reconstruction for structural
and functional QSM and other phase-critical imaging applications. Barbara Dymerska1
, Oliver Josephs1
, Benjamin James1
, Vahid Malekian1
, Nadine N. Residual phase singularities in E remain only in
regions lacking magnitude support for the respective coil (see B1), and therefore do not propagate to the
final phase image. 2 In vivo data acquisition
MORSE-PI was evaluated using data collected following ethical approval from local Research Ethics Committee
and informed consent from study participants.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Compliance with ethical standars
Funding: This research was funded in whole, or in part, by the Wellcome Trust [203147/Z/16/Z and 226793/Z/22/Z].
Conflict of interest: The authors have no competing interests to declare that are relevant to the content of this article. p. 8 1 Impact of cross-coil correlation on VRC spatial support
The strength of the correlation between coil sensitivities can be adjusted by varying 𝑛 in Equation (5). We
tested values of 𝑛 = -1, 0, 1, 2, 3 to evaluate the VRC magnitude support across the regions of interest, namely
the brain and cervical spinal cord in 3T and 7T structural scans.Where the paper says it
Ethical approval: All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. p. 7 Residual phase singularities in E remain only in
regions lacking magnitude support for the respective coil (see B1), and therefore do not propagate to the
final phase image. 2 In vivo data acquisition
MORSE-PI was evaluated using data collected following ethical approval from local Research Ethics Committee
and informed consent from study participants.Where the paper says it
Informed consent: Informed consent was obtained from all individual participants included in the study. p. 7 Residual phase singularities in E remain only in
regions lacking magnitude support for the respective coil (see B1), and therefore do not propagate to the
final phase image. 2 In vivo data acquisition
MORSE-PI was evaluated using data collected following ethical approval from local Research Ethics Committee
and informed consent from study participants.Where the paper says it
-
Reichenbach JR, Barth M, Haacke EM, Klarhöfer M, Kaiser WA, Moser E. High-Resolution MR Venography at 3.0 Tesla. J Comput Assist Tomogr. 2000;24(6):949.
-
Li J, Lin H, Liu T, et al. Quantitative susceptibility mapping (QSM) minimizes interference from cellular pathology in R2 estimation of liver iron concentration. p. 2
J Magn Reson Imaging. p. 0Where the paper says it
This contrast also underpins blood oxygenation level-dependent (BOLD) functional MRI, which detects changes in blood oxygenation via associated susceptibility variations4 . With the increasing availability of high-field MRI systems (≥3T), there has been rapid growth in high-resolution, phase-based imaging techniques sensitive to the magnetic properties of tissue, such as SusceptibilityWeighted Imaging (SWI)5 , Quantitative Susceptibility Mapping (QSM)6 , and functional QSM (fQSM)7 .
2018;48(4):1069-1079 doi:10.1002/jmri.26019.Where the paper says it
MORSE-PI – Flexible and artefact-free image reconstruction for structural and functional QSM and other phase-critical imaging applications. Barbara Dymerska1 , Oliver Josephs1 , Benjamin James1 , Vahid Malekian1 , Nadine N.
-
Wei H, Dibb R, Decker K, et al. Investigating magnetic susceptibility of human knee joint at 7 tesla. p. 2
Magn Reson Med. Published online January 1, 2017:n/a-n/a doi:10. p. 2Where the paper says it
Introduction Magnetic Resonance Imaging (MRI) is highly sensitive to subtle variations in magnetic susceptibility across the human brain, particularly when using T2*-weighted acquisitions with gradient-echo (GRE) or echo-planar imaging (EPI) readouts, and at high magnetic field strengths. These techniques enable the study of cerebral vasculature1 as well as regional variations in tissue myelin and iron — two major contributors to magnetic susceptibility contrast2,3 .
1002/mrm.26596.Where the paper says it
Denser array coils further exacerbate this issue by producing more spatially localised and variable sensitivity patterns. A well-established reconstruction method for 7T phase imaging is ASPIRE, which is a fast online (i.
