Scan-rescan reproducibility and interobserver reliability of T1 and T2 measurements of extraocular muscles using a multi-dynamic multi-echo MRI sequence

Abstract

Background

Quantitative assessment of extraocular muscles (EOMs) is crucial for the clinical management of thyroid-associated ophthalmopathy, and other orbital disorders. The multi-dynamic multi-echo (MDME) sequence is a synthetic MRI technique acquiring 2D multi-echo images via turbo spin echo or a combination of gradient echo and spin echo methods. The aim of this study was to evaluate the scan-rescan reproducibility and interobserver reliability of T1 and T2 quantification in the EOMs using MDME sequence.

Methods

In this prospective study, twenty-two healthy volunteers underwent two MDME scans (interval, 7–14 days) on a 3T MRI system with a 20-channel head-neck coil. Quantitative T1 and T2 maps were generated using post-processing software. Two observers independently outlined regions-of-interest within the EOMs to extract T1 and T2 relaxation times. Intraclass correlation coefficient (ICC), coefficient of variation (CV), and Bland-Altman analyses were employed to assess reproducibility and agreement.

Results

Nineteen subjects (mean age 40.3 ± 10.5 years) were finally analyzed. The median scan-rescan interval was 9 days [interquartile range (IQR), 7–14]. Scan-rescan reproducibility was good for T1 value (first scan: 933.4 [892.5–1102.5] ms, second scan: 956.4 [856.7–1069.9] ms, ICC = 0.694, bias = 28.1 ms, CV = 8.47%) and moderate for T2 value (first scan: 91.1 [86.5–99.7] ms, second scan: 97.2 [89.5–108.5] ms, ICC = 0.587, bias = − 3.5 ms, CV = 8.42%). Interobserver agreement was excellent (T1: ICC ≥ 0.968, absolute bias ≤ 11.4 ms, CV ≤ 1.92%; T2: ICC ≥ 0.927, absolute bias ≤ 1.3 ms, CV ≤ 2.05%).

Conclusion

The MDME sequence enables rapid and reproducible T1 and T2 quantification of EOMs with excellent interobserver reliability, demonstrating promising potential as a practical and feasible routine tool for quantitative assessment of orbital disorders, particularly thyroid-associated ophthalmopathy.

Clinical trial number

Not applicable.

Introduction

Quantitative assessment of extraocular muscle (EOM) inflammation plays a crucial role in the clinical management of thyroid-associated ophthalmopathy (TAO), also called Graves’ ophthalmopathy (GO) or thyroid eye disease (TED), and other orbital disorders [1–3]. Magnetic resonance imaging (MRI) is a non-ionizing, non-invasive modality that provides superior soft-tissue contrast, exquisite sensitivity to tissue water content, and multi-parametric quantification capabilities, offering distinct advantages over computed tomography [4, 5]. Currently, MRI is widely used to assess inflammation and fibrosis within the EOMs [1, 6]. It enables precise quantification of pathological changes, allowing clinicians to monitor microstructural alterations behind the eyeball and better correlate clinical manifestations with underlying pathology [7].

Among various MRI techniques, T1 and T2 mapping yield quantitative, tissue-specific values that reflect the intrinsic properties of tissues [8]. These techniques have proven effective in diagnosing, staging, and predicting treatment response in patients with active TAO, owing to their high sensitivity to inflammation and edema. T1 mapping, in particular, allows quantitative assessment of TAO activity and response to glucocorticoid therapy in affected patients, and may provide insights into the degree of extraocular muscle fibrosis [9]. It was reported that low T1 values in EOMs are associated with refractory diplopia and may help predict the response of diplopia to intravenous glucocorticoids [10]. Similarly, T2 mapping not only offers a quantitative means to evaluate inflammation and edema but also enables simultaneous assessment of both extraocular muscles and orbital adipose tissue [11], thereby improving the sensitivity for determining TAO activity [12].

