Reliability of Tibiofemoral Contact Area and Centroid Location in an Upright, Open MRI (UO-MRI)

Background: Biomechanical studies are often performed using conventional closed-bore MR, which has necessitated simulating weightbearing load on the joint. The clinical applicability of these biomechanical ndings is unclear because of the limitations of simulating weightbearing. Upright, open MRI (UO-MRI) can be used to assess knee joint mechanics, in particular contact area and centroid location. However, it is not clear how reliably measurements of contact area and centroid location can be made in upright weightbearing postures. Methods: Manual segmentation of cartilage regions in contact was performed and centroids of those contact areas were automatically determined for the medial (MC) and lateral (LC) tibiofemoral compartments. To assess reliability, inter-rater, test-retest, and intra-rater reliability were determined by intra-class correlation (ICC 3,1 ), standard error of measurement (SEM), smallest detectable change with 95% condence (SDC 95 ). Accuracy was assessed by using a high-resolution, 7T MRI as a reference and determined by measurement error (%). Results: Contact area and centroid location reliability (inter-rater, test-retest, and intra-rater) for sagittal scans in the MC demonstrated ICC 3,1 values from 0.95-0.99 and 0.98-0.99 respectively, and in the LC from 0.83-0.91 and 0.95-1.00 respectively. The smallest detectable change in contact area was 1.28% in the MC and 0.95% in the LC. Contact area and centroid location reliability for coronal scans in the MC demonstrated ICC 3,1 values from 0.90-0.98 and 0.98-1.00 respectively, and in the LC from 0.76-0.94 and 0.93-1.00 respectively. The smallest detectable change in contact area was 0.65% in the MC and 1.41% in the LC. Contact area segmentation was accurate to within 4.81% measurement error. Conclusions: Knee contact area and contact centroid location can be assessed in upright weightbearing MRI with good to excellent reliability and accuracy within 5%. The lower eld strength used in upright,

The strength of MR imaging is the ability to directly assess soft tissue structures such as cartilage. With the body in functional positions there is an opportunity to study the biomechanical behaviour of these structures. In biomechanical studies using conventional closed-bore MR, imaging cannot be performed during natural weightbearing. To address this, approaches include imaging cartilage in supine before and after a knee loading activity is performed 9,10 , positioning the participant supine in the scanner with an axial load applied to the foot (closed kinetic chain) 3,6,8 , and applying a torque to the shank while the participant lies supine 7 .
The reliability and accuracy for contact area and centroid location from studies with simulated loading have been estimated. The coe cient of variation (CV) for tibiofemoral contact area and centroid location, which indicates the extent of variability between multiple testing sessions, has ranged between 3.1-9.0% and 0.3-3.3%, respectively 3,7,8,11 . Determining contact area by combining MRI with biplanar radiography has shown a slightly larger standard error of measurement of 14±11% in a cadaveric validation study 12 .
There is emerging interest in open MR machines that allow scanning to take place with participants in functional positions like standing. Similar to the utility of standing X-ray in the clinical investigation and operative planning of knee osteoarthritis, standing MRI may have the potential to better characterize the biomechanical effect of tibiofemoral pathology involving soft tissues like ligaments, menisci, and cartilage. Injury to such structures is a risk factor in the development of osteoarthritis, and UO-MRI may aid in the research of these injuries before the eventual development of deforming bony changes.
Upright, open MRI (UO-MRI) addresses the limitations of simulated weightbearing in supine scanners by allowing joint imaging during weightbearing 7,13,14 . However, UO-MRI scanners have lower eld strengths than standard closed-bore scanners, which limits the image quality that can be obtained 15 . Thus, there is a need to establish the reliability of measurements in upright weightbearing postures obtained via the UO-MRI.
The aims of this study were: 1) to assess the reliability and accuracy of tibiofemoral cartilage contact area and centroid location acquired both sagitally and coronally 2) to describe the implementation of an UO-MRI protocol that permits acquisition of these measures in vivo under physiologic weight-bearing conditions.

