Accuracy and Precision of Quantitative Assessment of Cartilage Morphology by Magnetic Resonance Imaging at 3.0T

Transcription

1 ARTHRITIS & RHEUMATISM Vol. 52, No. 10, October 2005, pp DOI /art , American College of Rheumatology Accuracy and Precision of Quantitative Assessment of Cartilage Morphology by Magnetic Resonance Imaging at 3.0T Felix Eckstein, 1 H. Cecil Charles, 2 Robert J. Buck, 3 Virginia B. Kraus, 4 Ann E. Remmers, 3 Martin Hudelmaier, 1 Wolfgang Wirth, 1 and Jeffrey L. Evelhoch 3 Objective. Quantitative magnetic resonance imaging (MRI) of articular cartilage represents a powerful tool in osteoarthritis (OA) research, but has so far been confined to a field strength of 1.5T. The aim of this study was to evaluate the precision of quantitative MRI assessments of human cartilage morphology at 3.0T and to correlate the measurements at 3.0T with validated measurements at 1.5T. Methods. MR images of the knee of 15 participants with OA and 15 healthy control subjects were acquired using Siemens 1.5T and 3.0T scanners. Double oblique coronal scans were obtained at 1.5T with a 1.5-mm partition thickness, at 3.0T with a 1.5-mm partition thickness, and at 3.0T with a 1.0-mm partition thickness. Cartilage volume, thickness, and surface area of the femorotibial cartilage plates were quantified using proprietary software. Results. For 1.5-mm partition thickness at 1.5T, the precision error was 3.0% and 2.6% for cartilage volume and cartilage thickness, respectively. The error was smaller for a 1.5-mm partition thickness at 3.0T Supported by Pfizer. 1 Felix Eckstein, MD, Martin Hudelmaier, MD, Wolfgang Wirth, Dipl Inf, Paracelsus Private Medical University, Salzburg, Austria, and Chondrometrics, Ainring, Germany; 2 H. Cecil Charles, PhD: Duke University Image Analysis Laboratory, Durham, North Carolina; 3 Robert J. Buck, PhD, Ann E. Remmers, PhD, Jeffrey L. Evelhoch, PhD (current address: Amgen, Thousand Oaks, California): Pfizer Global Research and Development, Ann Arbor, Michigan; 4 Virginia B. Kraus, MD, PhD: Duke University Medical Center, Durham, North Carolina. Dr. Eckstein has received consulting fees or honoraria (more than $10,000 each) from Virtual Scopics and Pfizer. Dr. Charles has received consulting fees or honoraria (less than $10,000 each) from GlaxoSmithKline, Pfizer, Novartis, Bristol-Meyers Squibb, and the NIH. Address correspondence and reprint requests to Felix Eckstein, MD, Institute of Anatomy and Musculoskeletal Research, Paracelsus Private Medical University, Strubergasse 21, A5020 Salzburg, Austria. felix.eckstein@pmu.ac.at. Submitted for publication February 2, 2005; accepted in revised form June 30, (2.6% and 2.5%) and still smaller for a 1.0-mm partition thickness at 3.0T (2.1% and 2.0%). Correlation coefficients between values obtained at 3.0T and 1.5T were high (r > 0.96), with no significant deviation between the two field strengths. Conclusion. Quantitative MRI measurement of cartilage morphology at 3.0T (partition thickness 1 mm) was found to be accurate and tended to be more reproducible than at 1.5T (partition thickness 1.5 mm). Imaging at 3.0T may therefore provide superior ability to detect changes in cartilage status over time and to determine responses to treatment with structuremodifying drugs. Quantitative magnetic resonance imaging (MRI) of articular cartilage represents a powerful tool in cartilage and osteoarthritis (OA) research and shows great promise for evaluating the response to treatment with structure/disease-modifying drugs (1,2). Radiography is the currently accepted method for assessing structural changes in the joints, but it has some important limitations. Strong interest is therefore currently directed at developing reliable biomarkers that are noninvasive, reproducible, and accurate in terms of reflecting joint status and disease progression (1 3). One of the current obstructions in the evaluation of new pharmacologic compounds is the lack of reliable biomarkers for evaluating therapeutic efficacy within reasonable observation intervals (3). At a field strength of 1.5T, 3-dimensional highresolution MRI protocols and innovative image analysis methods have made it possible to quantitatively determine articular cartilage morphology with a high level of accuracy and reproducibility in healthy and OA knees (2,4 6). MR imaging of the cartilage at 3.0T offers either an increased signal- and contrast-to-noise ratio (S/CNR), a higher spatial resolution, or a shorter imaging time relative to MR imaging at 1.5T. Since increased 3132