Conventional multi-sequence protocols are often time-consuming and prone to motion artifacts. Traditional T2 mapping alone requires 6 min 24 s [13] to 8 min 36 s [14], while combined T1 and T2 mapping frequently exceeds 12 min, depending on MRI protocol settings. These limitations not only diminish patient compliance, especially among claustrophobic, pediatric, or elderly patients, but also impair the reliability of imaging outcomes. Synthetic MRI addresses these issues by using quantitative measurements of multiple tissue properties to generate various image contrasts from a single scan [15]. Unlike conventional MRI protocols where key imaging parameters such as repetition time (TR), echo time (TE), and inversion time (TI) are fixed, synthetic MRI allows for post-processing adjustment of these parameters through mathematical modeling, eliminating the need for pre-scan parameter definition [16, 17].

The multi-dynamic multi-echo (MDME) sequence is a synthetic MRI technique that acquires two-dimensional multi-echo images using either turbo spin echo or a combination of gradient echo and spin echo methods, both with and without radiofrequency pulse saturation; each image is obtained with various combinations of echo times and delay times to enable quantitative measurement of tissue relaxometry parameters [18]. It can simultaneously quantify relaxation times and generate multiple synthetic weighted images in a single scan. Its applications are particularly promising for the evaluation of brain diseases, such as multiple sclerosis, brain tumors, stroke, with additional utility extending other organs and sites, e.g., breast, spine, and shoulder [18–22]. Recently, Wang et al. reported that T1 and T2 relaxation times from synthetic MRI can distinguish active from inactive TAO patients, suggesting that quantitative synthetic MRI may assist clinicians in accurately detecting active TAO to enable timely intervention [23]. Despite the clinical potential, there is a scarcity of reported data on the scan-rescan reproducibility of EOMs assessments via the MDME technique. Therefore, the purpose of this study was to evaluate the scan-rescan and interobserver reproducibility of T1 and T2 measurements in the EOMs using an MDME sequence. It’s a technical reproducibility investigation conducted in healthy volunteers, intended as a methodological foundation for the conduct of future studies on TAO and orbital diseases applications.

Materials and methods

Participants

This prospective study was in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Shanghai General Hospital, and individual consent was obtained for each subject. Twenty-two consecutive healthy volunteers were prospectively recruited from March 2025 to June 2025 in the present study. All volunteers were in good health and had no history of ocular or neurological disorders. The following criteria were used to exclude subjects: implants in the body, claustrophobia, and poor image quality caused by motion artifacts or low signal-to-noise ratio(SNR).

While no formal a priori sample size calculation was performed, a theoretical sample size estimation assuming an expected ICC of 0.8, two raters, α = 0.05, and power = 0.80 indicated that approximately 8 participants would be sufficient for reproducibility assessment. This estimate provides supportive context for the study cohort and is comparable to sample sizes frequently reported in ICC-based reproducibility studies [24–29].

The demographic characteristics of each subject were documented, including age, sex, and the time interval between scan-rescan MRI examinations.

Magnetic resonance imaging

All MRI examinations were performed on a 3-Tesla whole-body scanner (MAGNETOM Vida, Siemens Healthcare, Erlangen, Germany) using a 20-channel head/neck coil for magnetic resonance signal reception. Participants were scanned in the supine position with their eyes centered in the magnet bore and kept closed throughout the examinations. To minimize head motion during image acquisition, two soft cushions were placed between the head and coil. Acquisition parameters for the MDME sequence were carefully selected and optimized based on established principles for T1 and T2 relaxation time mapping and clinical practicality [26, 30–33]. The TR and TE were chosen to maximize SNR while minimizing T1 and T2 weighting bias and ensuring full coverage of the relaxation decay curve for accurate quantitative fitting. The TIs were optimized to optimize T1-weighted signal contrast, suppress signals from non-target tissues, and enable robust T1 curve fitting. The numbers of dynamic phases and echo trains were optimized to balance temporal resolution, spatial resolution and scan duration in order to reduce motion-related artifacts, and the flip angle was set to minimize radiofrequency energy deposition while preserving sufficient signal for reliable quantitative estimation. Matrix size and slice thickness were selected to achieve adequate anatomical coverage and spatial resolution for clinical analysis, with bandwidth optimized to reduce chemical shift artifacts and improve signal stability. The main parameters of the sequence were as follows: TR, 4410 ms; TEs, 23, 101 ms; TIs, 174, 615, 2085, and 4290 ms; flip angle, 150°; field-of-view, 186 × 230 mm²; matrix, 208 × 320; slice thickness, 3.0 mm; slice gap, 0.75 mm; number of slices, 30; parallel acceleration factor, 3; echo train length, 10; receiver bandwidth, 120 Hz/pixel; average, 1; acquisition time, 4 min 32 s.

