{"id":"6844dbd0-0f5e-46ef-bfe7-2d1add2859a3","arxiv_id":"2501.07081","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"mIR-rt, a multi-inversion-recovery Look-Locker sequence with real-time GRE, enables myocardial T1 mapping at 5T and reports T1 around 1550 ms, showing MOLLI underestimates.","lead":"A new MRI sequence, mIR-rt, measures heart muscle T1 values at 5 Tesla using repeated inversion pulses and real-time imaging. It reports the first myocardial T1 values at 5T and argues the standard MOLLI sequence underestimates them.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Phantom accuracy is overstated: Table 1 normalized mIR-rt errors are 6-8%, contradicting the abstract's '<3%', so the quantitative accuracy claim is not supported as written.","rationale":"The reader's weakest_assumption focuses on steady state before the second IR. I considered this carefully and believe it is not the most load-bearing issue: the acquisition is fixed at 40 images between IR pulses, approximately 6.5 s, while the effective Look-Locker recovery time T1* for typical myocardial T1 values and the stated TR/flip angle is around 0.4-0.8 s. Thus the longitudinal magnetization is within a small fraction of steady state before the second IR, and the initial condition in Eq. [3] is likely valid. The reader's rationale also flags the abstract/Table 1 discrepancy, and that is the concern I find most load-bearing. The central claim of 'accurate' T1 mapping is supported in the paper by simulation and phantom results; the abstract quantifies this as '<3%' errors, but Table 1 shows phantom errors of 6-8%. Even though mIR-rt remains more accurate than MOLLI in the same table, the quantitative accuracy claim is overstated, and this should be corrected. This does not overturn the comparative conclusion or the reported 5T values, but it does require revision of the reported error magnitude. Therefore the existing CONDITIONAL verdict is appropriate, and no change in verdict is recommended.","tokens_in":13016,"tokens_out":9942,"duration_ms":100979,"concrete_test":"Recompute normalized phantom errors from Table 1 as accuracy/Ref: for mIR-rt, 76/1235 = 6.2%, 120/1485 = 8.1%, 125/1885 = 6.6%, and 66/950 = 6.9%; for MOLLI, 166/1235 = 13.4%, 231/1485 = 15.6%, 162/1885 = 8.6%, and 86/950 = 9.1%. If the abstract's '<3%' claim is retained, either the phantom accuracy metric must be redefined or the Table 1 entries must be corrected; otherwise the abstract and significance should state the observed 6-8% phantom errors rather than '<3%'.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that mIR-rt provides accurate myocardial T1 mapping at 5T and better accuracy than MOLLI. The accuracy portion rests heavily on phantom validation, but the abstract states 'errors less than 3%' for both simulation and phantom studies. Table 1 reports absolute accuracy deviations of 66, 76, 120, and 125 ms for phantom reference T1 values of 950, 1235, 1485, and 1885 ms. Normalized, these are 6.9%, 6.2%, 8.1%, and 6.6%, respectively, not below 3%. The simulation section does report sub-1% errors for T1 > 1000 ms, but the phantom data do not support the '<3%' claim. This is not purely cosmetic: the significance and conclusion rest on the method being 'accurate', and the abstract's quantitative support is contradicted by the paper's own table. The comparison with MOLLI is less damaged, since MOLLI's Table 1 errors are 8.6-15.6%, but the accuracy claim for mIR-rt must be restated using the reported phantom values. I did not elevate the reader's flagged steady-state assumption in Eq. [3]: with 40 images at 162.5 ms per image, the interval between IR pulses is about 6.5 s, while T1* for myocardium with the stated TR and flip angle is well under 1 s, so steady state is almost certainly reached before the second IR.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"Ge et al. present a multi-inversion recovery real-time spoiled GRE (mIR-rt) sequence for myocardial T1 mapping at 5T. The sequence applies two inversion pulses in a single breath-hold, acquires real-time GRE images continuously, retrospectively selects diastolic images, and fits a Look-Locker model with an inversion-efficiency parameter to produce T1 maps. The method is evaluated with Bloch simulations, phantom measurements against IR-SE references, and in vivo scans of 16 healthy volunteers, with comparison to MOLLI. The authors report native myocardial T1 values around 1526-1553 ms, significantly higher than MOLLI values around 1350 ms, and claim the technique is more accurate