All-in-One Cardiac MRI: Unified Non–Contrast-enhanced and Contrast-enhanced Imaging for a Continuous-Acquisition Examination

Abstract

Purpose

To evaluate a novel continuous-acquisition cardiac MRI method that integrates non–contrast-enhanced and contrast-enhanced imaging within a single 20-minute, free-breathing, non–electrocardiographically (ECG) gated scan.

Materials and Methods

A three-dimensional (3D) stack-of-stars T2-prepared fast low angle shot (FLASH) pulse sequence, combined with a 3D multitasking reconstruction and multiparametric mapping framework, was developed to jointly reconstruct coregistered precontrast and postcontrast images. A prospective study conducted from September 2021 to February 2022 involved 10 healthy human volunteers (mean age, 37 years ± 21 [SD]; eight male, two female) and four pigs with reperfusion injury. Quantitative and qualitative MRI measurements were verified in a digital phantom and in vivo. Statistical analysis included intraclass correlation coefficient, Bland–Altman analysis, and paired t tests to compare the proposed method against reference methods.

Results

Multitasking T1 measurements were highly correlated with two-dimensional (2D) inversion recovery spin-echo-T1 (R² > 0.99) and T2 measurements with 2D spin-echo-T2 (R² > 0.960) in the phantom study. In vivo, multitasking provided higher myocardial T1 values than did modified Look–Locker inversion recovery (MOLLI) (1383 msec ± 37.8 vs 1217 msec ± 16.3; P < .001) and higher myocardial T2 values than did T2-prep FLASH (43.9 msec ± 1.4 vs 40.0 msec ± 1.6; P < .001). There was no evidence of a difference between multitasking and MOLLI extracellular volume (ECV) values (27.5% ± 1.4 and 27.9% ± 1.8; P = .52). In pigs, the proposed method depicted increased precontrast T1 and ECV in ischemic injury regions, aligning well with reference measurements.

Conclusion

The novel 3D multitasking method enables a comprehensive, 20-minute, push-button cardiac MRI examination without ECG gating or breath holding, providing cardiac function, T1, T2, ECV, and late gadolinium enhancement measurements.

Keywords: Cardiac MRI, Cardiac MRI Multitasking, Continuous-Acquisition Cardiac MRI, All-in-One Cardiac MRI

Supplemental material is available for this article.

© RSNA, 2025

Summary

A novel continuous-acquisition cardiac MRI technique combines non–contrast-enhanced and contrast-enhanced imaging into a single 20-minute, free-breathing, non–electrocardiographically gated examination, yielding cardiac cine, precontrast T1 and T2, extracellular volume, and late gadolinium enhancement measurements.

Key Points

• ■ A single continuous-acquisition cardiac three-dimensional MRI multitasking method can unite both non–contrast-enhanced (cardiac cine, precontrast T1 and T2) and contrast-enhanced (extracellular volume [ECV], late gadolinium enhancement) information within a multidimensional reconstruction framework, paving the way for an all-in-one examination.

• ■ In healthy human participants, multitasking measured values within normal ranges, estimating higher myocardial T1 (mean, 1383 msec ± 37.8 [SD]) and T2 (43.9 msec ± 1.4) than did modified Look–Locker inversion recovery (1217 msec ± 16.3) and T2-prepared fast low angle shot (40.0 msec ± 1.6) and slightly lower left ventricular ejection fraction (mean difference, −3%) and similar ECV (27.5% ± 1.4 vs 27.9% ± 1.8; P = .52).

• ■ In an ischemic reperfusion injury pig model, multitasking depicted statistically significantly elevated T1 in the injured versus remote myocardium (1694.0 msec ± 138.4 vs 1294.0 msec ± 61.2) along with increased ECV (51.0% ± 4.9 vs 24.0% ± 1.7) and prolonged T2 (60.5 msec ± 15.3 vs 49.4 msec ± 11.3), mirroring reference‐sequence findings.

