{"id":"67202dc8-e2a0-4f8c-962a-f0ec6d7d6ab2","arxiv_id":"2603.04233","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A 3D real-time CMR method produces continuous beat-to-beat ventricular ejection-fraction measurements and reveals bimodal EF distributions during PVC episodes.","lead":"This paper tests a 3D real-time MRI method that continuously tracks heart motion without breath-holds or ECG gating, then uses the motion to measure the heart's pumping volume beat by beat. It reveals that in patients with premature ventricular contractions (PVCs), individual abnormal beats can have much lower ejection fractions than normal beats.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Per-beat PVC volume curves are never validated against an independent ground truth; the rank-16 / 4-Hz motion model could plausibly produce the observed bimodal EF distributions as an artifact.","rationale":"The paper is best read as a feasibility study, and the CONDITIONAL verdict already reflects the need for stronger validation. My concern reinforces the reader's weakest_assumption: the model's ability to capture high-frequency or otherwise out-of-subspace arrhythmic motion is not independently validated. The phantom experiment only covers a regular sinusoidal pattern, and the in-vivo PVC results are compared only in terms of mean EF, not beat-level volumes. The observation of bimodal EF distributions is the key novel finding, but without ground truth for per-beat volumes in PVCs, it could be an artifact of the low-rank/band-limited motion model. This is a correctness risk rather than an internal inconsistency. I do not think the paper should be rejected; rather, the CONDITIONAL verdict should stand until a targeted validation of PVC-like dynamics is performed. I also note the abstract/full-text inconsistencies in cohort size and phantom EF, which further erode confidence but are secondary to the missing per-beat validation.","tokens_in":12710,"tokens_out":4600,"duration_ms":52059,"concrete_test":"Simulate free-running 3D CMR k-space data from a digital or physical phantom with known ground-truth PVC-like beat patterns (irregular timing, reduced stroke volume, occasional premature beats) using the same OPRA trajectory and sequence parameters. Reconstruct with the proposed CMR-MOTUS pipeline and compare propagated beat-to-beat EDV, ESV, and EF against ground truth, specifically for PVC beats. If per-beat EF errors exceed approximately 5 EF points or the bimodal low-EF mode is not recovered, the central claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that the method provides accurate beat-to-beat ventricular volumes in arrhythmic patients—depends on the reconstructed motion fields faithfully representing PVC dynamics. This is not directly validated anywhere in the manuscript. The phantom validation uses a single periodic 60-bpm sinusoidal motion with a small ejection fraction (~22%), which does not reproduce the irregular, non-periodic, low-stroke-volume contractions seen in PVCs. The in-vivo comparison to 2D cine is limited to mean EF; the PVC group's mean EF is systematically lower than the 2D reference, and the authors attribute this to PVC episodes, but no per-beat volume or EF is checked against an independent standard. The rank-16 temporal subspace and the 4 Hz Butterworth low-pass filter in Eq. (2) and the Implementation section may be adequate for periodic cardiac motion, but their ability to represent abrupt PVC contractions is not tested. If a PVC event contains motion components outside this subspace, the reconstruction can fold errors into the motion fields, producing a spurious low-EF mode in the histograms. The reported ECG correlation is qualitative and does not quantify volumetric accuracy. Therefore the feasibility claim for arrhythmic patients is underdetermined by the present evidence.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper extends the CMR-MOTUS framework to 3D for free-running, ECG-free, continuous acquisition, jointly reconstructing a motion-corrected reference image and time-resolved motion fields expressed in a low-rank (rank-16) decomposition with temporal low-pass filtering. Ventricular volumes are obtained by propagating a single manual segmentation along the reconstructed motion fields, enabling beat-to-beat EF curves and histograms. The method is validated on a deformable cardiac phantom with static ground-truth acquisitions and applied to 4 healthy volunteers and 4 PVC patients (the abstract instead states 10+10). The authors report close phantom agreement (22.1±0.6% vs 21.9%), narrow beat-to-beat EF distributions in healthy subjects, bimodal EF distributions in PVC patients, and qualitative ECG correlation linking volume dips to PVC episodes. The central claim is that this workflow can quantify continuous volumetric function in arrhythmia where conventional binning and gating fail.","tokens_in":13038,"tokens_out":3645,"duration_ms":38152,"significance":"If the central