than MOLLI at 5T, representing the first myocardial T1 mapping at 5T.","tokens_in":13318,"tokens_out":6066,"duration_ms":57229,"significance":"The work addresses a real need: at 5T, longer T1 and incomplete magnetization recovery cause MOLLI to underestimate T1, and there is no established myocardial T1 mapping technique at that field strength. The paper's strengths include a physically motivated sequence design, systematic Bloch simulations over T1, flip angle, and inversion efficiency, phantom validation against a spin-echo reference, and reproducibility assessment with ICC. The explicit modeling of inversion efficiency is a useful contribution. However, the quantitative accuracy claim in the abstract is not supported by the phantom data, and the handling of inversion efficiency in vivo may introduce spatial bias. With revisions to the accuracy claims and the inversion-efficiency treatment, the paper would be a valuable contribution to ultra-high-field cardiac MRI.","major_comments":[{"comment":"The statement that phantom T1 values have 'errors less than 3%' is contradicted by Table 1. The absolute errors of mIR-rt are 66 ms (T1_ref = 950 ms), 76 ms (T1_ref = 1235 ms), 120 ms (T1_ref = 1485 ms), and 125 ms (T1_ref = 1885 ms), corresponding to normalized errors of 6.9%, 6.2%, 8.1%, and 6.6%, respectively. The 'less than 3%' claim is supported only by the simulation results in §4.1. Please revise the abstract and the main text to report the actual phantom accuracy or to explicitly restrict the '<3%' statement to the simulation study.","section":"Abstract and §4.2/Table 1"},{"comment":"The fitting model in Eq. [5] uses the inversion efficiency δ as a known constant. The methods text (§2.2) says δ is 'estimated by the ratio of image intensities immediately before and after the second inversion pulse first,' but the in-vivo analysis (§3.2) fixes δ to the average value 0.85 from five volunteers. This inconsistency matters because Fig. 5 shows marked B1 inhomogeneity in the lateral wall, and the authors attribute the local T1 reduction there to reduced inversion efficiency. A single scalar δ cannot capture such spatial variation; using a global δ will bias T1 estimates in regions where the local inversion efficiency differs from 0.85. Please either estimate δ per-pixel from the images flanking the second IR or provide a sensitivity analysis quantifying how T1 errors scale with δ deviations. This is central to the accuracy and uniformity claims for the proposed method.","section":"§2.2, §3.2, and Fig. 5"},{"comment":"The conclusion states that this study reports myocardial T1 values at 5T 'for the first time.' The introduction cites reference [14] on feasibility of cardiovascular MRI at 5T; if that reference or any other prior work has already reported myocardial T1 at 5T, the novelty claim needs qualification. Please verify the first-report claim against the existing literature and adjust the wording if needed.","section":"§6 (Conclusion)"}],"minor_comments":[{"comment":"The abstract reports apex/middle/base T1 values as 1553 ± 52, 1531 ± 53, and 1526 ± 60 ms, whereas Table 2 lists 1553 ± 54, 1530 ± 60, and 1532 ± 62 ms. Please harmonize the numbers.","section":"Abstract vs. Table 2"},{"comment":"The first sentence of §4.2 refers to 'Figure 3' for phantom T1 maps and bar graphs, but Figure 3 is already used in §4.1 for simulation error plots. Please renumber the figures so that each figure number is used only once.","section":"§4.2, Figure numbering"},{"comment":"The sentence 'the accuracy of the vertices was somewhat lower' appears to be a typo; it should refer to the apex region, not 'vertices.'","section":"Discussion, last paragraph"},{"comment":"The correction T1_corrected = T1 + 2Δt is stated without derivation or reference in the text. A brief explanation or citation would help readers understand the factor of 2.","section":"§2.2, Eq. [7]"},{"comment":"The ICC values are reported for only six subjects, and no confidence intervals are given. Given the small sample size, please report confidence intervals for the ICC estimates to allow assessment of reproducibility precision.","section":"§4.3, Table 3"},{"comment":"The paragraph explaining why the Wilcoxon p-values are identical across layers is unnecessary in the results section; a brief sentence would suffice, and the explanation interrupts the presentation of the in-vivo results.","section":"§4.3, identical p-values"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for this journal and addresses an important gap in ultra-high-field cardiac MRI. The overstatement of accuracy in the abstract and the inconsistency in the inversion-efficiency handling are the main concerns. The editor may also wish to verify the 'first time' claim independently, as the onus is on the authors to show no prior myocardial T1 mapping at 5T exists. The discussion paragraph defending identical p-values suggests the authors anticipated criticism; this is not a problem, but the explanation could be shortened."