Feasibility of Using an Inversion-Recovery Ultrashort Echo Time (UTE) Sequence For Quantification of Glenoid Bone Loss

Abstract

Objective

To utilize the 3D Inversion Recovery prepared Ultrashort-Echo-Time with Cones Readout (IR-UTE-Cones) MRI technique for direct imaging of lamellar bone with comparison to the gold standard of computed tomography (CT).

Materials and Methods

CT and MRI was performed on 11 shoulder specimens and 3 patients. Five specimens had imaging performed before and after glenoid fracture (osteotomy). 2D and 3D volume-rendered CT images were reconstructed and conventional T1-weighted and 3D IR-UTE-Cones MRI techniques were performed. Glenoid widths and defects were independently measured by two readers using the circle method. Measurements were compared with those made from 3D CT datasets. Paired-sample Student’s t tests and intraclass correlation coefficients were performed. In addition, 2D CT and 3D IR-UTE-Cones MRI datasets were linearly registered, digitally overlaid, and compared in consensus by these two readers.

Results

Compared with the reference standard (3D CT), glenoid bone diameter measurements made on 2D CT and 3D IR-UTE-Cones were not significantly different for either reader whereas T1-weighted images underestimated the diameter (mean difference of 0.18 cm, p=0.003 and 0.16 cm, p=0.022 for readers 1 and 2, respectively). However, mean margin of error for measuring glenoid bone loss was small for all modalities (range 1.46-3.92%). All measured ICCs were near perfect. Digitally registered 2D CT and 3D IR-UTE-Cones MRI datasets yielded essentially perfect congruity between the two modalities.

Conclusion

The 3D IR-UTE-Cones MRI technique selectively visualizes lamellar bone, produces similar contrast to 2D CT imaging, and compares favorably to measurements made using 2D and 3D CT.

INTRODUCTION

Glenoid bone loss in the setting of recurrent anterior shoulder instability is a difficult problem to treat. It has been demonstrated by several authors that percent bone loss of the glenoid is a major determining factor in the success of arthroscopic soft tissue procedures versus the need for glenoid bone grafting [1]. In a study by Burkhart and De Beer, it was shown that the failure rate of arthroscopic stabilization in patients with a significant glenoid bone defects was 67% and 89% in contact athletes [2]. Clinical assessment of the exact percentage of glenoid bone loss continues to be a key element in preoperative planning as it often drives surgical decision making, with larger defects requiring glenoid bone grafting. Thus, accurate assessment of the degree of bone loss is imperative.

Currently, three-dimensional (3D) computed tomography (CT) of the injured shoulder is largely considered to be the most reliable preoperative imaging method of assessing the percent bone loss [3]. In a cadaveric study, 3D CT was shown to be the most accurate with the highest inter- and intra-observer agreement for predicting percentage of glenoid bone loss when compared to two-dimensional (2D) CT, magnetic resonance imaging (MRI), and radiographs [4]. However, CT of the shoulder requires the patient to undergo an additional study leading to increased patient and healthcare time and costs as well as radiation exposure in usually young patients. Most patients with a history of shoulder instability are initially evaluated by radiography followed by MRI for soft tissue evaluation (including labrum and capsule), with CT ordered at a later time if there is suspicion of glenoid bone loss. It would be clinically beneficial to have an accurate and reliable MRI method to evaluate bone loss and eliminate the need for CT scans in younger patients.

Conventional MRI techniques are fundamentally limited in their assessment of bone due to the extremely short T2/T2* of lamellar bone [5]. Using conventional MRI sequences with longer echo times (TEs), the signal of short T2/T2* structures has decayed to background levels prior to acquisition. Rather, the contours of lamellar bone are inferred by visualizing fatty bone marrow or surrounding soft tissue, potentially resulting in imprecise localization. Techniques involving ultrashort echo time (UTE) can directly image lamellar bone, with newer methods incorporating fat and adjacent soft tissue suppression [5, 6]. These techniques produce images of cortical bone that are more similar in quality to that currently available by CT scan [7]. This yields itself well to the evaluation of patients with shoulder instability where MRI is performed early in their treatment course. Adoption of MRI as a primary method to assess glenoid bone defects could lead to less need for CT scans as well as lower healthcare costs and radiation exposure to patients. The purpose of this study is to utilize the 3D Inversion Recovery prepared Ultrashort-Echo-Time with Cones Readout (IR-UTE-Cones) MRI technique for direct imaging of lamellar bone with comparison to the gold standard of CT.

