Rapid online deep artifact suppression for real-time spiral bSSFP CMR with blipped-CAIPI simultaneous multi-slice imaging at 1.5 T

Daniel Knight; Einar Heiberg; Iulius Dragonu; Jennifer Steeden; Julius {\AA}kesson; Rebecca Baker; Ruta Virsinskaite; Tina Yao; Vivek Muthurangu

arxiv: 2605.26127 · v1 · pith:EUK7U4Z4new · submitted 2026-05-18 · ⚛️ physics.med-ph · cs.LG· eess.IV

Rapid online deep artifact suppression for real-time spiral bSSFP CMR with blipped-CAIPI simultaneous multi-slice imaging at 1.5 T

Julius {\AA}kesson , Iulius Dragonu , Einar Heiberg , Tina Yao , Rebecca Baker , Ruta Virsinskaite , Daniel Knight , Vivek Muthurangu

show 1 more author

Jennifer Steeden

This is my paper

Pith reviewed 2026-06-30 18:16 UTC · model grok-4.3

classification ⚛️ physics.med-ph cs.LGeess.IV

keywords real-time MRIsimultaneous multi-slicedeep artifact suppressionspiral bSSFPcardiac MRIcompressed sensingventricular volumesonline reconstruction

0 comments

The pith

Deep artifact suppression with a 3D U-Net enables online reconstruction of real-time SMS spiral bSSFP cardiac MRI in 30 seconds.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that a spiral bSSFP sequence with blipped-CAIPI simultaneous multi-slice acquisition can cover the short-axis heart in 15 seconds of free breathing. After k-space slice separation, a 3D U-Net removes residual artifacts in image space, completing reconstruction in 30 seconds versus nearly 25 minutes for compressed sensing. Image quality scores and quantitative metrics favor the deep-learning approach, and left and right ventricular volumes agree closely with standard breath-hold references. The work therefore converts an otherwise offline iterative method into a practical online pipeline for functional assessment.

Core claim

RT-SMS bSSFP with deep artifact suppression achieves 13-fold faster acquisition than breath-hold imaging and 50-fold faster reconstruction than compressed sensing, while delivering superior image quality and ventricular volume measurements that match breath-hold results within a few milliliters.

What carries the argument

The 3D U-Net applied for deep artifact suppression in image space after initial k-space slice separation.

If this is right

Full short-axis coverage becomes feasible in a single 15-second free-breathing scan.
Reconstruction finishes fast enough for immediate review during the exam.
Quantitative ventricular volumes remain within a few milliliters of breath-hold values.
Image quality metrics exceed those of compressed-sensing reconstruction on the same data.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same separation-plus-U-Net pipeline could be tested on other non-Cartesian trajectories that currently require slow iterative reconstruction.
If the network fails on arrhythmic patients, retraining on mixed healthy-plus-patient data would be a direct next step.
Higher slice acceleration factors become clinically realistic once reconstruction time is no longer the bottleneck.

Load-bearing premise

The network trained on ten healthy volunteers will maintain performance on patients who have pathology, implants, or irregular rhythms.

What would settle it

A test on a cohort of patients with known pathology showing either visibly worse image quality or ventricular volume biases exceeding 15 ml would falsify the claim of maintained diagnostic quality.

read the original abstract

Purpose: Real-time (RT) bSSFP MRI enables fast free-breathing cardiovascular imaging but requires 10-16 slices for functional assessment, resulting in prolonged scan times. Simultaneous multi-slice (SMS) imaging can reduce acquisition time but when combined with non-Cartesian trajectories, it relies on iterative reconstructions that preclude online use. This study investigates deep artifact suppression to facilitate rapid, online reconstruction of RT-SMS. Methods: A spiral bSSFP SMS RT sequence with two simultaneously acquired slices was implemented at 1.5 T. Reconstruction used slice separation in k-space, followed by deep artifact suppression in image space using a 3D U-Net. Ten healthy volunteers were imaged. RT-SMS image quality and reconstruction time were compared between deep artifact suppression and compressed sensing (CS) reconstructions. Left (LV) and right (RV) ventricular volumes at end diastole (EDV) and end systole (ESV) and LV mass (LVM) were compared between RT-SMS with deep artifact suppression and reference-standard breath-hold (BH) imaging. Results: The RT-SMS acquisition was ~13x faster than BH imaging (15 s vs 3 min 15 s). RT-SMS reconstruction using deep artifact suppression was ~50x faster than CS (30 s vs 24 min 55 s). Deep artifact suppression consistently outperformed CS in quantitative and qualitative image quality (p<0.001). Functional agreement between BH and RT-SMS with deep artifact suppression was good (LVEDV: -7.5 +/- 6.8 ml, LVESV: -0.9 +/- 4.2 ml, RVEDV: -6.4 +/- 8.4 ml, RVESV: 0.2 +/- 10.7 ml, LVM: -10.3 +/- 11.0 g). Conclusion: Online deep artifact suppression reconstruction for RT-SMS bSSFP CMR enables free-breathing short-axis coverage with a substantial reduction in acquisition and reconstruction time while maintaining diagnostic image quality.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gets online reconstruction working for RT-SMS spiral bSSFP at 1.5 T via 3D U-Net after k-space slice separation, but all numbers come from 10 healthy volunteers only.

