Measurement reproducibility of slice-interleaved T₁ and T₂ mapping sequences over 20 months: A single center study

Abstract

Background

Quantifying reproducibility of native T₁ and T₂ mapping over a long period (> 1 year) is necessary to assess whether changes in T₁ and T₂ over repeated sessions in a longitudinal study are associated with variability due to underlying tissue composition or technical confounders.

Objectives

To carry out a single-center phantom study to 1) investigate measurement reproducibility of slice-interleaved T₁ (STONE) and T₂ mapping over 20 months, 2) quantify sources of variability, and 3) compare reproducibility and measurements against reference spin-echo measurements.

Methods

MR imaging was performed on a 1.5 Tesla Philips Achieva scanner every 2–3 weeks over 20 months using the T1MES phantom. In each session, slice-interleaved T₁ and T₂ mapping was repeated 3 times for 5 slices, and maps were reconstructed using both 2-parameter and 3-parameter fit models. Reproducibility between sessions, and repeatability between repetitions and slices were evaluated using coefficients of variation (CV). Different sources of variability were quantified using variance decomposition analysis. The slice-interleaved measurement was compared to the spin-echo reference and MOLLI.

Results

Slice-interleaved T₁ had excellent reproducibility and repeatability with a CV < 2%. The main sources of T₁ variability were temperature in 2-parameter maps, and slice in 3-parameter maps. Superior between-session reproducibility to the spin-echo T₁ was shown in 2-parameter maps, and similar reproducibility in 3-parameter maps. Superior reproducibility to MOLLI T₁ was also shown. Similar measurements to the spin-echo T₁ were observed with linear regression slopes of 0.94–0.99, but slight underestimation. Slice-interleaved T₂ showed good reproducibility and repeatability with a CV < 7%. The main source of T₂ variability was slice location/orientation. Between-session reproducibility was lower than the spin-echo T₂ reference and showed good measurement agreement with linear regression slopes of 0.78–1.06.

Conclusions

Slice-interleaved T₁ and T₂ mapping sequences yield excellent long-term reproducibility over 20 months.

Introduction

Cardiovascular magnetic resonance (CMR) native T₁ and T₂ mapping have emerged as promising techniques for myocardial tissue characterization [1]. Studies have reported increased native T₁ times in the presence of myocardial fibrosis, inflammation, amyloids, and decreased T₁ in the presence of Anderson-Fabry disease, and iron overload [2]. Increased T₂ times have also been reported in the presence of edema or inflammation [3–5]. Assessing T₁ and T₂ measurement reproducibility is a necessary step toward their clinical utility as quantitative imaging biomarkers [6].

Various cardiac mapping techniques have been proposed for T₁ [7–11] and T₂ mapping [12–16]. The most widely used T₁ mapping sequence is the Modified Look-Locker inversion recovery (MOLLI) [7], which is based on sampling the inversion recovery of the longitudinal relaxation signal. Other types of T₁ mapping sequences, such as the Saturation recovery single-shot acquisition (SASHA), are based on sampling the saturation recovery curve [9]. A hybrid sequence combining inversion and saturation recovery curves, such as the Saturation pulse prepared heart rate independent inversion recovery (SAPPHIRE), has also been proposed [10]. The most widely used T₂ mapping sequences are based on T₂-preparation (T₂prep) [17–20] with balanced steady-state free precession (bSSFP) imaging [5, 12] or spoiled gradient echo (GRE) [14] acquired with at least 3 different echo times. Other types of T₂ mapping sequences are based on turbo spin echo (TSE) [15] or gradient spin echo (GraSE) [16].

In longitudinal studies, understanding technical variability is critical to determining if observed changes over time are biological and therefore clinically significant or only related to measurement variation [21]. Furthermore, higher reproducibility means fewer patients are necessary to achieve statistical significance in clinical trials, ultimately reducing study costs [22]. Several prior studies have investigated the reproducibility of various T₁ and T₂ mapping sequences, however they are test/retest studies carried out within several weeks [23–26]. Reproducibility studies using MOLLI and shortened MOLLI (ShMOLLI) have demonstrated that both sequences are highly reproducible [24, 26–29]. SASHA and SAPPHIRE were reported to have similar reproducibility as inversion recovery-based sequences [23]. The reproducibility of T₂ mapping of multi-echo-spin-echo T₂, T₂prep-bSSFP, and GraSE T₂ mapping sequences were also reported to be excellent [25].

The free-breathing slice-interleaved T₁ [30, 31] and T₂ [32] mapping techniques have been proposed and used in various clinical scenarios [33–37]. Slice-interleaved T₁ (STONE) acquires data for different slices within one inversion recovery curve to allow more accurate measurement with a bSSFP (STONE-bSSFP) [30] or spoiled gradient echo (STONE-GRE) [31]. Slice-interleaved T₂ uses slice-selective T₂prep with an interleaved slice acquisition scheme which permits increased time efficiency [32]. Slice-interleaved T₁ and T₂ mapping sequences provide highly reproducible measurements in test/retest studies of healthy subjects [22], however the long-term reproducibility (> 1 year) has not yet been studied. Long-term reproducibility of T₁ and T₂ measurements using slice-interleaved T₁ and T₂ mapping needs to be investigated prior to utilization of these sequences in longitudinal studies monitoring disease progression or treatment efficacy.

