Simultaneous myocardial T1, T2, T1ρ and fat fraction mapping with hybrid dual-echo cartesian acquisition and dictionary matching

Abstract

Background

T1, T2, T1ρ and fat fraction (FF) are important parameters for diagnosing various pathological changes in the myocardium. However, existing methods either require repeated breath-holds to quantify multiple parameters, or rely on complex image reconstruction to remove artifacts arising from highly undersampled acquisition. To address this technical gap, a free-breathing Dixon-MultiMap technique is proposed to allow for simultaneous myocardial T1, T2, T1ρ and FF mapping in a single scan with conventional single-shot Cartesian acquisition.

Methods

Dixon-MultiMap employs electrocardiogram-triggered, single-shot dual-echo Cartesian acquisition over 14 consecutive cardiac cycles. A hybrid dual-echo acquisition scheme is designed to acquire four-echo images within a short cardiac acquisition window in the first two cardiac cycles, which are used to obtain FF and B0 field maps with water-fat separation. The derived B0 map is used as prior information for dual-echo water-fat separation of the multi-contrast images acquired in the following cardiac cycles. Subsequently, dictionary matching is performed on the water images for multi-parameter mapping, accounting for spin-history B1+ inhomogeneities. Dixon-MultiMap was optimized through simulations and validated in phantoms, 13 healthy volunteers, and three patients with suspected hypertrophic cardiomyopathy.

Results

In phantoms, Dixon-MultiMap showed strong correlation with reference methods for measuring T1, T2, T1ρ and FF (R² > 0.96). In healthy subjects, Dixon-MultiMap achieved robust water-fat separation, yielding FF comparable to the conventional multi-echo method. Compared with separate breath-hold mapping techniques, Dixon-MultiMap yielded significantly higher T1 values (1348 ± 33 ms vs. 1159 ± 28 ms, p < 0.001), comparable T2 values (46.4 ± 1.7 ms vs. 45.4 ± 1.5 ms, p = 0.96), and a small but statistically significant increase in T1ρ values (55.8 ± 2.6 ms vs. 53.0 ± 1.8 ms, p = 0.01). Pathological changes detected by Dixon-MultiMap were consistent with those identified using conventional mapping techniques in the three patients.

Conclusion

The proposed Dixon-MultiMap allows for simultaneous myocardial T1, T2, T1ρ and FF mapping in a free-breathing acquisition of 14 cardiac cycles. Water-fat partial volume was resolved by water-fat separation in Dixon-MultiMap, which improved the accuracy of multi-parameter estimation.

Graphical abstract

1. Introduction

Quantitative cardiovascular magnetic resonance (CMR) imaging, with the ability to characterize myocardial tissue non-invasively, is increasingly recognized for diagnosing complex cardiomyopathies [1], [2]. Native myocardial T1 is shortened in the presence of fat and iron deposition, while it is prolonged by increased free water content or expansion of the extracellular space [3], [4], [5]. Elevated T2 values indicate increased myocardial free water, which is commonly observed in acute myocarditis, acute myocardial infarction and chronic inflammatory diseases [6]. Besides the well-recognized T1 and T2, the longitudinal relaxation time in the rotating frame, termed as T1ρ, is sensitive to low-frequency molecular interactions such as those between free water and macromolecules of collagen and proteoglycans [7], [8], [9]. Prolonged T1ρ values have been shown to correlate with increased collagen content, making it a useful marker for identifying myocardial fibrosis [10], [11], [12]. Multiple parameters provide complementary information about myocardial tissue, and their combined analysis can significantly enhance diagnostic sensitivity and accuracy [2], [13]. Conventional CMR parameter mapping techniques quantify a single parameter in a breath-hold acquisition [14], [15]. Measuring multiple parameters necessitates repeated breath-holds, which not only prolong scan time and increase patient discomfort, but also may lead to maps misregistration due to inconsistent breath-hold positions, complicating the multi-parameter analysis process.

To address these limitations, simultaneous cardiac multi-parametric mapping techniques have been widely investigated, which typically rely on magnetization preparation pulses to induce contrast changes followed by composite relaxometry model analysis for measuring multiple relaxation times jointly [8], [16], [17], [18], [19]. In particular, we developed the FB-MultiMap technique, enabling simultaneous quantification of myocardial T1, T2 and T1ρ with a short free-breathing acquisition of 16 heartbeats, achieving mapping quality comparable to conventional separate breath-hold mapping techniques [19].

In addition to relaxometry, cardiac fat also carries important information for diagnosing cardiac diseases. Fat fraction (FF) is an important parameter that reflects myocardial tissue composition and can be used to detect pathological changes in myocardial lipomatous metaplasia and epicardial fat accumulation, both of which have diagnostic and prognostic significance in cardiovascular diseases [20], [21], [22]. Furthermore, fat-related artifacts and water-fat partial volume may influence the accuracy of myocardial parametric mapping [23]. Therefore, simultaneous quantification of cardiac parameters and fat is required to mitigate partial volume effect and provide a more comprehensive characterization of myocardial tissue.

In MR imaging, accurate water-fat separation typically requires the acquisition of a B0 field map to correct for frequency shifts caused by B0 field inhomogeneity [24], relying on multi-echo acquisition to obtain sufficient phase information. The T1 and T2 mapping cardiac magnetic resonance fingerprinting (MRF) technique has been extended for additional fat quantification [25], [26], [27], using either radial sampling with three-echo Dixon readout or a rosette trajectory [28] that samples the k-space center multiple times during a readout to facilitate multi-echo reconstruction. However, multi-echo sampling leads to increased repetition time (TR) and only limited number of echoes can be acquired to avoid the issue of overlong acquisition window for the electrocardiogram-triggered mapping technique. Cruz et al. proposed to incorporate cardiac motion correction into the Dixon-MRF technique to facilitate using a long acquisition window of ∼480 ms for eight-echo acquisition, achieving simultaneous T1, T2, T2* and proton density fat fraction (PDFF) mapping [29]. However, these MRF-based methods typically adopt highly undersampled non-Cartesian sampling due to the limited breath-hold duration and rely on complex and computationally-intensive reconstruction to remove undersampling artifacts, which limits their scalability and clinical applicability [17].