2 QUANTITATIVE MRI ASSESSMENT OF CARTILAGE MORPHOLOGY 3133 Figure 1. Precision error, as determined by the root mean square (RMS) coefficient of variation (CV) percentage, for magnetic resonance imaging measurements of cartilage volume at 1.5T with a 1.5-mm partition thickness (PT), at 3.0T with a 1.5-mm partition thickness, and at 3.0T with 1.0-mm partition thickness. Double oblique coronal images of the knee joint (examples shown at the right) were obtained on all study subjects. Med medial; Tib tibia; Lat lateral; Fem femur. S/CNR or higher spatial resolution (e.g., lower partition thickness) may improve the extractable quantitative information from MR images, the objective of this study was to test the hypothesis that MRI at 3.0T displays superior reproducibility (precision) of quantitative cartilage imaging. However, since changes in relaxation times for 1.5T to 3.0T field strengths differ among the tissues in the knee (7), separation of cartilage from adjacent tissues may be potentially altered at 3.0T. A second objective of this study was therefore to crosscalibrate quantitative cartilage measurements at 3.0T with those analytically validated at 1.5T (2,4 6). PATIENTS AND METHODS Thirty female participants ages 45 years and older were recruited at Duke University Medical Center. Fifteen subjects had mild-to-moderate OA, and 15 subjects were healthy. Each healthy control subject was age-matched to an OA patient (within 5 years). The study protocol and informed consent documentation were approved by the Institutional Review Board, and the study was conducted in compliance with the ethical principles of the Declaration of Helsinki (revised in Edinburgh in 2000). Posteroanterior radiographs of the knee in fixed flexion were obtained with the use of a SynaFlex radiographic positioning frame (Synarc, San Francisco, CA) in all study subjects (8). Inclusion criteria for the OA patients were frequent symptoms in 1 knee, radiographic evidence of mildto-moderate knee OA (Kellgren/Lawrence grade 2 in 11 patients and grade 3 in 4 patients) (9), and a 2-mm medial tibiofemoral joint space width. The mean duration of OA was 14 years. Inclusion criteria for the control subjects were the complete absence of knee symptoms and complete absence of radiographic knee OA (Kellgren/Lawrence grade 0 bilaterally). The mean SD age of the OA patients was years. Their mean SD body mass index was kg/m 2, and their mean SD medial femorotibial joint space width was mm. The mean age of the control subjects was years. Their mean body mass index was kg/m 2, and their mean medial femorotibial joint space width was mm. Imaging was performed with a 1.5T Magnetom MRI scanner and a 3.0T Trio MRI scanner (Siemens, Erlangen, Germany), using a fast low-angle shoot sequence with selective water excitation. This sequence has previously been analytically validated in anatomic specimens and in patients with total knee replacement (4 6). The knee was placed inside a circularly polarized transmit receive birdcage extremity radiofrequency knee coil and was immobilized in a Vac-Fix fixation system (S&S Par Scientific, Houston, TX) during imaging. Double oblique coronal MR images were obtained (Figure 1) as described previously (10), with an orientation perpendicular to the tibial plateau and with the posterior ends of both femoral condyles being located in the same or in adjacent slices. If the posterior ends of the femoral condyles were more than 1 slice apart, the acquisition was repeated. Participants were removed from the magnet between scans. Six 3-dimensional fast low flip-angle scans were ob-

3 3134 ECKSTEIN ET AL tained in the signal knee of each participant: two at 1.5T with a partition thickness of 1.5 mm (20 msec repetition time, 7.6 msec echo time, 20 flip angle, 80 partitions, and 6 minutes 51 seconds acquisition time; two at 3.0T with a partition thickness of 1.5 mm (20 msec repetition time, 7.6 msec echo time, 12 flip angle, 80 partitions, and 6 minutes 51 seconds acquisition time); and two at 3.0T with a partition thickness of 1.0 mm (16 msec repetition time, 7.2 msec echo time, 12 flip angle, 120 partitions, and 8 minutes 44 seconds acquisition time). The flip angle was adapted at 3.0T to account for the increased spin-lattice relaxation times (T1). All other parameters were kept constant between protocols: 160-mm field of view, matrix, 0.31-mm in-plane resolution, 100% phase resolution, 