3D Zero Echo Time (ZTE) MRI versus 3D CT for Glenoid Bone Assessment

Abstract

Purpose:

To evaluate the 3D ZTE MRI technique and compare with 3D CT for the assessment of the glenoid bone.

Methods:

ZTE MRI using multiple resolutions and multislice CT was performed on six shoulder specimens before and after creation of glenoid defects and ten glenohumeral instability patients. Two musculoskeletal radiologists independently generated 3D volume rendered images of the glenoid en face. Postprocessing times and glenoid widths were measured. Intermodality and interrater agreement was assessed.

Results:

Intraclass correlation coefficients (ICCs) for intermodality assessment showed almost perfect agreement for both readers, ranging from 0.949-0.991 for the ex vivo study and 0.955-0.987 for the in vivo patients. Excellent interobserver agreement for both the ex vivo (ICCs ≥ 0.98) and in vivo (ICCs ≥ 0.92) studies was demonstrated. For the ex vivo study, Bland-Altman analyses for CT vs MRI demonstrated a mean difference of 0.6-1 mm at 1.0 mm³ MRI resolution, 0.3-0.6 mm at 0.8 mm³ MRI resolution, and 0.3-0.6 mm at 0.6 mm³ MRI resolution for both readers. For the in vivo study, Bland-Altman analyses for CT vs MRI demonstrated a mean difference of 0.6-0.8 mm at 1.0 mm³ MRI resolution, 0.5-0.6 mm at 0.8 mm³ MRI resolution, and 0.4-0.8 mm at 0.7 mm³ MRI resolution for both readers. Mean post-processing times to generate 3D images of the glenoid ranged from 32-46 seconds for CT and 33-64 seconds for ZTE MRI.

Conclusions:

3D ZTE MRI can potentially be considered as a new technique to determine glenoid width and can be readily incorporated into the clinical workflow.

Level of Evidence:

2, development of diagnostic criteria (consecutive patients with consistently applied reference standard and blinding).

INTRODUCTION

The glenohumeral joint is the most commonly dislocated joint, comprising around 50% of all major joint dislocations.¹ Trauma from falls, contact sports and motor vehicle accidents are the most common causes.^{2, 3} In patients with a history of shoulder dislocation and suspicion of glenoid bone loss, reliable evaluation of the glenoid bone is of great importance since it can affect surgical planning and the recommendation for bone grafting.^{4, 5} Amongst different imaging methods available, studies have shown that three-dimensional (3D) computed tomography (CT) is the best for assessing the glenoid bone ^6–8, demonstrating high accuracy and high inter- and intra-observer agreement, and is therefore the preferable preoperative study. However, considering that shoulder MRI remains the imaging modality of choice for soft tissue evaluation, many patients will need to undergo an additional CT examination to evaluate the glenoid bone, increasing costs, patient burden and exposing patients to ionizing radiation.

The assessment of both soft tissue and bone in a single examination for patients with shoulder dislocation would be highly desirable. Various MRI techniques have been studied to evaluate the glenoid bone, in order to try to replace CT and optimize preoperative planning. However, typical MRI sequences with longer echo times (TEs) are limited in their assessment of cortical bone. Essentially no signal remains at the time of acquisition and the cortical bone appears black.⁹ The contours of cortical bone are inferred by visualizing fatty bone marrow or surrounding soft tissues, potentially resulting in imprecise localization.¹⁰ In some studies, with MRI sequences employing TEs of several milliseconds, acceptable accuracy has been achieved, but typically long post-processing times are required to provide adequate 3D volume-rendered images, limiting their clinical utility.^{11, 12} Newer techniques applying ultrashort echo times (UTE), defined as TEs less than 1 millisecond, and zero echo times (ZTE) can help overcome these issues, since cortical bone can be directly imaged and the contrast is more similar to that obtained from CT studies ^{13, 14}, possibly helping decrease post-processing time. ZTE has been previously used to image the osseous structures of the head ^{15, 16}, shoulder ¹³, and pelvis ¹⁷.

