Introduction
Although 1.5 T MRI of the knee is standard in clinical practice and performs well in assessing internal joint derangement, limitations have to be considered. In particular, assessing abnormalities of the hyaline articular cartilage and lateral meniscus may be challenging [
1‐
3]. The current literature suggests that imaging at 3.0 T MRI offers clinical benefits compared to 1.5 T [
4,
5]. The stronger field strength increases signal-to-noise ratio (SNR), potentially allowing thinner sections and higher in plane spatial resolutions resulting in better visualization of anatomical and pathological structures [
6].
A number of in vitro studies have shown improved visualization of cartilage and ligament pathology at the knee and ankle using 3.0 T in comparison with 1.5 T MRI [
7,
8]. Studies have also demonstrated 3.0 T MRI to have high sensitivity and specificity in diagnosing knee meniscus, cartilage, and ligamentous pathology when compared with subsequent arthroscopy [
9‐
12]. A limitation of these studies, however, was that direct comparison with imaging at 1.5 T in the same subjects was not obtained and that previously published data from the literature obtained at 1.5 T were used as a reference in some of these studies. To the best of our knowledge, no studies have been performed directly comparing MR imaging at the knee at 1.5 and 3.0 T obtained in the same patients and correlating these findings with arthroscopy.
The goals of our study therefore were (a) to compare standard clinical MRI studies obtained at 1.5 and 3.0 T in the same patients concerning visualization of cartilage, menisci, and ligaments and (b) to assess these 1.5 and 3.0 T studies concerning abnormalities of cartilage using arthroscopy as a standard of reference.
Materials and methods
Patients
This Health Insurance Portability and Accountability Act-compliant (HIPAA) study had institutional review board approval. A retrospective search examining patient records from July 27, 2004 through July 25, 2008 was conducted. Twenty-six patients (eight males and 18 females) were identified who had both a comparative 3.0 and 1.5 T MRI scan of the knee. Comparative studies were completed within an average time interval of 101.66 ± 79.09 days. Nineteen of the 26 patients also had a knee arthroscopy of the same knee that were performed at our department of orthopedic surgery within an average time interval of 56 ± 40.9 days to the most recent MR study. The average age of the patients at the time of the first examination was 38.5 ± 11.3 years. Clinical indications for MRI of the knee included knee pain in 17 patients, pain and/or locking with suspected meniscal injury in five patients, and instability with suspected ligament tears in four patients. The main reasons for repeated imaging studies were (a) to obtain a 3 T study to better visualize cartilage lesions and (b) to provide a follow-up MRI after arthroscopy if no cartilage repair was performed.
MR imaging
MR studies were obtained using 1.5 and 3.0 T whole body scanners (Signa, GE Healthcare, Waukesha, WI, USA). Standard transmit/receive knee coils were used; most of the studies were obtained with quadrature knee coils (GE, Healthcare and MR Solutions, Brookfield, WI, USA; 18 at 3 T and 16 at 1.5 T) and more recently eight-channel phased-array knee coils (Invivo, Orlando, FL, USA; eight at 3 T and ten studies at 1.5 T) were used. Given that approximately the same number of the studies at each field strength was obtained with similar coils, no bias is expected due to coil design. The MR protocol consisted of a coronal T1-weighted and fat-saturated (fs) intermediate-weighted (iw) fast spin-echo (FSE) sequences, sagittal iw and fs iw FSE sequences, and axial fs iw FSE sequences. Detailed sequence parameters are listed in Table
1; sequence parameters had been adjusted to account for differences in relaxation at 3.0 T. Also, we reduced the number of excitations (NEX) from 3 to 2 and increased the echo train length at 3.0 T to maintain acquisition time. With the higher SNR at 3.0 T, it was also possible to increase the matrix size from 224 × 224 to 320 × 256 pixels.