MATERIALS AND METHODS

The institutional review board approved this Health Insurance Portability and Accountability Act–compliant study, which included an anonymized cadaveric portion and a human, in vivo portion. Written informed consent was obtained prior to patient participation in this study.

Specimens and Patients

11 shoulder specimens (6 females, 5 males; age range, 61-98 years; mean age, 86 years) were used for the cadaveric portion of this study. The specimens were frozen at −80°C and were thawed at room temperature for 24 hours prior to imaging. All shoulder specimens underwent the CT and MRI protocols while intact. Based on the CT, five shoulder specimens with the least amount of glenohumeral osteoarthritis were chosen, the scapulae were dissected free of surrounding soft tissues, and glenoid lesions were created with an osteotome. CT and MRI protocols were then repeated. In addition, 3 patients (1 female, 2 males; age range, 20-64 years; mean age, 38 years) with glenohumeral joint complaints were imaged, including one with severe glenohumeral arthritis and two with glenohumeral instability.

CT and MRI

All shoulders included in this study underwent both 64-slice CT (VCT, GE Healthcare, Milwaukee, WI) and 3T MRI (Signa HDx 3T or MR750 3T, GE Healthcare, Milwaukee, WI). The CT protocol consisted of 0.625 mm axial images using the following parameters: 120 kV, 597 mA, and pitch of 0.98. The axial images were used to reconstruct 2D sagittal oblique images of the glenoid with 2 mm thickness. The sagittal oblique plane was plotted based on both axial and coronal localizer planes in order to create an optimized en face view. In addition, 3D volume-rendered reconstructions were generated. The MRI protocol included conventional sequences, including axial and sagittal oblique T1-weighted (repetition time/echo time [TR/TE], 741/8.2 ms) images using the following parameters: field of view (FOV) of 15 cm, matrix of 384 × 320, slice thickness of 3 mm, 0.3 mm interslice gap, with acquisition time of approximately 2 minutes per sequence. For MRI, the sagittal oblique plane was also oblique based on axial and coronal localizing planes.

In addition to the conventional T1-weighted images, the 3D IR-UTE-Cones sequence was employed, which utilizes an adiabatic inversion recovery pulse followed by a prototype 3D UTE acquisition (Figure 1A). The 3D IR-UTE-Cones sequence employs a short rectangular pulse for signal excitation, which in combination with a time-efficient centric trajectory, allows for signal acquisition from rapidly decaying tissue components [8]. Bone images are obtained by inverting the longitudinal magnetization of the long T2 signal components (i.e., muscle and bone marrow fat) while saturating the signal from cortical bone [7, 9–12]. The Cones acquisition starts after an inversion time (TI) delay, which is used to null the long T2* components while permitting detection of recovered cortical bone signal (Figure 1B). By using the 3D UTE-Cones sequence, 3D volumetric UTE imaging can be obtained in an efficient way to maximize signal-to-noise ratio [7, 9] (Figure 1C). The following parameters were used: FOV of 16 cm, matrix of 192 × 192, slice thickness of 3-4 mm, and TE of 0.03 ms. TI, TR, and flip angle (FA) for cadaveric imaging was 55 ms, 134 ms, and 18° whereas it was 45 ms, 106 ms, and 16° for in vivo imaging. Selection of TI, TR, and FA varied between cadaveric and in vivo imaging largely due to temperature, which affects T1 relaxation [13]. The 3D IR-UTE-Cones technique was performed in the same imaging planes as the conventional sequences, including axial and sagittal oblique, with imaging time of approximately 3 to 4.5 minutes per sequence, depending on number of slices required to cover the anatomy.

Figure 1.

3D IR-UTE-Cones sequence. (A) Pulse sequence diagram showing an adiabatic inversion recovery preparation sequence followed by a 3D Cones acquisition with ultrashort echo time. (B) Magnetization scheme of three proton pools including bone, during the course of the inversion recovery preparation. (C) Coronal IR-UTE-Cones image shows selective visualization of lamellar bone, similar to computed tomography. IR, inversion recovery; RF, radiofrequency; G, gradient; DAW, data acquisition window; TE, echo time; FID, free induction decay; M, magnetization; TI, inversion time.