Various confounders can impact the accuracy and reproducibility of myocardial tissue characterization. Therefore, performance assessment of the myocardial tissue characterization techniques requires rigorous in-vivo or phantom validation. While in-vivo studies are the ideal experimental setting, a phantom study is necessary in cases where in-vivo experiments are not feasible or scenarios in which the reference standard can only be measured in a phantom setting. Phantom studies are also necessary for assessing long-term measurement variability when scanning volunteers for extended periods over multiple sessions is not feasible. T₁ or T₂ accuracy and temperature sensitivity, for example, can only be measured in the phantom setting. Although a phantom experiment may not address all relevant confounding factors of an in-vivo setting, it provides valuable information that may not be easily attainable from an in-vivo experiment.

The aim of this study was to carry out a single-center phantom study to 1) investigate the measurement reproducibility of slice-interleaved T₁ and T₂ mapping over 20 months, 2) quantify sources of variability, and 3) compare the performance of each in terms of reproducibility and measurement against reference spin-echo measurements.

Materials and methods

Experiments were performed using T₁ Mapping and ECV Standardization Program (T1MES) phantom [38]. This Food and Drug Administration (FDA)-cleared/Conformité Européene (CE)-marked MR phantom enables stable quality measures to study measurement variability over time. T1MES contains 9 vials (NiCl₂ doped agarose) covering the physiological ranges of T₁ and T₂ in the blood and myocardium pre- and post-Gadolinium-based contrast agents (GBCA; for a 1.5 T phantom: T₁:255ms to 1489ms, T₂:44ms to 243ms, referenced from the T1MES manual measured by slow inversion-recovery/spin-echo methods at 1.5T) (Fig 1A). The T1MES phantom volume is 2L with an inner dimension size of 197 × 122 × 122 mm, and the vials have a minimum diameter of 20 mm [38]. For T₁ mapping, all 9 vials were studied given that the phantom is designed to include all relevant T₁ ranges of myocardium and blood pre- and post-GBCA. For T₂ mapping we only studied vial ‘F’ (Fig 1A) which modulates “Medium” native myocardial T₁ and T₂ times at 1.5 T. Our T₂ mapping sequence is not designed to handle high T₂ values over 100 ms found in the blood, and all remaining vials had no variability (44–50 ms).

Fig 1. T1MES phantom used in this study, and the reproducibility study protocol.

a) The T1MES phantom used in this study consists of 9 vials of NiCl₂ doped agarose covering T₁ and T₂ ranges in the blood and myocardium before and after Gadolinium-based contrast agents. b) An imaging session was repeated every 2–3 weeks over 20 months (between-session reproducibility). Within each session, slice-interleaved T₁ and T₂ mapping sequences were repeated 3 times (between-repetition repeatability) for five slices (between-slice repeatability). SE T₁ and T₂ measurements and MOLLI were performed for comparison. STONE-bSSFP, slice-interleaved T₁ with balanced steady‐state free precession; STONE-GRE, slice-interleaved T₁ with spoiled gradient echo; SE, spin-echo.

Reproducibility is defined as the measurement precision between replicate measurements under varying conditions, and repeatability is defined as the measurement precision between replicate measurements under constant conditions [21]. In this study, we use ‘reproducibility’ when referring to measurement precision over multiple sessions, and ‘repeatability’ when referring to scanning in the same session. We defined a ‘session’ as a ‘single CMR imaging with identical image localization’.

The study design schematic is shown in Fig 1B. Reproducibility over several weeks was reported for between-session reproducibility. Images were acquired using STONE-bSSFP T₁ [30], STONE-GRE T₁ mapping [31], and slice-interleaved T₂ mapping [32] sequences. Within each session, imaging was repeated 3 times to allow repeatability assessment within each session and between repetitions. For multi-slice sequences, between-slice repeatability was also studied. Additionally, we acquired spin-echo (SE) T₁ and T₂ measurements and MOLLI in each imaging session for comparison; MOLLI was repeated 2 times.

CMR imaging

CMR imaging was performed using a 1.5 T scanner (Philips Achieva, Best, The Netherlands) with a 32-element cardiac phased-array receiver coil. The phantom was stored and scanned at room temperature in the scanner room. We assumed temperature and subsequently diffusion was uniform along vials in our study. Scanning was strictly performed according to the T1MES phantom user manual [38]. All acquisitions were performed with a simulated electrocardiogram (ECG) at a RR (interval time between two R-waves) period of 900 ms (heart rate 67 bpm). The positioning process was consistent for all sessions throughout the study. The book used to lift the phantom, large towel, coil, software version of the scanner, and air-flow setting of the scanner room remained constant throughout the study.