read the letter

The central point is that this pipeline makes real-time SMS spiral bSSFP feasible for online use at 1.5 T. They acquire two slices simultaneously with blipped-CAIPI, separate them in k-space, then run a 3D U-Net to clean up the images. That drops scan time to 15 s and reconstruction to 30 s, versus 3+ minutes for breath-hold and nearly 25 minutes for compressed sensing. In the ten volunteers the ventricular volumes line up reasonably with the reference (biases around 7 ml for LVEDV) and image quality scores beat CS.

The technical combination for online cardiac work at this field strength looks new relative to the cited methods. The speed numbers and the direct comparison to CS are useful concrete results.

The obvious gap is the test population. Every quantitative claim, every p-value, and every volume comparison rests on those ten healthy subjects. There is no patient data, no arrhythmia cases, no implants, and no external validation set. A data-driven artifact suppressor has no analytical bound on its failure modes, so the transfer to the actual clinical population remains an open question. The abstract and methods give no training details or cross-validation either.

This is for the cardiac MRI reconstruction group. Someone working on sequence acceleration or deep learning cleanup will find the timing and pipeline details worth looking at. A reader who needs evidence for routine clinical deployment will not.

It should go to peer review. The implementation is specific enough that referees can check the methods and ask for the missing validation steps. The work is not ready for clinical claims as written, but the core engineering is worth the time to review.

Referee Report

2 major / 2 minor

Summary. The paper implements a spiral bSSFP simultaneous multi-slice (SMS) real-time sequence at 1.5 T using blipped-CAIPI, with k-space slice separation followed by 3D U-Net deep artifact suppression. In ten healthy volunteers it reports ~13× faster acquisition than breath-hold (15 s vs 3 min 15 s), ~50× faster reconstruction than compressed sensing (30 s vs 24 min 55 s), superior image quality (p<0.001), and good functional agreement (e.g., LVEDV bias −7.5±6.8 ml) versus breath-hold reference, concluding that online deep suppression enables diagnostic-quality free-breathing short-axis coverage.

Significance. If the performance generalizes, the work would be significant for clinical CMR by demonstrating practical online reconstruction of RT-SMS bSSFP with large reductions in both acquisition and reconstruction time while preserving ventricular volume and mass metrics. The reported speedups and direct comparison to CS are concrete technical strengths.

major comments (2)

[Methods] Methods: All network training, testing, and quantitative/qualitative evaluations (image quality scores, LV/RV volumes, LVM) are performed exclusively on the ten healthy volunteers; no patient cohort, hold-out set with pathology, implants, or arrhythmia is described, directly undermining the transferability required for the central claim of 'diagnostic image quality' in the Conclusion.
[Results] Results: The reported functional agreement (LVEDV bias −7.5±6.8 ml etc.) and p<0.001 image-quality superiority are computed solely within the healthy-volunteer cohort; without external validation the generalization step remains load-bearing for the claim that the method 'enables … diagnostic image quality'.

minor comments (2)

[Abstract] Abstract: The statement that deep artifact suppression 'consistently outperformed CS' lacks the specific statistical test and degrees of freedom used to obtain p<0.001.
[Abstract] Abstract/Methods: No details are supplied on network architecture hyperparameters, training/validation split, data exclusion criteria, or how error propagation was handled for the reported volume biases.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's positive evaluation of the technical aspects of our work and the opportunity to clarify the scope of our study. We address the major comments regarding the volunteer cohort below.

read point-by-point responses

Referee: [Methods] Methods: All network training, testing, and quantitative/qualitative evaluations (image quality scores, LV/RV volumes, LVM) are performed exclusively on the ten healthy volunteers; no patient cohort, hold-out set with pathology, implants, or arrhythmia is described, directly undermining the transferability required for the central claim of 'diagnostic image quality' in the Conclusion.