To evaluate scan-rescan reproducibility of T1 and T2 quantification using the MDME sequence, all subjects underwent two imaging sessions separated by a time interval of no less than seven days. The scanner’s built-in AutoAlign (Head > Orbits) technique was utilized to automatically localize the orbital region, thereby minimizing slice positioning variability between the two scans.

Post-processing

All raw MDME data were sent to a dedicated workstation and further processed using home-built synthetic MR software. This post-processing tool automatically generates quantitative relaxation time maps of T1, T2, and proton density (PD), as well as synthetic qualitative weighted contrasts images of T1-weighted, T2-weighted (T2W), T1-weighted fluid attenuated inversion recovery (FLAIR), T2W FLAIR, short tau inversion recovery (STIR), phase-sensitive inversion recovery, and double inversion recovery sequences. The present study mainly focused on the quantification of T1 and T2 parametric maps for all subjects.

Image analysis

The T1, T2 maps, and synthetic T2W images were imported into ITK-SNAP software (version 4.2.0; www.itksnap.org) for quantitative analysis. Two observers (LC and HZ with 12 and 3 years of experience in MRI, respectively) independently performed the measurements and were blinded to subject identity and scan order.

Elliptical region-of-interests (ROIs; mean area 32.8 ± 4.5 mm²) were manually delineated on the synthetic coronal T2W images for the four EOMs (superior, inferior, medial, and lateral recti) of each eye. The analyzed imaging slice was positioned approximately 3 mm posterior to the globe. Visible vessels, tendinous portions, or voxels with partial volume from fat or orbit were excluded. The ROIs were then automatically copied onto the corresponding T1 and T2 maps (Fig. 1). For each subject, the mean T1 and T2 relaxation times of the eight extraocular muscles (four muscles for each eye) were calculated and recorded as the mean T1 and T2 values, respectively.

Fig. 1.

Example of scan–rescan analysis of T1 and T2 relaxation times in the extraocular muscle acquired using a multi-dynamic multi-echo sequence, with the imaging slice positioned approximately 3 mm posterior to the globe. The first row (a–d) and second row (e–h) show the first and second scans acquired 10 days apart. (a, e) T2-weighted images, (b, f) T1 maps, (c, g) T2 maps, and (d, h) T2-weighted images with ROI delineation. The same ROI was used for T1 and T2 quantification

Statistical analysis

Statistical analyses were performed using SPSS (v19.0; IBM Corp., Armonk, NY) and MedCalc (v18.2.1; MedCalc Software Ltd, Ostend, Belgium). The normality of quantitative variables was assessed using the Shapiro-Wilk test. Continuous variables are presented as mean ± standard deviation (SD) for normally distributed data or median (IQR) for non-normally distributed data, while categorical variables are expressed as percentages. The Wilcoxon signed-rank test was used to compare T1 and T2 values between scans and between observers. Coefficient of variation (CV), intraclass correlation coefficient (ICC), and Bland-Altman plot were used to evaluate the scan-rescan and interobserver reproducibility assessment. CV was calculated as CV = (SD / mean) × 100%, where mean and SD were derived from paired T1 or T2 measurements. For both scan–rescan and interobserver analyses, CV values are reported as mean ± SD. ICC were computed using a two-way mixed-effects model with absolute agreement and single-measure definition. Values were interpreted as follows: 0.00–0.20, poor concordance; 0.21–0.40, fair concordance; 0.41–0.60, moderate concordance; 0.61–0.80, good concordance; and 0.81–1.00, excellent concordance [34]; and Bland-Altman plots were generated to visualize the agreement by plotting the differences against the means of paired measurements. A p-value < 0.05 was considered statistically significant difference.