In contrast, the combination of single-shot Cartesian acquisition and dictionary matching enables simultaneous multi-parameter quantification with simple post-processing and is more readily adopted in clinical settings [19], [30]. However, the direct extension of the FB-MultiMap technique with multi-echo Dixon readout for simultaneous T1, T2, T1ρ and FF quantification presents challenges. Due to the short acquisition window in CMR imaging (∼200 ms) to minimize cardiac motion, extended-echo (>2) acquisition cannot fit within a single cardiac cycle [31]. Whereas, if the acquisition spans several cardiac cycles (multi-shot acquisition), respiratory motion can cause motion artifacts in the acquired images, especially in patients with poor breath-holding capacity.

In this study, we aim to incorporate water-fat separation into FB-MultiMap to reduce water-fat partial volume effects and simultaneously quantify fat fraction along with T1, T2, and T1ρ in a single free-breathing acquisition. Reliable water-fat signal separation and fat fraction quantification require acquiring images of multiple echo times (TE) to obtain adequate phase information for B0 field calibration [24]. To address the dilemma of extended-echo acquisition and limited cardiac acquisition window, we propose a novel hybrid dual-echo acquisition scheme, where four images with different TEs are obtained across two cardiac cycles with single-shot dual-echo acquisition in each cardiac cycle. The four-echo images are employed for B0 field calibration and fat fraction quantification. The estimated B0 map is then used to perform water-fat decomposition on the dual-echo multi-contrast images. Subsequently, parameters estimation is performed using the water images. The proposed technique was optimized using numerical simulations and validated in a T1, T2 and T1ρ phantom, a water-fat phantom, 13 healthy subjects, and three patients with suspected cardiomyopathy against conventional techniques.

2. Methods

An overview of the proposed water-fat multi-parametric mapping technique with dual-echo Dixon readout (Dixon-MultiMap) is provided in Fig. 1. This technique is characterized by electrocardiogram-triggered single-shot acquisition, water-fat separation, and water-only T1, T2, T1ρ mapping with dictionary matching. Details are provided in the following sections.

Fig. 1.

Sequence diagram (A) and post-processing steps (B) of a free-breathing simultaneous cardiac fat fraction (FF) and multi-relaxation parameter mapping technique. Single-shot dual-echo Cartesian acquisitions are performed over 14 consecutive cardiac cycles, with inversion recovery (IR), T2 preparation (T2-prep) and T1ρ preparation (T1ρ-prep) pulses applied to achieve T1, T2 and T1ρ sensitization, respectively. Interleaved dual-echo acquisitions are performed in the first two cardiac cycles without any preparation pulses, resulting in four-echo images for water-fat separation to obtain FF and B0 field maps. The derived B0 map was used as prior information for dual-echo water-fat separation of the following cardiac cycles, yielding multi-contrast water-only images. Subsequently, dictionary matching was performed to obtain water-only T1, T2, T1ρ and B1+ maps.

2.1. Sequence design

The proposed Dixon-MultiMap sequence consists of a 14-heartbeat single-shot Cartesian acquisition with spoiled gradient echo readout. A hybrid dual-echo acquisition strategy is employed for the first two cardiac cycles to obtain four images with different TEs for B0 field calibration and fat fraction measurement. To minimize the increase in TR so that single-shot dual-echo acquisition can be performed with reasonable cardiac acquisition window, the shortest TEs (TE1/TE3) were adopted for the dual-echo acquisition in the first heartbeat, while TEs (TE2/TE4) of the second heartbeat were designed to be evenly interleaved with TE1/TE3.

The subsequent cardiac cycles in Dixon-MultiMap are for multi-contrast image acquisition, where dual-echo Dixon readout same as in the first cardiac cycle is adopted and inversion recovery (IR), T2 preparation (T2-prep) and T1ρ preparation (T1ρ-prep) pulses are introduced for T1, T2 and T1ρ sensitization, respectively. The IR pulse is applied in the 3rd and the 9th cardiac cycles; T2-preps with two adiabatic refocusing pulses [15] are performed in the 6th to 8th cardiac cycles with durations of {35, 45, 55} ms; T1ρ-preps with composite spin-lock (SL) pulses (90°_x-SL_y-SL_-y-180°_-x-SL_-y-SL_y-90°_x) to enhance robustness against B1 and B0 field inhomogeneities [32], [33] are applied in the 12th to 14th cardiac cycles with spin lock durations of {16, 30, 50} ms. Similar to FB-MultiMap [19], Dixon-MultiMap is performed under free-breathing with diaphragmatic navigator (dNAV) at every heartbeat for slice tracking to mitigate potential through-plane respiratory motion, and in-plane motion is corrected retrospectively with a group-wise registration method [32]. All magnetization preparation pulses in Dixon-MultiMap are non-selective. To mitigate potential signal loss in the dNAV caused by the IR pulse, a slice-selective IR pulse was applied immediately before the non-selective IR pulse, with its imaging slab spatially aligned to the dNAV location at the right hemidiaphragm. This slice-selective pre-inversion effectively restores the longitudinal magnetization in the navigator region, preserving adequate signal for reliable motion tracking. Additionally, the dNAV acquisition was positioned immediately before the T2-preparation and T1ρ-preparation pulses within each cardiac cycle to avoid potential signal interference from these contrast preparation schemes.

2.2. Flip angle optimization

To balance the signal-to-noise ratio (SNR) of the images acquired in the first two cardiac cycles and the caused magnetization saturation for the following acquisition, a moderate flip angle (FA) of 8° was adopted for the initial two cardiac cycles. For the following 12 cardiac cycles with IR, T2 and T1ρ preparation pulses to introduce T1, T2 and T1ρ weightings, a small flip angle results in low SNR, while a larger flip angle necessitates calibration of the flip angle for the Look-Locker type acquisition. As was performed for the FB-MultiMap technique [19], an effective B1+, defined as the ratio between the actual flip angle and nominal flip angle, was measured to correct flip angle errors in Dixon-MultiMap. This effective B1+ reflects not only RF field inhomogeneities but also other factors affecting spin history, including imperfect inversion efficiency, slice profile effects, and through-plane motion. To improve B1+ sensitization, the flip angle after the second IR was varied. The flip angles after the first and second IR (FA1 and FA2 in Fig. 2A) were optimized using numerical simulation and phantom experiment.