Table 1
Standard protocol of the knee for 1.5 and 3.0 T MRI
1.5 | 1. sag fs iw FSE | 3,200 | 46 | 13 | 3 | 8 | 3 | 224 × 224 |
1.5 | 2. sag iw FSE | 2,000 | 20 | 13 | 3 | 4 | 3 | 224 × 224 |
1.5 | 3. cor fs iw FSE | 3,200 | 46 | 13 | 4 | 8 | 3 | 224 × 224 |
1.5 | 4. cor T1 FSE | 600 | 15 | 13 | 4 | 3 | 3 | 224 × 224 |
1.5 | 5. axial fs iw FSE | 3,700 | 46 | 13 | 4 | 8 | 3 | 224 × 224 |
3.0 | 1. sag fs iw FSE | 4,300 | 51 | 13 | 3 | 9 | 2 | 320 × 256 |
3.0 | 2. sag iw FSE | 3,300 | 13.7 | 13 | 3 | 8 | 2 | 320 × 256 |
3.0 | 3. cor fs iw FSE | 4,300 | 51 | 13 | 4 | 9 | 2 | 320 × 256 |
3.0 | 4. cor T1 FSE | 675 | 15.4 | 13 | 4 | 5 | 2 | 320 × 256 |
3.0 | 5. axial fs iw FSE | 4,300 | 51 | 13 | 4 | 9 | 2 | 320 × 256 |
MR image analysis
Three board-certified radiologists and a radiology resident with expertise in musculoskeletal MRI ranging from 3 to 23 years evaluated all images (CS, JZ, LS, TML) on PACS workstations (Agfa, Ridgefield Park, NJ, USA). At the time of analysis, the radiologists were blinded to the clinical history, previous reports, field strength, and sequence parameters. During the reading sessions, ambient light was kept at a minimum, and no time constraints were used. No more than 25 studies were reviewed at one time. Time between reading sessions was at least 96 h to prevent learning bias. Studies obtained at 1.5 and 3.0 T were presented to the radiologist in random order.
The radiologists reviewed all sequences of one MR study and were asked to grade image quality, using the following criteria: edge sharpness, amount of blurring, artifacts, contrast between fluid and cartilage, contrast between fluid and soft tissue, delineation of small ligamentous structures, and amount of noise. A four-level scale was used, in which four indicated optimal image quality and one substantially reduced image quality limiting diagnostic evaluation.
In addition, presence and absence of pathology were graded focusing on cartilage, meniscal, ligamentous, tendon, and bone marrow abnormalities. Cartilage lesions were graded according to Recht et al. based on a modified Noyes classification [
13]. Four grades were differentiated: grade 1 was defined as signal heterogeneity and/or swelling, grade 2 as a less than 50% defect, grade 3 as a more than 50% defect, and grade 4 as a full thickness defect. If a lesion was identified, radiologists were asked to assign a confidence score indicating whether they felt the lesion was a definite [
3], a probable [
2], or a questionable lesion [
1]. For meniscus pathology, radiologists were asked to identify lesions in the anterior and posterior horn as well as the body of both the medial and lateral meniscus. Abnormalities identified were graded as intra-substance signal abnormality, non-displaced vertical or horizontal tear, complex tear without deformity, and complex tears with deformity of the meniscus. A confidence score was also assigned, analogous to this used for the cartilage lesions.
The ligaments and tendons assessed included anterior cruciate ligament, posterior cruciate ligament, medial collateral ligament, lateral collateral ligament, popliteus, and patellar tendon. Abnormalities were assessed as grade 1 (mild), grade 2 (signal abnormality of the tendon or ligament suggesting partial tear), and grade 3 (complete tear) strain or sprain. A diagnostic confidence score was assigned as outlined previously for cartilage and menisci.
SNR measurements
Quantitative analysis was also performed calculating effective signal-to-noise-ratios (SNRE) for the sagittal fs iw FSE sequences with \( {\text{SNR}}_{\text{E}} =\frac{{\frac{{{\text{SI}}_{\text{Tissue}} }}{{{\text{SD}}_{\text{Background}} }}}}{{\sqrt {\text{scantime}} }} \), where SITissue was the signal intensity of cartilage, and SDBackground was the standard deviation of the background. Regions of interest were placed in the background and the trochlear cartilage.
