Assessment of cartilage-dedicated sequences at ultra-high-field MRI: comparison of imaging performance and diagnostic confidence between 3.0 and 7.0 T with respect to osteoarthritis-induced changes at the knee joint

Seven patients with mild osteoarthritis (OA) (43.57 ± 11.97 years, four male, three female) were recruited for the study. They showed clinical symptoms and mild radiographic signs of knee OA: four of them had grade 1 OA, the remaining presented with grade 2 OA based on the Kellgren–Lawrence (KL) score [14]. In addition, three male healthy controls (36.00 ± 4.24 years) without clinical evidence of knee OA were examined. Exclusion criteria for all subjects were any contraindications for MRI. Written informed consent was obtained from all subjects after the nature of the examinations had been fully explained. All exams were performed in accordance with the rules from the local Human Research Committee.

MRI imaging of the same knee joint of each subject was performed on a 3.0- and 7.0-Tesla system (Signa, GE Medical Systems, Waukesha, WI, USA) within 2 weeks. At both field strengths, a dedicated quadrature knee coil was employed (Clinical MR Solutions, Brookfield, WI, USA at 3.0 T and Nova Medical, Wilmington, MA, USA at 7.0 T) Transmit gain and shimming were manually adjusted.

The morphology of cartilage, menisci, ligaments and other knee structures was assessed on three sagittal cartilage dedicated pulse sequences: (1) a fat-saturated intermediate-weighted fast spin-echo sequence (IM-w FSE), (2) a T1-weighted 3D high-spatial-resolution volumetric fat-suppressed spoiled gradient-echo (SPGR) sequence, and (3) a 3D FIESTA-C (fast imaging employing steady-state acquisition with constructive interference in steady state, CISS) sequence. The CISS method [15‐18] is preferably applied at higher fields in order to reduce susceptibility artifacts due to magnetic field inhomogeneities.

The parameters used for the sequences are shown in Table 1 and are results of a preceding optimization study: the optimal parameters at 3.0 T to achieve the best image quality within a clinically acceptable time frame has been described [3]. Great care was taken to use a similar setting at 7.0 T and keep the parameters constant to allow comparison between the two field strengths. However, since T1 relaxation time increases and T2 relaxation time decreases at higher field strength [10, 11, 17], it was necessary to repeat the optimization procedure at 7.0 T. Using these relaxation times, Krug et al. [17] have previously optimized pulse sequences for cartilage, based on numerical simulations of the Bloch equations. The flip angle and TE of FIESTA and SPGR were optimized to yield maximum contrast between cartilage, joint fluid, and menisci. Since conventional fat suppression methods would disturb the steady state of FIESTA, we adapted a different method by changing TE. A prolonged TE due to full Fourier acquisition resulted in less bone marrow signal due to more off-resonance effects at bone and bone marrow transitions and thus better contrast [17]. Most of these optimizations can be made by simulations and directly translated to the scanner. However, at 7.0 T, further restrictions have to be considered and are mainly related to the increased power deposition, which affects the safety of the subject. Thus, additional optimizations had to be made for the FSE sequence experimentally taking in vivo conditions (mainly imaging time and the specific absorption rate SAR) into account. This was performed in the three healthy subjects without clinical findings of knee pathology and was conducted similar to the procedure described in [7, 19]). Optimizing the IM-w sequence proved to be challenging: FSE sequences are characterized by a series of rapidly applied 180° rephasing pulses which at higher field strength require the application of higher radiofrequency energy. Since the SAR has to be kept within the Food and Drug Administration (FDA) limits of 3.2 W/kg, the scanner automatically interrupted the data acquisition during this sequence and paused frequently for several minutes. In order to cover as much of the knee as possible with a minimal number of interruptions, TR was extended, and echo train length (ETL) was reduced. However, even with this compromise only 30 slices could be obtained.

All images were reviewed on PACS workstations (Agfa, Ridgefield Park, NJ, USA). Two radiologists with 19 and 5 years experience in musculoskeletal imaging assessed pathologic findings of the knee joint as well as diagnostic confidence within each sequence and image quality independently. They were blinded with regard to the field strength. In case of divergent results, an agreement was achieved with a consensus reading.

SNR and CNR were measured for each sequence for both field strengths by the two radiologists (TML and RS) in consensus. SNR for cartilage was measured as the ratio of the mean signal intensity and the SD of the background noise, measured in a region of almost no signal. For consistency, ROIs were placed at identical positions on each image. Values were obtained for cartilage at the lateral patella, at the lateral femoral condyle, and at the lateral tibia. The average of these three results was calculated. The CNR between cartilage and bone was defined as follows: (mean signal intensity_cartilage − mean signal intensity_bone)/SD_{background noise}. The ROI for bone was placed in the bone marrow of the femoral condyles. For the calculation of the signal intensity of cartilage, the same anatomical regions were used as described above. Again, the average of these results was calculated.