Fig. 2.

(A) Illustration of two variable flip angle strategies. The first strategy changes the flip angle after the second IR, resulting in two flip angles (FA1 and FA2). The second strategy changes the flip angle after the first T2-prep, second IR, and first T1ρ-prep, resulting in a total of four flip angles (FA1–1, FA1–2, FA2–1, and FA2–2). In both strategies, the flip angle for the first two cardiac cycles (FA0) remains unchanged. (B) The relative root mean-squared error (RRMSE) from numerical simulations of the dual flip angle scheme with different combinations of FA1 and FA2.

In the simulation, FA1 and FA2 were set ranging from 4° to 16° with a step size of 2° for estimating typical T1, T2, T1ρ and B1+ values in the myocardium of 1200 ms, 40 ms, 50 ms, and 0.8, respectively. The simulation for each flip angle combination was repeated 5000 times with added Gaussian noise (SNR=30) to simulate realistic imaging conditions. Then, candidate flip angle combinations were further assessed in phantom experiments. The relative root mean square error (RRMSE) for T1, T2 and T1ρ, as well as the mean RRMSE across the three parameters, was calculated to evaluate the performance of Dixon-MultiMap under different flip angle combinations. The RRMSE is calculated using the following formula:

$$\mathit{RRMSE}=\frac{\sqrt{\frac{1}{N}{\sum }_{i=1}^N{\left({\hat{x}}_i-x_{\mathit{true}}\right)}^2}}{x_{\mathit{true}}}$$

where ${\hat{x}}_i$ represents the estimated values, $x_{\mathit{true}}$ is the true or reference value, and $N$ is the total number of simulations.

2.3. Water-fat separation and parametric mapping

The GraphCut algorithm was adopted for water-fat separation of Dixon-MultiMap [34], [35]. As shown in Fig. 1B, the four-echo images acquired during the first two cardiac cycles were used to obtain fat fraction and B0 field maps. Then, the B0 field map derived from the four-echo data was used as prior information by initializing the B0 map that was then iteratively optimized during the dual-echo water-fat separation for the following cardiac cycles, thereby improving the accuracy of water-fat separation for the dual-echo multi-contrast images.

Parametric mapping was performed using dictionary matching. A subject-specific dictionary was generated according to the recorded R-wave intervals and trigger delays using the Bloch equation for ranges of T1, T2, T1ρ and B1+. To evaluate the impact of B1+ correction, a simplified post-processing version of Dixon-MultiMap (sDixon-MultiMap) was also implemented, where B1+ was assumed to be 1 in the dictionary simulation. To enhance the weightings introduced by the preparation pulses, centric ordering was adopted for Dixon-MultiMap. And the contrast of each cardiac cycle was determined by averaging the signal of the first 10 k-space lines, which represent the low-frequency components of k-space.

2.4. Phantom experiments

The proposed Dixon-MultiMap sequence was implemented and evaluated on a 3 T United Imaging scanner (uMR 890, United Imaging Healthcare, Shanghai, China) using 12-channel body and 24-channel spine coils. Phantom experiments were conducted on custom-made water-fat and water-only T1/T2/T1ρ phantoms.

The water-only T1/T2/T1ρ phantom, with relaxation parameters corresponding to the physiological ranges of cardiac tissue, was made of varying concentrations of agarose and gadolinium-based contrast agent. Phantom T1 values were calibrated using an inversion recovery spin echo (IR-SE) sequence with 11 inversion times ranging from 35 to 3000 ms and a TR of 7000 ms. Reference T2 values were measured using a multi-echo spin echo (ME-SE) sequence with 10 TEs ranging from 14 to 140 ms and a TR of 7000 ms. Reference T1ρ values were estimated using a T1ρ-prepared gradient echo (T1ρ-GRE) technique, with spin-lock times (TSLs) set to {2, 16, 30, 50, 80} ms and a spin-lock frequency of 350 Hz [36]. The reference scans were performed with a spatial resolution of 2 × 2 mm² and a slice thickness of 8 mm.

The water-fat phantom consisted of six vials containing mixtures of water solutions and peanut oil in different proportions, resulting in fat fraction ranging from 0% (pure water solutions) to 100% (pure peanut oil). Peanut oil was chosen due to its spectral similarity to the triglyceride protons found in human fat tissues [37]. The water solution was prepared following the method described by Hines et al., with adjustments of the gadopentetate dimeglumine concentration (0.11 mM/L) and agar content (4%) to achieve relaxation parameters similar to those of myocardium [38]. All vials were immersed in pure water. Reference measurements of FF were conducted using a multi-echo Dixon gradient echo (GRE) sequence with the following settings: Cartesian readout with mono-polar fly-back trajectory, six echoes with TE1/ΔTE = 1.24/2 ms, TR=13.8 ms, flip angle (FA) = 5°, bandwidth (BW) = 900 Hz/pixel, resolution = 2 × 2 mm, and slice thickness = 8 mm. The reference FF was estimated using the GraphCut algorithm, incorporating a pre-defined six-peak fat model, and including corrections for T2* decay and noise bias to ensure accuracy [24]. To evaluate the necessity and benefit of dual-echo water-fat separation for parametric quantification, the reference mapping sequence was modified to incorporate water-selective excitation pulses (1–3-3–1 binomial pulses) while maintaining all other sequence parameters identical to those used for water-only phantoms. This modification enabled the acquisition of reference water-only T1, T2, and T1ρ values for the water-fat mixed phantoms to validate Dixon-MultiMap parametric measurements.