Standard of reference
The standard of reference for the MR abnormalities was based on the arthroscopic findings in 19 subjects, which were reviewed with all available MR imaging and clinical findings. The arthroscopies were performed by one of three experienced orthopedic surgeons at UCSF Medical Center who specialize in sports medicine. Arthroscopies were performed after the orthopedic surgeons had reviewed the MRI studies. All articular surfaces were examined, and their lesions were graded using the Noyes scale, analogous to the grading system used for the cartilage lesions (grade 1 = softening, grade 2 = less than 50% defect, grade 3 = more than 50% defect, grade 4 = full thickness cartilage defect). The location and the type of meniscal defects were noted analogous to the classification used for the MRI. Ligament and tendon lesions were examined and scored as for the MRI studies.
Statistical analysis
A signed rank test, with a significance threshold of p < 0.05, was used to compare the differences in image quality measurements of all evaluated anatomical structures and confidence scores between 3.0 and 1.5 T. McNemar's tests were used to assess the differences in direct comparison of image quality. SNR differences of cartilage signal between 3.0 and 1.5 T were assessed using paired Student's t tests with a significance threshold of p < 0.05. Sensitivities, specificities, and accuracies with 95% confidence intervals were calculated to estimate the diagnostic performance for cartilage and meniscal lesions using the established standard of reference. All of the statistical computations were processed using JMP 7 (SAS Institute, Cary, NC, USA).
Discussion
The results of this study show modest, yet significantly increased accuracies and sensitivities in diagnosing cartilage abnormalities at 3.0 T compared to 1.5 T at the knee. Specificity for cartilage lesions, however, did not improve for 3.0 T. Also, there was a 10% increase in correct grading of cartilage lesions at 3.0 T compared to 1.5 T. Overall image quality and diagnostic confidence were graded consistently higher at 3.0 T than at 1.5 T for each individual radiologist.
Though a number of in vitro studies were performed comparing visualization of cartilage, ligaments, and menisci at 1.5 and 3.0 T [
7,
8,
14‐
16], there is a paucity of clinical studies in particular those comparing 1.5 and 3.0 T studies in the same subjects. Our results are consistent with previous in vitro studies demonstrating an increase in sensitivity and diagnostic performance observed at 3.0 T for cartilage lesion detection. Masi et al. [
8] were among the first to demonstrate improved diagnostic performance of 3.0 T versus 1.5 T MRI in cartilage lesions. In a porcine model, artificial cartilage lesions were created, and using intermediate-weighted fast spin-echo and spoiled gradient-echo sequences, these investigators were able to show improved accuracy of cartilage lesion detection. Similar to our study, they were also able to demonstrate improved performance in grading cartilage lesions. Link et al. [
16] compared different sequences in assessing cartilage abnormalities in vitro and concluded that while high-resolution intermediate-weighted fast spin-echo sequences showed substantial benefits in detecting cartilage lesions, standard clinical type fast spin-echo sequences did not demonstrate a major improvement in diagnostic performance. This may in part explain why only relatively minor improvement in sensitivity and accuracy in cartilage lesions were found in our study. Those investigators concluded that only thin section high-resolution sequences would exploit the full potential of 3.0 T imaging concerning cartilage lesion evaluation.
Kijowski et al. [
9,
17] recently reported results from a study analyzing 200 consecutive patients undergoing MRI of the knee using identical pulse sequences on either a 1.5 T (
n = 100) or 3.0 T (
n = 100) scanner who had subsequent arthroscopic knee surgery. These investigators found that 3.0 T MRI had higher specificity and accuracy but not higher sensitivity for detecting cartilage lesions at 3 T than at 1.5 T. Similar to our study, they also found MRI to be more accurate at grading articular surfaces at 3.0 T. These investigators, however, did not study the same patients at 1.5 and 3.0 T but two different patient populations, though their patient numbers in each group were high (
n = 100).