Results

Participants

All of the 22 volunteers completed the two scan sessions. Among them, three volunteers (13.6%) were excluded due to motion artifacts or inadequate SNR compromising image quality. Consequently, 19 volunteers (13 males, 68.4%; mean age 40.3 ± 10.5 years, range 24–60) were included in the final analysis (Table 1). The median scan-rescan interval was 9 [IQR, 7–14] days.

Table 1.

Demographic characteristics of the study cohort

Variables	Values
No. of subjects(n)	19
Mean age (years)	40.3 ± 10.5
Sex (male/female)	13/6
Scan-rescan interval (days)	9 [IQR, 7–14]

Scan-rescan reproducibility of T1 and T2 measurements

For T1 relaxation times, median values were 933.4 [892.5–1102.5] ms and 956.4 [856.7–1069.9] ms for the first and second scans, respectively, showing good agreement (ICC = 0.694) with a mean bias of 28.1 ms (Fig. 2a), and a mean CV of 8.47% ± 5.77% (Table 2). For T2 relaxation times, median values were 91.1 [86.5–99.7] ms and 97.2 [89.5–108.5] ms for the first and second scan, respectively, with moderate agreement (ICC = 0.587), a mean bias of -3.5 ms (Fig. 2b) and a mean CV of 8.42% ± 5.67% (Table 2). No statistically significant differences in T1 and T2 values were observed between the two scans (all p > 0.05).

Fig. 2.

Bland-Altman plots of scan-rescan reproducibility for T1 (a) and T2 (b) relaxation times. Dashed horizontal lines (wine red) indicate 95% limits of agreement, and solid center lines (blue) represent mean bias

Table 2.

Scan-rescan reproducibility of T1 and T2 measurements using the MDME sequence in extraocular muscles

Relaxation time	Scan 1 Median [IQR] (ms)	Scan 2 Median [IQR] (ms)	ICC (95% CI)	Bias (LoA, ms)	CV (Mean ± SD–%)
T1	933.4 [892.5–1102.5]	956.4 [856.7–1069.9]	0.694 (0.374 to 0.866)	28.1 (-259.3 to 315.5)	8.47 ± 5.77
T2	91.1 [86.5–99.7]	97.2 [89.5–108.5]	0.587 (0.219 to 0.812)	-3.5 (-32.3 to 25.3)	8.42 ± 5.67

CI, confidence interval; CV, coefficient of variation; ICC, intraclass correlation coefficient; IQR, interquartile range; LoA, limit of agreement;

MDME, multi-dynamic multi-echo; SD, standard deviation

Interobserver reproducibility of T1 and T2 measurements

For the first scan session, median T1 relaxation times were 951.4 [896.0–1092.5] ms for Observer 1 and 963.4 [893.0–1117.3] ms for Observer 2, showing excellent agreement (ICC = 0.971) with a bias of -8.0 ms and mean CV of 1.80% ± 1.76% (Table 3; Fig. 3a). Corresponding T2 values were 90.2 [86.5–100.0] ms and 92.8 [85.8–100.0] ms, also demonstrating excellent agreement (ICC = 0.932; bias 0.8 ms; CV 1.72% ± 3.32%) (Table 3; Fig. 3b).

Table 3.