Introduction

Cardiac MRI is a powerful imaging modality for visualizing and measuring the heart’s function, structure, and tissue composition. Diagnostic measurements, such as left ventricular ejection fraction (LVEF) from cardiac cine imaging and late gadolinium enhancement (LGE), are well established (1). Recent advancements in quantitative techniques permit measurements of T1, T2, and extracellular volume (ECV), allowing for identification of diffuse fibrosis, inflammation, or infiltration (2–5) and aiding in accurate diagnosis, risk stratification, and tailored therapeutic management for cardiovascular diseases (6–8).

Cardiac MRI examinations traditionally use multiple separate sequences to gather diagnostic elements (eg, cine, T1, T2, ECV, LGE), requiring breath holding to freeze respiratory motion and electrocardiographic (ECG) triggering to synchronize cardiac motion. To fit each acquisition within a breath hold, conventional cardiac MRI serially captures multiple two-dimensional (2D) sections instead of a single three-dimensional (3D) acquisition. This setup demands substantial expertise and results in overall examination times of 40–60 minutes, with potential respiratory misregistration between breath holds. Newer rapid protocols for free-breathing cines and LGE can shorten total examination time, but they still require multiple setups and exhibit intersection respiratory misalignment. The overall time and difficulty of cardiac MRI have resulted in access limited predominantly to specialized centers.

To address these gaps, researchers have developed "all-in-one" or "push-button" scans, which streamline the process by collecting multiple parameters simultaneously (9). MR fingerprinting (10–13) and other multiparametric mapping techniques (14–16) have led to progress on native multiparameter relaxation mapping without the added complexity of gadolinium-based contrast agent dynamics. Because LGE is crucial for clinical cardiac MRI, incorporating contrast enhancement is a critical gap in all-in-one protocols.

Initial reports of cardiac MRI multitasking (17), a multidimensional imaging framework, demonstrated the ability to handle contrast agent dynamics and different contrast phases (17,18) using a low-rank tensor image reconstruction approach. In this study, we introduce and evaluate a 3D cardiac MRI multitasking method that jointly reconstructs pre- and postcontrast images, resulting in a 20-minute push-button, free-breathing, non–ECG-gated, single-scan examination. This method addresses common clinical questions by providing cardiac cine, T1, T2, ECV, and LGE in one single scan, representing a substantial step toward an all-in-one examination. A method that jointly reconstructs pre- and postcontrast images to decrease examination time and complexity has not previously been well established in the literature.

Materials and Methods

The institutional review board of Cedars-Sinai Medical Center approved this study. All human participants provided written informed consent, and the study complied with the Health Insurance Portability and Accountability Act. Animal experiments were conducted with approval from the Institutional Animal Care and Use Committee, following ethical guidelines. One author was in industry during the study (F.H., Siemens Medical Solutions); three authors now work in industry since completing the study (X.M., Meta Platforms; T.C., Noah Medical Corporation; Z.C., United Imaging). Academic authors not in industry controlled the data and the information submitted for publication.

Participants

This prospective study, conducted from September 2021 to February 2022, involved 10 healthy volunteers and four pigs with reperfusion injury. Inclusion criteria for human participants included being healthy adults with no known history of cardiovascular diseases and no contraindications to MRI or gadolinium-based contrast agents.

