Jenny Folkesson

Segmenting Articular Cartilage Automatically Using a Voxel Classification Approach

We present a fully automatic method for articular cartilage segmentation from magnetic resonance imaging (MRI) which we use as the foundation of a quantitative cartilage assessment. We evaluate our method by comparisons to manual segmentations by a radiologist and by examining the interscan reproducibility of the volume and area estimates. Training and evaluation of the method is performed on a data set consisting of 139 scans of knees with a status ranging from healthy to severely osteoarthritic. This is, to our knowledge, the only fully automatic cartilage segmentation method that has good agreement with manual segmentations, an interscan reproducibility as good as that of a human expert, and enables the separation between healthy and osteoarthritic populations. While high-field scanners offer high-quality imaging from which the articular cartilage have been evaluated extensively using manual and automated image analysis techniques, low-field scanners on the other hand produce lower quality images but to a fraction of the cost of their high-field counterpart. For low-field MRI, there is no well-established accuracy validation for quantitative cartilage estimates, but we show that differences between healthy and osteoarthritic populations are statistically significant using our cartilage volume and surface area estimates, which suggests that low-field MRI analysis can become a useful, affordable tool in clinical studies

Identification of progressors in osteoarthritis by combining biochemical and MRI-based markers

IntroductionAt present, no disease-modifying osteoarthritis drugs (DMOADS) are approved by the FDA (US Food and Drug Administration); possibly partly due to inadequate trial design since efficacy demonstration requires disease progression in the placebo group. We investigated whether combinations of biochemical and magnetic resonance imaging (MRI)-based markers provided effective diagnostic and prognostic tools for identifying subjects with high risk of progression. Specifically, we investigated aggregate cartilage longevity markers combining markers of breakdown, quantity, and quality.MethodsThe study included healthy individuals and subjects with radiographic osteoarthritis. In total, 159 subjects (48% female, age 56.0 ± 15.9 years, body mass index 26.1 ± 4.2 kg/m2) were recruited. At baseline and after 21 months, biochemical (urinary collagen type II C-telopeptide fragment, CTX-II) and MRI-based markers were quantified. MRI markers included cartilage volume, thickness, area, roughness, homogeneity, and curvature in the medial tibio-femoral compartment. Joint space width was measured from radiographs and at 21 months to assess progression of joint damage.ResultsCartilage roughness had the highest diagnostic accuracy quantified as the area under the receiver-operator characteristics curve (AUC) of 0.80 (95% confidence interval: 0.69 to 0.91) among the individual markers (higher than all others, P < 0.05) to distinguish subjects with radiographic osteoarthritis from healthy controls. Diagnostically, cartilage longevity scored AUC 0.84 (0.77 to 0.92, higher than roughness: P = 0.03). For prediction of longitudinal radiographic progression based on baseline marker values, the individual prognostic marker with highest AUC was homogeneity at 0.71 (0.56 to 0.81). Prognostically, cartilage longevity scored AUC 0.77 (0.62 to 0.90, borderline higher than homogeneity: P = 0.12). When comparing patients in the highest quartile for the longevity score to lowest quartile, the odds ratio of progression was 20.0 (95% confidence interval: 6.4 to 62.1).ConclusionsCombination of biochemical and MRI-based biomarkers improved diagnosis and prognosis of knee osteoarthritis and may be useful to select high-risk patients for inclusion in DMOAD clinical trials.

Automatic segmentation of the articular cartilage in knee MRI using a hierarchical multi-class classification scheme

Osteoarthritis is characterized by the degeneration of the articular cartilage in joints. We have developed a fully automatic method for segmenting the articular cartilage in knee MR scans based on supervised learning. A binary approximate kNN classifier first roughly separates cartilage from background voxels, then a three-class classifier assigns one of three classes to each voxel that is classified as cartilage by the binary classifier. The resulting sensitivity and specificity are 90.0% and 99.8% respectively for the medial cartilage compartments. We show that an accurate automatic cartilage segmentation is achievable using a low-field MR scanner.

Automatic quantification of local and global articular cartilage surface curvature: Biomarkers for osteoarthritis?