100% slice resolution, number of excitations 1, 130-Hz pixel bandwidth, elliptical filter on, and asymmetric echo off. No significant drift in spatial linearity was observed during the data acquisition phase of the study, based on phantom measurements. All data were reviewed at the Duke Image Analysis Laboratory, anonymized, and CDs of the data were shipped to the image analysis center at Chondrometrics (Ainring, Germany), where images were processed using proprietary software. Segmentation involved tracing of the bone interface and the cartilaginous joint surface of the medial tibia, lateral tibia, medial femoral condyle, and lateral femoral condyle on a slice-by-slice basis. A specific region of interest was analyzed in femoral cartilage, between the intercondylar notch and the posterior intercondylar bone bridge, as described previously (10). All data for each patient (6 scans) were analyzed by 1 of 8 technicians who had formal training and thorough experience in cartilage segmentation. The initial (segmented) scan was uploaded on the screen during segmentation of the repeat scans within each protocol, but each protocol was segmented independently of the others. Quality control of all segmentations was performed by a single expert (FE), who reviewed all segmented slices of each data set. If required, quality control comments (text or drawings) were entered interactively with the software, and the technician who had performed the segmentations adjusted them accordingly. The segmentations were then used to compute the cartilage volume (numerical integration of segmented voxels), the cartilage surface area (triangulation), and the cartilage thickness over the cartilagecovered areas of subchondral bone (3-dimensional Euclidean distance transformation, with minimal distance from articular surface to bone interface), based on previously described methods (11,12). The precision (reproducibility) of each protocol was determined by computing the root mean square coefficient of variation percentage from the repeated (dual) acquisitions. Precision errors between the protocols were compared using a 1-sided Wilcoxon s signed rank test. The accuracy of the 3.0T acquisitions was evaluated by computing Pearson s correlation coefficients, random pairwise differences (%; average pairwise differences with elimination of the and signs), and systematic pairwise differences (%; average pairwise differences without elimination of the and signs) acquisitions at 1.5T. Systematic differences between techniques were tested using Hotelling s paired t-test. Table 1. Precision error associated with repeated measurements of cartilage volume, surface area, and mean cartilage thickness over the cartilage-covered area of bone of the femorotibial joint by magnetic resonance imaging at 1.5T and at 3.0T* Cartilage region, morphologic parameter 1.5 mm at 1.5T Partition thickness and field strength 1.5 mm at 3.0T 1.0 mm at 3.0T All regions (summarized) Cartilage volume Surface area Cartilage thickness Medial tibia Cartilage volume Surface area Cartilage thickness Lateral tibia Cartilage volume Surface area Cartilage thickness Medial femur Cartilage volume Surface area Cartilage thickness Lateral femur Cartilage volume Surface area Cartilage thickness * Precision error was determined by computing the root mean square coefficient of variation percentage. Precision errors between protocols were compared using a 1-sided Wilcoxon s signed rank test. Cartilage thickness represents the mean cartilage thickness over the cartilagecovered area of bone. A 1-sided Wilcoxon s signed rank test was applied here to averages of coefficients of variation within a knee. Coefficients of variation smaller (P 0.1) than at (borderline significance). Coefficients of variation significantly smaller (P 0.05) than at 1.5 mm at 1.5T. RESULTS For 1.5-mm thick partitions acquired at 1.5T, the precision error for all cartilage plates summarized was 3.0% for the cartilage volume and 2.6% for the cartilage thickness. The error was smaller for 1.5-mm thick partitions acquired at 3.0T (2.6% and 2.5%) (P 0.08 and 0.39, respectively) and still smaller for 1.0-mm thick partitions acquired at 3.0T (2.1% and 2.0%) (P and 0.04, respectively), with similar trends being observed in all knee cartilage plates (Table 1 and Figure 1). No differences in precision errors were observed between OA patients and control subjects (data not shown). Measurements at 3.0T correlated highly with those at 1.5T (Table 2), with correlation coefficients ranging from 0.90 to Random pairwise differences