There has been discussion about the best-suited sequence for cartilage imaging; gradient-echo sequences (FLASH, fast low-angle shot, and SPGR, spoiled gradient echo) and intermediate- and T2-weighted FSE sequences with and without fat saturation have been used previously [
3,
18,
19]. Currently, however, intermediate-weighted FSE sequences are the preferred sequences in a clinical setting and experience with these sequences encompasses multiple studies; sensitivities of 63–100% have been described using arthroscopy as a standard of reference [
20‐
22], which is consistent with the results found in our study. Concern may be raised by the fact that results at 3.0 T in diagnosing cartilage lesions are still relatively low. This, however, may be explained by the fact that imaging studies were obtained over a period of several years, and most of the imaging studies were performed with a quadrature knee coil. Also, we used standard clinical protocols. High-resolution imaging sequences obtained with thinner slice thickness and multi-channel phased-array coils would have probably improved the results. However, the goal of our study was to analyze the effect of 3 T imaging on standard knee imaging, not using dedicated research protocols. The relatively low results for correct grading of cartilage lesions found in our study on the other hand are expected, and similar results have been previously reported at 3.0 T [
8,
16].
It should be noted that using arthroscopy as a standard of reference has raised concerns [
23]. Differentiating between cartilage lesions that are less than or greater than 50% may be challenging with arthroscopy as the bone surface is not exposed. The same applies for meniscal lesions, where radiologists may diagnose lesions in the red zone of the meniscus, which could not be verified because of the difficulty examining this area with arthroscopy; surgical blind spots, including the posterior meniscocapsular junction [
24], and even incomplete arthroscopy may have skewed our false-positive numbers [
1,
25].
In addition to optimized sequences, the benefits of 3.0 T imaging also heavily depend on suitable coils, which was a limitation in the early phase of 3.0 T MRI. In an in vivo study, Lutterbey et al. [
26] compared the diagnostic performance of knee MRI performed with the body coil at 3.0 T and a dedicated knee coil at 1.5 T. These investigators found a significant decrease in image quality at 3.0 T with a visible and measurable signal loss, but stated that nearly all clinically important information could be obtained.
While most of the previous studies comparing 1.5 and 3.0 T imaging were performed at the knee, two previous studies focused on imaging of the ankle, comparing visualization of cartilage and ligament lesions at 1.5 and 3.0 T in human cadaver ankles [
7,
14,
26]. Bauer et al. [
7] found that compared with 1.5 T, 3.0 T MRI of the ankle joint improved diagnostic performance in assessing cartilage significantly, and there was a higher sensitivity for assessing ligamentous pathology. This was one of the first studies indicating that 3.0 T MRI may also improve visualization of ligamentous abnormalities. While in our study diagnostic confidence in assessing ligamentous and tendon lesions was also higher at 3.0 T than at 1.5 T, the number of lesions was not high enough to provide statistically meaningful data concerning diagnostic performance.
To the best of our knowledge, comparative studies assessing meniscal lesions at 1.5 and 3.0 T have not yet been performed. Schoth et al. compared visualization of anatomic structures at the knee and found superior subjective visualization at 3.0 T versus 1.5 T for the meniscus and ligaments; however, no dedicated evaluation of meniscal lesions with arthroscopic correlation was available [
27]. Magee et al. and Ramnath et al. [
11,
12] used 3 T MRI to study meniscal abnormalities and found promising results. Unfortunately, a direct comparison with imaging at 1.5 T was not available, and results were compared to those previously published in the literature at 1.5 T.
The greatest limitation of our study was our limited patient population, of which only 19 patients had arthroscopy as a standard of reference. An additional limitation is the use of standard protocols, which did not include high-resolution sequences. However, the goal of this study was to assess the performance of 3.0 T versus 1.5 T MRI in a clinical setting without using dedicated research protocols. Also, we acknowledge the differences in imaging performance between the readers; however, we aimed at having readers with different levels of training read the images and assess their performance in relation to field strength. The higher spatial resolution of the imaging sequences at 3.0 T compared to 1.5 T could have also contributed to the results; however, acquisition time was kept in the same range by reducing NEX and increasing ETL.
In conclusion, our study shows that 3.0 T MRI was superior for detecting and grading cartilage lesions compared to 1.5 T. Though a higher diagnostic confidence was found at 3.0 T, the false-positive rate was not decreased. Overall image quality at 3.0 T was rated superior to 1.5 T consistently by all four radiologists.