claim is established, the method would be a meaningful step toward 3D real-time volumetric assessment in arrhythmia, with interpretable motion fields enabling downstream deformation metrics from a single free-running scan. The explicit low-rank motion model and joint reconstruction are technically sound and, unlike black-box deep-learning approaches, provide interpretable motion fields. The manuscript also candidly discloses limitations (small cohort, 2-hour reconstruction, scanner/protocol heterogeneity). The phantom validation and healthy-volunteer results support the feasibility of the motion-field-propagation workflow for regular, periodic cardiac motion. However, the evidence for the specific arrhythmic/PVC claim is incomplete: the 2D reference analysis excludes PVC beats, the phantom does not reproduce PVC dynamics, and the low-rank/4-Hz motion model is not independently validated for these events. The abstract/full-text numerical inconsistencies further reduce confidence in the reported results.","major_comments":[{"comment":"The abstract reports 10 healthy volunteers and 10 PVC patients and a phantom EF of 17.86% versus 17.27%, while the Methods and Results sections report 4 healthy volunteers, 4 PVC patients, and phantom EF of 22.1±0.6% versus 21.9%. These are not minor typographical differences; they change the evidence base and the numerical result that is used as primary validation. The authors must reconcile the numbers and state which cohort and phantom values are correct.","section":"Abstract vs Methods/Results"},{"comment":"The in-vivo validation for PVC patients is not a validation of PVC-beat volumes. The 2D real-time cine reference was deliberately limited to dominant sinus contractions ('For PVC patients, the arrhythmic beats were not captured consistently in every slice; therefore, the analysis was limited to dominant sinus contractions'). Thus the lower mean EF measured by the proposed method in subjects 5, 7, and 8 is expected by construction and cannot distinguish an accurate inclusion of PVC beats from an artifact. The manuscript should either provide an independent beat-resolved gold standard for PVC beats or explicitly restrict the claim to feasibility, with the PVC results treated as illustrative rather than quantitatively validated.","section":"Methods, 'In Vivo Study' / Results, Figure 8"},{"comment":"The motion model constrains D = ΦΨᵀ with R=16 and the temporal basis is passed through a 4-Hz Butterworth low-pass filter. PVC contractions are short, non-periodic events that may contain components above 4 Hz or outside a rank-16 subspace; if so, the reconstruction can fold those components into errors that manifest as spurious low-EF beats. The phantom validation uses a 60-bpm sinusoidal, periodic, small-EF motion and does not exercise this regime. A concrete test would be to simulate or acquire a phantom motion waveform with a known PVC-like pattern (e.g., a premature, low-stroke-volume contraction) and verify that the reconstructed per-beat EF matches the injected waveform. Without such a test, the bimodal EF distributions in Figure 7 could be in part a model artifact.","section":"Equation (2) and Implementation (4-Hz Butterworth, rank-16)"},{"comment":"The claimed temporal correspondence between volume irregularities and PVC episodes is supported only by qualitative visual alignment of the ECG trace and volume curve. This does not quantify volumetric accuracy or even the specificity of the association. The authors should report a quantitative comparison, for example the detection rate of PVCs from the volume curve against the ECG, or the timing offset between ECG R-waves and reconstructed volumetric troughs/peaks. This would strengthen the argument that the observed low-EF peaks are physiological rather than reconstruction artifacts.","section":"Results, Figure 6"}],"minor_comments":[{"comment":"In Eq. (1), the sum is written as ∑_t but the domain is not fully specified; in Eq. (2), the norm subscript is shown as ℓ without specifying which ℓ (the text says ℓ1 for in vivo, ℓ2 for phantom). Please make the notation explicit.","section":"Equations (1)-(2)"},{"comment":"The text states flip angle = 6° for the phantom GRE sequence, but Table 1 lists flip angle = 10°. Please reconcile.","section":"Methods, Phantom Study"},{"comment":"The hyperparameter selection is described both as a parameter sweep on phantom data and as tuning on an additional volunteer's acquisition. Clarify which settings were used for in-vivo reconstructions and whether the 'additional volunteer' is part of the reported cohort.","section":"Data Analysis / Implementation"},{"comment":"The sentence 'The authors have identified several computational solutions...' is vague; either name the specific algorithmic optimizations or remove the sentence.","section":"Discussion"},{"comment":"The abstract states '10 healthy volunteers, and 10 patients with PVCs' while the limitations section and full text state 4+4. This consistency issue is critical and must be corrected as part of the major revision.