The purpose of this study is to evaluate the 3D ZTE MRI technique and compare with 3D CT for the assessment of the glenoid bone. It was hypothesized that the 3D MRI and 3D CT would offer similar results with high intermodality and interrater agreement.

MATERIALS AND METHODS

Institutional review board approval was obtained for this Health Insurance Portability and Accountability Act–compliant study, that included both ex vivo specimens and glenohumeral instability patients. The time frame for our study was from February 1, 2019 to Oct 28, 2019. Written informed consent was obtained prior to patient participation.

Cadaveric Study and Subject Recruitment

An ex vivo study was performed to ensure a range of clinically relevant glenoid widths and to optimize the imaging protocol for the in vivo study. Six shoulder specimens (5 males, 1 female; age range, 21-57 years; mean age, 43.5 years; 3 left/3 right shoulders) were used for the cadaveric portion of this study. The specimens were fresh-frozen at −80°C and thawed at room temperature for 24 hours prior to use. The soft tissues and humeri were removed with a scalpel, leaving just the scapulae. Scapulae specimens were embedded in plastic containers containing 4% agarose gel and imaged with the CT and MRI protocols listed below. Thereafter, the bony specimens were removed from the agarose gel and glenoid lesions were created with a band saw to mimic osseous Bankart-lesions with approximately 15-20% bone loss. Specimens were once again embedded in agarose gel and CT and MRI protocols were repeated.

For the in vivo study, ten consecutive patients presenting to our clinic with a history of glenohumeral joint instability were recruited. Patients were assessed by a fellowship-trained orthopedic surgeon specializing in sports medicine (HH, 30 years of experience). Patients were included if the ZTE MRI research exam could be performed within 6 months of their clinical CT shoulder exam. Patients who had contraindications to MRI were excluded.

CT and ZTE MRI Protocols

CT scanning was performed on a 64-slice scanner (VCT, GE Healthcare, Milwaukee, WI) with the following scan parameters: 0.625 mm slice thickness, reconstruction matrix 512 x 512 pixels, field of view 20-30 cm, 120 kVP, and pitch factor of 0.98. MR imaging was performed on a 3T system (MR750 3T, GE Healthcare, Milwaukee, WI) using a three-channel shoulder coil. The ZTE MRI sequence has been described ^{13, 17}, however the most recent vendor-provided version included rapid, fully automated on-line processing (on the scanner). The executed algorithms included signal bias correction and linear contrast inversion ¹⁷, yielding “CT-like” 2D images automatically after scanning (Figure 1). The following ZTE scan parameters were used: echo time (TE) 0, repetition time (TR) 0.8-1.1 ms, flip angle 1°, bandwidth ± 62.50 kHz, number of excitations 1-3, and field of view 20 cm. The ZTE sequence was always repeated three times to achieve the following isotropic voxel sizes: 1 mm³, 0.8 mm³, and 0.6-0.7 mm³ with corresponding scan times of approximately 3-4, 8-9, and 10-13 minutes, respectively. For the cadaveric study the highest resolution used was 0.6 mm³ and for the in vivo study the highest resolution used was 0.7 mm³since the in vitro experiments were performed prior to the in vivo experiments. When pilot protocol testing was initiated on volunteers, it was noted that remaining still for a sequence lasting ~13 minutes was uncomfortable and resulted in motion artifacts in some cases. However, all pilot volunteers were able to hold still for ~10 minutes and therefore we reduced the resolution to 0.7 mm³ to accommodate..

Figure 1:

Axial CT and ZTE MRI images obtained of the left shoulder of a 38-year-old man. (A) CT, (B) ZTE MRI 1.0 mm, (C) ZTE MRI 0.8 mm and (D) ZTE MRI 0.7 mm axial images all demonstrate high-contrast imaging of the osseous structures, including the glenoid and glenohumeral joint. ZTE, zero echo time.