Pulse Sequence

A continuous-acquisition 3D stack-of-stars fast low angle shot (FLASH) pulse sequence was developed based on previous cardiac MRI multitasking sequences (17,19,20) (Fig S1B) and implemented on a 3-T scanner (MAGNETOM Vida; Siemens). In the first half of the examination (precontrast), gradient-echo readouts with repetition time msec/echo time msec of 3.6/1.6 were continually acquired between interleaved inversion recovery and T2-prepared (T2-prep) inversion recovery (21) preparation pulses (T2-prep durations of 0, 30, 40, 50, and 60 msec) every 2.5 seconds, generating multiple T1 and T2 contrasts (Fig S1). Successive recovery periods alternated between two FLASH excitation flip angles (3° and 10°) to allow B1⁺-plane-motion–robust and through-plane-motion–robust T1 mapping (22). In the second half of the examination (postcontrast), the preparation scheme switched to inversion recovery only and a single flip angle (5°) with no other changes to timing, allowing for T1 mapping during dynamic contrast-enhanced (DCE) and postcontrast phase-sensitive inversion recovery (PSIR) imaging. Volumetric imaging was in short-axis orientation throughout, at 1.4 mm × 1.4 mm × 8 mm spatial resolution with whole-ventricle coverage (14 sections and an additional section-oversampling rate of 14.3%). Further details of the pulse sequence parameters are provided in Table 1.

Table 1:

Scan Parameters for the Proposed and Reference Cardiac MRI Methods

Parameter	3D Multitasking	2D MOLLI	2D T2-Prep FLASH	2D IR-SE T1	2D SE-T2
Field of view (mm)	220 × 220*	220 × 220	220 × 220	280 × 192	280 × 192
In-plane spatial resolution (mm)	1.4 × 1.4	1.4 × 1.4	1.4 × 1.4	1.4 × 1.4	1.4 × 1.4
Sections
No.
Phantom	14†	1	1	1	1
In vivo (short axis)		3	3
Thickness (mm)	8	8	8	5	5
Repetition time (msec)	3.6	2.7	3.3	5000	5000
Echo time (msec)	1.6	1.1	1.4	8	15, 25, 45, 70, 100, 140, 180, 250, 350
Readout type	FLASH	bSSFP	FLASH	SE	SE
Inversion scheme	Look–Locker IR; IR pulse spacing, 2.5 sec; 720 inversion times	MOLLI 5(3)3; 8 inversion times	NA	Standard IR; 8 inversion times (TI: 150, 300, 500, 800, 1200, 1600, 2000, 4500 msec)	NA
In vivo	20:20 min	11 heartbeats	9 heartbeats	NA	NA
Scan time	9:29 precontrast + 10:51 postcontrast
Flip angle (degrees)	3/10 (precontrast); ‡ 5 (postcontrast)	35	12	90	90
T2-prep durations (msec)	0,§ 30, 40, 50, 60	NA	0,║ 35, 55	NA	NA

Note.—bSSFP = balanced steady-state free precession, FLASH = fast low-angle shot, IR = inversion recovery, IR-SE = inversion recovery spin echo, MOLLI = modified Look–Locker inversion, SE = spin echo, T2-prep = T2-prepared, 3D = three-dimensional, 2D = two-dimensional.

^*The 3D multitasking scans were reconstructed to two times the nominal field of view (440 mm × 440 mm).

^† Section direction oversampling in k-space of 14.3% brought the total number of reconstructed sections to 16.

^‡ Precontrast pulse sequence alternated between flip angles for different inversion recovery periods.

^§ 0 msec = inversion recovery pulse with no T2-prepared pulse.

^║ 0 msec = no T2-prepared pulse.

Image Reconstruction and Analysis

All multitasking image reconstructions were performed on a Linux workstation with a 2.90-GHz Intel Xeon processor in MATLAB 2018a (MathWorks). Coregistered images from one 20-minute acquisition spanning the pre- and postcontrast periods (10 minutes before + 10 minutes after administration of contrast agent) were jointly reconstructed using a modified multitasking framework (17,19,20,23). Precontrast images comprised a five-way low-rank tensor with four time dimensions (cardiac and respiratory motions, T1 and T2 contrasts), and postcontrast images comprised a five-way tensor with four time dimensions (cardiac and respiratory motions, T1 recovery, and DCE). For each participant, a shared spatial subspace was enforced between both the pre- and postcontrast tensors for memory-efficient joint image reconstruction; each spatial subspace was fixed at order 56. Image reconstruction of the 56 spatial coordinate maps was regularized with a wavelet-sparsity term, using the same regularization strength across all participants. A total of 20 cardiac phases and six respiratory bins were used, resulting in a total of 120 motion states (20 × 6). The 10-minute postcontrast acquisition was time-resolved (30-second DCE temporal resolution) to track gadolinium settling, with the last time point used for postcontrast measures, such as ECV and LGE. Figure 1 shows the 20-minute 3D cardiac MRI multitasking protocol compared with an approximately 40-minute conventional clinical cardiac MRI examination.