T₁ mapping

The STONE-bSSFP sequence was acquired with the following parameters: 5 slices, in-plane resolution = 2.1 × 2.1 mm², slice thickness = 8 mm, slice gap = 4 mm, field-of-view = 280 × 280 mm², TR/TE/flip angle = 2.8 ms / 1.39 ms / 70°, a sensitivity encoding (SENSE) rate = 2, linear ordering, 10 linear ramp-up pulses and bandwidth = 1894 Hz, acquisition duration = 1 min 38 sec. Eleven inversion images were acquired with inversion times of ∞, 130, 1030, 1930, 2830, 3730, 350, 1250, 2150, 3050, and 3950 ms. The STONE-GRE sequence was acquired with the following parameters: 5 slices, in-plane resolution = 2 × 2 mm², slice thickness = 8 mm, slice gap = 4 mm, field-of-view = 280 × 280 mm², TR/TE/flip angle = 4.7 ms / 2.3 ms / 10°, a SENSE rate = 2.5, half-scan factor = 0.75, linear ordering, 10 linear ramp-up pulses and bandwidth = 383 Hz, acquisition duration = 1 min 38 sec. Eleven inversion images were acquired with inversion times of ∞, 109, 1009, 1909, 2809, 3709, 350, 1250, 2150, 3050, and 3950 ms. For both STONE-bSSFP and STONE-GRE sequences, the inversion preparation pulse was an adiabatic hyperbolic secant pulse with an 11 ms pulse duration. The radiofrequency (RF) excitation pulse was a slice-selective Sinc-Gauss pulse with a duration of 0.43 ms. Images were acquired without prospective slice tracking, and the order of slices was 1-4-2-5-3. The MOLLI 5b(3s)3b [39] sequence was acquired with the following parameters: single slices, in-plane resolution = 2 × 2 mm², slice thickness = 8 mm, field-of-view = 280 × 280 mm², TR/TE/flip angle = 2.6 ms / 1.30 ms / 35°, a SENSE rate = 2.5, linear ordering, 10 linear ramp-up pulses and bandwidth = 1786 Hz, acquisition duration = 8 sec. Eight inversion images were acquired with inversion times of 79, 979, 1879, 2779, 3679, 350, 1250, and 2150 ms. SE T₁ times were obtained using inversion-recovery SE acquisitions with 16 inversion times of 50, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000, 1250, 1500, 1750, 2000, and 3000 ms with the following imaging parameters: single slice, in-plane resolution = 1.2 × 1.2 mm², slice thickness = 8 mm, field-of-view = 140 × 140 mm², TR/TE/flip angle = 10 s / 11 ms / 90° and bandwidth = 510 Hz, acquisition duration = 5 hour 18 min.

T₂ mapping

The slice-interleaved T₂ mapping sequence was acquired with the following parameters: 5 slices, in-plane resolution = 2 × 2 mm², slice thickness = 8 mm, slice gap = 4 mm, slice ordering = 1-3-5-2-4, field-of-view = 280 × 280 mm², TR/TE/flip angle = 2.8 ms / 1.42 ms / 55°, a SENSE rate = 2.5, linear ordering, 10 linear ramp-up pulses and bandwidth = 1786 Hz, acquisition duration = 1 min 26 sec. Ten T₂prep images were acquired with T₂prep echo times of 0, 25, 35, 45, 55, 65, 75, 85, 95 ms, and ∞ was simulated with a saturation pulse. For T₂ mapping with 4 echo times, T₂prep images of 0, 25, 55, and ∞ were used for map reconstruction, and the results are reported as T₂ 4echo. A T_2-prep pulse consists of a tip-down slice-selective 90° pulse, followed by four non-selective 180° refocusing pulses that end with a closing tip-up slice-selective 90° pulse [32, 40]. SE T₂ times were obtained using a Carr-Purcell-Meiboom-Gill (CPMG) SE sequence with 32 TEs of 10, 20, 30, …, 320 ms. The imaging parameters were as follows: single slice, in-plane resolution = 1.16 × 1.16 mm², slice thickness = 8 mm, field-of-view = 140 × 140 mm², TR/TE/flip angle = 10 s / 10 ms / 90°, number of signals averaged = 4, bandwidth = 1029 Hz, acquisition duration = 1 hour 21 min.

Map reconstruction

Slice-interleaved T₁ and T₂ maps were reconstructed using both 2-parameter (2P) and 3-parameter (3P) fit models and all results were reported for both 2P and 3P maps. T₁ and T₂ maps were reconstructed offline using MATLAB (MathWorks Inc., Natick, Massachusetts, USA). STONE-bSSFP and STONE-GRE maps were estimated by voxel-wise curve-fitting of the signal with a 2-parameter ($S_{T_1,2P}$) and 3-parameter ($S_{T_1,3P}$) model of the inversion-recovery signal [30]. MOLLI and SE T₁ values were obtained using $S_{T_1,3P}$. For MOLLI, apparent T₁ values were corrected using Look-Locker correction based on the fitted parameters [7].

For slice-interleaved T₂ mapping, a 2-parameter ($S_{T_2,2P}$) and 3-parameter ($S_{T_2,3P}$) curve fitting model of the T₂ signal capturing the effect of imaging pulses on the magnetization was used [41]. SE T₂ values were estimated using $S_{T_2,2P}$. All parameters were estimated using a Levenberg-Marquardt optimizer [42].