Interobserver reproducibility of T1 and T2 measurements using the MDME sequence in extraocular muscles

Scans	Relaxation time	Observer 1 Median [IQR] (ms)	Observer 2 Median [IQR] (ms)	ICC (95% CI)	Bias (LoA, ms)	CV (Mean ± SD, %)
Scan 1	T1	951.4 [896.0–1092.5]	963.4 [893.0–1117.3]	0.971 (0.929 to 0.989)	-8.0 (-76.9 to 61)	1.80 ± 1.76
	T2	90.2 [86.5–100.0]	92.8 [85.8–100]	0.932 (0.833 to 0.972)	0.8 (-10.0 to 11.5)	1.72 ± 3.32
Scan 2	T1	964.5 [851.4–1088.4]	959.5 [861.9–1055.7]	0.968 (0.919 to 0.988)	11.4 (-57.6 to 80.4)	1.92 ± 1.40
	T2	96.9 [89.2–108.2]	95.0 [90.7–109.1]	0.927 (0.823 to 0.9701)	-1.3 (-10.3 to 7.7)	2.05 ± 2.52

CI, confidence interval; CV, coefficient of variation; ICC, intraclass correlation coefficient; IQR, interquartile range; LoA, limit of agreement; MDME,

multi-dynamic multi-echo; SD, standard deviation

Fig. 3.

Bland-Altman plots of interobserver agreement for (a) T1 and (b) T2 relaxation times (first scan session). Dashed horizontal lines (wine red) indicate 95% limits of agreement, and solid center lines (blue) represent mean bias

During the second scan, Observer 1 recorded T1 values of 964.5 [851.4–1088.4] ms, compared with 959.5 [861.9–1055.7] ms recorded by Observer 2, showing similarly excellent concordance (ICC = 0.968; bias 11.4 ms; CV 1.92% ± 1.40%) (Table 3; Fig. 4a). For T2 measurements, values were 96.9 [89.2–108.2] ms and 95.0 [90.7–109.1] ms, respectively, with similarly excellent agreement (ICC = 0.927; bias − 1.3 ms; CV 2.05% ± 2.52%) (Table 3; Fig. 4b). Interobserver comparisons of T1 and T2 measurements revealed no statistically significant differences (all p > 0.05).

Fig. 4.

Bland-Altman plots of interobserver agreement for (a) T1 and (b) T2 relaxation times (second scan session). Dashed horizontal lines (wine red) indicate 95% limits of agreement, and solid center lines (blue) represent mean bias

Sample size justification

Based on the observed minimum ICC of 0.587, a post-hoc power analysis (PASS 15) for the two-way mixed-effects ICC model (absolute agreement) with α = 0.05 and two raters indicated that the sample size of 15 participants achieved a statistical power of 0.802. This supports that the study cohort provided sufficient information to assess reproducibility.

Discussion

Quantitative MRI of the EOMs provides an objective approach for evaluating ocular inflammatory diseases. However, its routine clinical application has been limited by prolonged scan times. The MDME technique has been increasingly used for assessing various diseases, yet its utility in orbital disorders remains limited. In this study, we incorporated both T1 mapping and T2 mapping using an MDME sequence to acquire comprehensive microstructure information of the EOMs, thereby primarily demonstrating the potential of this technique for quantitative assessment of EOMs. Our results revealed that the MDME scan achieves moderate to good scan-rescan reproducibility and excellent interobserver reproducibility for T1 and T2 quantification in the EOMs. These findings support the potential clinical value of the MDME sequence in the management of orbital disorders, particularly TAO, while significantly reducing acquisition time.

As previously reported, both T1 [9, 35, 36] and T2 values [14, 37] serve as objective biomarkers for distinguishing disease activity and monitoring treatment response in patients with TAO. Elevated native T1 values are associated with EOM inflammation, while postcontrast T1 may potentially reflect fibrotic changes within the EOMs [9]. Similarly, the signal intensity ratio of T2 relaxation time has been independently associated with active TAO [12]. Advances in imaging technology and algorithmic design have enabled MRI to characterize and quantify soft-tissue microstructure with unprecedented speed and accuracy. T1 mapping, a technique that quantifies T1 relaxation times, has demonstrated strong utility in detecting fibrotic changes in various conditions, such as cardiomyopathy, liver fibrosis, and chronic pancreatitis [38–40]. This method has also been adapted for orbital imaging to support the clinical evaluation of TAO [41, 42]. Previous studies have shown that T1 values in the extraocular muscles are significantly lower in patients with inactive TAO compared to those with active disease. T2 mapping, which is sensitive to changes in water content within tissues [43], can indicate acute inflammation in muscle tissue. Increased T2 values provides key insights into pathological changes in the EOMs of patients with Myasthenia Gravis [44] as well as TAO [45].