Reference IDs
arXiv id 2606.21336
Page 1 pg1
Page 2 pg2
Page 3 pg3
Page 4 pg4
Page 5 pg5
Page 6 pg6
Page 7 pg7
Page 8 pg8
Page 9 pg9 (Note: Page numbers are based on the provided text structure, not the actual arXiv numbering)
Summary of MORSE-PI
MORSE-PI flexibly provides high SNR, fold-over-free and singularity-free phase images, as demonstrated for single-echo and multi-echo structural GRE and functional EPI scans. The method naturally lends itself to imaging techniques, such as structural and functional QSM, that necessitate high-quality phase images. p. 1 Abstract:
Purpose:
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and
other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo
acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR,
artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.Where the paper says it
Pruning redundant reasoning tokens
MORSE-PI is introduced to address the sensitivity of phase imaging applications like QSM to noise amplifications, phase singularities, and other artefacts in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo acquisitions. The method aims to produce high-SNR, artefact-free, and singularity-free phase images for both structural (SWI, QSM) and functional (fQSM) phasebased brain imaging.
How the training signal is built
MORSE-PI extends the previous approach, MORSE, by introducing a Virtual Reference Coil (VRC). The VRC is constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements using the noise covariance matrix. This VRC ensures robust signal support across the entire brain and is used to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR phase. The VRC is constructed from smooth MORSEderived coil sensitivities, rather than from individual coil images, having enhanced the inter-coil correlation. p. 4 11
, the original VRC approach often leads to phase singularities, which can move around
depending on the selection of the seed voxel or the weights used for the linear combination of the signal from
the separate coils. To improve the robustness of the VRC approach in MORSE-PI, we construct the VRC from smooth MORSEderived coil sensitivities, rather than from individual coil images, having enhanced the inter-coil correlation,Where the paper says it
Where the gains come from
The gains come from leveraging the noise covariance matrix to enhance the spatial correlation between neighbouring coil sensitivities prior to VRC calculation. This ensures that the VRC, calculated as the complex sum of coil sensitivities, has consistent support across the entire field of view. p. 7 The VRC sensitivity in step D is then
computed as the complex sum of the correlated coil sensitivities, using scalar phase matching as described
in Equation (6). This process ensures robust VRC support across the entire brain (see VRC magnitude image
in D). Therefore, having initially decorrelated the separate coil signals, i. ,
whitened the data, we subsequently use the lower triangular matrix to enhance the correlation between the
first order MORSE-derived coil sensitivities:
𝒃̃𝑟,1 = 𝑳
𝑛𝒃̂
𝑟,1
(5)
where 𝒃̂
𝑟,1 ∈ ℂ
𝑁coil is a vector with the original first order coil sensitivity estimates in physical coil space at a
given spatial location 𝑟. 𝑬̂𝑟
v
= 𝑼𝑟𝑺𝑟𝑽𝑟
𝐻 = ∑𝜎𝑟,𝑘𝒖𝑟,𝑘𝒗𝑟,𝑘
𝐻
𝑁𝑟𝑒𝑓
𝑘=1
(3)
The voxel-wise decomposition in Equation (3) produces 𝑁ref pairs, indexed by 𝑘, of right singular vectors
𝒗𝑟,𝑘 𝜖 ℂ
𝑁ref and left singular vectors 𝒖𝑟,𝑘 𝜖 ℂ
𝑁coil
, with singular values, 𝜎𝑟,𝑘𝜖 ℝ+. A subset of left singular
vectors, (𝑁order ≤ 𝑁ref) with singular values substantially greater than zero, is used as “higher-order”
sensitivity estimates accounting for situations such as chemical shifts, rapidly varying sensitivities (e.Where the paper says it
Where the paper says it
Where the paper says it
MORSE-PI demonstrates greater robustness to artefacts such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T and 7T. The comparison with alternative image reconstruction schemes showed that MORSE-PI produced singularity-free phase images and high-quality CLEAR-SWI and QSM reconstructions. Reproducibility assessments indicated that QSM results reconstructed using MORSE-PI exhibited reduced variability compared to those obtained with GRAPPA+ASPIRE, indicative of improved reproducibility.
