Comparison of 3D vs. 2D fast spin echo imaging for evaluation of articular cartilage in the knee on a 3T system scientific research

Purpose We sought to retrospectively compare the accuracy of a three-dimensional fat-suppressed, fast spin-echo sequences acquired in the sagittal plane, with multiplanar reconstructions to that of two-dimensional fat-suppressed, fast spin echo sequences acquired in three planes on a 3 T MR system for the evaluation of articular cartilage in the knee. Materials and methods Our study group consisted of all patients ( N = 34) that underwent 3 T MR imaging of the knee at our institution with subsequent arthroscopy over an 18-month period. There were 21 males and 13 females with an average age of 36 years. MR images were reviewed by 3 musculoskeletal radiologists, blinded to operative results. 3D and 2D sequences were reviewed at different sittings separated by 4 weeks to prevent bias. Six cartilage surfaces were evaluated both with MR imaging and arthroscopically with a modified Noyes scoring system and arthroscopic results were used as the gold standard. Sensitivity, specificity, and accuracy were calculated for each reader along with Fleiss Kappa assessment agreement between the readers. Accuracies for each articular surface were compared using a difference in proportions test with a 95% confidence interval and statistical significance was calculated using a Fisher's Exact Test. Results Two hundred and four articular surfaces were evaluated and 49 articular cartilage lesions were present at arthroscopy. For the patellofemoral surfaces, the sensitivity, specificity, and accuracy were 76.5%, 83%, and 78.2% for the 3D sequences and were 82.3%, 76%, and 82% respectively for the 2D sequences. For the medial compartment surfaces, the sensitivity, specificity, and accuracy were 81.1%, 65.1%, and 78.5% for the 3D sequences and were 82.5%, 48%, and 76.7% respectively for the 2D sequences. For the lateral compartment surfaces, the sensitivity, specificity, and accuracy were 89.3%, 39%, and 79.5% for the 3D sequences and were 94.7%, 18.8%, and 79.5% respectively for the 2D sequences. The accuracies were not significantly different between 3D and 2D sequences. Fleiss Kappa agreement values for the assessment of inter-observer agreement ranged from substantial for the patella and medial femur to moderate for the trochlea and fair for the medial tibia and lateral compartment. Conclusion There was no significant difference in accuracy for the evaluation of articular cartilage of a single three-dimensional, fast spin echo sequence with multi-planar reformatted images vs. two-dimensional, fast spin echo sequences acquired in all three imaging planes in the knee. Keywords MRI Knee Articular cartilage 1 Introduction Magnetic resonance (MR) imaging provides a noninvasive means for morphological and biochemical assessments of articular cartilage [1] . Morphological data have been the mainstay of current clinical analysis of articular surfaces with two dimensional fast spin echo techniques being the most commonly used imaging sequences [2–4] . Three dimensional imaging of articular cartilage is not a new technique, and was initially described with a spoiled gradient echo technique [5] . This technique has not been widely used as the long acquisition times and limited soft tissue contrast for other structures within the knee make them less efficient. Newer 3D GRE imaging techniques such as 3D double-echo steady-state (DESS) have improved on the SPGR sequences with faster acquisition times and provide an arthrographic effect from water excitation [6] . Additional techniques such as true fast imaging with steady-state precession (FISP) also attempt to acquire isotropic data in one plane and use multiplanar reformations in an attempt to decrease scanning time [7–9] . Three dimensional T2/intermediate acquisitions have initially been plagued by (echo related) signal loss and echo blurring from long echo times needed for the acquisitions. Newer techniques such as CUBE (GE Medical Systems) and SPACE (Siemens Medical Systems) have circumvented these problems using flip angle modulation which allows for decreased blurring; and parallel imaging, which greatly reduces the number of