Methods
This study was approved by the UBC Clinical Research Ethics Board (H18-01459). All participants provided informed, written consent (Appendix B).

Participants
A sample of 5 patients from a larger comparative cohort study volunteered for reliability analysis. The cohort study was a convenience sample of 18 patients with prior ACL rupture. Patients were recruited through posted noti cations and targeted e-mails (Appendix A). The ve patients selected for this study were the only patients who consented to the complete the scanning process, a three-hour procedure, on two separate dates, thus allowing for test-retest reliability analysis.
Inclusion criteria for the cohort study were: 1) adult participants between the ages of 18-50 years old with unilateral, isolated ACL ruptures; 2) intact cartilage and evidence of complete ACL rupture on MRI; 3) reported ACL rupture within the last 5 years and if reconstructed, done within 1 year from injury; and 4) have completed a full rehabilitation program and returned to regular sport or recreational activities.
Exclusion criteria were: 1) associated ligament rupture other than the ACL (though incomplete MCL ruptures were not excluded); 2) known knee osteoarthritis diagnosed by a physician; 3) presence of other joint disease; 4) incompletely rehabilitated injury, de ned as a range of motion less than 0-130 degrees, quadriceps atrophy, or persistent mechanical symptoms; 5) individuals prohibited from undergoing MRI based on the MRI screening form (Appendix C); 6) history of fainting, or evidence of change in orthostatic blood pressure; 7) prior or subsequent knee surgery other than diagnostic arthroscopy; 8) history of corticosteroid injection to either knee; and 9) bilateral ACL rupture or ACL re-rupture.
Demographic data from participants were collected including age, height, body mass, date of injury, time from injury to surgery (if applicable), and time from injury to study participation.

Imaging
Participants were scanned standing in a 0.5T upright, open MRI (MROpen, Paramed, Genoa, Italy). All scans were done in the morning, participants were instructed not to do any impact exercise prior to scanning, and participants were seated for 30 minutes prior to scanning, during which time questionnaires were administered. Participants wore compression socks to minimize venous pooling in the lower extremities during standing scans. Participants then stood for 15 minutes prior to acquiring standing scans to ensure a cartilage deformation equilibrium had been reached. Each participant wore a chest harness suspended from an aluminum ceiling track safety-rated to 450 lbs (Handicare, Concord, ON) as a precautionary measure in case the participant fainted during upright scanning. No weight was borne through the bars or the harness. Standing scans of the ACL-injured leg were acquired with the knees in full extension, with the participant instructed to stand comfortably and distribute their weight equally between legs. Three support bars (shins, buttocks, and hands) were placed to help the participant remain still during scanning. We obtained sagittal and coronal images with a double echo steady state T2 sequence (Table 1) using a commercial 2-channel knee coil (ParaMed) suspended around the knee.
The sequence was optimized to provide excellent cartilage signal quickly enough to minimize the effects of patient movement and fatigue while standing. The data was denoised by an optimized blockwise nonlocal means denoising lter 16 , and the component DESS images were subsequently t to a signal model with a global T1 estimate of 0.5 17 . Two trained raters, A.M.S and D.J.S, with two years and three years' experience respectively, performed segmentation for all data sets. Both raters were trained in knee joint segmentation by a post-doctoral fellow with 10 years of experience in segmenting MSK data. Prior to the study both raters established a set of general guidelines for segmentation. All data sets were anonymized, and a numerical code was assigned to each patient. Raters identi ed tibiofemoral contact regions by manually tracing regions with no visible separation between cartilage surfaces on each image slice using the Editor module in 3D Slicer 18 (http://www.slicer.org) in both the coronal and sagittal planes ( Figure 1.A). Raters selected voxels of cartilage that were in direct contact and did not contain any contribution from other structures (e.g. meniscus or synovial uid). Volumes were created that represented medial and lateral contact areas, each with a known number of voxels ( Figure 1.B). We multiplied the number of voxels in contact by their axial dimensions (length and width) to calculate contact areas for the medial and lateral compartments.
To account for differences in size between subjects, the cartilage contact area measurement in the axial plane measurement was normalized by taking the ratio (%) of the contact area over the maximum axial cross-sectional area of the tibial plateau.
The centroid location was calculated as the geometric center generated from the contact area segmentations in the medial and lateral compartments (Figure 1.B). A validated joint coordinate system was employed to locate contact area centroids within a consistent coordinate frame 19 . Centroid location was reported in mm and also quanti ed as a percentage on the tibial plateau in the medial (0%) to lateral (100%) and posterior (0%) to anterior (100%) directions to account for differences in size between participants. The coordinate system was established based on speci c bony landmarks, allowing for description of the position and orientation of the tibia relative to the femur in three dimensions 19 . Reference bony landmarks were established from supine scout scans of the hip, knee, and ankle, with the scan position relative to each other noted from the difference in UO-MRI scan table position. Positions and orientations of the coordinate systems in the upright posture were determined by registering supine images of the tibia and femur to corresponding upright images using Analyze 12.0 (AnalyzeDirect, Inc., Overland Park, KS).