Figure 1:

The timeline of the conventional series of breath-held, electrocardiographically (ECG) triggered cardiac MRI examinations and the 20-minute, push-button, free-breathing, non–ECG-gated, motion-resolved whole-ventricle three-dimensional (3D) cardiac MRI (CMR) multitasking examination. The 40-minute duration of the conventional protocol represents the total examination time, including ECG setup, scouts, actual examination time, breath-hold recovery, intersequence planning, and a wait time after contrast agent injection. ECV = extracellular volume, LGE = late gadolinium enhancement.

Imaging Experiments

Phantom study

An International Society for Magnetic Resonance in Medicine (ISMRM)/National Institute of Standards and Technology (NIST) phantom (24) was imaged on a 3-T scanner (MAGNETOM Vida). The layer with the most relevant T1 and T2 values (T1∈[200,2500] msec; T2∈[20,800] msec) was analyzed. The proposed multitasking acquisition was applied, along with four reference methods: modified Look–Locker inversion recovery (MOLLI) 5(3)3 (25,26), T2-prep FLASH, inversion recovery spin-echo (IRSE-T1, and T2-weighted spin-echo (SE-T2). Detailed examination parameters are shown in Table 1. Regions of interest (ROIs) were manually delineated by an author (X.M., an MRI physicist with 10 years of experience) in the central portion of each vial, leaving clearance from the vial walls to minimize partial-volume effects.

In vivo human study

Ten human volunteers were imaged on a 3-T scanner (MAGNETOM Vida) with an 18-channel body coil. The 3D multitasking pulse sequence imaged the whole ventricle, using scan parameters from the phantom study. An automated contrast agent injector was used to administer a gadolinium-based contrast agent dose of 0.2 mmol/kg at 2 mL/sec; the injection was manually triggered approximately 30 seconds after the start of the DCE data acquisition. Reference pre- and postcontrast MOLLI T1 maps and precontrast T2-prep FLASH T2 maps (1.4 × 1.4 × 8.0 mm³) were acquired at both systole and diastole during end-expiration breath holds at basal, mid, and apical short-axis sections, using scan parameters from the phantom study. Full short-axis stacks of ECG-gated, breath-held, 2D balanced steady-state free precession (bSSFP) cine scans (1.3 × 1.3 × 8.0 mm³, precontrast) and 2D PSIR scans with segmented FLASH readouts (1.4 × 1.4 × 8.0 mm³, postcontrast) were also acquired. T1, T2, and ECV maps were segmented and analyzed using CVI42, version 5.12.4 (Circle Cardiovascular Imaging) (27). Endocardial and epicardial contours were manually delineated with conservative borders to minimize partial-volume effects (by X.M.). The software then automatically divided the myocardium into the 16 American Heart Association segments for analysis. Cine images were also contoured and analyzed in CVI42 to extract left ventricular end-diastolic volumes (LVEDV), left ventricular end-systolic volumes (LVESV), LVEF, and left ventricular mass measurements for comparison (by A.C.K., an imaging cardiologist with 8 years of experience).

In vivo animal study

To investigate PSIR and ECV quality in myocardium with ischemic injury, 3D multitasking was performed in pigs with ischemia-reperfusion injuries. Four farm pigs (female, 30–35 kg) were imaged on a 3-T scanner (Biograph mMR; Siemens). A balloon dilation catheter was inserted into the left anterior descending coronary artery in each pig and inflated to achieve 100% occlusion of coronary blood flow for 90 minutes, followed by reperfusion. MRI was performed 3 days after reperfusion.

