Osteoarthritis is the most common joint disorder, characterized by an imbalance between synthesis and degradation of the articular cartilage with destruction of the joint [1]. For patients with mild to severe diseases, increasing numbers of surgical and non-surgical treatment modalities, such as analgetic treatment, application of hyaluronic acids and growth factors, cartilage transplants, osteochondral transfers, microfracturing, corrective osteotomies or partial and total knee replacements, have gained popularity [2–5]. In regard to this wide range of therapeutic options, it has become increasingly difficult to recommend adequate treatment. In clinical practice, a detailed assessment of disease severity includes an exact grading of the cartilage. Therefore, a reliable non-invasive visualization of cartilage disorders becomes important and may be an additional support in the decision, which therapeutic options should be suggested. Several studies comparing the value of cartilage diagnostics on MRI to intra-operative findings present very different results. In a large part of the studies, diagnostic values of MRI were assessed by collapsing several grades of cartilage disorders into a disease positive and a disease negative status [6–15]. This simplification mostly does not correspond to surgeons' demands of an exact staging of cartilage disorders. Additionally, only a few comparative clinical studies have focused on degenerative cartilage diseases [16–22]. These studies, which compare MRI grading and arthroscopic grading of the cartilage in patients with knee osteoarthritis, used different statistical methods and demonstrated various results [16–22]. Blackburn et al. showed a "moderate" correlation of MRI grading and arthroscopic grading (Pearson correlation coefficient r = 0.4) [17]. In contrast, two other studies demonstrated a "highly significant" correlation using the Spearman rank correlation test (P > 0.0003 and P > 0.0001, respectively) [18, 20]. When intra- and inter-observer agreements were assessed, the kappa values ranged between "slight" and "very good" agreement [17, 19, 21]. The diagnostic values reported in clinical studies and cadaver studies also present a wide range, with sensitivities ranging from 31% to 100% [16, 20, 22]. The aim of this study was to determine whether MRI is a reliable method for a detailed assessment of the articular cartilage in patients with advanced and clinically relevant osteoarthritis or not. For this purpose, MR images and intra-operative findings in a representative sample of patients were evaluated.

According to the inclusion criteria, advanced cartilage disorders were predominant in our study. Of 216 joint surfaces, 85 (39%) were assessed as grade 0, 10 (5%) as grade 1, 31 (14%) as grade 2, 50 (23%) as grade 3 and 40 (19%) as grade 4 lesions during arthroscopy.