Glenoid Evaluation

Measurements were independently done by two musculoskeletal radiologists (with 10 and 8 years of experience) on randomized and anonymized CT and MRI images to establish intermodality and interrater agreement. The axial images were used to reconstruct 3D volume-rendered of the glenoid. 3D CT and 3D MRI images were exported into an external viewing program (OsiriX MD DICOM viewer, version 11, Pixmeo, Bernex, Switzerland). Glenoid widths were evaluated using images of the en face glenoid on the 3D rendered images. The width was defined as the largest anteroposterior measurement in the lower two-thirds of the glenoid, measured perpendicular to the long axis.¹⁸ Post-processing times to generate the surface-rendered 3D glenoid images from the original 2D images for both CT and ZTE MRI were also measured for the in vivo cases.

Statistical Analysis

Descriptive statistics and paired t-tests were performed to assess post-processing times for ZTE MRI at the different resolutions compared with CT. Intraclass correlation coefficients (ICC) were calculated for the 3D ZTE MRI images with varying resolutions compared to the reference standard images (3D CT) for each reader. Interrater agreement was also assessed with ICCs. Agreement interpretation was considered according published standards: 0.01 poor agreement, 0.01 to 0.20 slight agreement, 0.21 to 0.40 fair agreement, 0.41 to 0.60 moderate agreement, 0.61 to 0.80 substantial agreement, and 0.80 to 1.0 almost perfect agreement.¹⁹ Bias and limits of agreement values (1.96 X SD_diff were calculated with the methods of Bland and Altman.²⁰ P-values less than 0.05 were considered statistically significant. Statistical analyses were performed using the SPSS software package (version 21; SPSS, Chicago, IL, USA).

RESULTS

Ex Vivo Study

All 3D ZTE MR and 3D CT images that were obtained on ex-vivo specimens yielded high quality reconstructions (Figure 2). ICCs for intermodality and interobserver assessment showed almost perfect agreement on the ex vivo study for both readers and all three MRI resolutions (ICCs ≥ 0.95; 95% CI, 0.60-1.0; all P values < 0.001) (Table 1). Bland-Altman analysis revealed a bias (mean difference) of 1 mm or less for all measurements of both readers (Table 2). For CT vs MRI at 1.0 mm³ resolution, there was a bias of 0.6 mm (limits of agreement [LoA]: −0.7, 1.9) for Reader 1 and bias of 1.0 mm (LoA: −1, 3.0) for Reader 2. For CT vs MRI at 0.8 mm³ resolution, there was a bias of 0.3 mm (LoA: −0.7, 1.3) for Reader 1 and bias of 0.6 mm (LoA: −1, 2.2) for Reader 2. For CT vs MRI at 0.6 mm³ resolution, there was a bias of 0.3 mm (LoA: −1, 1.6) for Reader 1 and bias of 0.6 mm (LoA: −1.5, 2.7) for Reader 2. Figures 3 shows the Bland-Altman plots demonstrating high inter-modality agreement for each reader when comparing 3D CT and 3D ZTE MRI at the three different resolutions. Based on the Bland-Altman plots, no apparent differences in bias were noted for the glenoids pre- or post-defect creation. Similarly, no apparent differences were noted based on glenoid lesion size (e.g., the smallest glenoid measurements did not demonstrate larger differences).

Figure 2:

Glenoid width measurement method shown in an ex-vivo right scapula (21-year-old donor). (A) 3D CT, (B) 3D ZTE MRI 1.0 mm, (C) 3D ZTE MRI 0.8 mm and (D) 3D ZTE MRI 0.6 mm images with en face view of the glenoid shows the width measurement, defined as the largest anteroposterior measurement in the lower two-thirds of the glenoid, measured perpendicular to the long axis. ZTE, zero echo time.

Table 1.