The scan-rescan CV values for EOM T1 and T2 quantification were < 9%, representing technical variability that is substantially smaller than the pathological T1 and T2 relaxation time shifts observed in TAO [46, 47]. Previous studies have reported pronounced pathological T1 changes (> 20% relative to healthy controls) in active TAO EOMs, with longitudinal T1 shifts associated with disease activity and treatment response [9, 42]. Similarly, T2 measurements variability similarly remains small relative to pathological T2 shifts in active TAO [46, 47]. In addition, Bland-Altman plots showed that the limits of agreement (LoAs) for T1 and T2 measurements between repeated scans and observers were narrow, with minimal mean bias. Although small random deviations were present within the LoAs, the magnitudes of these deviations were negligible for clinical interpretation, and thus the LoAs were considered clinically acceptable for the present study’s clinical context. Therefore, despite moderate scan-rescan variability, the MDME sequence provides sufficient reliability to differentiate pathological from normal EOM tissues. In clinical context, the variability observed in this healthy cohort serves as an important reference: T1 and T2 alterations detected in TAO patients that exceed this healthy baseline can be interpreted as pathological changes associated with disease activity, rather than technical measurement noise. This suggests that our MDME-based quantification method is robust for clinical TAO assessment, with the capacity to distinguish true disease-related changes from minor technical variability.

The ICC values of scan-rescan reproducibility for T1 and T2 quantification using MDME ranged from 0.587 to 0.694, which fell into the lower segment of the “moderate to good” range. Several factors inherent to the MDME sequence and our imaging acquisition protocol may have contributed to this moderate scan-rescan variability [30, 32, 33]. First, a parallel imaging acceleration factor of 3 was applied to balance scan efficiency and reduce motion artifacts, such acceleration can introduce minor signal inhomogeneities across repeated scans, which may affect the consistency of T1 and T2 fitting. Second, intrinsic SNR limitations of the MDME sequence for quantitative T1 and T2 mapping should be noted: the trade-off between MDME sampling and scan duration resulted in moderate SNR, and subtle signal fluctuations between scans may propagate to small deviations in relaxation time quantification and thus lower ICC values. Third, modeling constraints specific to the MDME sequence represent a key contributing factor: the MDME sequence relies on a model-based fitting algorithm for T1 and T2 quantification, which requires iterative fitting of relaxation curves across multiple TIs and TEs; minor scan-rescan variations in TI and TE timing calibration and magnetization preparation can introduce small systematic biases in curve fitting, further contributing to scan-rescan variability. Fourth, minor scan-rescan positional misalignment may have contributed moderately to the observed EOM measurement variability, even with strict standardization of head position, eye closed, and slice alignment. Even small deviations in head posture (e.g., subtle axial rotation or vertical tilt) or minor shifts in eye closed can alter the anatomical orientation of the EOMs, leading to small differences in muscle cross-sectional area, thickness, T1 and T2 measurements on repeated scans [46, 47]. Besides, the lower scan-rescan reproducibility of T2 relaxation time (moderate) compared with T1 (good) for EOM quantification may be attributed to the use of only two echoes for T2 calculation, resulting in reduced fitting robustness, together with the lower SNR and greater motion sensitivity at longer echo times.

A direct comparison between MDME-derived T1 and T2 values and conventional gold-standard reference mapping techniques was not included in this study, as the primary focus was scan-rescan reproducibility in healthy volunteers. Nevertheless, previous studies have shown strong quantitative agreement and high accuracy of MDME-derived T1 and T2 measurements compared with conventional reference mapping methods [16, 30]. Future studies incorporating direct methodological comparisons in both healthy volunteers and TAO patients would further strengthen the evaluation of MDME accuracy for EOM quantification.

