Introduction
Magnetic resonance (MR) imaging is generally regarded as the best available noninvasive method for evaluating injury and repair of the articular cartilage of the knee [
1‐
3]. Currently, most evaluation of knee cartilage is done with two-dimensional (2D) acquisition techniques, such as turbo spin-echo (TSE) sequences, as they provide excellent tissue contrast and high in-plane spatial resolution [
4]. However, 2D sequences have relatively thick slices (≥2 mm) and small gaps (0.2 mm) between slices, which can obscure pathology because of partial volume averaging [
4,
5].
Three-dimensional (3D) imaging has the great potential of acquiring volumetric data sets with isotropic resolution, providing thin (0.5–0.6 mm) contiguous slices through the knee joint, thereby reducing partial volume artefacts. Multiplanar reformats along any user-defined imaging plane can be generated from the source data without loss of resolution [
4‐
6].
Techniques for morphological 3D MR imaging of cartilage have changed rapidly in the past 2 decades. Gradient-echo (GRE) sequences were the first 3D sequences used for cartilage imaging [
4,
7,
8]. They are classically divided into dark fluid sequences (e.g. spoiled gradient echo or FLASH) and bright fluid sequences (e.g. double echo steady-state or DESS) based upon the signal intensity of synovial fluid [
4,
8]. These traditional GRE methods have been shown to be highly accurate for cartilage lesions in various studies and are still considered the standard of reference for high-resolution 3D isotropic imaging of cartilage [
5‐
8]. A water-excited DESS sequence has been used in various knee osteoarthritis trials and is currently being used in the Osteoarthritis Initiative to assess the articular cartilage of the knee joint [
9]. Many different methods of steady-state free-precession-based imaging (SSFP), a new variant of the GRE method, are also available for imaging cartilage, and all have higher cartilage signal compared with conventional GRE methods [
10]. However, major disadvantages of 3D GRE imaging include their sensitivity to susceptibility artefacts and their suboptimal tissue contrast [
4‐
8].
The increased availability of high-field 3-T imaging systems combined with improved coil technology facilitate the use of novel isotropic 3D sequences in clinical practice. Recently, 3D TSE sequences (e.g. SPACE) were developed for 3-T scanners. The important advantage of the 3D TSE acquisitions is their capability of mimicking the contrast properties of conventional 2D TSE intermediate-weighted acquisitions, allowing for better tissue contrast and higher conspicuity of cartilage lesions [
5,
11].
In the knee, the superficial and middle layers of articular cartilage have T2 values of ~50 ms, while the deep cartilage layer has much shorter T2 values, in the range of 1 to 2 ms or less [
12]. Therefore, with a practical minimum echo time (TE) of about 10 ms for a spin-echo (SE) sequence and of about 2 ms for a GRE sequence, conventional MR imaging lacks the ability to display signal near the osteochondral junction. Through the use of ultrashort TE (UTE) sequences, now becoming available on clinical 3-T scanners, the minimum TE can be considerably reduced and the signal detected before it has totally decayed. As a result, signal can now be acquired from these previously ‘invisible’ tissues at the bone-cartilage interface, allowing for direct visual assessment [
12,
13].
Controversy remains as to which is the single best 3D sequence for clinical cartilage imaging of the knee at 3 T [
14,
15]. Also, clinical UTE imaging of the knee has been performed in healthy volunteers [
16,
17], but studies investigating the role of morphological UTE imaging in patients with cartilage lesions are lacking. Thus, the main purpose of the present study was to compare standard and novel 3D cartilage MR sequences at 3 T. It was hypothesised that 3D TSE acquisitions compare favourably to the standard of reference GRE sequences for clinical cartilage imaging. In addition, we sought to explore how the normal UTE signal patterns change in normal and abnormal cartilage seen at conventional MR.
Discussion
The most important findings of the present study are that, first, morphological cartilage imaging of the knee can be reliably performed using 3D TSE MRI, showing good image quality and high accuracy when compared to the standard of reference sequences. Second, although the newly available UTE sequence can distinguish between superficial and deep cartilage layers, it cannot be used as a single sequence to assess the articular cartilage at present.
The most common 3D sequences used in clinical practice to evaluate the articular cartilage of the knee are GRE-based methods, e.g. FLASH or DESS sequences. These sequences are available on most MR imaging systems and have been successfully used to evaluate cartilage. In addition, these sequences can be used to perform cartilage volume measurements in osteoarthritis research studies [
4‐
8]. Although dark-fluid sequences have lower contrast between cartilage and fluid than bright-fluid sequences [
4,
7], there was no significant difference in the performance of both sequences to evaluate cartilage in our study.