Intraclass Correlation Coefficients (Ex Vivo Study, n=12)

Intermodal Assessment
	Reader 1		Reader 2
Measurement	ICC (95% CI)	p-value	ICC (95% CI)	p-value
CT vs MRI 1.0 mm	0.983 (0.872-0.996)	<0.001	0.949 (0.600-0.988)	<0.001
CT vs MRI 0.8 mm	0.991 (0.958-0.998)	<0.001	0.975 (0.866-0.993)	<0.001
CT vs MRI 0.6 mm	0.988 (0.955-0.997)	<0.001	0.965 (0.869-0.990)	<0.001
Interobserver Assessment
Measurement	ICC (95% CI)	p-value
CT	0.991 (0.971-0.997)	<0.001
MRI 1.0 mm	0.983 (0.873-0.996)	<0.001
MRI 0.8 mm	0.988 (0.925-0.997)	<0.001
MRI 0.6 mm	0.987 (0.942-0.996)	<0.001

CI, confidence interval; CT, computed tomography; ICC, intraclass correlation coefficient; MRI, magnetic resonance imaging

Table 2.

Intermodal Bland-Altman Analysis (Ex Vivo Study, n=12)

	Reader 1		Reader 2
Measurement	Bias (mm)	1.96 x SDdiff (mm)	Bias (mm)	1.96 x SDdiff (mm)
CT vs MRI 1.0 mm	0.6	1.3	1.0	2.0
CT vs MRI 0.8 mm	0.3	1.0	0.6	1.6
CT vs MRI 0.6 mm	0.3	1.3	0.6	2.1

CT, computed tomography; ICC, intraclass correlation coefficient; MRI, magnetic resonance imaging

Figure 3:

Ex-vivo study: Bland-Altman plots of the differences between 3D CT and 3D ZTE MRI for glenoid width measurements. (A), (C), and (E) represent plots at 1, 0.8, and 0.6 mm ZTE resolution for Reader 1, respectively, and (B), (D), and (F) represents those for Reader 2. The solid lines represent the mean differences (bias), and the dashed lines indicate the 95% limits of agreement. It is expected that the limits include 95% of differences between the two measurement methods. Each glenoid at a particular condition is represented by an individual point. Black dots: pre-lesion glenoids. Hollow circles: post-lesion glenoids. ZTE, zero echo time.

In Vivo Study

Ten patients were included, and none were excluded. Mean age of the study group was 35±7.3 years (age range, 23-47 years; 9 males and 1 female; 4 left, 6 right shoulders). Glenoid bone loss was present in 7 patients. All 3D ZTE MR and 3D CT images that were obtained in patients yielded high quality reconstructions (Figure 4). For Reader 1, post-processing times required to generate proper 3D images of the glenoid for CT and ZTE MRI at 1.0, 0.8, and 0.7 mm were 32.1 ± 7.9, 36.0 ± 7.2, 32.6 ± 4.1, and 42.1 ± 16.4 seconds, respectively. For Reader 1, reconstruction times with ZTE MRI were not significantly different compared with CT (1.0 mm, p = 0.136; 0.8 mm, p = 0.824; 0.7 mm, p = 0.311). For Reader 2, post-processing times required to generate proper 3D images of the glenoid for CT and ZTE MRI at 1.0, 0.8, and 0.7 mm were 45.5 ± 5.1, 63.0 ± 11.1, 60.0 ± 11.2, and 64.4 ± 13.5 seconds, respectively. For Reader 2, reconstruction times with ZTE MRI were significantly longer compared with CT (1.0 mm, p = 0.006; 0.8 mm, p = 0.005; 0.7 mm, p = 0.008).

Figure 4:

Glenoid width measurement method shown in a left scapula of a 29-year-old man. (A) 3D CT, (B) 3D MRI 1.0 mm, (C) 3D MRI 0.8 mm and (D) 3D MRI 0.7 mm images with en face view of the glenoid shows the width measurement, defined as the largest anteroposterior measurement in the lower two-thirds of the glenoid, measured perpendicular to the long axis. ZTE, zero echo time.