The objective of this study was to quantitatively assess the surface curvature of the articular cartilage from low‐field magnetic resonance imaging (MRI) data, and to investigate its role in populations with varying radiographic signs of osteoarthritis (OA), cross‐sectionally and longitudinally. The curvature of the articular surface of the medial tibial compartment was estimated both on fine and coarse scales using two different automatic methods which are both developed from an automatic 3D segmentation algorithm. Cross‐sectionally (n = 288), the surface curvature for both the fine‐ and coarse‐scale estimates were significantly higher in the OA population compared with the healthy population, with P < 0.001 and P ≪ 0.001, respectively. For the longitudinal study (n = 245), there was a significant increase in fine‐scale curvature for healthy and borderline OA populations (P < 0.001), and in coarse‐scale curvature for severe OA populations (P < 0.05). Fine‐scale curvature could predict progressors using the estimates of those healthy at baseline (P < 0.001). The inter‐scan precision was 2.2 and 6.5 (mean CV) for the fine‐ and coarse scale curvature measures, respectively. The results showed that quantitative curvature estimates from low‐field MRI at different scales could potentially become biomarkers targeted at different stages of OA. Magn Reson Med 59:1340–1346, 2008.

Combining binary classifiers for automatic cartilage segmentation in knee MRI

We have developed a method for segmenting tibial and femoral medial cartilage in MR knee scans by combining two k Nearest Neighbors (kNN) classifiers for the cartilage classes with a rejection threshold for the background class. We show that with this threshold, two binary classifiers are sufficient compared to three binary classifiers in the traditional one-versus-all approach. We also show that the combination of binary classifiers produces better results than a kNN classifier that is trained to partition the voxels directly into three classes. The resulting sensitivity, specificity and Dice volume overlap of our method are 84.2%, 99.9% and 0.81 respectively. Compared to state-of-the-art segmentation methods, our method outperforms a fully automatic method and is comparable to a semi-automatic method.

Locating articular cartilage in MR images

Accurate computation of the thickness of the articular cartilage is of great importance when diagnosing and monitoring the progress of joint diseases such as osteoarthritis. A fully automated cartilage assessment method is preferable compared to methods using manual interaction in order to avoid inter- and intra-observer variability. As a first step in the cartilage assessment, we present an automatic method for locating articular cartilage in knee MRI using supervised learning. The next step will be to fit a variable shape model to the cartilage, initiated at the location found using the method presented in this paper. From the model, disease markers will be extracted for the quantitative evaluation of the cartilage. The cartilage is located using an ANN-classifier, where every voxel is classified as cartilage or non-cartilage based on prior knowledge of the cartilage structure. The classifier is tested using leave-one-out-evaluation, and we found the average sensitivity and specificity to be 91.0% and 99.4%, respectively. The center of mass calculated from voxels classified as cartilage are similar to the corresponding values calculated from manual segmentations, which confirms that this method can find a good initial position for a shape model.

312 AUTOMATIC KNEE CARTILAGE VOLUME QUANTIFICATION COMPARED TO JOINT SPACE WIDTH: BIOMARKERS OF LONGITUDINAL PROGRESSION?