Accuracy
We assessed the accuracy of contact area measurement by comparing our method in the UO-MRI to reference measurements of contact area made in a 7T MR scanner (Bruker Biospin, Ettlingen, Germany) for two cartilage preparations at two load levels. We created two cartilage contact preparations by dissecting a bovine knee and extracting medial and lateral tibial and femoral blocks using a handsaw. The block dimensions were approximately 30 mm by 30 mm in the anteroposterior direction and mediolateral direction and were approximately 20 mm in the axial (compressive) direction. The bony side of each osteochondral block was a xed to polycarbonate tissue mounts with cyanoacrylate glue. Care was taken to extract osteochondral blocks in an orientation that approximately matched and were oriented on tissue mounts in a manner that maximized contact of the attest part of the mating joint surfaces. The preparations were immersed in phosphate-buffered saline and positioned in an MRcompatible compression chamber such that axial compression could be applied by rotating a Delrin plunger (2 mm thread) within the capsule of the compression chamber. The samples were positioned such that opposing cartilage surfaces were touching but not compressed, and images were acquired. An axial load was then applied until cartilage compression could be visualized, and the specimen was rescanned. Five minutes were allowed to pass in between cartilage compression and re-scanning in order to permit the cartilage to equilibrate. The displacement of the plunger was marked on the outside of the chamber so that the process could be repeated. On completion, the load was removed, and the cartilage given time to re-equilibrate. The process was performed rst on the UO-MRI and then at the 7T MRI with imaging parameters listed in Table 1. In a previous study, intra-observer repeatability of segmentation of loaded tibial and femoral cartilage images in this 7T scanner was within 2.3 and 3.3 voxels for cartilage depth, 95% of the time 20 .

Statistics
Inter-rater, test-retest, and intra-rater reliability statistics were calculated for tibiofemoral contact area and centroid location. Inter-rater reliability was obtained for two raters who individually segmented and calculated contact areas for each scan. Test-retest reliability was established by scanning each participant twice, with approximately one month between scans, with one rater (D. J. S.) segmenting both scans. Intra-rater reliability was obtained for one rater (A. M. S.) segmenting the contact areas for each sample 3 times, each 2 weeks apart. We calculated the intra-class correlation coe cient for xed raters (ICC 3,1 ) using the methods described by Shrout and Fleiss 21 , the standard error of measurement (SEM), and the smallest detectable change with 95% con dence (SDC 95 ). ICCs less than 0.5 indicated poor reliability; 0.5 to 0.75 moderate reliability; 0.75 to 0.9 good reliability; and greater than 0.9 excellent reliability. All metrics were obtained for both coronal and sagittal scans.
We assessed contact area accuracy by nding mean absolute error (MAE) for contact areas measured using low-resolution 0.5T UO-MRI and those measured for the same region and load using high-resolution 7T MRI from images obtained in the sagittal plane.