In our study, the diagnostic values for MRI assessment of cartilage lesions were relatively low at all grades of disease. The assessment of inter-observer agreement between readers 1 and 2 revealed mostly moderate and good results (weighted kappa = 0.51-0.75). For the intra-observer agreement, when comparing MRI grading to arthroscopic grading, slightly poorer results with fair, moderate and good values (weighted kappa = 0.32-0.65) were demonstrated. Previous clinical studies focused on patients with osteoarthritis showed different values for the agreement of MRI grading and arthroscopic grading of cartilage damage. Drapé et al. demonstrated very good intra- and inter-observer agreements (weighted kappa = 0.91 and 0.64, respectively). Based on this data, MRI was proposed as an outcome measure of cartilage lesions in clinical trials treating osteoarthritis [19]. In contrast to this study, McNicholas et al. noted mainly "slight" or "fair" intra- and inter-observer agreements for the evaluation of cartilage damage in patients with knee osteoarthritis. In regard to this data, reservations about the use of MRI in the assessment of disease severity were stated [21]. Similarly to a study of Blackburn et al. [17], our results, lie in the mid range, demonstrating moderate to good kappa values (Table 1). There are several possibilities to explain the differences between these studies and our study. Regarding patient selection, the inclusion criteria for clinically relevant osteoarthritis were clearly defined in the study of Drapé et al. and Blackburn et al. [17, 19], whereas no inclusion criteria were set in the study of McNicholas et al. [21]. Furthermore, it has to be mentioned that MRI techniques used in the studies were quite different. Drapé et al. used a 0.2-T musculoskeletal dedicated MR unit (Artoscan; Esaote Biomedica) with two successive 3D gradient-echo sequences and 1.4 mm slice thickness [19], whereas McNicholas et al. used a FISP 3 D gradient echo sequence on a 1.0 T scanner (Siemens) and a dedicated surface coil without further description of slice thickness [21]. Similarly to our study, Blackburn et al. used proton density-, T1- and T2-weighted spin-echo sequences with 4 mm slice thickness on a 1.5 T scanner and a standard extremity coil [17]. The comparable MRI techniques used in the study of Blackburn et al. and our study as well as comparable inclusion criteria are possibly the cause for similar results. A better diagnostic performance reported in the study of Drapé et al. could be explained by a relatively thin slice thickness. Furthermore, the performance reported in this study could be explained by the use of two 3D gradient-echo sequences, which were thought to be advantageous for the detection of cartilage lesions [6, 11, 15, 31, 32]. Even so, the MR sequence best suited for the detection of chondral abnormalities is still under debate [33, 34]. The fact that MR images were available to the surgeon at arthroscopy could further influence our results. At arthroscopy, the MRI visualization of osteoarthritis, which is often combined with additional findings, such as osteophytes, bone marrow edema, sclerosis, cysts, etc. could affect the intra-operative grading of the cartilage, even if the MRI grading of the hyaline cartilage was not present at the time of arthroscopy. Interestingly, intra-observer agreements were poorest in the patellofemoral compartment. This could be explained by the so-called "magic angle effect", which influences the visualization of the cartilage at certain orientations of collagen fibers corresponding to the magic angle of 55° [35]. This phenomenon, often occurring at the articular poles of the patella, can influence the accurate interpretation of these areas [36]. A poor inter-observer agreement at arthroscopy could also impair the validity of our study, where several orthopedic surgeons were involved. Studies about inter-observer agreement at arthroscopy demonstrate fair and moderate inter-observer agreement, particularly for the patellofemoral cartilage [37, 38], but also sufficient reproducibility [39]. In this context, it has to be mentioned that a comparable intra-operative cartilage assessment is an important objective within most orthopedic departments, especially regarding patients with osteoarthritis. Therefore, a standardized cartilage grading and mapping, as described in the methods chapter, is routinely performed within our clinic. This consistent grading of the cartilage, which has been asserted for years, possibly leads to familiar use of this classification with comparative results within our clinic. However, appropriate appraisement of these studies, which compare the grading of cartilage among surgeons of several countries, clinics, etc., seems to be difficult. Nevertheless, a study of Acebes et al., which evaluated the agreement between arthroscopic and histopathological grading, shows much better results. This study revealed that the arthroscopic method is a valuable tool in clinical research to score chondropathies in the medial femorotibial compartment in knee osteoarthritis [40].

Bachmann et al. studied degenerative cartilage changes in trauma patients with a mean age of 29 years. A significant tendency to underestimate the severity of damage on MR images was reported [16]. In further studies comparing MRI- and arthroscopic findings, degenerative cartilage disorders were undergraded more often than overgraded [17, 18]. Broderick et al. concluded, that this tendency to underestimate cartilage disorders must be taken into account when clinical and research procedures are being planned [18]. According to our study protocol, only patients with clinically relevant osteoarthritis were included. This leads to a relatively high frequency of advanced cartilage disorders with numerous possibilities for underestimation. In our series, undergrading during MRI assessment was noticed for both observers in 23% and 25% of cases, respectively; and overgrading was assessed in 20% and 28% of cases, respectively. Thus, even though advanced cartilage disorders were relatively frequently compared to the studies mentioned above, this tendency could not be confirmed in our study.