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"To the editor: The abstract/full-text discrepancies (10+10 vs 4+4; 17.86% vs 22.1% phantom EF) are serious and need resolution before any further consideration. The core methodological idea is promising and the phantom/healthy validation is reasonable, but the PVC-specific claim is under-supported: the 2D reference excludes PVC beats, no independent beat-level ground truth exists, and the motion model constraints could plausibly create the observed bimodality. These are fixable in principle (correcting the text, adding a PVC-like phantom/simulation test, quantifying ECG-volume correspondence), so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this if you work on real-time cardiac MRI or arrhythmia imaging. The method is a real step forward, but the paper has a reporting inconsistency you need to know about before trusting any number in it. The abstract says 10 healthy volunteers and 10 PVC patients with phantom EF 17.86% vs ground truth 17.27%; the full text says 4 and 4, with phantom EF 22.1±0.6% vs 21.9%. Those are different experiments. That discrepancy alone would make me hold off citing it until the authors sort out which is which.\n\nWhat's genuinely new: the extension of CMR-MOTUS from 2D to 3D, with a single motion-corrected reference image instead of a time-varying one, and joint reconstruction of the reference image and a rank-16 motion field at ~16–20 Hz from a free-running Cartesian acquisition. The OPRA trajectory is a smart choice. The phantom validation with static ground-truth images is a solid check: the reconstructed EF agrees to within about a tenth of a percent. Propagating one manual segmentation through the motion fields is a clean workflow, and the ECG traces lining up with the volume dips in PVC patients is good qualitative evidence that the low-EF mode is real physiology, not just a reconstruction artifact.\n\nThe soft spots are real but not disqualifying for a feasibility study. The per-beat PVC volumes are never checked against an independent standard — the 2D cine reference explicitly excludes arrhythmic beats, so the mean-EF disagreement in Figure 8 is expected, not a validation. The 4 Hz Butterworth filter and rank-16 temporal subspace could smooth or distort very short PVC contractions, and the paper doesn't test whether that happens. The hyperparameters were tuned for visual quality, and code/data are only 'available upon request,' which limits reproducibility. The cohort is tiny and the acquisition protocols vary across scanners and contrast agents. The authors list some of these limitations themselves, which is to their credit.\n\nAs a feasibility claim, the central argument holds up. I'd send this to peer review without hesitation, but the referees should insist the abstract and full text be reconciled and that the PVC dynamics be validated against something more than a qualitative ECG overlay before publication. Right now it's a promising method with a sloppy write-up.","headline":"A genuinely new 3D real-time CMR method, but the abstract and full text disagree on cohort size and phantom EF — fix that before trusting the numbers.","tokens_in":13546,"tokens_out":3198,"would_cite":false,"duration_ms":32499,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single free-running 3D CMR acquisition can be reconstructed into continuous beat-to-beat ventricular volumes and ejection fractions, exposing functional heterogeneity in arrhythmia that standard gated imaging averages away.","keywords":["free-running CMR","real-time 3D MRI","motion fields","premature ventricular contractions","beat-to-beat ejection fraction","volumetric quantification","CMR-MOTUS","arrhythmia imaging"],"falsifier":"Drive a beating phantom with known ground-truth volumes through an abrupt, short contraction (e.g., less than 150 ms duration) and check whether the rank-16, 4 Hz reconstruction recovers the true volume dip; if the recovered EF change is attenuated, the method cannot be trusted for high-frequency arrhythmic beats.","tokens_in":12642,"feed_emoji":"🫀","tokens_out":4071,"duration_ms":41542,"temperature":0.7,"pith_summary":"The paper claims that by jointly reconstructing a motion-corrected reference image and time-resolved 3D motion fields from a single free-running MRI scan—no ECG gating, no breath-holding—one can propagate a single segmentation through every cardiac cycle and obtain continuous beat-to-beat ventricular volumes and ejection fractions. In healthy volunteers this gives narrow EF distributions, while patients with premature ventricular contractions show broad, sometimes bimodal EF distributions where the low mode corresponds to PVC beats. If true, conventional gated or binned CMR, which averages many heartbeats, misses clinically relevant functional heterogeneity, and a single 2-minute scan could become a per-beat functional histogram.","feed_headline":"Free-running 