Multitasking was acquired using the same parameters and setup as those used in the human study. Full short-axis stacks of reference methods were acquired: 2D cardiac cine (1.4 × 1.4 × 6.0 mm³), 2D segmented PSIR-FLASH scans (1.3 × 1.3 × 6.0 mm³), pre- and postcontrast 2D MOLLI T1 maps (1.8 × 1.8 × 6.0 mm³), and precontrast T2-prep FLASH maps (1.8 × 1.8 × 6.0 mm³).

Myocardial ROIs were manually delineated in MATLAB (by X.M.). LGE areas on the 2D PSIR-FLASH images guided the definition of injured and remote territories.

Statistical Analysis

Statistical analysis included linear regression analysis, intraclass correlation coefficient (ICC), Bland–Altman analysis, and paired t tests (at a significance level of .05) to compare the proposed method against reference methods in the phantom and healthy volunteers. T1, T2, and ECV values were quantified and compared at the injury ROI and the remote ROI for both the multitasking and reference methods. Hematocrit values were not collected before the animal examination, so we used 40% for the ECV quantification (mean hematocrit values of pigs) (28,29). Statistical analyses were performed using SPSS Statistics 29 (IBM).

Results

Phantom Study

Phantom T1 and T2 maps from the proposed 3D multitasking and 2D reference methods are shown in Figure 2. Table 2 summarizes ICC and paired t test results comparing multitasking measurements and the clinically available cardiac mapping sequences MOLLI and T2-prep FLASH against reference standard IRSE-T1 and SE-T2 measurements. Multitasking and both IRSE-T1 and SE-T2 showed excellent agreement, with ICC greater than 0.96 for both T1 and T2. MOLLI and T2-prep FLASH also showed excellent agreement with IRSE-T1 and SE-T2 measurements (ICC > than 0.99 for both T1 and T2).

Figure 2:

International Society for Magnetic Resonance in Medicine (ISMRM)/National Institute of Standards and Technology (NIST) phantom. (A) Photograph of the ISMRM/NIST phantom. (B) Gray-scale image of the relevant layer of vials in the ISMRM/NIST phantom. (C) T1 and T2 maps of the relevant ring of vials in the ISMRM/NIST phantom using the multitasking method and two sets of reference methods (modified Look–Locker inversion recovery [MOLLI], T2-prepared [T2-prep] fast low-angle shot [FLASH] methods; inversion recovery spin-echo T1 [IRSE-T1] and T2-weighted spin-echo [SE-T2] methods). Eight vials with T1 less than 2000 msec were are used for T1 analysis, and five vials with T2 less than 120 msec were used for T2 analysis.

Table 2:

Intraclass Correlation Coefficients and Corresponding P Values for Comparison between Cardiac Mapping Measurements and Reference Standard Measurements across Relevant Vials with ISMRM/NIST Phantom

Variable	ICC	P Value for Bias (Paired t Test)
Multitasking T1 vs IRSE-T1	>0.99	.59
MOLLI T1 vs IRSE-T1	>0.99	.002*
Multitasking T2 vs SE-T2	0.99	.95
T2-prep FLASH T2 vs SE-T2	0.99	.44

Note.—FLASH = fast low-angle shot, ICC = intraclass correlation coefficient, IRSE = inversion recovery spin echo, ISMRM = International Society for Magnetic Resonance in Medicine, MOLLI = modified Look–Locker inversion, NIST = National Institute of Standards and Technology, SE = spin echo, T2-prep = T2-prepared.

^*Statistical significance.