4 QUANTITATIVE MRI ASSESSMENT OF CARTILAGE MORPHOLOGY 3135 Table 2. Accuracy of measurements of cartilage volume, surface area, and mean cartilage thickness over the cartilage-covered area of bone of the femorotibial joint by magnetic resonance imaging at 3.0T compared with 1.5T* Cartilage region, morphologic parameter Pearson s correlation coefficient, r Average random pairwise differences, % Average systematic pairwise differences, % All regions (summarized) Cartilage volume Surface area Cartilage thickness Medial tibia Cartilage volume Surface area Cartilage thickness Lateral tibia Cartilage volume Surface area Cartilage thickness Medial femur Cartilage volume Surface area Cartilage thickness Lateral femur Cartilage volume Surface area Cartilage thickness * Cartilage thickness represents the mean cartilage thickness over the cartilage-covered area of bone. None of the systematic differences were significant at P 0.05, by Hotelling s paired t-test. ranged from 3.1% to 6.7% for all cartilage plates summarized (Table 2). Systematic differences between the measurements at 3.0T and 1.5T were small and were not statistically significant. DISCUSSION The in vivo precision (reproducibility) of quantitative MRI is critical when using the technique to monitor disease progression and response to therapy. Since the magnitude of change (cartilage loss in OA) per se should be independent of the specific methodology, one of the most important factors in determining the sample size and/or trial duration necessary to demonstrate a desired effect of a therapeutic compound with statistical confidence is the magnitude of the precision error of the technique applied. To our knowledge, this is the first study to examine a potential reduction in precision errors of quantitative measurements of cartilage morphology by MRI at 3.0T. The ways in which the higher field strength of 3.0T can be exploited involve a shorter imaging time, an increased S/CNR, or an increased spatial resolution of the images. In the present study, we evaluated whether an increase in S/CNR or a reduction of the partition thickness (with constant in-plane resolution) at a field strength of 3.0T is more efficient in increasing the precision of quantitative MRI measurement of cartilage morphology. Precision errors found at 1.5T were in the same range as those previously reported for coronal images at 1.5T (10,11,13,14). Imaging of the cartilage at 3.0T yielded somewhat smaller precision errors than at 1.5T, and the reduction in partition thickness was more effective than a gain in S/CNR. Improvements were made in measuring cartilage thickness rather than surface area, likely because the boundaries along the bone interface and the articular surface profited more from the higher resolution (reduction in partial volume effects) than the relatively sharp boundaries at the sides of the cartilage plates. Our findings also show that values obtained at 3.0T are highly consistent with those previously validated at 1.5T. Hence, if the flip angle is adapted to account for differences in relaxation times, no differences in the relative signal characteristics of cartilage and the surrounding tissues occur at 3.0T that obstruct the validity of quantitative cartilage imaging. Random

5 3136 ECKSTEIN ET AL differences between 3.0T and 1.5T were not larger than those between 1.5T and an ex vivo gold standard (2,4 6). This study had several limitations. Only a few patients and controls were studied, and the study population was confined to women. However, the 15 OA patients spanned a wide range of OA changes, with some displaying almost no apparent cartilage changes and 3 displaying areas of bone erosion. Another limitation is that only 2 repeat scans were acquired for each protocol and only a limited number of protocols were tested in order to keep the aggregate acquisition time within reasonable limits. We cannot therefore rule out the possibility that using even thinner partitions at 3.0T would have increased the precision further or that a reduction in the partition thickness at 1.5T would have also involved improvements in precision errors compared with the standard 1.5-mm partition thickness protocol. In addition, MR scanners produced by only 1 vendor were used, and the results may not necessarily apply to scanners from other manufacturers. A general limitation is that fewer 3.0T scanners are