To evaluate heart rate dependency, Dixon-MultiMap scans were performed at simulated heart rates ranging from 40 bpm to 120 bpm, with increments of 20 bpm. The imaging parameters were as follows: Cartesian readout with mono-polar fly-back trajectory, field of view (FOV) = 320 × 260 mm²; pixel size = 2 × 2.5 mm²; slice thickness = 8 mm; TR1/TE1/TE3 = 4.27 ms/1.17 ms/2.75 ms for the 1st and 3rd to 14th cardiac cycles; TR2/TE2/TE4 = 5.06 ms/1.96 ms/3.54 ms for the 2nd cardiac cycle; readout bandwidth = 900 Hz/pixel; inversion recovery time = 100 ms; partial echo factor = 0.875; parallel imaging acceleration factor ≈ 2, resulting in 48 k-space lines to be acquired for each cardiac cycle. It is noted that the slightly longer TR in the second cardiac cycle results in a little longer cardiac acquisition window of this heartbeat than the other cardiac cycles (242 ms vs. 205 ms). However, considering the heart is almost quiescent during diastole, the cardiac motion caused by the small acquisition window difference is negligible. Furthermore, registration was performed on the multi-contrast images to correct any potential mis-alignments before parametric mapping.

To reduce computation cost, the dictionary used in the phantom studies was generated based on the actual parameter ranges of the phantom. Specifically, the dictionary contained approximately 6037,000 combinations of T1, T2, T1ρ and B1+: T1 ranged from [10:10:1800] ms, T2 and T1ρ ranged from [10:2:160] ms, and B1+ ranged from [0.4:0.05:0.55, 0.55:0.02:0.8, 0.8:0.05:1.2]. Combinations where T2 > T1 or T1ρ > T1 were excluded from the dictionary simulation. Bloch simulations were implemented using MATLAB (The MathWorks, Natick, Massachusetts), and the dictionary generation was performed on a workstation (Intel Xeon Gold 6226 R 2.9 GHz processor, 384 Gb RAM, Intel Corp. Santa Clara, California) with parallel computing, taking approximately 5.4 min.

2.5. In vivo experiments

The in vivo study was approved by the local institutional review board. Thirteen healthy volunteers (3 females, age: 23 ± 4.3 years, heart rate: 70 ± 5.7 bpm) and three patients with suspected hypertrophic cardiomyopathy (Table 1) participated in the imaging experiments after providing written informed consent.

Table 1.

Clinical characteristics of patients with suspected hypertrophic cardiomyopathy.

Patient	Sex	Age (year)	Height (cm)	Weight (kg)	BMI (kg/m2)	Heart rate (bpm)
#1	Male	68	172	80	27	58.1
#2	Male	68	169	72.5	25.4	73.8
#3	Female	74	162	55	21	77.4

The Dixon-MultiMap was compared against the following separate breath-hold mapping methods: MOdified Look Locker Inversion recovery (MOLLI) 5(3)3 for T1 mapping [14], T2-prep bSSFP for T2 mapping [15], T1ρ-prep bSSFP for T1ρ mapping [36] with FOV = 320 × 280 mm²; pixel size = 2.08 × 1.67 mm²; slice thickness = 8 mm; acquisition window = 225 ms. The T2-prep bSSFP sequence acquired three T2-weighted images with T2-prep durations of {0, 35, 55} ms, while the T1ρ-prep bSSFP sequence acquired four T1ρ-weighted images with TSLs of {2, 16, 30, 50} ms and spin-lock frequency of 350 Hz. To ensure sufficient signal recovery, three idle cardiac cycles were added between each readout for T2-prep bSSFP and T1ρ-prep bSSFP sequences. The 6-echo Dixon GRE sequence with electrocardiogram-triggered multi-shot acquisition was adopted for FF mapping, the parameter settings of which were identical to those in the phantom experiments. Three short-axis slices positioned at the basal, mid and apical levels of the left ventricle were acquired for all mapping techniques. Other imaging parameters, including matrix size, slice thickness and readout bandwidth, were consistent with those used in the phantom studies of the Dixon-MultiMap method.

Additionally, to evaluate the parameter estimation variability caused by respiratory motion under free-breathing, Dixon-MultiMap was also performed under breath-hold considering that 14-heartbeat breath-hold can be easily tolerated by healthy subjects. To assess the effect of water-fat separation, dictionary matching was also performed with the first echo image of the multi-contrast acquisition in addition to the water images obtained with water-fat separation.

The dictionary for the in vivo studies contained approximately 2,330,000 combinations of T1, T2, T1ρ and B1+ values: T1 ranged from [500:100:800, 810:10:1700, 1710:100:2000] ms, T2 and T1ρ ranged from [10:10:30, 31:2:70, 71:5:91, 91:10:121] ms and B1+ ranged from [0.4:0.05:0.5, 0.5:0.03:1.1, 1.1:0.05:1.2]. It took approximately 1.3 min to simulate the subject-specific dictionary, and about 10 s to extract parametric maps for each slice using dictionary matching.

2.6. Image analysis

For the phantom experiments, the mean value for each tube was calculated, and the estimations from Dixon-MultiMap with and without B1+ correction were compared with the reference methods using Pearson correlation.

For in vivo image analysis, regions of interest (ROIs) were manually drawn on the parameter map according to the 16-segment model of the American Heart Association (AHA) [39], and the mean and standard deviation (SD) of each segment were calculated to evaluate the accuracy and precision of parameter estimation in healthy subjects. The measurements were compared among four methods: separate breath-hold mapping, breath-hold Dixon-MultiMap (BH-Dixon-MultiMap) with the first echo, BH-Dixon-MultiMap with water images, and free-breathing Dixon-MultiMap (FB-Dixon-MultiMap) with water images. The comparisons were performed using one-way analysis of variance (ANOVA) with Bonferroni post-hoc correction. A p-value less than 0.05 was considered statistically significant. For fat quantification evaluation, FF was measured from the interventricular septum, pericardial fat, and subcutaneous fat for all subjects with obvious pericardial fat and compared between FB-Dixon-MultiMap and the six-echo GRE.

Parametric measurements from water Dixon-MultiMap and first echo Dixon-MultiMap were compared within regions affected by water-fat partial volume effects to assess potential bias. Pixels that are subject to partial volume effects around the heart were identified using Dixon-MultiMap fat fraction maps by selecting regions where FF ranged from 0.3 to 0.7.