Results
Descriptive characteristics for the 5 participants included in the reliability analysis are reported in Table 2. There were 4 female participants and 1 male; 3 had undergone ACL reconstruction and 2 had not. Mean absolute contact areas were 452 mm 2 (±103) and 314 mm 2 (±41) for medial and lateral compartments, respectively. Mean normalized contact areas were 13.7% (±2.6) and 9.7% (±1.6) for medial and lateral compartments, respectively.
For scans acquired in the sagittal plane, contact area ICC 3,1 values (including inter-rater, test-retest, and intra-rater reliability) ranged from 0.94 to 0.99 in the medial compartment, and 0.83 to 0.91 in the lateral compartment (Table 3). From the test-retest data, contact area SDC 95 was 1.28% in the medial compartment and 0.95% in the lateral compartment. Qualitatively, contact regions were very similar between raters (Figure 2), and centroid location demonstrated high reliability (Table 4). SDC 95 for medial centroid locations in the X and Y direction were 3.39% and 4.94% (1.89 mm and 2.29 mm), respectively. SDC 95 for lateral centroid locations in the X and Y direction were 4.41% and 3.85% (3.31 mm and 1.42 mm), respectively.  In the accuracy analysis, data from one sample (medial compartment unloaded) was discarded due to a technical error. During scanning of this sample at the 7T MRI, the eld of view did not include the full bovine specimen, and thus did not include the full contact area of the sample. Unfortunately, this error was discovered during image post-processing after scanning had concluded. The remaining areas obtained in the 0.5T UO-MRI for the lateral compartment unloaded, medial compartment loaded, and lateral compartment loaded were: 120 mm 2 , 271 mm 2 , and 254 mm 2 respectively; areas measured using the 7T MRI were 126 mm 2 , 258 mm 2 , and 240 mm 2 respectively. This produced a mean contact area measurement error of 4.8% and an MEA of 11 mm 2 .