Data analysis

A region-of-interest (ROI) was manually contoured once for each vial, and identical ROIs were programmatically applied to all slice-interleaved T₁, T₂ and MOLLI maps throughout all experiments. A graphical illustration of the ROI is shown in S1 Fig. The mean area of the elliptical ROIs of each vial was 73 mm². A separate ROI was manually contoured once and identical ROIs were used for all SE T₁ and T₂ maps throughout all sessions. The mean area of the elliptical SE ROIs of each vial was 90 mm². A linear translation of ROIs less than 1cm in the imaging plane directions was applied in case of offsets from the isocenter. The measurement was defined for each vial as the mean T₁ or T₂ in each ROI and was acquired separately for all slices, repetitions, sessions, vials, and sequences. Data analysis was performed using MATLAB (MathWorks Inc., Natick, Massachusetts, USA).

Statistical analysis

To investigate T₁ and T₂ measurement drift over 20 months, a linear regression was performed for each vial over sessions, and the regression slope and 95% confidence interval (CI) of the slopes were reported. We carried out three analyses to assess the reproducibility and repeatability of the observed slice-interleaved T₁ and T₂ measurements via coefficients of variation, variance component decompositions, linear regressions, and Bland-Altman plots.

Estimation of coefficient of variation

The coefficient of variation (CV), defined as the ratio of the standard deviation to the mean multiplied by 100, was performed to assess reproducibility between sessions, repeatability between repetitions and within a session, and repeatability between slices and within single repetitions. CV was reported as the mean ± standard deviation and visualized by bar plots. To further study variability in T₁ mapping due to different T₁ times, a CV scatter plot for each vial, sorted from shortest T₁ to longest T₁ time, and a Spearman correlation between the CV and T₁ time (vials) was reported. For T₂ mapping, between-session reproducibility CV was estimated for a single vial and therefore no standard deviation among sessions was reported. CV was considered excellent at 0–5%, and good at 5–10%.

Variance decomposition analysis

We considered the observed T₁ and T₂ measurements as random variables whose variability originates from experimental factors and measurement errors. We considered temperature, session, repetition, and slice as the experimental factors and studied how much T₁ and T₂ variability is due to each of these factors. Variance component decomposition analysis [43] yielded an estimation of variance components for each factor. The mean square variance and the variance component to total variance ratio was multiplied by 100, yielding the variability percentage of the respective experimental factor. The analysis was performed for each vial, and we reported the averaged variance and variance ratio of all vials respectively.

Performance analysis against the spin echo

For T₁ mapping, a t-test was performed to assess between-session reproducibility differences between reference SE T₁ measurements and MOLLI vs. slice-interleaved T₁ sequences. Measurement comparison analysis of each sequence to the SE was also performed by using the Pearson correlation between the SE and each sequence. Linear regression was performed and slopes between the sequences and the 95% CI of the slopes were reported. Finally, Bland-Altman analysis was performed to study measurement bias between the two sequences, and the percentage of data points outside of the 95% limits of agreement (mean ± 2 standard deviations) was reported.

For T₂ mapping, the relative CV percentage difference between slice-interleaved T₂ and SE T₂ was reported to assess differences in between-session reproducibility. A measurement comparison analysis to the SE was performed using the Pearson correlation, linear regression, and Bland-Altman analysis. Since only one vial was used for T₂ mapping analyses, slice-interleaved T₂ was averaged over all slices/ repetitions for each of the 37 sessions and compared to the SE T₂ measurement of 37 sessions.

For all analyses, type-I error was set to 0.05. All statistical analyses were performed with SAS software (SAS Institute Inc., Cary, North Carolina, USA).

Results

Thirty-seven imaging sessions were performed from March 7, 2016 to October 31, 2017 The interval between successive sessions was 17±4 days. One session was excluded from the analysis due to incomplete acquisition of the SE T₁ sequence. The isocenter cross marker of the phantom bottle enabled consistent positioning of the phantom throughout the study. Linear translations of ROIs were applied in 6 sessions with the offsets from the isocenter of 2.19±1.20 mm. Examples of T₁ and T₂ weighted images of each sequence are shown in S2 Fig. The temperature of the scanner room over the 20 months duration of experiments was 20.22±1.12°C (range 18–22°C). No measurement drift was observed in vials with low T₁ (<1000 ms) over the 20 month study duration; increased T₁ measurements were observed in vials with high T₁ (>1000 ms) (S3 Fig; S1 Table). No drift in the T₂ measurements was observed over the 20 month study duration (S4 Fig; S2 Table).

T₁ Mapping

Estimation of coefficient of variation

Excellent reproducibility between sessions, and excellent repeatability between repetitions and slices of slice-interleaved T₁ mapping sequences were observed with a CV less than 2% (Fig 2). There was a positive association between the T₁ value and the CV, with longer T₁ times corresponding to higher variability (Fig 3). The Spearman correlation between the T₁ of each vial and the variability of each sequence was as follows: SE T₁ = 0.88, MOLLI = 0.37, STONE-bSSFP 2P = 0.48, STONE-bSSFP 3P = 0.48, STONE-GRE 2P = 0.60, and STONE-GRE 3P = 0.60.