3. Results

3.1. Flip angle optimization

No FA1/FA2 combinations could minimize the RRMSE for all parameters, with T1 favoring the flip angle combination of 6°/14°, while T2 and T1ρ respectively favored moderate FA1 and FA2 values, consistent with previous findings in FB-MultiMap [19] (Fig. 2B). The RRMSE of T1 and B1+ was primarily influenced by the flip angle after IR and before T2- and T1ρ-prep, while the RRMSE of T2 and T1ρ was respectively more affected by the flip angles in the T2- and T1ρ-prepared acquisitions. Based on these observations, a four-flip-angle strategy was proposed, in which the flip angles were also altered after the first T2-prep pulse and the first T1ρ-prep pulse (FA1–1, FA1–2, FA2–1 and FA2–2 in Fig. 2A). The four-flip-angle scheme was determined to be 6°/10°/14°/8° based on the RRMSE maps. The derived four-flip-angle strategy demonstrated superior performance compared to representative constant-flip-angle and dual-flip-angle schemes in phantom experiments (Additional file 1: Tables S1 and S2).

3.2. Phantom experiments

Dixon-MultiMap estimates were highly consistent with reference values at a simulated heart rate of 80 bpm, achieving correlation coefficients of R² > 0.98 across all parameters (Fig. 3A-B) with minimal variations across different heart rates (Fig. 3C). Notably, Dixon-MultiMap achieved reliable fat quantification for the tubes with low to high fat fraction. Compared to the reference method, Dixon-MultiMap exhibited a tendency to underestimate longer T1 values (>1000 ms) and slightly overestimate longer T1ρ values (>110 ms). Without B1+ correction, sDixon-MultiMap exhibited significant deviations in the measured T2 and T1ρ values compared to the reference method, and its T1 estimation accuracy was also slightly reduced compared to Dixon-MultiMap with B1+ correction. Water-only Dixon-MultiMap demonstrated superior agreement with the reference method for all three parameters compared to single echo MultiMap, which showed substantial deviations in vials containing water-fat mixtures across varying FF (0–50%) (Additional File 1: Fig.S1). These results demonstrate that water-fat separation is essential for accurate parametric quantification in the presence of partial volume effects.

Fig. 3.

Phantom results. (A) Example phantom T1, T2, T1ρ and fat fraction maps estimated by Dixon-MultiMap with simulated heart rate of 80 bpm. (B) Correlation of Dixon-MultiMap and sDixon-MultiMap (simplified post-processing without B1+ correction) T1, T2 and T1ρ measured at heart rate of 80 bpm with the reference values. (C) T1, T2 and T1ρ estimated with Dixon-MultiMap at simulated heart rates from 40 to 120 bpm with the reference values showing in the legend.

3.3. In vivo imaging

Both the conventional 6-echo GRE method and free-breathing Dixon-MultiMap produced water-phase and fat-phase images without water-fat swap artifacts across the full FOV, with visually similar FF maps demonstrating comparable water-fat separation performance between the two techniques (Fig. 4). FF measurements from free-breathing Dixon-MultiMap showed strong correlation with 6-echo GRE across four ROIs in all healthy subjects (R² = 0.96, Fig. 5B), with good agreement across low FF (septum: 3.14% vs. 2.69%), moderate FF (pericardial fat: 74.8% vs. 80.0%), and high FF regions (subcutaneous fat: 84.9% vs. 90.5%, Fig. 5C). Detailed FF measurements are provided in Table S2 of the online supplemental materials.

Fig. 4.

Water images, fat images and fat fraction (FF) maps on the full FOV obtained with the conventional six-echo gradient echo (GRE) method (top row) and the Dixon-MultiMap technique (bottom row).

Fig. 5.

(A) Illustration of the four regions of interest (ROIs) for measuring fat fraction (FF), including the septum (red), two separate pericardial (green and blue), and subcutaneous fat regions (yellow). The septum ROI is drawn on the water image, while the other ROIs are drawn on the fat image. (B) Comparison between Dixon-MultiMap FF and 6-echo gradient echo (GRE) reference FF measured in the four ROIs for all 13 healthy subjects, showing a high determination coefficient (R² = 0.9605). (C) Violin plots comparing FF measurements between Dixon-MultiMap and 6-echo GRE in three tissue regions with varying fat contents for all 13 healthy subjects. The median and quartiles are shown as dashed and dotted lines, respectively. *** indicate p < 0.001.

The FB-Dixon-MultiMap technique demonstrated comparable mapping quality to conventional separate breath-hold mapping methods and BH-Dixon-MultiMap, without visible motion artifacts as shown for a representative healthy subject in Fig. 6. Parametric mapping of Dixon-MultiMap with water images yielded superior visual mapping quality compared to that with the first echo by effectively reducing water-fat partial volume effects at the myocardium-pericardial fat interface (zoomed-in views shown on the left side of each map in Fig. 6). The mean Dixon-MultiMap parameters were overall homogeneous across all segments, with segment-wise mean and SD calculated by averaging across all healthy subjects as shown in the bullseye plots in Fig. 7. The SD of Dixon-MultiMap measurements was similar between the breath-hold and free-breathing acquisition for all segments. Water-fat separation reduced the measurement SD of Dixon-MultiMap.

Fig. 6.

T1, T2, T1ρ and Fat Fraction maps at two short-axis slices of a representative healthy subject, obtained using traditional separate breath-hold mapping techniques, breath-hold (BH)-Dixon-MultiMap with the first echo image, BH-Dixon-MultiMap with water images, and free-breathing (FB)-Dixon-MultiMap with water images. Zoomed-in views (shown on the left side of each map) highlight regions susceptible to water-fat partial volume effects, illustrating the superior image quality of water-only Dixon-MultiMap compared to first-echo-only acquisition. Mean ± SD values across the entire left ventricular myocardium are displayed on each parametric map.

Fig. 7.

