Morphological MR imaging of the articular cartilage of the knee at 3 T—comparison of standard and novel 3D sequences

This prospective study was approved by the Institutional Review Board of the Antwerp University Hospital according to the ICH Good Clinical Practice rules (registration no. B300201421002). Written informed consent was obtained from all of the participants after the nature of the examinations had been fully explained. From 1 March to 31 May 2014, a total of 20 subjects were prospectively enrolled in this study, including 10 healthy volunteers (8 males and 2 females; mean age 25 years; range 23–32 years; mean body mass index 22.6 kg/m²) and 10 patients (2 males and 8 females; mean age 49 years; range 27–61 years; mean body mass index 26.4 kg/m²). All volunteers were medical or paramedical personnel (e.g. medical students, residents, and nurses), and none of them had known musculoskeletal disease or prior knee symptoms. A patient was enrolled in the study if they (1) had a clinical suspicion of a cartilage abnormality at the knee (pain and stiffness for more than 6 months, n = 5; anterior knee pain aggravating when climbing stairs, n = 3; recent episodes of pain and functional limitation in nonprofessional marathon runners, n = 2), (2) had no history of prior knee surgery, and (3) consented to undergo a 3-T MRI of the knee with dedicated cartilage imaging sequences. Subjects with contra-indications to the use of MRI (e.g. aneurysm clip, cardiac pacemaker, claustrophobia) were not included in the study. The right knee was imaged in all study participants.

All images were acquired with a 3-T MR imaging system (Magnetom SKYRA, Siemens, Erlangen, Germany) by using a transmit/receive 15-channel knee coil (Quality Electrodynamics, Mayfield, OH, USA). The MR protocol consisted of five different isotropic 3D sequences: water-excited (we) 3D Fast-Low Angle Shot (FLASH), we Double-Echo Steady-State (DESS), we True Fast Imaging with Steady-State Free Precession (TrueFISP), fat-saturated (fs) intermediate-weighted 3D Turbo Spin-Echo Sampling Perfection with Application optimised Contrasts using variable flip angle Evolutions (SPACE), and a prototype 3D Ultrashort Echo-Time (UTE) sequence. All sequences had a sagittal plane of acquisition and used a 0.6 mm³ voxel volume, except for the 3D UTE sequence, which employed a 0.8 mm³ voxel size. Detailed parameters for all imaging sequences are provided in Table 1. In our study, a dual echo 3D UTE sequence was used consisting of a 60-μs nonselective RF pulse followed by 3D radial ramp sampling from the centre to the surface of a sphere. In order to achieve the shortest possible TE, data acquisition started already during ramp-up time of the readout gradient. Frequency-based fat suppression was available as an additional option. Difference images were formed to suppress signals from long T2 components by subtracting the later echo time (TE = 3.92 ms) images from the first image (TE = 60 μs) (Fig. 1). The scan time of the 3D UTE sequence was 6 min 48 s, which is feasible for routine clinical use.

All images were independently reviewed during separate sessions by two readers [with 25 years (reader A) and 10 years (reader B) of experience in musculoskeletal radiology at the time of the study]. Images were obtained without any annotation. At the time of review, the readers were blinded to the patient’s clinical history and clinical findings. A total of 100 MRI data sets (5 data sets per subject for a total of 20 study subjects) were reviewed in 5 separate sessions at least 96 h apart to prevent recall bias. Each session consisted of 20 data sets that were presented randomly except that different data sets of the same patient were never reviewed in the same session. The readers used sagittal source data to create multiplanar reformatted (MPR) images in the axial and coronal plane. Subjective image quality was assessed by using the following criteria: edge sharpness, artefacts, amount of noise, tissue contrast, and reformat quality. The quality score consisted of an ordinal 5-point Likert-scale and was defined as (1) poor, unacceptable for diagnostic purposes; (2) adequate but poorer than average quality; (3) average quality of a diagnostic acceptable image; (4) above average quality; (5) best quality. The readers graded each of the six articular surfaces of the knee as follows: (1) normal; (2) low-grade cartilage defect (<50 % of total cartilage thickness); (3) high-grade cartilage defect (>50 % of total cartilage thickness). Each reader was asked to assign a confidence level to the diagnosis as follows: (1) definitely normal; (2) probably normal; (3) probably abnormal; (4) definitely abnormal. Presence of bone marrow oedema was also recorded using the same 4-point Likert scale. On UTE images, the readers prospectively evaluated the morphology of the deep osteochondral tissues to predict the status of the overlying cartilage: a distinct, continuous high linear signal at the cartilage/bone interface was considered normal (indicating normal cartilage) and thinning, interruption, or complete absence of that signal was considered abnormal (indicating cartilage lesion), according to previously described criteria by Bae et al [12, 18]. Finally, the UTE morphology of the osteochondral tissue was retrospectively assessed in consensus and with knowledge of the findings on conventional MR sequences.

The standard of reference for the MR abnormalities was based on image analysis by two senior radiologists in consensus with all available MR sequences as well as clinical findings. Both radiologists had more than 25 years of experience in musculoskeletal radiology at the time of the study and were not involved in the blinded study readings.

Statistical analysis was performed by a statistician with use of R software (version 3.1.1.; R Foundation for Statistical Computing, Vienna, Austria). Differences between the MR sequences were considered to be statistically significant if p < 0.05. The Friedman test and pairwise Wilcoxon tests with Bonferroni-Holm correction for multiple testing were used to compare the performance and subjective image quality gradings of all sequences [19]. Interreader agreement was determined using the weighted kappa (к) statistic with к value ≥ 0.8 indicating excellent agreement. To increase statistical power for the performance analysis, calculations were made for all articular surfaces, all grades of severity, and both readers combined. Therefore, a bootstrapping procedure with 1000 samples at the level of the patients was used to calculate standard errors. Finally, the proportion of definite MR diagnoses (i.e. definitely normal or definitely abnormal) was calculated for each sequence.