Scatterplots and Bland–Altman plots of the T1 and T2 values in biologically relevant vials (0 ≤ T1 ≤ 2000 msec; 0 ≤ T2 ≤ 120 msec) are shown in Figure 3. Multitasking T1 measurements were highly correlated (R² > 0.99) with the 2D IRSE-T1 reference. The bias and 95% limits of agreement were 13.2 msec ± 139.8 for multitasking T1 versus IRSE-T1 and −68.7 msec ± 97.1 for MOLLI T1 versus IRSE-T1. Multitasking T2 measurements were also highly correlated (R² > 0.960) with the 2D SE-T2 reference. The bias and limits of agreements were 2.5 msec ± 10.5 for multitasking T2 versus SE-T2 and 2.5 msec ± 10.5 for T2-prep FLASH versus SE-T2. The only bias that was statistically significant was MOLLI versus IRSE-T1 (P = .002), consistent with MOLLI’s known property of underestimating T1 (25).

Figure 3:

Scatterplots and Bland–Altman plots of T1 and T2 values in the phantom. (A) Scatterplots show that both multitasking and modified Look–Locker inversion recovery (MOLLI) showed high R² values against inversion recovery spin-echo T1 (IRSE-T1) in the linear regression. Bland–Altman plots show that the T1 bias (solid line) and limits of agreement (dashed lines) were 13.2 msec ± 139.8 for multitasking versus IRSE-T1 and −68.7 msec ± 97.1 for MOLLI versus IRSE-T1. (B) Both multitasking and T2-prepared (T2-prep) fast low-angle shot (FLASH) showed high R² value against T2-weighted spin-echo (SE-T2). Bland–Altman plots show that the T2 bias (solid line) and limits of agreement (dashed lines) were 2.5 msec ± 10.5 for multitasking and SE-T2 and 2.5 msec ± 10.5 for T2-prep FLASH versus SE-T2.

In Vivo Human Study

Images from 10 participants (mean age, 37 years ± 21 [SD]; eight male, two female) were analyzed. Figure 4 shows representative T1, T2, and ECV mapping results and PSIR images from a healthy participant. Reconstruction times varied between 16 and 24 hours depending on shared workstation use. Summary statistics for segment-wise (n = 16 values, each averaged over all participants at a specific American Heart Association segment) and participant-wise (n = 10 values, each averaged over the whole myocardium within a specific participant) T1, T2, and ECV values are provided in Figures S2 and S3, respectively.

Figure 4:

Three short-axis precontrast T1 and T2, extracellular volume (ECV), and phase-sensitive inversion recovery (PSIR) images (base, mid, apex) from the proposed three-dimensional multitasking and reference methods in a healthy 22-year-old female participant. FLASH = fast low angle shot, MOLLI = modified Look–Locker inversion recovery.

For participant-wise T1, T2, and ECV values, multitasking and MOLLI T1 values were 1383 msec ± 37.8 and 1217 msec ± 16.3 (P < .001), respectively. Multitasking T2 and T2-prep FLASH T2 values were 43.9 msec ± 1.4 and 40.0 msec ± 1.6 (P < .001), respectively. There was no evidence of a difference between multitasking ECV and MOLLI ECV values (27.5% ± 1.4 and 27.9% ± 1.8; P = .52). The bias and limits of agreements for both participant-wise and segment-wise T1, T2, and ECV comparisons can be found in the Bland–Altman plots in Figure S3.

LGE was not present in any of the healthy volunteers. Figure 5 and Table S1 summarize the LVEF, LVEDV, LVESV, and LV mass calculated from the black-blood multitasking cines and bright-blood bSSFP cines. Multitasking black-blood volumes produced lower estimates for both LVEDV and LVESV (P < .001) than the bright-blood volumes and slightly lower estimates of LVEF (mean difference, −3%; P = .01). LV mass measurements were not significantly different (P = .07).

Figure 5:

Bland–Altman plots compare left ventricular ejection fraction (LVEF), left ventricular end-diastolic volume (LVEDV), left ventricular end-systolic volume (LVESV), and left ventricular mass between multitasking and reference cines in 10 healthy human participants. The dashed lines represent 95% limits of agreement, and the solid lines represent mean bias.