Bullseye plots of different mapping methods, including conventional separate breath-hold mapping techniques, breath-hold (BH)-Dixon-MultiMap with the first echo, BH-Dixon-MultiMap with water images and free-breathing (FB)-Dixon-MultiMap with water images. The mean and standard deviation (SD) of T1, T2 and T1ρ averaged across all healthy subjects are shown for the 16 segments in the three short-axis slices. The overall mean values across all segments are displayed below each bullseye plot.

No significant differences were observed in parameter estimations between BH- and FB-Dixon-MultiMap across the three short-axis slices of all healthy subjects (Fig. 8). However, the mean and SD of water T1 values from the FB-Dixon-MultiMap were significantly higher than those from MOLLI (1348 ± 33 ms vs. 1159 ± 28 ms, p < 0.001). The water T2 values from Dixon-MultiMap were comparable to those from the T2-prep bSSFP method (46.4 ± 1.7 ms vs. 45.4 ± 1.5 ms, p = 0.96), while the water T1ρ values were higher than those from the T1ρ-prep bSSFP method (55.8 ± 2.6 ms vs. 53.0 ± 1.8 ms, p = 0.01). Without resorting to water-fat separation, BH-Dixon-MultiMap with the first echo image resulted in significantly larger T1 and T1ρ SD than the other three mapping methods, and significantly larger T2 SD compared with breath-hold T2-prep bSSFP.

Fig. 8.

Violin plots showing the mean and standard deviation (SD) of myocardial T1, T2 and T1ρ values estimated using conventional separate mapping techniques, breath-hold (BH)-Dixon-MultiMap with the first echo, BH-Dixon-MultiMap with water images, and free-breathing (FB)-Dixon-MultiMap with water images measured from all healthy subjects. The black dashed lines indicate the median values, and the dotted lines represent the first and third quartiles. Mean ± SD values are displayed below each violin plot. * indicates statistically significant differences (p < 0.05); ** and *** indicate p < 0.01 and p < 0.001, respectively.

The mean water Dixon-MultiMap parametric values within the water-fat partial volume regions (T1:1374 ± 457 ms, T2: 49.0 ± 15.4 ms, T1ρ: 57.0 ± 20.4 ms) were closer to those measured in the left ventricular myocardium (T1: 1348 ± 33 ms, T2: 46.4 ± 1.7 ms, T1ρ: 55.8 ± 2.6 ms) than single echo MultiMap (T1: 1862 ± 920 ms, T2: 64.4 ± 32.8 ms, T1ρ: 45.2 ± 27.3 ms) for a representative healthy subject and across all subjects (Additional file 1: Fig.S2A-C). The parametric value distributions were displayed as both point clouds (top) and ellipses (bottom), with each ellipse centered on the mean and horizontal and vertical radii representing the SD of T1, T2 and T1ρ measurements (Fig.S2B-C). The larger measurement deviations and higher standard deviations of single echo MultiMap demonstrate that water-fat separation in Dixon-MultiMap significantly mitigates water-fat partial volume effect and enhances quantitative accuracy and precision of parameter estimation.

Septal water T1 values in a 68-year-old male patient with suspected hypertrophic cardiomyopathy and extensive LGE enhancement (Patient #1) were markedly elevated compared to healthy volunteers (1607 ± 57 ms vs. 1348 ± 33 ms), and water T1ρ values were also elevated (59.4 ± 4.1 ms vs. 55.8 ± 2.6 ms), while T2 values showed no significant difference from healthy subjects (Fig. 9). These findings agreed with conventional mapping techniques (septal T1: 1282 ± 75 vs. 1159 ± 28 ms; septal T2: 46.7 ± 6.3 vs. 45.4 ± 1.5 ms; septal T1ρ: 56.1 ± 5.2 vs. 53.0 ± 1.8 ms). Dixon-MultiMap yielded multi-parametric and FF maps with high image quality in two additional patients with suspected hypertrophic cardiomyopathy and mild LGE enhancement in the interventricular septum (Patients #2 and #3 in Fig. 10). Water-only parametric maps demonstrated visual superiority over single-echo parametric maps by suppressing partial volume effects from fat.

Fig. 9.

Parameter maps at two slices obtained with the proposed Dixon-MultiMap method (water and first echo, respectively) and conventional techniques including MOLLI, T2-prep bSSFP, T1ρ-prep bSSFP, and 6-echo GRE, along with slice-matched LGE images in one HCM patient (Patient #1). Water-only parametric values and first echo parametric values are in good agreement with conventional measurements. The water-only T1 map demonstrates more uniform signals in the right ventricle compared to the first echo T1 map, indicating effective suppression of partial volume effects. Quantitative values (mean±SD) measured in the septum are displayed on each parametric map.

Fig. 10.

Parameter maps obtained with the proposed Dixon-MultiMap method (water) and conventional techniques including MOLLI, T2-prep bSSFP, T1ρ-prep bSSFP, and 6-echo GRE, along with slice-matched LGE images in two HCM patients (Patient #2 and #3). Example ROIs drawn in the scar (red) areas are shown on the conventional T2 maps, and the mean ±SD values measured in these ROIs are displayed on each parametric map.

4. Discussion

A free-breathing simultaneous cardiac multi-relaxation parameter and fat fraction mapping technique using single-shot Cartesian acquisition and dictionary matching was developed and validated in this study. A hybrid dual-echo acquisition scheme was proposed to avoid excessively long TR while providing extended-echo images for fat quantification and B0 field calibration. A variable excitation flip angle strategy was designed to ensure accurate T1, T2 and T1ρ parameter estimations in the presence of flip angle errors. In phantom experiments, the optimized sequence achieved strong correlation with reference measurements (R² > 0.98 for all parameters) and exhibited minimal variation across different heart rates (Fig. 3). In healthy subjects, the free-breathing Dixon-MultiMap achieved similar mapping precision to the conventional separate breath-hold mapping techniques, with comparable inter-segment variability (SD) between breath-hold and free-breathing acquisitions across all myocardial segments (Fig. 7). In preliminary validations in three patients, Dixon-MultiMap successfully detected pathological changes consistent with findings from conventional separate breath-hold mapping techniques.

The Dixon-MultiMap technique was built upon our previously proposed FB-MultiMap technique to maintain the advantages of efficient free-breathing acquisition and simple post-processing with single-shot Cartesian acquisition [19]. The further development was aimed for water-fat separation which typically requires multi-echo data to estimate B0 field to correct for frequency shifts caused by B0 field inhomogeneity [24]. However, extended-echo acquisition leads to significantly increased TR and acquisition window which cannot fit in a single cardiac cycle, while multi-shot acquisition may need scan time that exceeds the breath-hold duration and is not compatible with free-breathing acquisition. To address this challenge, we proposed a hybrid dual-echo acquisition strategy, in which four echoes are acquired across two cardiac cycles, with single-shot dual-echo acquisition in each cardiac cycle. Fat fraction measurements from Dixon-MultiMap showed strong correlation with the conventional 6-echo GRE in both phantom (R²>0.99) and in vivo imaging (R²=0.96), demonstrating reliable water-fat separation with the four-echo acquisition (Fig. 4, Fig. 5). It is noted that the echo times in Dixon-MultiMap were primarily set for inducing minimal TR increase. Although further optimization of the TEs may be possible [40], it leads to increased TR and prolonged cardiac acquisition window. Additionally, T2* decay was not accounted for during the water-fat separation of Dixon-MultiMap considering the small TE values cause negligible signal loss due to T2* decay. Furthermore, mono-polar flyback readout was adopted instead of bipolar readout to mitigate eddy-current-induced phase errors that may compromise fat fraction quantification accuracy. While bipolar readouts facilitate shorter TR, the eddy currents generated by gradient polarity reversals would necessitate additional correction strategies to maintain accurate water-fat separation [41].

The Dixon-MultiMap technique was designed to facilitate water-fat separation before dictionary matching of the multi-contrast water images. The four-echo acquisition was performed at the beginning of the Dixon-MultiMap scan to avoid the contrast changes between the two sets of dual-echo images introduced by the preparation pulses. Otherwise, dictionary-based water-fat separation may be needed along with the multi-parametric mapping, which requires a dictionary with significantly increased size and is impractical to implement. Furthermore, the entire signal evolution including the acquisition of the first two cardiac cycles was simulated for generating the dictionary for multi-parameter mapping. With a moderate flip angle of 8°, the acquisition of the first two cardiac cycles induced a small magnetization saturation of 7.3% for the tissue with typical myocardial T1 (1200 ms).

Different from FB-MultiMap [19], the spoiled gradient echo readout was adopted in Dixon-MultiMap. Although the SNR is lower than that of balanced steady-state free precession, it is more robust to B0 field inhomogeneity, which is crucial for CMR at high-field scanners. Due to the different readout scheme, the variable-flip-angle scheme was redesigned for Dixon-MultiMap to correct flip angle errors. Based on the simulation and phantom experiments, a four-flip-angle strategy (6°/10°/14°/8°) was determined. Although the optimization process did not involve all possible flip angle combinations due to the prohibitive computation cost, the adopted variable flip angle combination effectively enhanced B1+ sensitization, enabling flip angle error measurement and correction during dictionary matching. This strategy improved T2 and T1ρ estimation accuracy compared to constant and dual-flip-angle schemes in phantom experiments (Additional file 1: Tables S1-S2). Using low excitation angles could mitigate the impact of flip angle error, which however, results in low acquisition SNR and reduced measurement precision.

The key innovation of Dixon-MultiMap lies in its hybrid dual-echo acquisition scheme, which addresses fundamental challenges in simultaneous multi-parametric mapping and fat quantification during free-breathing cardiac imaging. First, our hybrid acquisition scheme acquires four echoes across only two cardiac cycles using single-shot dual-echo readout in each cycle, solving the challenge of obtaining multi-echo data for FF quantification within the limited cardiac acquisition window (<250 ms). This is a substantial advancement over conventional multi-echo Dixon methods that require either extended TR incompatible with single-shot cardiac imaging or multi-shot acquisition unsuitable for free-breathing scans. Second, the four-echo data from the initial two cycles enable robust B0 field calibration and fat fraction quantification, with the derived B0 map serving as prior information for dual-echo water-fat separation of subsequent multi-contrast images. Third, by performing dictionary matching on water-only images rather than mixed water-fat signals, Dixon-MultiMap effectively mitigated water-fat partial volume effects in multi-parameter mapping, as demonstrated by the closer agreement of parametric values within partial volume regions to myocardial measurements compared to single-echo MultiMap (water T1: 1374 ± 457 ms vs. 1862 ± 920 ms; Additional file 1: Fig.S2).

In phantom validation, Dixon-MultiMap achieved strong correlation with reference values for fat fraction measurements (R² >0.99) across FF ranging from 0 to 100%, and accurately measured T1, T2 and T1ρ in the water-only multi-parameter phantom (R² > 0.98 for all parameters, Fig. 3). In contrast, sDixon-MultiMap yielded T2 and T1ρ measurements that deviated significantly from the reference values, which highlights the importance of effective B1+ correction in Dixon-MultiMap. This effective B1+ correction accounts for not only RF field inhomogeneities but also other spin history effects, including imperfect inversion efficiency, slice profile variations, and through-plane cardiac motion, all of which contribute to flip angle errors if left uncorrected. Similar to the FB-MultiMap technique [19], Dixon-MultiMap exhibited a tendency to underestimate T1 values >1000 ms, which is attributed to the assumption of perfect inversion during the dictionary simulation while the actual inversion efficiency can be below 100%. Reducing the assumed inversion efficiency from 1.0 to 0.95 substantially mitigates the T1 underestimation, while having minimal impact on T2 and T1ρ estimations (Fig.S3). This analysis demonstrates that imperfect inversion efficiency is a primary contributor to the observed T1 underestimation in phantom experiments. Direct measurement of inversion efficiency through a brief calibration scan and incorporating it as a known parameter in the Bloch simulation could potentially mitigate the bias in dictionary generation [42]. Additionally, Dixon-MultiMap showed slight overestimation of T1ρ values > 110 ms. The spin-lock durations of T1ρ-prep were set for measuring moderate T1ρ (∼50 ms) corresponding to myocardium. If long T1ρ values are targeted, the longest spin-lock duration should be increased accordingly.