phase encoding steps required for a large 3D data set [10] . It is the hope that improvements in imaging techniques will transcend technical problems with 3 dimensional T2/PD weighted sequences and be able to capture the tissue contrast similar to the 2D FSE techniques which have become the mainstay of musculoskeletal imaging. If T2/PD isotropic data can be adequately generated, this will significantly decrease imaging time as patient will only have to be scanned in one plane, significantly improving the efficiency of musculoskeletal examinations, improving patient comfort and decreasing motion related artifacts from prolonged scan times. The goal of this study was to retrospectively evaluate the diagnostic accuracy of three-dimensional fat-suppressed fast spin-echo pulse sequence with multiplanar reformatted images, as compared to that of two-dimensional fat-suppressed fast spin-echo sequence acquired in all three imaging planes in the detection of cartilage lesions in the knee in vivo on a 3 T magnet, while using arthroscopic findings as the standard of reference. 2 Materials and methods 2.1 Patient selection and medical records review This study was performed in compliance with the Health Insurance Portability and Accountability Act (HIPAA) and with approval from our Human Investigation Committee. We reviewed the records of patient logs of knee MR examinations on our 3 T system who subsequently underwent knee arthroscopies. We identified 34 patients over an 18 month period who met the inclusion criteria. There were 21 males and 13 females with an average age of 36 years. Indications for MRI evaluation and subsequent arthroscopy were not limited to articular cartilage injury and included the diagnoses of meniscal tears, ACL tear, loose body, ostochondritis dissecans as well as isolated articular cartilage injury. Inclusion criteria included having a 3 T MR at our institution with a standard protocol in which 3D FSE T2 weighted imaging was performed and subsequent arthroscopy. Exclusion criteria were prior arthroscopic surgery, significant metal artifact which limited imaging, uninterpretable examinations and arthrographic studies. 2.2 MR imaging technique All patients underwent MR imaging of the knee on 3-T system software version 14.0 HDX (GE Medical Systems, Milwaukee, WI) and were imaged in a dedicated high resolution, 8 channel knee coil (Invivo Corp, Orlando, FL). The 3D FSE sequence parameters were similar to those previously described by Gold et al. [10] . The 3D FSE sequence was acquired in the sagittal plane with following imaging parameters: TR/TE 2500/38, 15 FOV, matrix 256 × 224, 1.2 slice thickness with a slice zip of 2 to generate 0.6 mm images, ASSET factor of 2, frequency selective fat suppression and a pixel bandwidth of 162.77. The approximate imaging time was 5 min and 11 s. The interpolated voxel size was 0.7 cm × 0.6 cm × 0.6 cm. Coronal and axial reformatted images were generated on a Synapse workstation (Fuji Medical Systems, Stamford, CT). Two dimensional sequence parameters were standard sequences we use to evaluate articular cartilage at our institution. They were acquired in all 3 planes (axial, sagittal and coronal) with the following parameters: TR/TE 3500/80; 15 FOV, matrix 384 × 224; 3 mm slice thickness with 1 mm inter-slice gap; ETL 14; ASSET factor 1.5, frequency selective fat-suppression, 2 Excitations and a pixel bandwidth of 122.07. The approximate imaging time of each sequence was 2 min and 56 s. 2.3 MR image analysis The MRI images were interpreted independently by three fellowship-trained and experienced musculoskeletal radiologists (HM, LDK, AHH). Six knee articular surfaces were evaluated: patella, trochlea, medial femoral condyle, lateral femoral condyle, medial tibial plateau, and lateral tibial plateau, using the Modified Noyes classification [11] . This classification consists of five grades ranging 0–4 as follows: grade 0 is normal articular cartilage ( Fig. 1 ), grade 1 is intact cartilage with softening or deformation at arthroscopy or signal change on MR imaging, grade 2 is surface cartilage disruption involving less than 50% of the total thickness ( Fig. 2 ), grade 3 is surface cartilage disruption involving greater than 50% of the total thickness, and grade 4 is exposed subchondral bone. The different sequences were evaluated at different sittings on different days separated by 4 weeks to prevent bias. The radiologists were blinded to results of the knee arthroscopy, though they did have knowledge that the patients had undergone arthroscopy. 