Fig 2. Reproducibility between sessions, and repeatability between repetitions and slices of slice-interleaved T₁ mapping sequences were assessed using coefficients of variation (CV).

Slice-interleaved T₁ mapping sequences showed excellent between-session reproducibility (CV: SE T₁ = 1.1±0.5%, MOLLI = 1.2±0.6%, STONE-bSSFP 2P = 0.8±0.4%, STONE-GRE 2P = 0.8±0.4%, STONE-bSSFP 3P = 1.0±0.3%, STONE-GRE 3P = 1.0±0.4%), between-repetition repeatability (CV: MOLLI = 0.5±0.6%, STONE-bSSFP 2P = 0.3±0.1%, STONE-GRE 2P = 0.3±0.1%, STONE-bSSFP 3P = 0.6±0.2%, STONE-GRE 3P = 0.6±0.2%), and between-slice repeatability (CV: STONE-bSSFP 2P = 0.3±0.2%, STONE-GRE 2P = 0.3±0.1%, STONE-bSSFP 3P = 0.6±0.3%, STONE-GRE 3P = 0.5±0.3%).

Fig 3. Coefficients of variations (CV) shown as scatter plots for each vial.

Vials are sorted from shortest T₁ to longest T₁ time (reference T₁ from the T1MES manual measured by slow inversion-recovery/spin-echo methods at 1.5T: 255, 300, 430, 458, 562, 803, 1090, 1333, and 1489 ms). Vials with higher T₁ time show higher variability.

Variance decomposition analysis

The sources of variability for slice-interleaved T₁ mapping sequences are summarized in Table 1. The main source of variability was temperature when reconstructed with a 2-parameter fit model, and slice location/ orientation when reconstructed with a 3-parameter fit model. Repeated measurements within the same session at the same slice location did not contribute to variability (variance decompositions less than 1%).

Table 1. Sources of variability in T₁ mapping defined by variance decomposition analysis.

T1 Mapping, variance [ms2] (variance ratio [%])
	SE T1	MOLLI	STONE-bSSFP 2P	STONE-bSSFP 3P	STONE-GRE 2P	STONE-GRE 3P
Temperature	151.8 (26.5)	78.9 (30.8)	91.1 (52.8)	108.2 (39.4)	98.3 (51.0)	116.4 (38.3)
Session	99.0 (73.5)	64.0 (32.2)	29.6 (28.5)	35.5 (19.2)	27.3 (26.0)	30.2 (20.4)
Repetition	N/A	52.5 (37.0)	0.0 (0.0)	0.0 (0.1)	0.00 (0.1)	0.0 (0.3)
Slice	N/A	N/A	9.0 (18.8)	22.1 (41.4)	9.9 (22.9)	26.9 (40.9)

In slice-interleaved T₁ mapping, the main source of variability is temperature when reconstructed with a 2-parameter fit model, and slice when reconstructed with a 3-parameter fit. Variability due to repetition is minimal with variance decompositions less than 1%. SE, spin-echo; STONE-bSSFP, slice-interleaved T₁ with balanced steady‐state free precession; STONE-GRE, slice-interleaved T₁ with spoiled gradient echo.

Performance analysis against the spin echo

Between-session reproducibility and comparison of slice-interleaved T₁ mapping sequences against SE T₁ and MOLLI are summarized in Table 2. Slice-interleaved T₁ mapping sequences provided superior between-session reproducibility compared to SE T₁ when reconstructed with a 2-parameter fit model (p<0.05). There were no statistically significant differences between the slice-interleaved T₁ and the reference when reconstructed with a 3-parameter fit model (p>0.05). Slice-interleaved T₁ mapping sequences provided superior between-session reproducibility compared to MOLLI (p<0.05).

Table 2. Between-session reproducibility and the comparison of slice-interleaved T₁ mapping sequences against SE T₁ and MOLLI.

	SE T1	MOLLI	STONE-bSSFP 2P	STONE-bSSFP 3P	STONE-GRE 2P	STONE GRE 3P
Between-session Reproducibility (CV, %)	1.1±0.5	1.2±0.6	0.8±0.4	1.0±0.3	1.0±0.4	1.0±0.4
p-value (vs. SE T1)	N/A	N/A	0.005	0.377	0.001	0.117
p-value (vs. MOLLI)	N/A	N/A	0.011	0.031	0.010	0.024

Slice-interleaved T₁ mapping sequences provided superior reproducibility compared to SE T₁ when reconstructed with a 2-parameter fit model, and no statistically significant difference when reconstructed with a 3-parameter fit model. Slice-interleaved T₁ mapping sequences provided superior reproducibility compared to MOLLI.