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper is the first to report myocardial T1 mapping at 5T, with native values around 1550 ms, and it proposes a practical sequence (mIR-rt) that avoids MOLLI's underestimation at that field. The design is sensible: multi-inversion Look-Locker with real-time spoiled GRE, retrospective diastolic selection, and phase-sensitive reconstruction. The validation path is thorough—Bloch simulations across T1, flip angle, and inversion efficiency, plus a phantom study against IR-SE and a 16-subject in-vivo reproducibility assessment. The simulations show the method is robust to inversion efficiency and flip angle for the long T1 values relevant to 5T, which is the key technical argument. The steady-state assumption before the second inversion pulse is likely fine: with ~6.5 s between pulses and sub-second T1* for myocardium, the spins have certainly settled. I would not hang much criticism there.\n\nThe real soft spot is the accuracy claim. The abstract says phantom errors are “less than 3%,” but Table 1 gives absolute errors of 66, 76, 120, and 125 ms on reference T1s of 950, 1235, 1485, 1885 ms, which are 6–8% normalized. That is a contradiction, not a cosmetic one, because the conclusion leans on accuracy. It has to be restated with the actual numbers. Also, the abstract values and Table 2 differ slightly (1553±52 vs 1553±54 for apex, etc.); minor but should be fixed. More substantively, the in-vivo claim that mIR-rt is “better accuracy” than MOLLI is not validated against any in-vivo gold standard; the data show a difference, not which one is right. The fixed inversion efficiency taken from 5 volunteers and applied to all subjects is a limitation, though the simulation insensitivity to InvE at long T1 makes it a moderate concern, not fatal.\n\nThe citation pattern and internal logic are sound. This is a well-executed feasibility study with a genuine first result. The accuracy overstatement is fixable, and the method deserves referee time. I would send it to review with a request to reconcile the abstract with Table 1 and to soften the in-vivo accuracy language.","headline":"First 5T myocardial T1 values and a plausible Look-Locker GRE sequence, but the accuracy claims in the abstract outrun the phantom table.","tokens_in":13884,"tokens_out":1487,"would_cite":true,"duration_ms":16959,"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 real-time multi-inversion GRE sequence maps myocardial T1 at 5T without the underestimation seen in MOLLI.","keywords":["myocardial T1 mapping","5T MRI","Look-Locker","inversion recovery","real-time spoiled GRE","MOLLI","multi-inversion recovery","T1 mapping accuracy"],"falsifier":"Run mIR-rt on a phantom or volunteer with the interval between the two inversion pulses shortened in steps, for example 40, 30, and 20 images, while keeping all other settings fixed; if the measured $T_1$ shifts systematically as the interval shrinks, the steady-state assumption behind the second recovery curve is violated. A complementary check is to trigger the second inversion before and after the signal has visibly plateaued and compare the fitted $T_1$ values.","tokens_in":12820,"feed_emoji":"🫀","tokens_out":8886,"duration_ms":78105,"temperature":0.7,"pith_summary":"The paper sets out to establish that a Look-Locker-based sequence called mIR-rt, which samples magnetization recovery with continuous real-time spoiled GRE after multiple inversion pulses, can measure myocardial $T_1$ at 5T accurately. The authors argue that the standard MOLLI sequence systematically underestimates $T_1$ at 5T because it assumes full recovery between inversions and an ideal inversion pulse, assumptions that fail for the longer $T_1$ values found at ultra-high field. If the claim holds, mIR-rt supplies the first accurate myocardial $T_1$ values at 5T, with native myocardium near 1550 ms rather than the roughly 1350 ms reported by MOLLI, and gives clinicians a way to quantify diffuse myocardial disease at a field strength that is only beginning to be used for cardiac imaging.","feed_headline":"Heart T1 at 