Image Evaluation

2D CT, 2D T1-weighted MRI, and 3D IR-UTE-Cones MRI images were assessed using a clinical picture archiving and communication system (PACS) (Impax version 6.6.1, Agfa Healthcare, Ridgefield Park, NJ). 3D volume-rendered CT images were measured using OsiriX (version 5.8, Pixmeo, Bernex, Switzerland). Cases were randomized and two readers independently made measurements, including a fellowship-trained musculoskeletal radiologist (E.Y.C. with 6 years of experience) and a fellowship-trained orthopaedic surgeon (J.W. with 1 year of experience). Using images of the glenoid en face, a best-fit circle was placed along the inferior portion of the glenoid and the diameter was measured using electronic calipers. For the bone loss cases, a vertical line bisecting the glenoid along its long axis was drawn from the supraglenoid tubercle and the center of the best fit circle overlaid this vertical line. The width of the glenoid defects was measured and percentage of bone loss was calculated as the width of the defect divided by the diameter of the best-fit circle. For the reference standard measurement of glenoid bone loss, both the 3D CT of the intact and fractured glenoid was utilized for maximal confidence (Figure 2). Specifically, the best-fit circle drawn on the intact glenoid was digitally copied to the fractured glenoid and the glenoid defect was then measured.

Figure 2.

Reference standard method used to calculate glenoid bone loss. (A) A best-fit circle was placed along the inferior portion of the glenoid on the 3D CT image of the intact glenoid and the diameter was measured. (B) The circle and diameter measurements were digitally copied to the 3D CT image of the fractured glenoid, the width of bone loss was measured (red line), and the measured defect was divided by the diameter of the best-fit circle.

In addition, for all cases, registration of the 2D CT and 3D IR-UTE-Cones MRI datasets was performed after binary thresholding and linear registration via the FSL FLIRT program (FMRIB’s Linear Registration Tool, fsl.fmrib.ox.ac.uk/fsl/fslwiki/FLIRT). Registered images were digitally overlaid and qualitatively evaluated in consensus by the two readers to assess congruity.

Statistical Analysis

Descriptive statistics were performed. For the shoulders without glenoid bone loss, paired-sample Student’s t tests and intraclass correlation coefficients (ICC) were used to compare the measurements made on 2D CT, 2D T1-weighted MRI, and 3D IR-UTE-Cones MRI images to the reference standard images (3D CT) for each reader. For the cases with glenoid defects, the percentage of bone loss calculated for each modality was compared with the reference standard of 3D CT (using both pre- and post-lesion datasets). The absolute value of the difference between the imaging measure and reference standard assessment was calculated and mean and standard deviation were performed. Based on the difference between the imaging measure and reference standard assessment, 95% confidence intervals were also performed. To assess reliability between readers, paired-sample Student’s t test and ICCs were also used to compare all measurements made for each modality and sequence. An ICC of 0.01 was considered poor agreement, 0.01 to 0.2 was considered slight agreement, 0.21 to 0.4 was considered fair agreement, 0.41 to 0.6 was considered moderate agreement, 0.61 to 0.8 was considered substantial agreement, and 0.8 to 1.0 was considered almost perfect agreement [14]. P-values less than 0.05 were considered significant. All statistical analyses were performed using the SPSS software package (version 21; SPSS, Chicago, IL, USA).

RESULTS

Figure 3 shows representative images of T1-weighted, IR-UTE-Cones, 2D CT, and 3D CT in a 30-year-old patient with glenohumeral instability. Confidence in lamellar bone margin was lacking on the T1-weighted images, whereas the margins were well-identified on IR-UTE-Cones and CT imaging.

Figure 3.

30-year-old patient with glenohumeral instability. (A) T1-weighted image shows poor definition of glenoid cortical bone contour and measurement of 2.56 cm was obtained. (B) IR-UTE-Cones image shows excellent depiction of bone contours and measurement of 2.79 cm was obtained, which is nearly perfect when compared with 3D CT measurement. (C) 2D CT image also shows excellent depiction of contours with 2.74 cm glenoid diameter. (D) 3D CT image was used as the reference standard.

Table 1 shows the mean (± standard deviation [SD]) of the intact glenoids as measured on each modality for each reader with comparison statistics. Compared with the reference standard (3D CT), measurements made on 2D CT and IR-UTE-Cones were not significantly different for either reader whereas T1-weighted images underestimated the diameter (mean difference of 0.18 cm, p=0.003 and 0.16 cm, p=0.022 for readers 1 and 2, respectively). Despite this, ICCs for 2D CT, T1-weighted images, and IR-UTE-Cones images were near perfect when compared with 3D CT.

Table 1.

Comparison of 2D CT, T1-weighted MRI, and IR-UTE-Cones MRI with 3D CT as a reference standard.