Slice-interleaved T₁ mapping showed good agreement to the SE measurement with Pearson correlation coefficients of 1.00 (p<0.001) for all STONE-bSSFP 2P, STONE-GRE 2P, STONE-bSSFP 3P, and STONE-GRE. MOLLI also showed good agreement to the SE with Pearson correlation coefficients of 1.00 (p<0.001). All sequences showed good correlation to SE measurements with regression slopes as follows: MOLLI = 0.94 (95% CI: 0.936–0.945), STONE-bSSFP 2P = 0.95 (95% CI: 0.949–0.958), STONE-GRE 2P = 0.96 (95% CI: 0.957–0.961), STONE-bSSFP 3P = 0.97 (95% CI: 0.971–0.976), STONE-GRE 3P = 0.99 (95% CI: 0.987–0.992) (Table 3).

Table 3. Linear regression analysis of slice-interleaved T₁ mapping sequences against reference SE T₁ measurements.

	Regression Slope (Standard Error)	95% Confidence Interval
MOLLI	0.9407 (0.0023)	0.9362, 0.9452
STONE-bSSFP 2P	0.9536 (0.0024)	0.9489, 0.9583
STONE-bSSFP 3P	0.9737 (0.0011)	0.9715, 0.9759
STONE-GRE 2P	0.9591 (0.0011)	0.9569, 0.9613
STONE GRE 3P	0.9894 (0.0011)	0.9872, 0.9916

All sequences show strong agreement with the reference measurements with regression slopes of 0.9–1.0 and tight 95% confidence limits.

Bland-Altman analysis results for all vials are shown in Fig 4, and the result per each vial is shown in S3 Table. STONE-GRE 3P showed very close T₁ values to the SE with an underestimation less than 1 ms. T₁ bias between SE and other T₁ mapping sequences were as follows: MOLLI = -29.6 ms, STONE-bSSFP 2P = -27.9 ms, STONE-bSSFP 3P = -10.2 ms, STONE-GRE 2P = -25.7 ms. The % of data points outside the 95% limits of agreement were as follows: MOLLI = 3.7%, STONE-bSSFP 2P = 3.4%, STONE-bSSFP 3P = 5.6%, STONE-GRE 2P = 5.3%, STONE-GRE 3P = 5.3%.

Fig 4. Bland-Altman analyses of slice-interleaved T₁ mapping sequences against SE measurements.

The mean difference (bias) is presented as the red line, and the 95% limits of agreement (mean ± 2 standard deviations) are presented as dashed lines. Each data point represents one study time point which was averaged for all repetitions and slices within each session. The T₁ mapping sequences show underestimation compared to the reference measurement. STONE-GRE 3P shows strongest agreement with the reference measurement with an underestimation less than 1 ms.

T₂ mapping

Estimation of coefficient of variation

High reproducibility between sessions, and high repeatability between repetitions and slices of slice-interleaved T₂ mapping sequences were observed with a CV less than 7% (Fig 5).

Fig 5. Reproducibility between sessions, and repeatability between repetitions and slices of slice-interleaved T₂ mapping sequences were estimated using coefficients of variation (CV).

Slice-interleaved T₂ mapping had good between-session reproducibility (CV: SE T₂ = 2.5%, slice-interleaved T₂ 2P = 3.5%, slice-interleaved T₂ 3P = 6.3%, slice-interleaved T₂ 4-T₂preps 2P = 4.7%, slice-interleaved T₂ 4-T₂preps 3P = 6.7%), between-repetition repeatability (CV: slice-interleaved T₂ 2P = 2.7±0.2%, slice-interleaved T₂ 3P was 6.0±0.2%, slice-interleaved T₂ T₂preps 2P = 4.2±0.2%, slice-interleaved T₂ T₂preps 3P was 6.5±0.2%), and good between-slice repeatability (CV: slice-interleaved T₂ 2P = 2.9±0.2%, slice-interleaved T₂ 3P = 6.5±0.3%, slice-interleaved T₂ T₂preps 2P = 4.5±0.4%, slice-interleaved T₂ T₂preps 3P was 7.0±0.4%).

Variance decomposition analysis

The sources of variabilities are summarized in Table 4. The main source of variability was the slice location/ orientation, which represents variability due to spatial location and B₀, B₁ field inhomogeneity. The second source of variability was the temperature. Variability in repeated measurements was minimal with variance decompositions of 0%.

Table 4. Sources of variability in T₂ mapping defined by the variance decomposition analysis.

T2 Mapping, variance [ms2] (variance ratio [%])
	SE T2	Slice-interleaved T2 2P	Slice-interleaved T2 3P	T2 4echo 2P	T2 4echo 3P
Temperature	1.8 (83.5)	2.2 (35.1)	1.4 (12.5)	2.0 (21.4)	1.3 (10.4)
Session	0.4 (16.5)	0.6 (9.7)	0.3 (3.0)	0.5 (5.0)	0.3 (2.4)
Repetition	N/A	0.0 (0.0)	0.0 (0.0)	0.0 (0.0)	0.0 (0.0)
Slice	N/A	3.4 (55.3)	9.3 (84.5)	7.0 (73.6)	10.9 (87.2)

In slice-interleaved T₂ mapping, the main source of variability is slice, representing different spatial locations and different B₀ and B₁ field inhomogeneity. The variability in repeated measurements is minimal with variance decompositions of 0%.