The MDME sequence’s capability to acquire quantitative parameters simultaneously within a 5-minute scan addresses a major limitation of conventional MRI protocols. For patients with TAO, many of whom may experience discomfort from orbital pressure or claustrophobia, shorter scan times minimize motion artifacts and also enhance patient cooperation. This advantage is particularly important in the context of longitudinal monitoring and follow-up. The compatibility of the MDME sequence with synthetic MRI reconstruction further enhances its clinical utility. It allows clinicians to access both quantitative relaxation parameters and conventional anatomical contrasts within a single scan, thereby overcoming the traditional trade-off between quantitative data and diagnostic image quality. For instance, a radiologist can identify enlargement of the EOMs on synthetic T2W images and then consult corresponding MDME-derived T2 values to objectively assess the severity of edema—all acquired from a single scan.

The current study has several limitations. First, the study cohort consisted exclusively of healthy volunteers, future studies are needed to validate the MDME sequence in TAO cohorts across different activity stages to evaluate T1 and T2 reproducibility in pathological EOM. Second, the sample size was relatively small. Although it is consistent with published ICC reproducibility studies and supported by post-hoc power analysis with adequate power (> 0.80), ICC confidence intervals are inherently sensitive to sample size. Consequently, a larger sample may yield narrower CIs and more precise reproducibility estimates. Future studies with a larger cohort are warranted to further validate our findings. Third, it was a single-center study conducted on one vendor scanner with a 20-channel head-neck coil, which may introduce vendor- and hardware-specific biases. Multi-center, multi-vendor studies with standardized protocols and dedicated orbital coils should be conducted to assess the reproducibility. Fourth, this study only focused on global EOM reproducibility in a small cohort rather than muscle-specific patterns. Average T1 and T2 values across all eight rectus muscles per subject improve precision but may obscure systematic differences between individual muscles. Fifth, intra-observer reproducibility was not evaluated and will be assessed in future work. Lastly, all ROIs were drawn on a single slice for each muscle. EOMs are not homogeneous along their length, thus single-slice sampling may introduce bias. Further studies should involve the development and validation of automated ROI segmentation tools to minimize observer-dependent variability.

Conclusion

In conclusion, the MDME sequence provides reproducible T1 and T2 measurements of the EOMs, effectively combining scan efficiency, measurement reproducibility, and multi-parametric capability. This approach addresses major limitations associated with conventional EOM MRI. Although further studies involving patients with TAO and multi-center validation are warranted to fully establish its clinical utility, the current results support its feasibility and potential as a practical routine tool, pending validation in patients with TAO and other orbital disorders. It is particularly promising in clinical scenarios requiring rapid and standardized quantitative assessment of the EOMs, such as staging TAO activity, monitoring treatment response, and facilitating long-term follow-up in patients with orbital disorders.

Abbreviations

CT

Computed tomography

CV

Coefficient of variation

EOM

Extraocular muscle

FLAIR

Fluid attenuated inversion recovery

ICC

Intraclass correlation coefficient

LoA

Limit of agreement

MDME

Multi-dynamic multi-echo

MRI

Magnetic resonance imaging

PD

Proton density

STIR

Short tau inversion recovery

SNR

Signal-to-noise ratio

TAO

Thyroid-associated ophthalmopathy

TE

Echo time

TI

Inversion time

ROI

Region-of-interest

TR

Repetition time

Author contributions

RC, JL, and QW conceived and designed the study. JL, ZW, and LC wrote the manuscript. HX, XW, HS and YS performed the study. ZW acquired MR images. LC and HZ performed the image analysis. JL, LC, HZ, and MC analyzed the data. RC is the principal investigator of this work and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Funding

This work was supported by the National Key Research and Development Program of China: 2024YFA1802701; National Natural Science Foundation of China (NSFC): 82270826.

Data availability

The authors declare that the data supporting the findings of this study are available within the paper. The original data of this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was performed in accordance with the second Helsinki declaration and was approved by the Medical Ethics Committee, Shanghai General Hospital (2023SQ098). All participants provided written informed consents before voluntary participation.

Consent for publication

Informed consent for publication was obtained from all authors.

Competing interests

The authors declare no competing interests.