Dixon-MultiMap demonstrated reliable water-fat separation across the full FOV in vivo, with FF measurements showing strong correlation with the reference 6-echo GRE method (R² = 0.96, Fig. 5). Dixon-MultiMap showed slight overestimation at low FF (septum: 3.14% vs. 2.69%) and underestimation at high FF (pericardial/subcutaneous fat: 74.8–84.9% vs. 80.0–90.5%) in healthy subjects (Fig. 5C). These systematic differences likely result from the reduced number of echoes. Further reducing the number of echoes, for example, using only two echoes for FF measurement may worsen the systematic bias. Beyond fat quantification, water-fat separation improved multi-parameter estimation precision. Water-only Dixon-MultiMap exhibited reduced measurement variability compared to single-echo acquisition, with lower SD across all parameters (T1: 55.7 ms vs. 63.8 ms; T2: 3.7 ms vs. 4.1 ms; T1ρ: 5.5 ms vs. 5.9 ms Fig. 7, Fig. 8). Partial volume artifacts from pericardial fat in single-echo MultiMap were effectively suppressed in water-only parametric maps (Fig. 6). This improvement is attributed to the improved SNR achieved by combining the two echoes to reconstruct water images, as well as the mitigation of variation introduced by water-fat partial volume.

To enable free-breathing acquisition, the respiratory motion correction strategy proposed in FB-MultiMap was adopted in Dixon-MultiMap, where the diaphragm navigator was used for slice tracking to correct potential through-plane motion during acquisition, and residual in-plane motion was corrected with group-wise multi-contrast image registration [19]. The motion correction strategy mitigated the B0 field variation during the free-breathing acquisition as validated in Additional File 1 Section 5, and facilitated the shared B0 strategy for water-fat separation of the dual-echo multi-contrast images. The free-breathing Dixon-MultiMap produced parameter maps with measurement precision comparable to breath-hold acquisition (mean SD across all segments: T1: 33 ms vs. 26 ms; T2: 1.7 ms vs. 1.8 ms; T1ρ: 2.6 ms vs. 2.1 ms), with no visible motion artifacts (Fig. 6, Fig. 7). No significant differences were observed in mean parametric values between free-breathing and breath-hold acquisitions (Fig. 8), demonstrating effective respiratory motion correction. However, it is noted that the adopted empirical slice tracking factor may not be optimal for all subjects, and the pre-trained subject-specific respiratory motion model [8], [43], [44] holds the promise of further improving motion correction in free-breathing Dixon-MultiMap. Moreover, under extreme deep breathing conditions, the diaphragm navigator slice tracking accuracy may be compromised, potentially leading to through-plane motion which cannot be corrected during retrospective registration and may degrade the mapping quality of Dixon-MultiMap (Additional file 1: Section 6).

5. Limitations

This study has several limitations. First, the design of Dixon-MultiMap was primarily inspired by the FB-MultiMap technique without undergoing thorough sequence optimization. Exploring more complex preparation pulse configurations and variable flip angle schemes may further improve the accuracy and precision of Dixon-MultiMap. Second, due to the constraint of the cardiac acquisition window, the resolution of Dixon-MultiMap is slightly lower than conventional mapping techniques. Higher spatial resolution can be achieved by increasing the undersampling factor or reducing the phase-encoding FOV, which however, requires more complex reconstruction techniques to remove aliasing artifacts [45], [46]. Third, the proposed technique has only been validated in healthy subjects. The clinical value of Dixon-MultiMap should be further evaluated in patients with fat infiltration. Finally, the feasibility of Dixon-MultiMap has been demonstrated at a 3 T scanner. Further optimization and validation may be needed for the proposed technique to be applied at lower or higher field strength such as 1.5 T or 5 T scanners.

6. Conclusion

In this study, we developed a novel free-breathing quantitative cardiac MR technique, which allows for simultaneous T1, T2, T1ρ and fat fraction quantification in a single scan without complex image reconstruction and post-processing. Water-fat partial volume was resolved by water-fat separation, contributing to improving the multi-parametric mapping quality. Good accuracy and precision were observed for Dixon-MultiMap in both phantom and in vivo experiments. Dixon-MultiMap offers a practical solution for comprehensive myocardial tissue characterization. Future studies in a larger cohort are warranted to further establish its clinical value, particularly in patients with myocardial fat infiltration.

Declarations

Ethics approval and consent to participate: The study was approved by the institutional review board at ShanghaiTech University, Shanghai, China. All participants provided written informed consent.

Authors' contributions

Each author made significant contributions to the study. Zhenfeng Lyu and Haikun Qi were responsible for the concept and design of this study. Zhenfeng Lyu, Huanghong Zhang and Haotian Hong collaborated in the sequence optimization, data acquisition and image analysis. All authors contributed to drafting and revising the manuscript.

Funding

This work was supported in part by the Explorer Program of the Science and Technology Commission of Shanghai Municipality under Grant 23TS1400300, in part by the National Natural Science Foundation of China under Grant 82572351, in part by the High Technology Research and Development Center of the Ministry of Science and Technology of China under Grant SQ2022YFC2400133.

CRediT authorship contribution statement

Haikun Qi: Writing – review and editing, Resources, Project administration, Methodology. Zhenfeng Lyu: Writing – review and editing, Writing – original draft, Visualization, Validation, Software, Project administration, Methodology, Data curation. Hongzhang Huang: Software, Methodology. Haotian Hong: Software, Methodology. Mengsu Zeng: Methodology, Data curation. Peng Hu: Methodology. Yumin Zhong: Methodology. Junpu Hu: Software, Methodology. Yali Wu: Methodology, Data curation. Xianling Qian: Methodology, Data curation. Liwei Hu: Methodology.

Appendix A. Supplemental material

Supplemental material

.

Availability of data and materials

The datasets used and/or analyzed in the current study will be available from the corresponding author upon reasonable request.