Performance analysis against the spin echo

Slice-interleaved T₂ mapping yielded lower between-session reproducibility than SE T₂ (3.5 vs. 2.5% for slice-interleaved T₂ 2P vs. SE T₂; 6.3 vs. 2.5% for slice-interleaved T₂ 3P vs. SE T₂; 4.7 vs. 2.5% for T₂ 4echo 2P vs. SE T₂; 6.7 vs. 2.5% for T₂ 4echo 3P vs. SE T₂).

Slice-interleaved T₂ mapping showed good correlation with Pearson correlation coefficients of 0.92 for slice-interleaved T₂ 2P, 0.91 for slice-interleaved T₂ 3P, 0.93 for T₂ 4echo 2P, and 0.91 for T₂ 4echo 3P (p <0.001 for all). Slice-interleaved T₂ mapping showed good correlation to SE T₂ with regression slopes as follows: slice-interleaved T₂ 2P = 1.06 (95% CI: 0.91–1.21), slice-interleaved T₂ 3P = 0.78 (95% CI: 0.66–0.90), T₂ 4echo 2P = 1.04 (95% CI: 0.90–1.18), and T₂ 4echo 3P = 0.78 (95% CI: 0.66–0.90) (Table 5).

Table 5. Linear regression of slice-interleaved T₂ mapping sequences against the reference SE T₂ measurements.

	Regression Slope (Standard Error)	95% Confidence Interval
Slice-interleaved T2 2P	1.0594 (0.0767)	0.9091, 1.2097
Slice-interleaved T2 3P	0.7830 (0.0606)	0.6642, 0.9018
T2 4echo 2P	1.0436 (0.0707)	0.9050, 1.1822
T2 4echo 3P	0.7828 (0.0623)	0.6607, 0.9049

Slice-interleaved T₂ sequences show good agreement with reference measurements with regression slopes of 0.8–1.1.

Bland-Altman analysis of slice-interleaved T₂ mapping sequences showed different estimation biases depending on the fitting model (Fig 6). The slice-interleaved T₂ showed overestimation when reconstructed with a 2-parameter fit model (slice-interleaved T₂ 2P = 10.6 ms, T₂ 4echo 2P = 6.3 ms), and an underestimation when reconstructed with a 3-parameter fit model (slice-interleaved T₂ 3P = -6.4 ms, T₂ 4echo 3P = -6.3 ms) against the SE T₂ (Fig 6). The % of data points outside the 95% limits of agreement were as follows: slice-interleaved T₂ 2P = 5.6%, slice-interleaved T₂ 3P = 8.3%, T₂ 4echo 2P = 5.6%, T₂ 4echo 3P = 5.6%.

Fig 6. Bland-Altman plots of slice-interleaved T₂ mapping sequences against the reference SE T₂ measurements.

Bland-Altman analyses of slice-interleaved T₂ mapping shows an overestimation when the map is reconstructed with a 2-parameter fit model, and an underestimation when reconstructed with a 3-parameter fit model. Each data point represents one study time point which was averaged for all repetitions and slices within session. The mean difference (bias) is presented as the red line, and the 95% limits of agreement (mean ± 2 standard deviations) are presented as dashed lines.

Discussion

In this study, we demonstrate highly reproducible long-term measurements of slice-interleaved T₁ and T₂ mapping with a CV less than 2% for T₁ and less than 7% for T₂. Reproducible measurements are essential to detect subtle changes in T₁ and T₂ times due to pathological processes. In particular, assessing long-term measurement stability is necessary for confidently differentiating variability due to disease progression or treatment efficacy over an extended period in a longitudinal study. The current phantom study reports rigorous long-term technical performance of slice-interleaved T₁ and T₂ mapping sequences to better understand baseline variations under controlled conditions.

Regular phantom-based quality control is recommended to ensure stability of a CMR system. Our study reports baseline long term variability which can be used to assess the stability of a CMR system for quality control, and to establish normal and clinical values with expected ranges of variability due to technical confounders. In the long-term time span, factors such as scanner performance can result in systematic differences compared to a shorter time interval. Furthermore, phantom-based quality control allows for T₁ or T₂ accuracy assessment and temperature sensitivity measurements as monitored for each session in our study.

T₁ and T₂ measurements with identical imaging parameters can still vary across session, repetition, and slice due to various factors. Our study showed that the main source of variability in T₁ mapping was temperature when reconstructed with a 2-parameter fit model, and slice when reconstructed with a 3-parameter fit model. Temperature impacts diffusion coefficients [44], which can in turn impact T₁ and T₂. In vials with longer T₁ times where a concentration of Ni²⁺ is low, T₁ becomes more sensitive to temperature due to the temperature sensitivity of the T₁ of water in gel [38]. Imperfect inversion pulses due to field inhomogeneity can be modeled using a 3-parameter fit model [9, 30, 45]. In turn, variability due to slice, representative of B₀, B₁ inhomogeneity, becomes dominant. For T₂ mapping, slice was the main source of variability, which may be associated with differences in B₀ and B₁ field inhomogeneity. Variability in repeated measurements was negligible.