Discussion
We assessed in vivo inter-rater, test-retest, and intra-rater reliability of tibiofemoral contact area and centroid location measurements for UO-MRI scans in both sagittal and coronal planes. We evaluated the accuracy of our contact area measurements by comparing measurements made using the UO-MRI to measurements made in a high resolution 7T MRI for a bovine knee model. All measures of contact area reliability, including inter-rater, test-retest, and intra-rater, ranged from good to excellent for coronal and sagittal scans. Qualitatively, there was close correspondence between contact regions identi ed by different readers (Figure 2). The accuracy analysis found an overall mean absolute error of 11 mm 2 between areas found from 7T MRI and from the UO-MRI. Our results suggest that sagittal or coronal scans are similarly well-suited to evaluate cartilage contact and centroid location in the tibiofemoral joint, with slightly higher repeatability values resulting from sagittal plane acquisition and evaluation.
Our assessment of SDC 95 , the smallest amount of change that provides 95% con dence that a true change has occurred and is not due to inherent measurement error, may provide useful information for planning research studies that compliments the more widely-used ICC values. For example, our nding of SDC 95 of 2-2.5 mm for changes in contact location (using sagittal plane images) is smaller than the 4.2 mm difference reported between knees with ACL rupture and healthy knees estimated using a biplanar radiography/MRI image registration approach 22 , which suggests that our UO-MRI approach can effectively detect differences in centroid location due to ACL de ciency. Similarly, previous investigations have estimated that tibiofemoral contact changes by as much as 94.8 mm 2 medially and 56.3 mm 2 laterally 23 . Our largest estimated SEM was 18.9 mm 2 , indicating that UO-MRI may also be effective at detecting such differences.
Our measures of contact area and centroid location reliability in weightbearing MR are comparable to those from 3T conventional closed-bore scans despite using a lower resolution scanner. For inter-rater reliability, our nding for contact area ICC in the medial compartment of 0.95 is consistent with ndings in 3T MRI (0.90) 24 . Our nding for contact area ICC in the lateral compartment for inter-rater reliability (0.83) was also in the excellent range for an ICC value, although it had a much wider con dence interval range (0.06 to 0.98) and was lower than ndings in 3T MRI (0.92) 24 . The inter-rater contact location ICCs (0.99 medially and 0.95 laterally) were also similar to those found in 3T (0.99 medially and 0.91 laterally) 24 . For intra-rater reliability our ndings for contact area ICC were 0.99 medially and 0.91 laterally, which was again consistent with 3T MRI ndings of 0.97 both medially and laterally 24 . Our intrarater contact location ICCs (0.99 medially and 0.98 laterally) were similar to those found in 3T (1.00 medially and 0.91 laterally) 24 . No previous study has evaluated the test-retest reliability of contact area and centroid location in vivo, although one cadaveric study examined the patellofemoral joint using a 1.5T magnet and found a test-retest ICC value of 0.98, which is comparable to our results 25 . The slightly higher variation in test-retest reliability in the current study is likely due to slight differences in participant posture and positioning between test dates, which may be easier to control in a cadaveric study. The testretest reliability measures will be of value in experimental design, especially for studies requiring testing on more than one day. Our accuracy results, which found a mean contact area measurement error of 4.8%, suggest higher accuracy for our method than the results from a cadaver study using a silicone casting technique reference standard, which found a standard error of measurement of 14% 12 . This may be because the reference method of the current study (high eld MRI) is different from the reference method for the previous study (silicone casting). The absolute values of our contact areas were slightly higher than previously reported values 6,23 , though the ratio of lateral to medial contact area were similar. This may have been partially due to cartilage creep, as our participants stood in a weightbearing position for 15 minutes prior to standing. Additionally, differences in tibiofemoral contact area in the same knees have been observed depending on the MRI sequence used 8 . The T2 DESS sequence that we used greatly enhances cartilage, which may have accounted for our high contact area values.
The primary strength of this study is that it provides a comprehensive assessment of the role of the intraand inter-individual differences in raters, and repeated scans, on the reliability of tibiofemoral contact measures. The good to excellent reliability results are supported by an accuracy assessment. Incorporation of both sagittal and coronal plane assessment and reporting of SDC 95 may be useful in protocol development for future studies. Given the advantages for ecological validity with the UO-MRI approach for these assessments compared to traditional supine MRI, we feel that our ndings have important implications for the study of knee joint mechanics and function in future UO-MRI studies.
The ndings should be considered in light of some limitations. First, reliability was assessed in ACLruptured knees only. The cartilage of these participants may not be representative of cartilage in uninjured knee joints. The effect of this limitation is unknown however, since we performed the scans relatively soon after injury it is unlikely that enough time had passed for cartilage degeneration to signi cantly affect our segmentation process. Second, the accuracy assessment was an imperfect reference standard. The number of samples used in the accuracy assessment was low, and because it was an ex vivo sample, there was no possibility of noise via subject movement or blood ow. The sample was also immersed in phosphate buffered saline, which can increase SNR, and may have in uenced our results. Additionally, the bovine osteochondral blocks used may not have adequately represented human tibiofemoral contact behaviour. We chose 7T MRI as it was the highest resolution possible with which we could ensure similar loads by using the same loading rig. The lengthy scan time and cost of the 7T scanner hindered our ability to process more samples for accuracy assessment; similarly, we were not able to establish the reliability of measuring contact area in the 7T MRI before we used it as the reference standard. Third, the low resolution of the UO-MRI due to the low magnetic eld strength may have reduced variability. This should be taken into account when planning future experiments. Finally, some of our intra-class correlation results were bounded by wide con dence intervals. More raters and tests would improve the certainty of our estimates.

Conclusions
In conclusion, knee contact area and contact centroid location can be assessed in upright weightbearing MRI with good to excellent reliability. The lower eld strength used in upright, weightbearing MRI does not compromise the reliability of tibiofemoral contact area and centroid location measures. Our results will be useful for planning future experiments in the UO-MRI.
DRW was the senior author and was involved with study planning and manuscript drafting. All authors read and approved the nal manuscript.