In Vivo Animal Study

Three short-axis sections of ECV maps and LGE images in two farm pigs with ischemic myocardial injuries are displayed in Figure 6. T1, T2, and ECV measurements from the ischemic injury region and the remote region of the myocardium are shown in Figure 7 and summarized in Table S2. Both multitasking and reference sequences depicted increases in precontrast T1 and ECV within the ischemic injury region of the myocardium. In the remote region, multitasking and reference methods measured T1 values of 1294.0 msec ± 61.2 and 1181.2 msec ± 35.8 (P = .07), T2 values of 49.4 msec ± 11.3 and 44.3 msec ± 6.7 (P = .18), and ECV values of 24.0% ± 1.7 and 23.9% ± 1.7 (P = .95). In the injured region, multitasking and reference methods yielded T1 values of 1694.0 msec ± 138.4 and 1528.8 msec ± 63.8 (P = .03), T2 values of 60.5 msec ± 15.3 and 57.1 msec ± 11.5 (P = .32), and ECV values of 51.0% ± 4.9 and 52.1% ± 4.1 (P = .10). The bias and limits of agreements for T1, T2, and ECV comparisons in both injured and remote regions are shown in the Bland–Altman plots in Figure 7. LGE was clearly present in all pigs because of the ischemic myocardial injury (Fig 6).

Figure 6:

Postcontrast phase-sensitive inversion recovery (PSIR) and extracellular volume fraction (ECV) images in three short-axis sections of two farm pigs with ischemic injuries in the myocardium, generated by both three-dimensional multitasking and reference methods (modified Look–Locker inversion recovery [MOLLI] T1 and two-dimensional segmented PSIR fast low-angle shot).

Figure 7:

Scatterplots (top) and Bland–Altman plots (bottom) depict the precontrast T1 and T2 and extracellular volume (ECV) differences between the injured and remote myocardium regions in the four farm pigs with ischemic myocardial injuries, measured by multitasking and reference modified Look–Locker inversion recovery and T2-prepared fast low-angle shot. In the Bland–Altman plots, the dotted line denotes the mean difference (bias) and the dashed lines denote the limits of agreement.

Discussion

This study presents a free-breathing, non–ECG-gated, 3D multitasking framework to unify non–contrast-enhanced and contrast-enhanced cardiac MRI, with the goal of delivering whole-ventricle T1, T2, ECV, cine, and PSIR maps in a single 20-minute continuous-acquisition "push-button" cardiac MRI examination. In the phantom study, multitasking precontrast T1 and T2 measurements showed excellent agreement (ICC > 0.96) with the reference standard IRSE-T1 and SE-T2 measurements. No statistically significant biases were found between the multitasking and reference measurements. In humans, multitasking measured higher myocardial T1 values than did MOLLI and exhibited more variability over the population. The higher multitasking T1 values are consistent with MOLLI’s known behavior to underestimate T1 behavior (30), which was observed again in the present phantom study. The slightly higher myocardial T2 values with multitasking than with T2-prep FLASH is consistent with previous 3D multitasking observations (19,20), and still within the normal reported range (31–33). The animal study revealed similar trends, although only the multitasking T1 and MOLLI T1 difference in the injured region was statistically significant (P = .03), possibly due to the smaller sample size in the animal study. Multitasking produced lower LV blood volume estimates, presumably because we analyzed black-blood contrast with fully nulled blood rather than the bright-blood approach of bSSFP cine; however, these underestimates largely canceled out in LVEF measurements, resulting in a difference of only −3%, which would not be considered clinically significant in most cases.

For the in vivo human study, the ECV maps obtained from the multitasking method resembled MOLLI ECV maps and measurements and were all within the normal range (34). No significant biases were found between the multitasking and reference methods. For the animal study, multitasking and reference ECV values also agreed well in both ischemic injured and remote regions of the myocardium, and no significant biases were found. PSIR images from multitasking and conventional methods showed no LGE in healthy volunteers and showed similar LGE extent to each other in animals, although the multitasking PSIR images in particular appeared blurrier than reference images, indicating a potential area for improvement.