currently available, and these scanners are more costly. The precision errors reported here represent a segmenting of the data sets by the same observer at the same point in time. The rationale for this approach was that the same approach is taken in longitudinal studies, with blinding of the reader to the order of acquisition (baseline followup) (14). However, the 3.0T scans were segmented de novo, without displaying the previously segmented 1.5T scans. The 1.5T images were processed first to avoid the possibility that segmentations of the standard 1.5T acquisitions were driven by mental maps of the previously segmented (and potentially superior) 3.0T scans. This assumption was retrospectively verified by the finding of lower precision errors at 3.0T. In conclusion, this study demonstrates that quantification of human cartilage morphology by MRI at 3.0T (partition thickness 1.0 mm) is accurate and tends to be more reproducible than at 1.5T (partition thickness 1.5 mm). Although the current gold standard (1.5T) yields satisfactory reproducibility for performing scientifically accurate followup studies, the implementation of 3.0T coronal quantitative MRI with a partition thickness of 1 mm may result in an improved ability to detect longitudinal changes in cartilage status and to determine treatment responses to structure-modifying drugs. ACKNOWLEDGMENTS We are grateful for the invaluable assistance of Janet Huebner, MS (Duke University), Maureen Ainslie, MS, RT (Duke University), Peggy Asbury, RN (Pfizer), staff of the Center for Advanced MR Development, staff of the Duke University Image Analysis Laboratory, and the Pfizer A Team in conducting this study. In addition, we would like to thank Dr. Susanne Maschek, Dr. Barbara Wehr, Sabine Mühlsimer, Linda Jakobi, Gudrun Goldmann, Manuela Kunz, Astrid Grams, and Regina Feurer (Chondrometrics) for data segmentation. REFERENCES 1. Gray ML, Eckstein F, Peterfy C, Dahlberg L, Kim YJ, Sorensen AG. Toward imaging biomarkers for osteoarthritis. Clin Orthop 2004;427 Suppl:S Eckstein F, Glaser C. Measuring cartilage morphology with quantitative magnetic resonance imaging. Semin Musculoskelet Radiol 2004;8: Peterfy CG. Role of MR imaging in clinical research studies. Semin Musculoskelet Radiol 2001;5: Burgkart R, Glaser C, Hyhlik-Durr A, Englmeier KH, Reiser M, Eckstein F. Magnetic resonance imaging based assessment of cartilage loss in severe osteoarthritis: accuracy, precision, and diagnostic value. Arthritis Rheum 2001;44: Burgkart R, Glaser C, Hinterwimmer S, Hudelmaier M, Englmeier KH, Reiser M, et al. Feasibility of T and Z scores from magnetic resonance imaging data for quantification of cartilage loss in osteoarthritis. Arthritis Rheum 2003;48: Graichen H, Eisenhart-Rothe R, Vogl T, Englmeier KH, Eckstein F. Quantitative assessment of cartilage status in osteoarthritis by quantitative magnetic resonance imaging: technical validation for use in analysis of cartilage volume and further morphologic parameters. Arthritis Rheum 2004;50: Gold GE, Han E, Stainsby J, Wright G, Brittain J, Beaulieu C. Musculoskeletal MRI at 3.0T: relaxation times and image contrast. AJR Am J Roentgenol 2004;183: Peterfy C, Li J, Zaim S, Duryea J, Lynch J, Miaux Y, et al. Comparison of fixed-flexion positioning with fluoroscopic semiflexed positioning for quantifying radiographic joint-space width in the knee: test-retest reproducibility. Skeletal Radiol 2003;32: Kellgren JH, Lawrence JS. Radiological assessment of osteoarthrosis. Ann Rheum Dis 1957;16: Glaser C, Burgkart R, Kutschera A, Englmeier KH, Reiser M, Eckstein F. Femoro-tibial cartilage metrics from coronal MR image data: technique, test-retest reproducibility, and findings in osteoarthritis. Magn Reson Med 2003;50: Stammberger T, Eckstein F, Englmeier KH, Reiser M. Determination of 3D cartilage thickness data from MR imaging: computational method and reproducibility in the living. Magn Reson Med 1999;41: Hohe J, Ateshian G, Reiser M, Englmeier KH, Eckstein F. Surface size, curvature analysis, and assessment of knee joint incongruity with MRI in vivo. Magn Reson Med 2002;47: Hyhlik-Durr A, Faber S, Burgkart R, Stammberger T, Maag KP, Englmeier KH, et al. Precision of tibial cartilage morphometry with a coronal water-excitation MR sequence. Eur Radiol 2000; 10: Eckstein F, Heudorfer L, Faber SC, Burgkart R, Englmeier KH, Reiser M. Long-term and resegmentation precision of quantitative cartilage MR imaging (qmri). Osteoarthritis Cartilage 2002;10: