Quantitative assessment of cartilage status in osteoarthritis by quantitative magnetic resonance imaging: Technical validation for use in analysis of cartilage volume and further morphologic parameters
Quantitative diagnostic tools for osteoarthritis (OA) are important for evaluating the treatment response to structure-modifying drugs. This study was undertaken to test the technical validity (accuracy) of quantitative magnetic resonance imaging (qMRI) for reliable determination of the total bone interface area, percentage of cartilaginous (denuded) joint surface area, and cartilage thickness in OA.
High-resolution MRIs of femorotibial and patellar cartilage were acquired in 21 patients prior to total knee arthroplasty, using a T1-weighted gradient-echo sequence with water excitation. After segmentation of original bone interface areas (before disease onset) and the actual cartilage layer, the percentages of cartilaginous joint surface area, cartilage thickness, and cartilage volume were determined using proprietary software. During surgery, the patella and the medial and lateral tibia were resected. Results obtained with qMRI were compared with those obtained by direct image analysis of surface area, cartilage thickness, and cartilage volume of the surgically removed tissue.
Pairwise differences between results obtained with qMRI and morphologic analysis were ±4.6% for percentage of cartilaginous surface area, ±8.9% for cartilage thickness, and ±9.1% for cartilage volume. Correlation coefficients ranged from 0.92 (thickness) to 0.98 (volume).
Quantitative MRI permits technically accurate and differential assessment of increases in eroded joint surface area and reductions in cartilage thickness in OA. The surrogate validity of these parameters requires testing in longitudinal studies. These parameters may be advantageous over determination of cartilage volume alone when diagnosing OA, exploring its progression, or testing responsiveness to new therapies.
Osteoarthritis (OA) is one of the most common chronic diseases and represents a tremendous burden on the aging population and on the economics of health care systems (1, 2). Although OA has for some time been considered a “boring” degenerative disease that is almost inevitably associated with aging, the recent declaration of the Bone and Joint Decade (3) is meant to increase awareness of the disease and to promote advances in the understanding of its pathophysiology, diagnosis, and treatment. Population-based epidemiologic studies have identified genetic and environmental risk factors in OA (4–7) and have revealed that symptoms correlate poorly with structural changes observed on radiographs (4, 8). However, radiography has the potential disadvantages that cartilage cannot be delineated directly (9), results are valid only for the medial (but not lateral) femorotibial compartment (10), and there may be interactions with processes other than cartilage loss, such as meniscal subluxation (11). The current view of OA pathophysiology is therefore based on indirect evidence of joint destruction (joint space and osteophytes), although articular cartilage has been identified as the key tissue in the disease process.
New techniques have been proposed for the surgical treatment of OA, such as subchondral microfracture (12), autologous cartilage transplantation (mosaic plasty ), autologous chondrocyte implantation (14), matrix-assisted chondrocyte implantation, and others (15). In addition, strong interest is currently directed at developing orally administrable compounds that offer the ability to delay or stop the structural changes involved in the OA process (16). Two recent studies (17, 18) have suggested that an orally administered dietary supplement, glucosamine sulfate, is capable of producing structure modification in OA. These conclusions have been based on measurements of joint space narrowing on radiographs, but it has been argued that the observed effects may have been due to a change in joint position on standing extended radiographs, following symptomatic relief (19). This ongoing discussion highlights the need for an efficient, validated, and robust technique for quantitative assessment of cartilage status in OA that can be used in clinical trials to assess disease progression and response to treatment within reasonable time intervals. In fact, one of the most severe obstructions in the process of validation of new drugs has been the lack of reliable markers for therapeutic efficacy in vivo.
Magnetic resonance imaging (MRI) has the distinct advantage that it can delineate articular cartilage (the target tissue) directly and noninvasively. Three-dimensional (3-D) image protocols with high spatial resolution and innovative image analysis methods have made it possible to determine articular cartilage morphology in the human knee joint with high technical validity (accuracy) and precision (20–23). However, most studies have focused on the investigation of cartilage in normal joints, and only small-scale technical validation studies have been conducted in OA patients (20, 21, 24). In this context, total knee arthroplasty (TKA) provides a unique opportunity for validating measurements in vivo, because the patient can be imaged prior to surgery, and the excised tissue analyzed after the operation. Moreover, a wide range of cartilage destruction is observed, with some compartments demonstrating only mild, but others more severe, structural changes.
Previous studies have confirmed that cartilage volume can be measured with satisfactory validity in patients with OA (20, 21, 24) if appropriate MRI protocols are used. The 3-D capabilities of MRI and current image analysis techniques further permit differentiation between the process of reduction in cartilage thickness of remaining cartilage fragments and a decrease in cartilaginous joint surface area or increase in denuded (eroded) joint surface area.
The objective of this study was to comprehensively test the technical validity of quantitative measures of cartilage morphology in OA patients prior to TKA. Technical validity refers to the degree to which a measurement (and variation in measurements between subjects) corresponds to the true value. Technical validity must be distinguished from surrogate validity, which is the ability to measure a surrogate marker associated with clinical outcome; the current study addresses the former. Specifically, we tested the hypothesis that quantitative MRI (qMRI) can be used to accurately measure the percentage of cartilaginous or denuded joint surface area, as well as the thickness of remaining cartilage fragments.
PATIENTS AND METHODS
Patients and MRI.
The study included 21 patients (4 men and 17 women; mean ± SD age 70.6 ± 7.7 years [range 58–86]) with a clinical indication for TKA. Written informed consent was obtained from the patients, and the study protocol was approved by the local ethics committee.
MRI was performed prior to TKA, using a clinical 1.5T magnet (Magnetom Symphony; Siemens, Erlangen, Germany) and a circular polarized transmit-receive extremity coil. To obtain high-contrast and high-resolution images of the cartilage, a T1-weighted spoiled 3-D gradient-echo sequence was used (fast low-angle shoot sequence with selective water excitation; radiofrequency amplitude ratios 1-2-1, repetition time 19 ms, echo time 8.6 ms, flip angle 20°, bandwidth 130 Hz/pixel) (Figure 1). The partition thickness was 1.5 mm and the in-plane resolution was 0.31 mm (field of view 160 mm, matrix 5122 pixels, phase resolution 100%, slice resolution 75%). One coronal data set of the femorotibial joint and 1 axial data set of the femoropatellar joint were acquired.
Digital image analysis of the MRI data.
After MRI, the data were transferred digitally to a workstation (Octane Duo; Silicon Graphics, Mountain View, CA). An interactive B-spline snake algorithm was used for segmentation of the medial and lateral tibial cartilage from the coronal images, and the patellar cartilage from axial images. This algorithm combines image-based and model-based approaches of image segmentation and has been shown to yield higher interobserver agreement than manual segmentation (25). The algorithm was applied on a slice-by-slice basis, for delineating the boundary between cartilage and subchondral bone. This was achieved by delineating the line of steepest gradient between the hyperintense cartilage and the hypointense bone in the T1-weighted, fat-suppressed gradient-echo image. The same procedure was performed for the articular surface, at the line of steepest gradient between the hyperintense cartilage and the hypointense joint cavity. In each slice, performance of the algorithm was carefully controlled by an experienced operator. Manual correction was performed at regions of low contrast, if required. Additionally, the original bone interface area (prior to disease) including cartilaginous and denuded bone interface areas but excluding peripheral osteophytes, was segmented, (Figure 2).
The size of the original bone interface and the area of cartilage coverage were determined using a triangulation algorithm that takes into account the curvature of the joint surface in a direction perpendicular to the images (26). The cartilage volume was obtained by numerical integration of the voxels assigned to the cartilage plates (Figure 2). The mean thickness of cartilage was determined 3-dimensionally, independent of the original section orientation, using a 3-D Euclidean distance transformation algorithm (22). The reproducibility (precision) of these computations has been determined previously in long-term studies of healthy volunteers (27) as well as short-term studies of OA patients, with joint repositioning (21). In OA, precision errors (standard deviations of repeated measurements) have been shown to range from 56 μl in the medial to 59 μl in the lateral tibia (volume) and from 55 μm in the lateral to 77 μm in the medial tibia (mean cartilage thickness). These errors were similar to those reported for healthy volunteers (48 and 66 μl, and 52 and 39 μm, respectively) (28).
Validation of surface, volume, and thickness measurements.
To test the technical validity of the measurements, the tibial plateau (including the medial and lateral tibial cartilage) and the patella were surgically removed during the operation, before insertion of a PFC prosthesis (DePuy Orthopedics, Warsaw, IN). One lateral tibial plateau and 2 patellae were lost, leaving 60 joint surfaces available for validation. In the first step, the original bone interface area was covered with thin aluminum foil that was fitted to the curved surface. The foil was then removed and unfolded into 1 plane, placing release cuts where necessary. The size of the foil was then measured using a calibrated independent image analyzing system (Kontron KS 400 with software release 3.0; Kontron, Munich, Germany). The same process was repeated for the cartilaginous joint surface area. The percentage of cartilaginous joint surface area was calculated as a percentage of this area to the total original bone interface area of each cartilage plate. The percentage of denuded (eroded) joint surface area was defined as 100 minus the cartilaginous joint surface area.
The 60 surfaces were then divided into 2 subsets for further study. In the first subset, the cartilage was surgically removed from the joint surface at the bone interface at the level of the calcified cartilage layer (20, 21). The volume of the tissue was determined according to Archimedes' principle (29), using a high-precision laboratory scale (PM 100; Mettler Toledo, Giessen, Germany). In the second subset, 10 anatomic sections were obtained from the remaining surfaces at equal intervals with a diamond band saw (Exakt Trennschleifsystem; Exakt, Norderstedt, Germany), using a coronal section orientation for the tibia and an axial orientation for the patella. Direct measurements of cartilage thickness in the remaining cartilage fragments were obtained from 5 equally distributed points throughout both sides of each section (100 points in each plate), using the same Kontron image analysis system. Measurement points located in denuded joint areas were discarded from analysis. Morphologic analyses were performed by one investigator (RvE-R) who was blinded to the qMRI analysis results.
The MRI values were correlated with those determined by direct morphologic measurement by 1) computing the absolute (random) pairwise difference (mean of individual differences with the + and − signs ignored), 2) computing the systematic pairwise difference (mean of individual differences with the + and − signs retained), 3) testing the systematic differences for statistical significance (paired t-test), and 4) computing the standard error of the estimate and the linear correlation coefficient (Pearson) to assess linear dependency of these parameters.
Figure 3 shows the correlation of values calculated using qMRI-based measures with those obtained from direct morphologic measurements. Comparison between surfaces indicated that accuracy errors tended to be smallest in the patella and highest in the medial tibia, but errors were relatively small for all surfaces (Table 1).
Table 1. Validation of quantitative magnetic resonance imaging in osteoarthritis patients in vivo, versus direct morphologic analysis*
Absolute difference, %
Systematic difference, %
Standard error y/x, %
ToBiOrg = original total bone interface area including cartilaginous and denuded (eroded) joint surface areas, but not peripheral osteophytes; ToBiCr = total bone interface area covered by articular cartilage (in cm2); ToBiCr% = % of total bone interface area covered by articular cartilage (note that percentage of denuded area = 100 − ToBiCr%).
P < 0.05.
The patella displayed 100% cartilaginous surface area in all cases.
The random differences for the surface areas (original bone interface and percent cartilaginous surface) were ∼8% for the patella and the medial and lateral tibia, with the qMRI-based values being significantly lower (P < 0.01) than those obtained by direct image analysis (Table 1). Pairwise random errors in estimates of the percentage of cartilage (or denuded) surface ranged from ±3.6% (lateral tibia) to ±5.5% (medial tibia), but there was no systematic over- or underestimation by qMRI (Table 1). Pairwise differences in cartilage volume ranged from ±6.6% (patella) to ±11.5% (medial tibia), with a slight qMRI overestimation in the patella (5.1%; P < 0.05) and lateral tibia (3.6%; P not significant) and a slight underestimation in the medial tibia (−3.1%; P not significant). The mean cartilage thickness of the remaining cartilage fragment displayed random differences between ±4.3% in the patella and ±12.3% in the medial tibia, with no systematic difference between qMRI and direct morphologic analysis (Table 1). The high correlation coefficients (r ≥ 0.92, except for the cartilage thickness in the medial tibia) and relatively small standard errors confirmed a high linear relationship between values derived from qMRI and those measured morphologically.
In this study we demonstrated that MRI-based quantitative measurement of the original bone interface area (including both cartilaginous and denuded joint areas, but not peripheral osteophytes), cartilaginous joint surface areas, and thickness of the remaining cartilage fragments displayed a high degree of technical accuracy in a relatively large sample of patients who underwent TKA. The standard errors and absolute, pairwise differences between qMRI results and those obtained with independent methods were generally <10%. The systematic (and statistically significant) difference between qMRI and direct image analysis is most likely due to enlargement of the foil when it is flattened for the purpose of direct measurements of areas with image analysis. Although release cuts were made to minimize this, enlargement of the foil during unfolding cannot be entirely eliminated.
Errors in estimating the percentage of the cartilage surface and cartilage thickness tended to be higher for the medial tibia compared with the lateral tibia and patella. This may be partly due to the more advanced stage of OA in the medial femorotibial compartment, since the majority of TKA patients had varus OA. Similarly, two previous studies in unselected cadavers also showed higher between-method deviations in results in the medial tibia compared with the lateral tibia and patella (30, 31), with higher errors potentially being due to the higher degree of congruity and larger contact area of the medial femorotibial compartment (26).
Recent studies have demonstrated an annual loss of cartilage volume of ∼5% in the tibia (32) and patella (33) of patients with OA. Given the satisfactory level of technical accuracy observed in this study, qMRI may be used to distinguish the effect of cartilage thinning of remaining cartilage fragments from that of a decrease in cartilaginous joint surface area (increase in denuded joint surface area), both of which may lead to a loss of cartilage volume. This may help to identify how cartilage loss occurs in OA, and whether different types of insults produce similar or different quantitative outcomes. Because a reduction in cartilaginous area may be accompanied by swelling of the remaining cartilage (34), separating these two variables is potentially more effective than using cartilage volume alone.
A limitation of the current study is that only the technical validity of the measurements, but not the surrogate validity, was addressed. The surrogate validity of these parameters requires further testing in longitudinal studies, exploring their efficacy over measurement of cartilage volume alone, or over radiographic measurement of joint space width. Changes in radiographic joint space width occur when the cartilage thins at the joint surface area where the measurement is made. Analysis of the percentage of cartilaginous (or denuded) joint surface area with qMRI, however, is site-independent and may potentially represent a more objective quantifiable parameter in OA. Disadvantages of qMRI, however, include the cost of image acquisition, limited access to MRI scanner time, limited availability of dedicated software applications, and the fact that MRIs are acquired under non–weight bearing conditions.
In a recent cross sectional study (35), we have demonstrated that T scores (differences between a patient value and the mean value of young healthy adults of the same sex, divided by the standard deviation in young healthy adults) show better discrimination between patients and healthy subjects when MRI-based measurements of cartilage volume are normalized to the total original bone interface area. This normalization was shown to be superior in relation to normalization of body weight and height. The current analysis shows that measurement of the original bone interface area in patients with advanced OA is not only reproducible and useful, but also technically valid, despite the presence of peripheral osteophytes. In the context of cross-sectional analysis, the percentage of denuded joint area may also be potentially used as an inclusion/exclusion criterion for clinical trials. Cartilage volume, in contrast, is problematic as an inclusion criterion because larger individuals with larger bones display larger cartilage volume (36), independent of OA changes.
In conclusion, this study demonstrated that qMRI permits quantification of the original bone interface area, the percentage of cartilaginous (denuded) surface area, and the thickness of the remaining cartilage fragments with high technical validity (accuracy), if adequate MR sequences and image analysis tools are used. Future longitudinal studies must test the surrogate validity of these parameters and determine whether this analysis is advantageous over measurement of radiographic joint space narrowing or of cartilage volume alone.