2.4 Surgical data Knee arthroscopies were performed by 5 orthopedic surgeons with specialty training in sports medicine. Time from imaging to surgery ranged from 3 days to 354 days, with mean and median times of 69 and 48 days respectively. Standard arthroscopy techniques were used and all 6 articular surfaces were inspected. Operative reports from all cases were reviewed and the articular cartilage surface descriptions were recoded under the Modified Noyes classification. 2.5 Statistical analysis The articular cartilage defects were dichotomized for the purpose of analysis into no defects (Modified Noyes Grades 0–1) and defects (Modified Noyes Grades 2–4). Sensitivity, specificity, and accuracy were calculated for each reader for each of the six articular surfaces with arthroscopic data used as the gold standard. Fleiss Kappa test was used to assess the interobserver agreement between each articular surface. Agreement was graded as almost perfect (1.00–0.81), substantial (0.80–0.61), moderate (0.60–0.41), fair (0.40–0.21) and slight (0.20–0.0). Accuracies for each articular surface were compared using a difference in proportions test with a 95% confidence interval and statistical significance was calculated using a Fisher's Exact Test. 3 Results Two hundred and four articular surfaces were evaluated and 49 articular cartilage lesions were present at arthroscopy ( Table 1 ). The most frequent articular defects were identified in the medial femoral surface ( Fig. 3 ). The mean of sensitivities, specificities and accuracies for 3D patellar evaluation was 80%, 86% and 82% respectively. Two dimensional patellar evaluation demonstrated sensitivities, specificities and accuracies of 86%, 92% and 88% respectively. The mean sensitivities, specificities and accuracies for 3D trochlea evaluation was 73%, 80% and 74% respectively. Two dimensional trochlear evaluation demonstrated sensitivities, specificities and accuracies of 79%, 60% and 76% respectively ( Table 2 ). In the medial compartment, the mean sensitivities, specificities and accuracies for 3D vs. 2D femoral evaluation is as follows: 3D; 86%, 84% and 85% vs. 2D; 84%, 76% and 81%. For the medial tibia, the mean sensitivities, specificities and accuracies for 3D vs. 2D tibia evaluation is as follows: 3D; 77%, 47% and 72% vs. 2D; 81%, 20% and 72% ( Table 2 ). In the lateral compartment, the mean sensitivities, specificities and accuracies for 3D vs. 2D femoral evaluation is as follows: 3D; 87%, 50% and 78% vs. 2D; 95%, 21% and 78%. For the lateral tibia, the mean sensitivities, specificities and accuracies for 3D vs. 2D tibia evaluation is as follows: 3D; 92%, 28% and 81% vs. 2D; 94%, 17% and 81% ( Table 2 ). The accuracy was used as a number that combined both sensitivity and specificity to compare the 3D vs. the 2D sequences. Each reader's scores for individual articular surface were compared for non-inferiority using a difference in proportions test and a Fisher's exact test and 95% confidence intervals were calculated. Additionally, the three reader's scores were lumped together for each articular surface and 3D and 2D accuracies were compared in a similar fashion. On both of these calculations there was no significant difference in accuracy between the 3D FSE sequence with multiplanar reformations and the 2D sequences acquired in the planes ( Table 3 ). The intraobserver agreement for the patella ranged from substantial for the 3D sequences to almost perfect for the 2D sequences. The trochlea had Fleiss Kappa values in the moderate range for both the 3D and 2D sequences. Evaluation of the medial femur demonstrated substantial agreement for both groups while the medial tibia showed only fair agreement for both groups. Agreement in the lateral femur was fair for both groups and in the lateral tibia fair for the 3D sequence but only slight for the 2D sequence ( Table 4 and Fig. 6 ). 