Reader, Modality	Mean ± SD (cm), p-value	ICC
Reader 1
3D CT	2.93 ± 0.27
2D CT	2.97 ± 0.28, 0.111	0.963
T1-weighted	2.75 ± 0.32, 0.003	0.875
IR-UTE-Cones	2.88 ± 0.25, 0.080	0.968
Reader 2
3D CT	2.94 ± 0.28	-
2D CT	2.98 ± 0.32, 0.265	0.962
T1-weighted	2.78 ± 0.39, 0.022	0.907
IR-UTE-Cones	2.92 ± 0.30, 0.618	0.950

Note: n=14 (intact specimens and in vivo cases)

Table 2 shows the mean (± SD) of the measured % glenoid bone loss for each modality. Using the reference standard of pre- and post-lesion 3D CT, mean % glenoid bone loss was measured at 18.28 ± 3.79 and 19.04 ± 4.46 by Reader 1 and Reader 2, respectively. If only the post-lesion 3D CT were used for measurement, the maximum change in accuracy expected in terms of percentage was −0.7% to 4.0% (Reader 1) and 2.0% to 4.4% (Reader 2) with 95% confidence. If only the post-lesion 2D CT were used for measurement, the maximum change in accuracy expected would be 0.5% to 6.9% (Reader 1) and 0.7% to 2.2% (Reader 2) with 95% confidence. If only the post-lesion T1-weighted MRI were used for measurement, the maximum change in accuracy expected would be −0.5% to 6.8% (Reader 1) and 0.4% to 6.5% (Reader 2) with 95% confidence. If only the post-lesion IR-UTE-Cones MRI were used for measurement, the maximum change in accuracy expected would be −0.6% to 6.5% (Reader 1) and 1.2% to 6.6% (Reader 2) with 95% confidence.

Table 2.

Measured Glenoid Bone Loss

Reader, Modality	Measurement (%)	Error (%)	Change in Accuracy (95% CI)
Reader 1
Reference Standard	18.28 ± 3.79	−
3D CT	18.60 ± 4.86	1.63 ± 1.87	−0.7% to 4.0%
2D CT	16.25 ± 6.48	3.68 ± 2.60	0.5% to 6.9%
T1-weighted	18.99 ± 5.81	3.13 ± 2.95	−0.5% to 6.8%
IR-UTE-Cones	19.72 ± 5.98	2.96 ± 2.88	−0.6% to 6.5%
Reader 2
Reference Standard	19.04 ± 4.46	−
3D CT	18.43 ± 4.18	3.22 ± 0.95	2.0% to 4.4%
2D CT	18.35 ± 4.60	1.46 ± 0.60	0.7% to 2.2%
T1-weighted	15.77 ± 2.85	3.27 ± 2.60	0.4% to 6.5%
IR-UTE-Cones	15.12 ± 4.66	3.92 ± 2.16	1.2% to 6.6%

Note: Data are reported as mean ± SD. Reference standard measurements were made using both pre- and post-lesion 3D CT datasets.

Table 3 shows inter-observer reliability statistics. Measurements made by Readers 1 and 2 were not significantly different for 3D CT, 2D CT, and IR-UTE-Cones measurements, but T1-weighted MRI measurements were different between readers (p=0.033). ICCs were near perfect for all modalities.

Table 3.

Inter-observer reliability

Modality	p-value	ICC
3D CT	0.151	0.997
2D CT	0.226	0.991
T1-weighted	0.033	0.957
IR-UTE-Cones	0.051	0.983

Registration and overlay of the 2D CT and 3D IR-UTE-Cones MRI datasets yielded essentially perfect congruity between the two modalities (Figure 4).

Figure 4.

Registered images from a shoulder specimen. (A) IR-UTE-Cones image shows high-contrast lamellar bone imaging of the glenoid en face. (B) Registered sagittal CT image after binary thresholding. (C) Overlaid CT image (in green) on the MRI image shows near-perfect correlation of lamellar bone contours

DISCUSSION

In this study, we utilized the 3D IR-UTE-Cones MRI sequence to evaluate the glenoid in both specimens and in vivo. This MRI sequence differs from existing conventional sequences in its ability to directly obtain signal from lamellar bone while simultaneously suppressing fat and surrounding soft tissues. The resulting contrast is similar to 2D CT. The UTE sequence, which utilizes TEs short enough to capture lamellar bone signal, was first described nearly three decades ago [15], but only in recent years has advances in hardware and software technology allowed for high quality imaging in clinically compatible imaging times. UTE sequences have been successfully implemented on every major MRI vendor system [5], but have not yet been offered as a product sequence. We believe that demonstrating this exciting application for assessment of glenoid bone stock will be important in pushing the product into the clinics.