5T is about 1550 ms, not 1350","feed_subtitle":"A multi-inversion real-time GRE sequence corrects the underestimation of the standard MOLLI heart scan at ultra-high field.","key_machinery":"The central object is the two-curve Look-Locker fitting model that the sequence builds around. In that model, the signal during a continuous low-flip-angle GRE readout relaxes with an apparent time constant $T_1^*$ rather than $T_1$ itself: $M(t)=M_\\infty-(M_\\infty-M(0))\\exp(-t/T_1^*)$, where $M_\\infty$ is the GRE steady-state signal. mIR-rt generates two such curves with two inversion pulses; the second curve is written with the initial condition $M_2(0)=-\\delta M_\\infty$, which turns the inversion efficiency $\\delta$ into a quantity measurable from the images just before and after that pulse. The fit then returns $A$, $B$, and $T_1^*$, and $T_1$ follows from $T_1=T_1^*(B/A-1)/\\delta$ with a small dummy-time correction. This second-inversion design is the load-bearing mechanism because it removes the need for an extra calibration scan and converts a four-unknown fit into a three-parameter one.","core_discovery":"The central claim is that mIR-rt measures myocardial $T_1$ at 5T accurately, with errors under 3% in simulation and phantom against reference values, while MOLLI under-reads $T_1$. The sequence applies an inversion pulse, then continuously acquires diastolic real-time spoiled GRE images; after magnetization reaches the steady state of the GRE readout, a second inversion pulse increases the number of fitted samples and also lets the inversion efficiency $\\delta$ be estimated from the image pair just before and after that pulse. Fitting follows the Look-Locker model with $M(t)=M_\\infty-(M_\\infty-M(0))\\exp(-t/T_1^*)$, and $T_1$ is recovered as $T_1=T_1^*(B/A-1)/\\delta$ with a dummy-time correction. In 16 healthy volunteers the native myocardial $T_1$ values were $1553\\pm52$ ms (apex), $1531\\pm53$ ms (middle), and $1526\\pm60$ ms (base), significantly higher than MOLLI values near 1350 ms, and the paper reports these as the first myocardial $T_1$ values at 5T.","pith_inferences":["A natural extension the authors do not pursue is to test whether the same steady-state-before-second-IR logic transfers to 7T cardiac $T_1$ mapping, where recovery times are even longer and the need for inversion-efficiency calibration scans is greater.","The reported roughly 200 ms gap between mIR-rt and MOLLI suggests that earlier 5T results obtained with MOLLI-type sequences should be re-examined rather than treated as physiological variation.","The method's weaker reproducibility at the apex and base, compared with the middle slice, points to motion and partial-volume effects rather than the fitting model as the current limiting factor, so combining mIR-rt with 3D whole-heart acquisition or learned motion correction could close that gap.","A testable clinical extension would be to apply mIR-rt to patients with suspected myocardial amyloidosis or iron overload: if the 5T baseline $T_1$ is longer, disease-related changes may produce a larger dynamic range than at 1.5T or 3T."],"forward_implications":["If mIR-rt is accurate at 5T, future 5T cardiac studies can use roughly 1550 ms as the native myocardial $T_1$ reference range for healthy young adults, and departures from it can be tested as markers of disease.","MOLLI-based 5T $T_1$ values near 1350 ms would be understood as underestimates rather than true tissue values, so published or clinical 5T MOLLI results would need recalibration.","Because the method corrects for inversion efficiency from the second inversion pulse, accurate 5T $T_1$ mapping does not require an additional proton-density or long-delay calibration acquisition.","The sequence's fixed 10.2 s acquisition time and heart-rate-independent sample count make it practical for a single breath-hold.","The same steady-state-before-second-inversion logic should apply to other long-$T_1$ tissues at ultra-high field, where MOLLI's incomplete-recovery assumption is most damaging."],"supporting_citations":[{"why":"Defines the MOLLI sequence used as the accuracy comparator and the source of the underestimation being corrected.","marker":"[4]"},{"why":"Supplies the Look-Locker snapshot-FLASH model and the $M_\\infty \\approx T_1^*/T_1$ relation from which the fitting equations start.","marker":"[10]"},{"why":"Shows that myocardial $T_1$ and inversion efficiency behave differently at ultra-high field (7T), motivating the 5T adaptation.","marker":"[7]"},{"why":"Represents the saturation-recovery