4 Discussion Articular cartilage imaging is paramount in clinical knee MR imaging. A plethora of different MR imaging techniques are being used to evaluate hyaline cartilage from standard 2D and 3D techniques for cartilage surface morphology to newer quantitative techniques such as T2 mapping, T1 rho, and delayed gadolinium-enhanced MRI of cartilage (dGEMRIC) which provide a look at cartilage, repair, and biochemistry. However, 2D FSE MR sequences remain the standard for clinical musculoskeletal imaging, particularly in evaluating cartilage lesions. Three-dimensional sequences offer another technique that could potentially supplant conventional 2D sequences currently used in the clinical realm. Three dimensional isovoxel true fast imaging with steady state precession with water excitation has been shown in recent studies to have comparable diagnostic performance to conventional 2D MR imaging for evaluation of the anterior cruciate ligament tears, meniscal tears and articular cartilage lesions [8,9] . Although in a larger study, Kijowski et al. described potential limitations of T2/T1 weighted tissue contrast of this sequence showing significantly lower sensitivities for the evaluation of lateral meniscal tears and bone marrow edema lesions [12] . Because of recent advances in hardware and software 3D FSE imaging has become an option for routine clinical imaging of the knee. This technique, with isotropic or relatively isotropic imaging allows imagers to scan in one plane and employ multiplanar reconstructions in additional desired planes with no or little loss of anatomic information. The sequence, due to its intermediate weighting, is more versatile than other 3D cartilage sequences and has been show to adequately evaluate additional structures in the knee such as menisci and ligaments. We have evaluated the diagnostic accuracy of this relative new technique for the evaluation of hyaline cartilage in the knee on a 3 T system with a dedicated 8 channel knee coil. We have shown that there is no significant difference in accuracy between our 3D FSE sequence with multiplanar reconstructions and 2D FSE sequences acquired in three planes for the evaluation of articular cartilage in the knee. These results are in line with preliminary studies on similar sequences by Gold et al. [10] . Our accuracies are comparable with a recent study [13] evaluating articular cartilage in the knee using the 3D FSE technique [79% vs. 83% respectively]. Our sensitivities were somewhat higher [82% vs. 72% respectively] and our specificities were lower [62% vs. 82%]. The differences in our numbers may reflect our smaller patient population. Our specificities in the lateral compartment were lower than those for the medial and patellofemoral compartment with both the 3D and 2D techniques. Additionally, the readers had the lowest agreement scores in the lateral compartment ranging from fair to slight. This is related to a high number of false positives and readers overcalling articular defects in the lateral compartment and also explaining the higher sensitivities ( Fig. 4 ). Prior studies have also demonstrated poorer results in the lateral compartment, possibly related to the convex surface of the lateral tibial plateau which can be associated with imaging artifacts [2,5] . There were several limitations to our study. First we had a relatively small patient population ( N = 34) with relatively few articular defects ( N = 49). Our study was also performed in a retrospective fashion and subsequently all of our readers were aware that all patients reviewed underwent arthroscopic surgery which could potentially lead to bias. Also, the 3D FSE sequence that was performed was anisotropic, with an interpolated voxel size was 0.7 cm × 0.6 cm × 0.6 cm. Using this anisotropic data set, there is potential for data lost with multiplanar reformations. Although even with this potential data loss, the accuracy of articular cartilage was similar to that the 2D FSE sequences ( Fig. 5 ). Another inherent limitation to our study is the assumption that diagnostic arthroscopy of the knee and concomitant grading of articular cartilage damage with the scale from the Modified Noyes classification system are completely accurate and reproducible. They are neither. A study