Pathologic findings at the knee joint were analyzed using a modified WORMS score [13]. The 15 previously described compartments were merged to a total of seven compartments: the medial and lateral tibia, trochlea, medial and lateral femur, as well as medial and lateral patella. Flattening or depression of the articular surfaces was not evaluated. Cartilage signal and morphology, marginal osteophytes, medial and lateral meniscal integrity, synovitis/effusion, intraarticular loose bodies and periarticular cysts/bursitis, anterior and posterior cruciate ligament integrity, as well as medial and lateral collateral ligament integrity were scored as suggested by the original WORMS score. To characterize the extent of the regional involvement of bone marrow edema and subchondral cysts, the original partitions of the femoral condyles and tibial plateaus in anterior, central, and posterior regions were used. Any other pathologies not included in a WORMS category score were also documented.

To better quantify small cartilage lesions, the volume of cartilage inhomogeneity (WORMS score = 1) and the volume of cartilage defects (WORMS score ≥ 2) were calculated on the sagittal IM-w FSE images as previously described [3]. Briefly, in lesions with a WORMS score of 1, the size of the signal inhomogeneity was quantified by multiplying its largest diameter with the number of slices that visualized it (slice thickness in mm including section gap). The volume of focal cartilage defects (WORMS score ≥ 2) was approximated by the following equation:

The depth of the lesion was determined by dividing the surrounding normal cartilage into three layers with equal thickness. If the maximum lesion depth did not exceed the thickness of the superficial layer it was assigned as 1/3. It became 2/3 or 3/3 if the maximal lesion depth comprised also the medium or the lowest layer, respectively.

The shape factor was 1 if more than 50% of the lesion had the largest depth (assessed in the section with the largest diameter), otherwise a shape factor of 0.5 was assigned. Extent of bone marrow edema pattern was assessed on the IM-w FSE images by multiplying the two largest diameters with the number of sections, where it was visualized (slice thickness including gap).

The precision of the two radiologists in the quantification of the extents of cartilage signal inhomogeneities, cartilage defects, and bone marrow edema at both field strengths was expressed by calculating the percentage coefficient of variation (CV%) as previously outlined by Gluer et al [20].

Both radiologists were asked to rate the confidence level for each compartment regarding the presence or absence of cartilage abnormalities and BMEP for each sequence at both field strengths. The following grading system was used: 1 “definitely no lesion,” 2 “most likely no lesion,” 3 “questionable lesion,” 4 “most likely a lesion,” 5 “definitely a lesion.” In addition, they used this grading system to rate the confidence level for meniscal lesions and integrity of the cruciate ligaments.

Based on the above definitions, a “diagnostic confidence score” for cartilage abnormalities and BMEP was calculated. The objective for this score was to provide a global measure of the diagnostic confidence for each sequence at each field strength in order to permit comparisons. It considers the certainty of the diagnosis by assigning different weight factors to the confidence levels. The score was calculated for each compartment including all studied subjects:

The more subjects with a high diagnostic certainty are found (“definitely no lesion” and “definitely a lesion”) the higher the score for the specific sequence in the examined compartment will be. This value is reduced by one count for each subject with a dubious diagnostic certainty (“questionable lesion”) in this compartment. It is also reduced but to a smaller amount (by 0.5 count) when the diagnosis was made with a high probability but not without doubt (subjects with “most likely a lesion” and “most likely no lesion”). In summary, the higher the score, the higher the global diagnostic confidence of the sequence.

Overall image quality of the cartilage-dedicated sequences was graded according to a four-level scale. For this grading, sharpness of edges, amount of blurring and other artifacts, contrast between fluid and cartilage, contrast between fluid and soft tissue, and amount of noise were analyzed subjectively by each of the two radiologists independently. In case of divergent results, a consensus was found in a second session. Images graded as 1 had poor quality (limited diagnostic information, severe artifacts), while images graded as 4 had excellent image quality (e.g., good diagnostic quality, minor or no artifacts).

Data were initially assessed for normality with the Kolmogorov–Smirnov test. T tests for paired samples were used to compare the SNR and CNR calculations as well as the overall image quality scores for the individual sequences and field strengths. T tests for paired samples were also used to compare the size of cartilage abnormalities and BMEP between the two field strengths.

Differences in the frequency of observed cartilage abnormalities and BMEP between the two field strengths were evaluated with the Chi² test. Differences in the distribution of the WORMS (sub-)scores and the sequence confidence ratings between 3.0 and 7.0 T were assessed with the Wilcoxon signed-rank test. Results were adjusted with Bonferroni correction for multiple comparisons.