Authors: We agree that all training, testing, and evaluations were performed exclusively on healthy volunteers and that this limits direct evidence of performance in patients. The study was designed as a technical feasibility demonstration of the online reconstruction pipeline under controlled conditions. In the revised manuscript we will add an explicit Limitations section stating that generalization to patients with pathology, implants or arrhythmia remains to be demonstrated, and we will revise the Conclusion to indicate that diagnostic-quality imaging was shown in healthy volunteers. revision: yes
Referee: [Results] Results: The reported functional agreement (LVEDV bias −7.5±6.8 ml etc.) and p<0.001 image-quality superiority are computed solely within the healthy-volunteer cohort; without external validation the generalization step remains load-bearing for the claim that the method 'enables … diagnostic image quality'.

Authors: The reported biases, limits of agreement and image-quality p-values are derived solely from the healthy-volunteer data. We will update the Results and Discussion to make this scope explicit and will cross-reference the new Limitations section so that the generalization claim is appropriately qualified. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical comparisons on healthy-volunteer cohort are independent of network outputs

full rationale

The manuscript reports an empirical MRI reconstruction study. Acquisition times (15 s vs 3 min 15 s), reconstruction times (30 s vs 24 min 55 s), image-quality scores, and ventricular volume biases (e.g., LVEDV −7.5 ± 6.8 ml) are measured quantities obtained from standard clinical analysis pipelines applied to the reconstructed images. No equations, first-principles derivations, or “predictions” are presented that reduce by construction to parameters fitted on the same data. No self-citations are invoked as load-bearing uniqueness theorems. The limitation that all quantitative results derive from ten healthy volunteers is a scope issue, not a circularity issue; the reported metrics remain externally verifiable against breath-hold references.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The U-Net itself contains millions of learned weights, but these are not enumerated.

pith-pipeline@v0.9.1-grok · 5968 in / 1215 out tokens · 18260 ms · 2026-06-30T18:16:23.713909+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages · 1 internal anchor

[1]

Adam: A Method for Stochastic Optimization

Price AN, Cordero‐Grande L, Malik SJ, Hajnal JV. Simultaneous multislice imaging of the heart using multiband balanced SSFP with blipped‐CAIPI. Magnetic Resonance in Med. 2020;83(6):2185-2196. doi:10.1002/mrm.28086 14. Yang Y, Meyer CH, Epstein FH, Kramer CM, Salerno M. Whole‐heart spiral simultaneous multi‐slice ﬁrst‐pass myocardial perfusion imaging. Ma...

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1002/mrm.28086 2020
[2]

Principles and applications of balanced SSFP techniques

Schefﬂer K, Lehnhardt S. Principles and applications of balanced SSFP techniques. Eur Radiol. 2003;13(11):2409-2418. doi:10.1007/s00330-003-1957-x 27. Campbell‐Washburn AE, Varghese J, Nayak KS, Ramasawmy R, Simonetti OP. Cardiac MRI at Low Field Strengths. Magnetic Resonance Imaging. 2024;59(2):412-430. doi:10.1002/jmri.28890

work page doi:10.1007/s00330-003-1957-x 2003

[1] [1]

Adam: A Method for Stochastic Optimization

Price AN, Cordero‐Grande L, Malik SJ, Hajnal JV. Simultaneous multislice imaging of the heart using multiband balanced SSFP with blipped‐CAIPI. Magnetic Resonance in Med. 2020;83(6):2185-2196. doi:10.1002/mrm.28086 14. Yang Y, Meyer CH, Epstein FH, Kramer CM, Salerno M. Whole‐heart spiral simultaneous multi‐slice ﬁrst‐pass myocardial perfusion imaging. Ma...

work page internal anchor Pith review Pith/arXiv arXiv doi:10.1002/mrm.28086 2020

[2] [2]

Principles and applications of balanced SSFP techniques

Schefﬂer K, Lehnhardt S. Principles and applications of balanced SSFP techniques. Eur Radiol. 2003;13(11):2409-2418. doi:10.1007/s00330-003-1957-x 27. Campbell‐Washburn AE, Varghese J, Nayak KS, Ramasawmy R, Simonetti OP. Cardiac MRI at Low Field Strengths. Magnetic Resonance Imaging. 2024;59(2):412-430. doi:10.1002/jmri.28890

work page doi:10.1007/s00330-003-1957-x 2003