addressed the accuracy and reproducibility of the Outerbridge classification system for chondral damage in the knee [14] . Six cadaveric knees first underwent videotaped arthroscopy and then direct arthrotomy to determine chondral damage. Those researchers found an overall accuracy rate of 68% for all observers. Predictably, low-grade lesions were diagnosed with less accuracy than high-grade lesions. The kappa value between the arthrotomy grade and the arthroscopy grade was 0.602. This kappa value correlates with fair to good agreement. The intraobserver kappa value was 0.80, which correlates with excellent agreement. The point was well made that diagnostic arthroscopy is not infallible. In conclusion, the results of our study demonstrate similar diagnostic accuracy between 3D FSE with MPR's and 2D FSE with images acquired in all three imaging planes, for the evaluation of the knee on 3.0 T. This study adds to the increasing pool of data suggesting that 3D imaging in one plane with multiplanar reformations may be able to obviate 2D scanning in all three planes, thereby increasing scanning efficiency and patient comfort, although additional larger studies are needed. Conflict of interest All authors declare that they have no conflict of interest. References [1] T.R. McCauley D.G. Disler MR imaging of articular cartilage Radiology 209 1998 629 640 [2] H.G. Potter J.M. Linklater A.A. Allen J.A. Hannafin S.B. Haas Magnetic resonance imaging of articular cartilage in the knee. An evaluation with use of fast-spin-echo imaging J Bone Joint Surg Am 80 September (9) 1998 1276 1284 [3] A.H. Sonin R.A. Pensey M.E. Mulligan S. HAtem Grading of the articular cartilage of the knee using fast spin-echo proton density-weighted M.R. imaging without fat suppression AJR 179 2002 1159 1166 [4] M.A. Bredella P.F. Tirman C.G. Peterfy Accuracy of T2-weighted fast spin-echo MR imaging with fat saturation in detecting cartilage defects in the knee: comparison with arthroscopy in 130 patients AJR Am J Roentgenol 172 April (4) 1999 1073 1080 [5] D.G. Disler T.R. McCauley C.G. Kelman Fat-suppressed three-dimensional spoiled gradient-echo MR imaging of hyaline cartilage defects in the knee: comparison with standard MR imaging and arthroscopy AJR Am J Roentgenol 167 July (1) 1996 127 132 [6] P.R. Knuesel C.W. Pfirrmann H.P. Noetzli MR arthrography of the hip: diagnostic performance of a dedicated water-excitation 3D double-echo steady-state sequence to detect cartilage lesions AJR Am J Roentgenol 183 December (6) 2004 1729 1735 [7] S.R. Duc P. Koch M.R. Schmid W. Horger J. Hodler C.W. Pfirrmann Diagnosis of articular cartilage abnormalities of the knee: prospective clinical evaluation of a 3D water-excitation true FISP sequence Radiology 243 May (2) 2007 475 482 [8] S.R. Duc C.W. Pfirrmann M.R. Schmid Articular cartilage defects detected with 3D water-excitation true FISP: prospective comparison with sequences commonly used for knee imaging Radiology 245 October (1) 2007 216 223 [9] S.R. Duc C.W. Pfirrmann P.P. Koch M. Zanetti J. Hodler Internal knee derangement assessed with 3-minute three-dimensional isovoxel true FISP MR sequence: preliminary study Radiology 246 February (2) 2008 526 535 [10] G.E. Gold R.F. Busse C. Beehler Isotropic MRI of the knee with 3D fast spin-echo extended echo-train acquisition (XETA): initial experience AJR Am J Roentgenol 188 May (5) 2007 1287 1293 [11] F.R. Noyes C.L. Stabler A system for grading articular cartilage lesions at arthroscopy Am J Sports Med 17 July–August (4) 1989 505 513 [12] R. Kijowski D.G. Blankenbaker J.L. Klaers K. Shinki A.A. De Smet W.F. Block Vastly undersampled isotropic projection steady-state free precession imaging of the knee: diagnostic performance compared with conventional MR Radiology 251 April (1) 2009 185 194 [13] R. Kijowski K.W. Davis M.A. Woods Knee joint: comprehensive assessment with 3D isotropic resolution fast spin-echo MR imaging – diagnostic performance compared with that of conventional MR imaging at 3.0 T Radiology 252 August (2) 2009 486 495 [14] M.L. Cameron K.K. Briggs J.R. Steadman Reproducibility and reliability of the Outerbridge classification for grading chondral lesions of the knee arthroscopically Am J Sports Med 31 2003 83 86