The development of balanced SSFP imaging, a variant of the GRE technique, was promising to improve 3D MRI of the musculoskeletal system. Several studies have shown excellent synovial fluid-cartilage contrast with these sequences. They are also useful in the imaging of other internal structures of the knee, such as the ligaments and menisci, a capability that makes it an attractive option for use in clinical practice [
10,
20]. Although our study results confirm good contrast properties between cartilage and synovial fluid, the TruFISP sequence yielded more severe susceptibility artefacts and poorer overall image quality than did the conventional GRE methods.
Despite these developments, GRE imaging displays image contrast characteristics different from those of the TSE pulse sequences commonly used in assessment of joints. Recently, 3D TSE sequences (e.g. SPACE) have been used to assess the knee joint with high spin-echo contrast resolution and isotropic spatial resolution, improving conspicuity of cartilage lesions [
5,
21]. Therefore, 3D TSE retains the advantages of 2D TSE while also addressing its limitations. The 3D TSE sequence typically uses variable flip angle modulation to constrain T2 decay over an extended echo train. This allows for intermediate-weighted images of the knee with bright fluid to be acquired with minimal blurring [
11]. Our study results are in concordance with prior studies demonstrating superior performance of 3D SPACE in assessing cartilage lesions [
14,
21]. A disadvantage of the SPACE sequence is its long acquisition time (10 min 4 s). This limitation, however, is counterbalanced by the fact that SPACE, due to its favourable tissue contrast, allows for comprehensive knee joint assessment. A single acquisition of 3D SPACE may therefore replace multiple 2D FSE acquisitions, which increases the time efficiency when used in a clinical knee MR protocol [
21].
The osteochondral junction has been implicated in the pathogenesis of OA [
22] and cartilage repair [
23]. Thus, direct visualisation of these tissue components is clinically relevant. To date, MR imaging has, however, not been capable of assessing the deep radial and calcified layers of cartilage. These deep layers of cartilage have short T2 characteristics (<1 ms), and conventional pulse sequences are unable to acquire data in this range. UTE sequences are designed to target tissue components with very short T2 and allow signal to be detected in the deep layer of cartilage [
24]. In a recent study, Bae et al. [
18] compared UTE MRI and histology of experimental preparations and determined that the presence of the deepest layer of uncalcified cartilage and the calcified cartilage, but not the subschondral bone, results in this linear signal intensity in UTE MRI.
The term UTE imaging has generally been applied to techniques using shorter RF excitation pulses and faster readout methods than conventional methods to produce images with very short TEs, typically in the range of 8–250 μs [
24]. A number of UTE techniques focussing on the method of image acquisition have been developed [
25,
26]. These include both 2D and 3D sequences. They are typically combined with some form of long T2 component reduction in order to isolate the signal from the short T2 components and thus demonstrate change in disease [
24].
Although a characteristic pattern of high linear signal near the osteochondral junction and low signal in the superficial cartilage layer could be observed in all subjects, sensitivity of our 3D UTE sequence in the detection of cartilage lesions was significantly lower compared to conventional sequences. Poor performance was probably related to blurring artefacts in the radial UTE acquisition together with field heterogeneity or potential gradient delays. Even at retrospective analysis, focal or diffuse abnormalities in the linear signal were found in only 10 of the 20 cartilage lesions. Most of these had high-grade cartilage lesions diagnosed on conventional MR sequences. In ten cartilage lesions, no abnormality in the UTE morphology of the osteochondral junction could be found. Of note, seven of these ten were apparent high-grade cartilage lesions on conventional MR sequences. This finding may suggest that these lesions were rather low-grade defects and overestimated on conventional MR images. Larger studies with arthroscopic correlation are needed to explore how the normal UTE signal pattern changes with disease and injury.
As with conventional MR techniques, UTE can also be used for quantitative mapping of tissues. A recent in vitro study [
25] has evaluated UTE T2* and T1rho values of the patellar osteochondral junction in cadaveric samples and found that these measurements were useful for non-invasive assessment of the deep calcified layer of cartilage, including understanding the involvement of this tissue component in osteoarthritis.
There were several limitations to this study. First, and most important, the small number of patients limited the statistical analysis. However, in this era of limited resources and cost savings in health care, a larger study including more patients would not be possible in our busy clinical practice. Second, there was no arthroscopic correlation available. Great care was taken to obtain the best possible standard of reference using all available MR sequences and consensus readings. High accuracy of these sequences for detecting cartilage lesions is well documented in the literature [
4‐
6]. Furthermore, arthroscopy also has limitations and should be considered an imperfect reference standard for grading of cartilage defects [
27]. Third, all readers were employed in centres equipped with the sequences tested in our study; this may have introduced an interpretation bias, since the readers were very probably able to identify the sequence from the overall appearance of the blinded MR images. Fourth, in our study, we evaluated only subjective image quality, without quantitative analysis of the signal-to-noise (SNR) or contrast-to-noise (CNR) ratio. However, quantitative analysis of images acquired with parallel imaging requires that the “difference method” be used [
28]. This would have been at the cost of doubling the total imaging time leading to motion artefacts again disturbing the methodological improvement.