3D CMR tracks beat-to-beat ejection fraction","feed_subtitle":"In PVC patients, per-beat EF splits into a bimodal distribution that gated cine averages away.","key_machinery":"The load-bearing object is the 3D CMR-MOTUS reconstruction: an alternating minimization that alternates between estimating a time-independent motion-corrected reference image q and a low-rank deformation field D = Phi Psi^T, where Phi holds 16 spatial basis components and Psi the temporal basis, with isotropic total-variation spatial regularization, l1 data consistency to tolerate inflow effects, and a 4 Hz Butterworth low-pass filter on the temporal components. The OPRA variable-density Cartesian trajectory provides temporally incoherent sampling without non-Cartesian gridding. Segmentation propagation turns the motion fields into beat-resolved volume curves.","core_discovery":"The paper extends CMR-MOTUS to 3D: from roughly two minutes of continuously acquired, ungated, free-breathing Cartesian data, it jointly reconstructs a motion-corrected reference image and a rank-16, time-smooth 3D motion field at about 16-21 Hz. By propagating one manual segmentation of the reference image through the motion fields, the authors obtain continuous volume curves and per-beat EF. Phantom EF agrees with ground truth (22.1% +/- 0.6% vs 21.9%), healthy volunteers show narrow EF distributions, and PVC patients show broader, sometimes bimodal distributions whose low-EF mode aligns with PVC episodes on simultaneously recorded ECG. The claim is that this makes beat-to-beat volumetric","pith_inferences":["The in-vivo cohort mixes field strengths, sequences, and contrast agents, so the apparent difference between healthy and PVC distributions could be partly acquisition-related; a matched homogeneous cohort is needed before reading the histograms as purely physiological.","A likely failure mode beyond what the paper tests: abrupt, high-frequency contractions near or above the 4 Hz Butterworth cutoff may be smoothed, so the method could understate the true volume drop in very short-coupled PVCs.","The single-reference-image assumption means any slow contrast change (e.g., late gadolinium wash-in) would be folded into motion; a time-dependent contrast model may be necessary for non-steady-state contrast protocols.","If EF distribution metrics prove reproducible, an 'EF burden' index (e.g., percentage of beats below a threshold) might become a more sensitive outcome measure than mean EF for device and ablation trials."],"forward_implications":["In arrhythmic patients, a single ~2-minute free-running scan could yield per-beat EF histograms instead of one averaged number, changing how PVC burden and treatment response are assessed.","Because the motion fields are explicit, the same acquisition can be reused to compute strain, desynchrony, and regional wall-motion metrics without re-scanning.","The approach removes the breath-hold and gating requirements of standard CMR cine, making volumetric function assessment feasible for patients who cannot hold their breath or have highly irregular rhythms.","The bimodal EF distribution offers a concrete, quantitative measure of the hemodynamic cost of PVC episodes, which standard 2D analysis cannot capture because arrhythmic beats appear in only some slices."],"fun_headline_variants":["3D real-time CMR captures beat-to-beat ejection fraction","Free-running CMR reveals bimodal EF in PVC arrhythmia","Ungated 3D CMR tracks continuous ventricular volumes","Beat-by-beat cardiac function from free-running 3D CMR"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The model assumes every acquired frame is a smooth, low-rank deformation of one time-invariant reference image; contrast changes or arrhythmic motion components outside the rank-16 and 4 Hz subspace will be misattributed to motion and bias the propagated volumes.","fun_headline_variants_meta":{"raw":{"variants":["3D real-time CMR captures beat-to-beat ejection fraction","Free-running CMR reveals bimodal EF in PVC arrhythmia","Ungated 3D CMR tracks continuous ventricular volumes","Beat-by-beat cardiac function from free-running 3D CMR"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000315,"raw_usage":{"total_tokens":1681,"prompt_tokens":862,"completion_tokens":819,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":606,"completion_tokens_details":{"reasoning_tokens":757}},"tokens_in":606,"tokens_out":819,"duration_ms":7624,"temperature":1.0,"reasoning_tokens":757,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T18:51:24.666922+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Drive a beating phantom with known ground-truth volumes through an abrupt, short contraction (e.g., less than 150 ms duration) and check whether the rank-16, 4 Hz reconstruction recovers the true volume dip; if the recovered EF change is attenuated, the method cannot be trusted for high-frequency arrhythmic beats.","supporting_citations":[],"review_version":1}