The multitasking approach streamlined the cardiac MRI workflow considerably. A conventionally acquired cardiac MRI examination with whole-heart 12-section cine and three-section (base, mid, apex) T1 and T2 maps would require 18 breath holds (10–20 minutes). After gadolinium-based contrast agent injection and an approximately 10-minute wait for contrast agent settling, three-section T1 maps for ECV and 12 sections of LGE would require another 15 breath holds (10–20 minutes). These examinations lead to total imaging time of 30–60 minutes, not counting localization, ECG setup, and repeats for failed breath holds. The 3D multitasking approach applied a 20-minute single-continuous scan without breath holds or ECG reliance. This study focused on adult volunteers, but the multitasking method’s freedom from breath holds or ECG reliance may also be beneficial in pediatric imaging, where motion can be especially challenging. Further extension to fetal imaging would introduce additional complexities, but the prospect of assigning separate temporal dimensions to maternal and fetal motion could potentially enable adaptation to fetal cardiac MRI.

The multitasking approach requires further development and validation before clinical application. The current acquisition protocol and general multitasking framework in principle support first-pass perfusion imaging (17) because DCE data were acquired during gadolinium-based contrast agent injection. However, this has not yet been fully integrated into image reconstruction. Enhancing map and image sharpness for clinical use will require technical improvements, such as better motion identification and binning, integration of advanced motion compensation, and/or more efficient low-dimensional image reconstruction (such as manifold models). The current 16–24-hour reconstruction and data fitting time per 3D acquisition must be reduced to minutes or supported by immediate in-line intermediate previews. New nonuniform fast Fourier transform toolboxes for fast data consistency (35) and deep neural networks for fast denoising and artifact reduction have each been shown to improve reconstruction times by orders of magnitude for similar problems (36). If deep learning is used, it must also be integrated into in-line scanner reconstruction pipelines and rigorously validated for generalizability and portability across multiple centers (37). In this study, cine evaluations used only black-blood contrast because of the simplicity of selecting the blood null point from the available contrasts; however, standard cine measurements and analysis software rely on bright-blood contrast. Achieving cine equivalence may require a bSSFP contrast-matching algorithm from existing T1 and T2 contrasts or synthesizing a target contrast from embedded T1 and T2 weighting information, as has previously been demonstrated with a brain multitasking method (38). In addition, repeatability tests were not conducted because of the limitation of administering a 0.2-mmol/kg contrast agent injection only once during a cardiac MRI examination. Finally, the present study solely evaluated the feasibility of the multitasking method in 10 healthy volunteers and four injured farm pigs; thus, there is a need for a larger clinical study of patients with cardiovascular diseases and a comparison to existing clinical practices in terms of diagnostic accuracy and patient outcomes. This should include a performance evaluation across a broad range of cardiovascular conditions, such as patients with arrhythmias or more complex abnormalities.

In conclusion, this study provides a proof-of-concept contrast-enhanced cardiac MRI multitasking approach for various free-breathing, ECG-free imaging measurements (eg, T1, T2, ECV, cine, and LGE). Multitasking provided coregistered precontrast T1 and T2, ECV, and PSIR images in 14 short-axis sections without the need for ECG signal or breath holding, during 20 minutes of continuous acquisition. The integration of contrast-enhanced imaging into continuous-acquisition multiparameter mapping is pivotal: Gadolinium contrast-enhanced images are diagnostic cornerstones, representing critical reference standards in clinical cardiac MRI diagnosis. Fully exploiting the workflow benefits of this contrast-enhanced multitasking framework and further enhancing spatial resolution may enable a "push-button" cardiac MRI examination that comprehensively quantifies cardiovascular health. A simplified cardiac MRI examination on the foundation of multitasking has the potential to expand the accessibility of cardiac MRI.