While MRI has previously been used to assess the glenoid bone, conventional T1-weighted and fluid-sensitive sequences do not allow precise delineation of bone margins since no signal can be directly achieved from lamellar bone. Some studies have suggested that quantitative measurements made on conventional MRI sequences are similar to those made on CT [16–19], whereas others have shown a higher performance of CT [4, 20, 21]. More recently, 3D MRI sequences were implemented with a Dixon fat-suppression technique which enabled soft tissue subtraction and increased contrast from lamellar bone [22–24]. However, as no signal is directly obtained from the lamellar bone, this type of image manipulation results in artifactual signal increase of all black pixels (e.g., zero signal air is also displayed as high signal in addition to lamellar bone). Moreover, post-processing times were reported to be approximately 10-25 minutes for experienced 3D lab technologists [22, 23], which is prohibitory for most clinical practices. With the 3D IR-UTE-Cones sequence, no post-processing is required as images can be directly visualized and measured using any standard PACS software.

We have shown that the 3D IR-UTE-Cones sequence allows for direct and selective visualization of lamellar bone which aligns nearly perfectly when registered to 2D CT. When used to evaluate the width of the intact glenoid, the 3D IR-UTE-Cones sequence performed equally well to 3D and 2D CT, whereas measurements made on T1-weighted images significantly underestimated the width. This may be due to lack of confidence in estimating the cortical margin (Figure 3A) and/or over-reliance on the hyperintense marrow fat (Figure 5).

Figure 5.

Axial images of a shoulder specimen. (A) T1-weighted MRI image shows the posterior glenoid contour to appear rounded (arrow) due to over-reliance on marrow contour and poor visualization of lamellar bone. (B) 2D CT image at the same location shows that the posterior glenoid is pointed (arrow) due to an osteophyte. (C) IR-UTE-Cones MRI image shows the same findings as 2D CT.

With regards to measurement of glenoid bone loss, we used the reference standard of 3D CT datasets before and after lesion creation. This was chosen as the gold standard since knowledge of the uninjured glenoid diameter allowed us to be most confident in both the numerator and denominator of the glenoid bone loss equation. We were surprised at the relatively poor performance of all the modalities for measuring glenoid bone loss, including the post-lesion 3D CT images. Upon retrospective review of all of our glenoid bone loss cases, we found the worst performance in glenoids which did not demonstrate a completely circular shape. In these cases, the loss of bone anteriorly coupled with the remaining non-circular margin resulted in incorrect estimation of the native bone contour (Figure 6). Other authors have noted similar limitations when using the best-fit circle method [18, 25]. However, we found that measuring glenoid diameters using IR-UTE-Cones and 2D CT were very accurate when compared to our reference standard of 3D CT and reliable between raters. We emphasize that the IR-UTE-Cones technique and CT corresponded nearly perfectly when digitally registered and therefore measurement errors were not due to modalities, but limitations in the measurement technique. Also of note is that all of our glenoid bone loss cases fell between 12-25%, which Bishop et al. showed to be the range with the worst inter-observer performance (kappa < 0.4 using 3D CT) [4].

Figure 6.

Potential source of error in non-circular glenoids. (A) Glenoid bone loss measurement based on best-fit circle obtained from the pre-lesion 3D CT [same image as Figure 2B]. Width of fragment measured 3.9 mm and 14.7% bone loss was calculated [0.39/2.66]. (B) When only the post-lesion 3D CT was used for measurement, a best-fit circle resulted in a fragment measurement of 4.4 mm and 16.4% bone loss calculation [0.44/2.68].

Our study has several limitations. First, our study utilized a limited sample size. However, we believe that the data demonstrates that this new MRI technique is feasible and is quite comparable to CT. Second, it is important to highlight that although the IR-UTE-Cones is a 3D MRI sequence, it is highly anisotropic. Unlike CT which utilizes sub-millimeter slice thicknesses, the signal-to-noise of lamellar bone remains too low to utilize these resolutions with the 3D IR-UTE-Cones sequence. Therefore, if 3D volume rendered imaging is desired, alternative methods are required. Finally, this new sequence is not yet available in routine clinical practice. However, we performed this study using two different MRI machines (including one over 12 years old) and the UTE sequence has been successfully employed on machines from every major vendor. This sequence only requires a software package to be installed without any hardware modifications or post-processing steps. We feel that demonstrating the utility of this novel technique is the most important step for manufacturers to incorporate this sequence into software package upgrades.

In conclusion, in this pilot study, we have utilized a new MRI technique (3D IR-UTE-Cones) which selectively visualizes lamellar bone and produces similar contrast to 2D CT imaging. This sequence can be added to conventional MRI examinations with minimal increases in imaging times and may eventually eliminate the need for pre-surgical CT examinations.