alternative that avoids MOLLI's bias but has lower SNR, framing the accuracy-precision tradeoff.","marker":"[8]"},{"why":"Provides the dynamic parallel-imaging reconstruction (TGRAPPA) that makes real-time GRE sampling fast enough for the sequence.","marker":"[17]"},{"why":"Supplies the phase-sensitive inversion-recovery reconstruction used to restore signal polarity for robust $T_1$ fitting.","marker":"[18]"},{"why":"Gives the dummy-time correction $T_{1,\\mathrm{corrected}}=T_1+2\\Delta t$ applied to the fitted values.","marker":"[19]"},{"why":"Provides the standardized T1MES phantom whose reference values are used to judge accuracy in the phantom study.","marker":"[22]"},{"why":"Supplies the inversion-recovery spin-echo protocol used to establish reference $T_1$ values in the phantom.","marker":"[23]"},{"why":"Supplies the strategy of measuring average inversion efficiency in a small volunteer subset and fixing it in the fit.","marker":"[30]"}],"fun_headline_variants":["5T T1: 1550 ms, not 1350","First accurate heart T1 at 5T: ~1550 ms","New 5T sequence corrects MOLLI's T1 underestimation","T1 at 5T: 1550 ms, not MOLLI's 1350","Real-time GRE gives true T1 at 5T: under 3% error"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method depends on the spins actually reaching the steady state of the GRE readout before the second inversion pulse fires; if the fixed interval between inversions is too short for a subject's heart rate, the initial condition for the second recovery curve is wrong and the fitted $T_1$ is biased.","fun_headline_variants_meta":{"raw":{"variants":["5T T1: 1550 ms, not 1350","First accurate heart T1 at 5T: ~1550 ms","New 5T sequence corrects MOLLI's T1 underestimation","T1 at 5T: 1550 ms, not MOLLI's 1350","Real-time GRE gives true T1 at 5T: under 3% error"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000669,"raw_usage":{"total_tokens":3135,"prompt_tokens":1118,"completion_tokens":2017,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":734,"completion_tokens_details":{"reasoning_tokens":1911}},"tokens_in":734,"tokens_out":2017,"duration_ms":16701,"temperature":1.0,"reasoning_tokens":1911,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:49:00.531992+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run mIR-rt on a phantom or volunteer with the interval between the two inversion pulses shortened in steps, for example 40, 30, and 20 images, while keeping all other settings fixed; if the measured $T_1$ shifts systematically as the interval shrinks, the steady-state assumption behind the second recovery curve is violated. A complementary check is to trigger the second inversion before and after the signal has visibly plateaued and compare the fitted $T_1$ values.","supporting_citations":[{"cited_title":"Figure 3(a) illustrates the T1 errors in the estimations obtained by mIR-rt under different InvEs and T 1 values","cited_arxiv_id":null,"evidence_quote":"Defines the MOLLI sequence used as the accuracy comparator and the source of the underestimation being corrected."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Look-Locker snapshot-FLASH model and the $M_\\infty \\approx T_1^*/T_1$ relation from which the fitting equations start."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows that myocardial $T_1$ and inversion efficiency behave differently at ultra-high field (7T), motivating the 5T adaptation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Represents the saturation-recovery alternative that avoids MOLLI's bias but has lower SNR, framing the accuracy-precision tradeoff."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the dynamic parallel-imaging reconstruction (TGRAPPA) that makes real-time GRE sampling fast enough for the sequence."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the phase-sensitive inversion-recovery reconstruction used to restore signal polarity for robust $T_1$ fitting."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the dummy-time correction $T_{1,\\mathrm{corrected}}=T_1+2\\Delta t$ applied to the fitted values."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the standardized T1MES phantom whose reference values are used to judge accuracy in the phantom study."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the inversion-recovery spin-echo protocol used to establish reference $T_1$ values in the phantom."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the strategy of measuring average inversion efficiency in a small volunteer subset and fixing it in the fit."}],"review_version":1}