ICCs for intermodality and interobserver assessment showed almost perfect agreement on the in vivo study for both readers and all three MRI resolutions (ICCs ≥ 0.93; 95% CI, 0.70-1.0; all P values < 0.001) (Table 3). Bland-Altman analysis revealed a bias of 0.8 mm or less for all measurements of both readers (Table 4). For CT vs MRI at 1.0 mm³ resolution, there was a bias of 0. 8 (LoA: −1.6, 3.2) for Reader 1 and bias of 0.6 (LoA: −1.1, 2.3) for Reader 2. For CT vs MRI at 0.8 mm³ resolution, there was a bias of 0.6 (LoA: −1.5, 2.7) for Reader 1 and bias of 0.5 (LoA: −0.9, 1.9) for Reader 2. For CT vs MRI at 0.7 mm³ resolution, there was a bias of 0.8 (LoA: −0.8, 2.4) for Reader 1 and bias of 0.4 (LoA: −1, 1.8) for Reader 2. Figure 5 shows the Bland-Altman plots demonstrating high inter-modality agreement for each reader, when comparing 3D CT and 3D ZTE MRI at the three different resolutions.

Table 3.

Intraclass Correlation Coefficients (In Vivo Study, n=10)

Intermodal Assessment
	Reader 1		Reader 2
Measurement	ICC (95% CI)	p-value	ICC (95% CI)	p-value
CT vs MRI 1.0 mm	0.955 (0.774-0.990)	<0.001	0.956 (0.782-0.990)	<0.001
CT vs MRI 0.8 mm	0.976 (0.899-0.994)	<0.001	0.985 (0.914-0.996)	<0.001
CT vs MRI 0.7 mm	0.983 (0.841-0.997)	<0.001	0.987 (0.933-0.997)	<0.001
Interobserver Assessment
Measurement	ICC (95% CI)	p-value
CT	0.926 (0.700-0.982)	<0.001
MRI 1.0 mm	0.951 (0.780-0.989)	<0.001
MRI 0.8 mm	0.953 (0.808-0.988)	<0.001
MRI 0.7 mm	0.958 (0.814-0.991)	<0.001

CI, confidence interval; CT, computed tomography; ICC, intraclass correlation coefficient; MRI, magnetic resonance imaging

Table 4.

Intermodal Bland-Altman Analysis (In Vivo Study, n=10)

	Reader 1		Reader 2
Measurement	Bias (mm)	1.96 x SDdiff (mm)	Bias (mm)	1.96 x SDdiff (mm)
CT vs MRI 1.0 mm	0.8	2.4	0.6	1.7
CT vs MRI 0.8 mm	0.6	2.1	0.5	1.4
CT vs MRI 0.7 mm	0.8	1.6	0.4	1.4

CT, computed tomography; ICC, intraclass correlation coefficient; MRI, magnetic resonance imaging

Figure 5:

In vivo study: Bland-Altman plots of the differences between 3D CT and 3D ZTE MRI for glenoid width measurements. (A), (C), and (E) represent plots at 1, 0.8, and 0.7 mm ZTE resolution for Reader 1, respectively, and (B), (D), and (F) represents those for Reader 2. The solid lines represent the mean of all differences (bias), and the dashed lines indicate the 95% limits of agreement. It is expected that the limits include 95% of differences between the two measurement methods. Each study participant is represented by an individual point. ZTE, zero echo time.

DISCUSSION

In our study we found that the 3D MRI and 3D CT techniques offer similar results with high intermodality and interrater agreement (ICCs ≥ 0.94) with bias of 1 mm or less for all measurements for both readers. Furthermore, our results indicate that the new ZTE MRI technique is very promising with high quality 2D images that can be obtained with clinically compatible imaging times, facilitating rapid post-processing times to generate the 3D reconstructions.