Purpose: For clinical studies, diagnostic and prognostic biomarkers are needed to select a population at the target stage of osteoarthritis (OA) with a high risk of progression; and an efficacy biomarker is needed to quantify the treatment effect. Currently, diagnostic and prognostic markers are available, but the development of progression biomarkers has proved to be challenging. The aim of this study was to evaluate whether a fully automatic cartilage volume quantification method is suitable as a biomarker for quantification of longitudinal progression of knee OA. For perspective, the results are compared to joint space width (JSW) quantification. Methods: A study population was prospectively selected with 159 subjects with age 21 to 81 years (mean 56), BMI 19 to 38 (mean 26), and 48% female. Radiographs were acquired in a load-bearing semi-flexed position using the SynaFlex. MRI scans with near-isotropic voxels were acquired from a Turbo 3D T1 sequence on a 0.18T Esaote scanner (40° FA, TR 50 ms, TE 16 ms, scan time 10 min, resolution 0.7 x 0.7 x 0.8 mm3). Radiographs and MRI were acquired for both left and right knees at baseline (BL), after one week for a subgroup of 31 knees, and at follow-up (FU) after 21 months. After exclusion of 25 knees used for training of the computer-based method, 288 knees were in the study at BL and 245 knees at FU. Kellgren and Lawrence (KL) score and JSW were evaluated from the radiographs in the medial tibio-femoral compartment and tibial and femoral cartilage volume was quantified in the medial compartments by a fully automatic framework. JSW and volume were normalized by the tibial plateau width. At BL, the distribution of KL scores was (145,88,30,24,1) for KL 0-4. At FU, 25 knees had progressed from healthy to OA (KL>0) and 101 had remained healthy. Results: At BL, the mean total cartilage volume was 6851 mm3 with a scan-rescan CV of 3.6% (since the method is fully automatic, the intra-scan CV was zero). The volume quantification allowed diagnostic separation at BL of healthy from OA (p<0.001) as well as from early OA (KL 1, p<0.01), see Figure 1. The BL volume predicted progression with borderline significance (p=0.08). Finally, the measured cartilage loss was higher for progressors than non-progressors (p<0.01), see Figure 2 (right). For comparison, JSW provided diagnostic separation of healthy from OA (p<0.001) and from early OA (p<0.01) but allowed neither prognostic (p=0.3) nor progression separation (p=0.4, Figure 2 left). Conclusions: Since JSW is an integral part of the KL score, the diagnostic ability was expected. However, the results indicated that the use of JSW as outcome measure in longitudinal studies is questionable. Cartilage volume was suitable as diagnostic marker and borderline suitable as prognostic. More importantly, the volume quantification showed increased cartilage loss for the OA progressors compared to the non-progressors (p<0.01). Thereby, the fully automatic computer-based method may be Figure 1. Left: Volume allowed separation of healthy (KL 0) from OA (KL>0). Right: There was a clear linear trend of reduced cartilage volume with increasing KL score.

P266 AUTOMATIC CARTILAGE VOLUME ESTIMATION FROM LOW-FIELD KNEE MRI: A LONGITUDINAL STUDY

Results: The phantom values obtained with VNA and fit with the TOPPCAT program compare favorably to the VFIRFT approach (Figure 1). The results from OA subjects are shown in Table 1 and compare favorably to literature values obtained with variations on inversion recovery techniques. We have included values from muscle as a reference and indication of upper limits of dynamic range of the technique.

P276 AUTOMATIC QUANTIFICATION OF CARTILAGE THICKNESS FROM MRI FOR MONITORING PROGRESSION OF OSTEOARTHRITIS–A LONGITUDINAL STUDY

Purpose: During progression of knee osteoarthritis (OA) the cartilage breakdown causes gradual thinning of the articular cartilage sheets. The aim of this study was to investigate whether cartilage thickness measurements from an automatic, computerized framework for cartilage quantification from low-field MRI are suitable for use in clinical studies. This was evaluated at baseline in terms of inter-scan precision and ability to separate healthy from knees with a degree of osteoarthritis. After 21 months, the longitudinal changes were compared to the precision and the ability to separate healthy from OA was evaluated. Methods: A randomized population of both male and female subjects was prospectively selected such that there was an even distribution among male and female and across the ages from 21 to 80 (mean 56) with BMI from 20 to 38 (mean 27). Both left and right knees and both healthy and knees with varying degree of osteoarthritis (OA) as defined by the Kellgren and Lawrence score at baseline (KL) were used giving a total of 215 knees in the study. MR scans were acquired using a sagittal Turbo 3D T1 sequence on a 0.18T Esaote C-Span scanner giving near-isotropic voxels with slice thickness of 0.8mm. Scans were acquired at baseline, after one week for a subgroup of 31 knees, and then again after 21 months for all knees. The thickness of the medial tibial cartilage compartment was measured at baseline and after 21 months using a fully automatic framework for morphometric cartilage analysis based on supervised learning and a statistical cartilage sheet shape model. We measured the mean cartilage thickness across the entire area of the bone including denuded regions which are measured with zero thickness. For baseline measurements, the cartilage thickness was normalized by the width of the medial tibial plateau. Results: The precision of the thickness measurements was 0.08