Regarding the patients included in this study, a detailed assessment of degenerative cartilage disorders was important to recommend adequate treatment. For these patients, MRI was expected to be an additional decision support. In the literature, only few studies evaluate diagnostic relevance of MRI in patients with degenerative cartilage diseases. Review of these studies revealed a high variability of diagnostic values. Kawahara et al. demonstrated the best results, showing a sensitivity of 32% for grade 1, 72% for grade 2, 94% for grade 3 and 100% for grade 4 disorders [20]. MR scans were performed on a 0.5 T unit. Similarly to our study, MR images were obtained with fast spin-echo imaging. In a further own study using fast spin-echo sequences on a 3-Tesla MRI unit, sensitivity was 26% for grade 1, 63% for grade 2, 64% for grade 3 and 77% for grade 4 lesions [22]. In contrast to the present study, only patients with at least grade 2 lesions and a negative trauma anamnesis were included. Further inclusion criteria were not defined. Thus, the present study better reflects patients with clinically relevant degenerative disorders of the cartilage, where decision-making sometimes appears difficult in clinical practice. Bachmann et al. reported a sensitivity of 14% for grade 1, 32% for grade 2, 94% for grade 3 and 100% for grade 4 lesions [16]. A disadvantage of this study is that mainly athletes with a mean age of 29 years suffering from an acute knee disorder were included. Therefore, a comparison to elderly patients with clinically relevant osteoarthritis could be difficult. Our diagnostic values for the detection of grade 1 and 2 lesions are reasonably within the ranges of the aforementioned studies, whereas the values for grades 3 and 4 were relatively poor. Especially the diagnostic values for the detection of grade 3 lesions, showing a sensitivity of only 30% and 36%, were disappointing. Specificities and negative predictive values, which were relatively good for all other grades, also showed poor results for grade 3 lesions. The slice thickness of 3 and 4 mm, used in our MRI protocol, might possibly affect the ability of MRI to detect a small grade 3 lesion, whereas the visualization of a grade 2 disorder, which is frequently extended over a large articular surface, is less affected. Furthermore, the thickness of the hyaline cartilage, which was reported to measure between 2 and 3 mm [41], could complicate a correct MRI cartilage diagnosis when a relatively broad slice thickness is chosen. Interestingly, in the study of Kawahara et al., which demonstrated the best diagnostic values, MR images were obtained with a slice thickness of 5 mm. Using fast spin-echo sequences, protocols chosen in this study were similar to our study [20]. Thus, poor diagnostic values in our study could not be solely explained by the slice thickness of 3 and 4 mm. Because our sensitivities show relatively low values at all grades of cartilage disease, it has to be assumed that the application of MRI for a precise grading of degenerative cartilage disorders is limited (Table 2). As a consequence, it has to be claimed that arthroscopy should not be replaced by MRI when a grading of the cartilage is crucial for the definitive planning of a therapeutic procedure. In these cases, a diagnostic arthroscopy should be suggested even if MR images are available. This point of view could be mirrored by the treatment procedures of some patients included in this study. In our series, arthroscopy was performed in five patients before uni- or bicondylar arthroplasties were implanted during the same surgery. In these cases, we had reservations about the reliability of MRI findings even though MR images were subjectively of quite good quality.

Regarding therapeutic decisions, controversial positions on surgical proceedings should be mentioned. On the one hand, several studies demonstrate that arthroscopic treatments in knee osteoarthritis may delay more extensive surgery such as replacement arthroplasty [42, 43]. On the other hand, a large evaluation of over 14.000 arthroscopic debridement procedures for knee osteoarthritis revealed that almost 10% of patients required total knee replacement within one year. Because rates of arthroplasty were high particularly in elderly patients, an overutilization of arthroscopic treatments in this patient group was discussed [44]. In this context, the unequivocal value of MRI for the visualization of further typical findings of osteoarthritis, such as osteophytes, bone marrow edema, subchondral sclerosis, cysts, etc. has to be mentioned. For instance, it has been shown that osteophytes and joint effusions detected at MRI were significantly associated to clinical features such as pain and stiffness [45–48]. Certainly, such MRI findings should be considered and adjusted to clinical investigation. Possibly, further aspects of MR imaging could support the individual decision-making regarding treatments in patients with knee osteoarthritis. With respect to the difficulties for an exact assessment of disease severity in osteoarthritis patients encountered herein, further studies on this complex topic must further query the value of MRI in clinical practice.

In patients with osteoarthritis, the value of MRI for a precise grading of the cartilage is limited. When the assessment of the cartilage is crucial for a definitive decision regarding therapeutic options in patients with osteoarthritis, our data suggest that arthroscopy should not be generally replaced by MRI.