Non-invasive imaging of the glenoid bone is important for patient care, and the capability of one imaging modality to assess both soft tissues and bone lesions is highly desirable from cost and convenience perspectives. It is widely accepted that 3D CT is the best imaging modality for the purpose of glenoid bone evaluation ^{6, 21}, although many authors have shown promising results with different 3D MRI techniques.^{11, 12, 22–24} “ A major limitation in clinical implementation of these previously utilized 3D MRI techniques is post-processing. Gyftopoulos et al. used a 3D gradient echo technique (fast low-angle shot [FLASH] with Dixon-based water-fat separation; Siemens, Munich, Germany), however post-processing times averaged 25 minutes.¹¹ Stillwater et al. also used a 3D gradient echo technique (volumetric interpolated breath-hold examination [VIBE] with water excitation; Siemens, Munich, Germany), with post-processing times ranging from 10-15 minutes.¹² With conventional 3D MRI techniques such as FLASH and VIBE, the signal intensity and contrast of bone is insufficiently distinct compared with overlying soft tissues and air surrounding the patient to enable rapid segmentation with typical 3D volume rendering techniques. With the new ZTE MRI technique, however, “CT-like” images are automatically generated on the scanner. The high-contrast bone images are well suited for 3D volume rendering using commonly employed ray casting algorithms ²⁵ without extensive image manipulation. In our study, although CT images required less time to post-process, both CT and MRI required on average less than one minute to generate proper glenoid images for geometric measurements. As many scanner vendors and commercial picture archiving and communication systems (PACS) used in hospitals and clinics include 3D volume rendering software based on ray casting algorithms, 3D post processing of the ZTE images may be readily incorporated into clinical workflow.

We investigated various isotropic resolutions of the ZTE MRI technique ranging from 0.6 to 1 mm for the ex vivo study and 0.7 to 1 mm for the in vivo study and found mean differences of 1 mm or less for all resolutions compared with 3D CT. This mean difference corresponds to approximately 1-pixel and roughly 3% of a typical glenoid width, however our 95% limits of agreement extend to approximately 3 mm or roughly 10% of a typical glenoid width. Although an error of 3% is comparable to the range of errors (2.2-3.5%) described for 3D CT when compared with ruler measurements on digital photographs²⁶ and is likely not clinically significant²⁷, a 10% error may be large enough to change clinical decision making. Although our study demonstrates high intermodality and interrater agreements, the true magnitude of error between ZTE and CT should be investigated in future, larger studies. Overall, our results did not demonstrate an apparent difference between MRI obtained at 0.6-0.7 mm, 0.9 mm, or 1 mm isotropic resolutions. The limits of resolution remain to be elucidated as Yanke et al. found no significant differences between volume rendered MRI based on 2D images with 2.0-3.5 mm slice thicknesses compared with 3D CT.²² However, ZTE with 1.0 mm isotropic resolution currently only requires 3-4 minutes of imaging time and a faster sequence may not be necessary since this is comparable with the total time it takes to obtain a shoulder CT (scout localizer plus axial imaging of the volume).

Although the ZTE MRI technique is not yet available in all clinics, in recent years the sequence has been utilized by numerous centers around the world on scanners from multiple manufacturers.^{13, 28} The technique has been commercialized by at least one vendor²⁹ and is available as a software upgrade without necessary hardware modifications. The fully automated, on-line processing of “CT-like” images from the ZTE dataset is not yet widely available but will likely be incorporated into future versions. Currently, users of the technique can simply invert the images on the PACS to generate similar “CT-like” images or process the images offline (requiring <1-minute computer computation time).

Limitations

This study is not without limitations. First is the relatively small sample of cadaveric specimens and patients included in this study, which is typical of inter-modality assessment studies of this type.^{6, 11, 12, 21–23} Second, MRIs were acquired within 6 months of the patient CT exams. None of the patients had surgery between the CT and MRI, however it is possible that the patients had sustained further injury in this time interval. Third, only linear glenoid widths were measured in our study, although several other measurements could have been used. However, not all of our patients had glenoid bone loss or bipolar lesions and therefore estimation of these abnormalities were not always possible. Fourth, only the glenoid bone was evaluated in this study. Specifically, post-processing times would undoubtedly increase if separate 3D reconstructions of both the glenoid and humeral head were required (for application of the glenoid track concept³⁰, for instance).

Conclusion

3D ZTE MRI can potentially be considered as a new technique to determine glenoid width and can be readily incorporated into the clinical workflow.