MORSE-PI is a highly flexible reconstruction method that delivers high-quality magnitude and phase images at both 3T and 7T for various acquisitions. It does not require additional reference body coil or multi-echo reference scans. p. 2 , on the
MRI console) approach that typically produces high-quality phase images from multi-echo GRE data without
requiring additional reference scans or a body coil14. However, ASPIRE requires at least two echoes, and its
phase image quality strongly depends on the underlying reconstruction of individual coil images, which may
be affected by artefacts from parallel imaging methods such as GRAPPA. bias field removal) and Laplacian phase unwrapping, and two options adjusted to the resolution of our
data: high-pass filter of size [2 2 0] voxels and Minimal Intensity Projection (MIP) across 14 axial slices (for
depicting vessel networks). The QSM pipeline included phase unwrapping and field map calculation using ROMEO30
, hybrid masking using
ROMEO30 and SPM31 segmentation32
, background field removal via PDF33
, and dipole inversion using STARQSM34
.Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Virtual Reference Coil (VRC) construction enhancement: The system can construct a VRC from
smooth MORSE-derived coil sensitivities, rather than from individual coil images
to ensurespatial homogeneity of the VRC
and avoid regions without support like ear canals or large vessels. -
Cross-coil correlation control: The AI system can leverage the noise covariance matrix to enhance spatial correlation between neighboring coil sensitivities by applying the inverse of the pre-whitening matrix, controlled by a power n (n=1 being sufficient), to ensure
VRC magnitude and phase support across the entire brain and cervical spinal cord.
-
Singularity-free phase output for challenging acquisitions: The improved system can generate
singularity-free images
from single-echo 3D EPI at 7T, which currently requires additional reference scans using existing methods. -
Superior QSM reconstruction quality: The system can produce
high SNR, fold-over-free and singularity-free phase images
for structural and functional QSM applications at both 3T and 7T, outperforming GRAPPA + ASPIRE in QSM due to lower noise floors. -
Robustness against acquisition artifacts: The reconstruction is shown to be unaffected by
local noise amplification
observed in GRAPPA reconstructions, leading tosubstantially lower noise floor
across the brain and higher quality CLEAR-SWI and QSM outputs.
Abstract
Phase imaging applications such as QSM are highly sensitive to noise amplifications, phase singularities, and other artefacts, particularly in challenging scenarios such as ultra-high field (7T), under-sampled or single-echo acquisitions. We present a novel image reconstruction method, MORSE-PI, designed to produce high-SNR, artefact-free, and singularity-free phase images for both structural and functional phase-based brain imaging. MORSE-PI extends our previous approach, MORSE, by introducing a Virtual Reference Coil (VRC). The VRC is constructed as a linear combination of coil sensitivity maps, with correlations enhanced between coil elements using the noise covariance matrix. Such a VRC ensures robust signal support across the entire brain and is used to correct phase offsets in the MORSE-derived coil sensitivity estimates, resulting in artefact-free, high SNR phase. Compared to GRAPPA with ASPIRE phase correction, MORSE-PI demonstrates greater robustness to artefacts such as noise amplification and aliasing, and shows improved reproducibility in structural imaging at both 3T and 7T. Unlike ESPIRiT and GRAPPA combined with adaptive coil combination methods, MORSE-PI yields singularity-free phase maps. MORSE-PI enables high-SNR reconstructions even for the most challenging scenarios, such as single-echo EPI at 7T. Its efficient, containerised implementation using the Gadgetron framework supports deployment on the MRI scanner console during measurements. MORSE-PI offers a flexible and computationally efficient solution for generating high-SNR, artefact- and singularity-free phase images in both single- and multi-echo GRE and EPI acquisitions. This makes it particularly well-suited for structural and functional QSM, as well as other phase-based MRI applications. Its robustness and rapid computational time facilitate efficient deployment on scanners across field strengths.
Sources
Related papers
- BART Online Open-Source Sequence Toolbox for Computational MRI
- Constrained estimation of rotational invariants of the cumulant expansion (RICE) for rapid tensor-valued diffusion MRI
- Improved Myelin Water Fraction Estimation and Uncertainty Mapping in Multicomponent T2 Relaxometry via NNLS Bootstrap Aggregation
- Radiation damage to normal mammalian tissue in vivo with laser-driven protons at ultra-high instantaneous dose rate
- Fast ungated five-dimensional cardiac MRI on a 1.5 T MR-linac for MRI-guided radiotherapy
- BEAM3R: Beam's-eye-view architecture with Mamba-3 for implicit dose reconstruction