In this phantom study, we used in-vivo protocols currently used in our laboratory to mimic a clinically-relevant setting. Identical in-plane resolution was used to maintain similar TEs for similar performance. We used extra padding around the phantom to create distance from the RF coils to approximate coil geometry and proximity when imaging the human heart. Slice-interleaved T₂ mapping was acquired with 10 T₂prep echo times, previously evaluated in-vivo [20, 22, 32].

The variability observed in the current study shows a similar CV magnitude range as that shown in in-vivo reproducibility studies. Recent shorter-term reproducibility studies in T₁ mapping yield CV magnitude ranges similar to our results, where the CV of ShMOLLI and MOLLI are reported as 2% for 35 patients undergoing repeated measurements the following day [46]. Slice-interleaved T₁ and T₂ show between-day CVs of 2.1% and 6.3%, respectively, in 11 healthy subjects on a 2-day test/retest study [22]. Higher variation is expected in the in-vivo study performed in a longer time span over multiple sessions due to patient-related artifacts such as respiratory and cardiac motion.

We performed long-term between-session reproducibility assessment including SE measurements. Even though SE is typically used as the reference, no study has evaluated its measurement variability over an extended period. Between-session reproducibility of SE measurements was excellent, and slice-interleaved T₁ mapping sequences showed superior between-session reproducibility compared to SE. In particular, STONE-GRE 3P had excellent agreement to SE with similar reproducibility and an underestimation of only < 1ms. Considering the long scan time of SE sequences (typically 5–6 hours for T₁ and 1–2 hours for T₂), alternative sequences for the reference measurement are desirable.

We performed Bland-Altman analyses in each individual vial and for all vials to reflect the unique dependence of T₁ on the bias. Longer T₁ times corresponded with higher T₁ error as previously reported [45]. We studied the measurement variability of slice-interleaved T₁ and T₂ maps reconstructed using both 2-parameter and 3-parameter fit models. For all sequences, higher reproducibility and repeatability was achieved when reconstructed with the 2-parameter fit model; however, the measurement bias was smaller when reconstructed with the 3-parameter fit model. This is in line with previous studies showing higher accuracy but lower precision when fitted with additional parameters [45]. Previous study demonstrated higher precision and reproducibility is achieved by increasing the number of T₂prep echo times from 3 to 14, where the effect nearly saturates above 10 echo times in both phantom and in-vivo studies [47]. Our result shows higher reproducibility for T₂ mapping with 9 T₂prep images compared to 4 T₂prep images as previously reported. We observed higher variability in T₂ mapping, which may be due to lower SNR of the T₂prep sequence due to field inhomogeneities and spoiling gradient.

Our study has several limitations. We studied slice-interleaved T₁ and T₂ mapping sequences on a single MRI scanner at a field strength of 1.5 T. The T1MES phantom used in this study is not optimally designed for studying T₂ mapping; therefore, T₂ analysis was carried out in a single vial with similar myocardial T₁ and T₂ values. A phantom with a different T₂ range needs to be developed to study T₂ reproducibility. The CPMG SE used as a reference of T₂ measurements may be susceptible to stimulated-echo related bias. Our data shows 10.6±1.5% T₂ difference compared to the T₂ measurements by slow SE acquired with 8 TEs from 10–640 ms [38]. We did not study the impact of SNR, although with a relatively large region of interest in the current study, the impact may be negligible. Respiratory and cardiac motion could degrade T₁ and T₂ mapping reproducibility and were not simulated in our phantom study. Future long-term reproducibility studies in humans are warranted to enhance our understanding of measurement variability in a more clinically relevant setting.

Conclusions

Slice-interleaved T₁ and T₂ mapping sequences demonstrate highly reproducible measurement with a coefficient of variation less than 2% for T₁, and 7% for T₂ ranges of < 100 ms measured beyond one year. Slice-interleaved T₁ mapping offers superior reproducibility than both MOLLI and SE T₁ when reconstructed with a 2-paremeter fit model, and slice-interleaved T₂ mapping shows lower reproducibility than SE T₂. All sequences demonstrate strong agreement with reference SE measurements.

Supporting information

S1 Fig. Graphical illustration of the selected ROI on top of a weighted image.

(DOCX)

S2 Fig. Representative examples of T₁ and T₂ maps and weighted images for all sequences.

(DOCX)

S3 Fig. T₁ measurements over 20 months in all 9 vials.

(DOCX)

S4 Fig. T₂ measurements over 20 months in vial ‘F’.

(DOCX)

S1 Table. T₁ measurements over 20 months in all vials.

(DOCX)

S2 Table. T₂ measurements over 20 months in vial ‘F’.

(DOCX)

S3 Table. Bland-Altman analyses performed per each vial for T₁ mapping.

(DOCX)

Data Availability

All DICOM files are available from the Harvard Dataverse database (doi:10.7910/DVN/ATXC5B).

Funding Statement

This work was supported by the National Institutes of Health [grant number R01HL129157, R01HL129122; https://www.nih.gov/] to RN; and the American Heart Association [grant number 15EIA22710040; https://www.heart.org/] to RN. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.