Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2022 Feb 13;40(11):2597–2608. doi: 10.1002/jor.25293

Predicting osteoarthritis onset and progression with 3D texture analysis of cartilage MRI DESS: 6‐Year data from osteoarthritis initiative

Ari Väärälä 1, Victor Casula 1,2,✉, Arttu Peuna 1,2,3, Egor Panfilov 1, Ali Mobasheri 1,4,5,6, Marianne Haapea 2,7, Eveliina Lammentausta 1,7, Miika T Nieminen 1,2,7
PMCID: PMC9790756  PMID: 35152476

Abstract

In this study, we developed a gray level co‐occurrence matrix‐based 3D texture analysis method for dual‐echo steady‐state (DESS) magnetic resonance (MR) images to be used for knee cartilage analysis in osteoarthritis (OA) studies and use it to study changes in articular cartilage between different subpopulations based on their rate of progression into radiographically confirmed OA. In total, 642 series of right knee DESS MR images at 3T were obtained from baseline, 36‐ and 72‐month follow‐ups from the OA Initiative database. At baseline, all 214 subjects included in the study had Kellgren‐Lawrence (KL) grade <2. Three groups were defined, based on time of progression into radiographic OA (ROA) (KL grades ≥2): control (no progression), fast progressor (ROA at 36 months), and slow progressor (ROA at 72 months) groups. 3D texture analysis was used to extract textural features for femoral and tibial cartilages. All textural features, in both femur and tibia, showed significant longitudinal changes across all groups and tissue layers. Most of the longitudinal changes were observed in progressors, but significant changes were observed also in controls. Differences between groups were mostly seen at baseline and 72 months. The method is sensitive to cartilage changes before and after ROA. It was able to detect longitudinal changes in controls and progressors and to distinguish cartilage alterations due to OA and aging. Moreover, it was able to distinguish controls and different progressor groups before any radiographic signs of OA and during OA. Thus, texture analysis could be used as a marker for the onset and progression of OA.

Keywords: cartilage, DESS, magnetic resonance imaging (MRI), osteoarthritis (OA), texture analysis

1. INTRODUCTION

Magnetic resonance imaging (MRI) is an optimal modality for noninvasive evaluation of articular cartilage thickness, integrity, and quality in osteoarthritis (OA) studies, because of its superior soft‐tissue contrast, and multiplanar capability. 1 Compared to radiographic detection of joint space narrowing, MRI is more sensitive to cartilage loss. 2 Earlier studies have shown that quantitative MRI (qMRI), like T2 and T1ρ mapping, can be used to detect changes within the cartilage extracellular matrix. 3 , 4

Gray level co‐occurrence Matrix (GLCM) based texture analysis provides a way to extract textural features from an image. These features are statistical measures of the arrangement of pixel intensities in a region and provide information on spatial heterogeneity, 5 which tends to increase in biological tissue as structural disorganization caused by an underlying pathology progresses. Although originally designed for analyzing aerial photographs, 5 , 6 this image processing method has been taken into use in the field of medical image processing. 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 Texture analysis studies have been applied in 2D, i.e., separately on single slices, using quantitative T2 and T1ρ maps and those have shown that the method can detect signs of OA in cartilage. 10 , 21 , 22 Furthermore, texture analysis studies applied on brain T1 and T2 weighted images, and knee dual‐echo steady‐state (DESS) images, have shown the potential in analyzing nonquantitative images. 23 , 24 Texture analysis can provide information about localized variations in cartilage collagen matrix, thus providing the ability to differentiate healthy subjects from subjects at risk for OA progression. 13 , 17 In earlier 3D texture analysis studies, 19 GLCMs were constructed across multiple slices (the third dimension) using the original 3D volumetric data without transformations and rotations of the pixel coordinates. However, this approach has limited directions for neighboring pixels, and it is not possible to follow the geometry of the cartilage and its laminar structure. Neither 2D nor 3D texture analysis approach has been employed for cartilage laminar analysis. Hence, in this study we propose a new method developed to perform more complex textural analyses of the cartilage in 3D. The new method extracts textural features for three different subsets of cartilage layers and the full thickness cartilage.

The 3D DESS is a combination of T1 and T2 weighted images, where the signals from two consecutive echoes, the FID‐signal of a FISP sequence (fast imaging steady precession) and the echo‐signal of a PSIF (reversed FISP), are separately acquired and the combined. The PSIF part of the sequence provides a high T2 contrast, and the FISP part provides representative morphological images with T1/T2 ratio dominated contrast. 25 All in all, DESS provides images with high contrast between cartilage and fluids, it has the advantage to combine morphological and quantitative analysis of cartilage from the same dataset with high resolution, and the imaging time is relatively short. 25 , 26 , 27 , 28 The possibility of isotropic resolution of DESS and the optimal contrast for cartilage suit well for automated segmentation. 29 Previously, the GLCM‐based 3D texture analysis method has not been applied to clinical MRI images of articular cartilage, and only one study used the method on MRI of joints to assess vertebral trabecular bones using sagittal T1‐weighted MR images. 30 For 3D texture analysis, isotropic coverage creates less distortion in coordinate transformations and rotations. Compared to qMRI methods, the DESS sequence is available for most clinical MRI scanners and sites, and it allows complete coverage of knee cartilage typically in less than 6 min. Thus, the 3D texture analysis method can be applied directly to DESS images without the need for mono‐exponential fitting like in the case of T1ρ and T2.

There are currently no established imaging biomarkers able to diagnose OA at an early stage and predict its progression rate, which can vary considerably between patients. 31 Generally, the radiological progression of OA is slow and can take several years or even decades. 32 , 33 , 34 , 35 , 36 , 37 , 38 However, it has been reported that in some cases, progress can be rapid and within 12 months, subjects with normal knee radiographs can develop radiographic OA (ROA), 39 or incident OA subjects can progress to advanced‐stage ROA. 40

The purpose of this study was to develop a 3D texture analysis method to be used for knee cartilage analysis in OA studies and clinical trials. Currently, there are no validated methods to predict the development of ROA from morphological images or any imaging markers. OA progresses at different rates and the longitudinal clinical Osteoarthritis Initiative (OAI) dataset enables us to study this, by creating different groups based on their time of progression into ROA. Earlier, 2D texture analysis for qMRI has shown promising results in finding differences between regions of interest (ROI) of healthy and degenerated cartilage before radiographic signs of OA. We hypothesize that 3D texture analysis of DESS images, given their higher resolution and 3D nature, would be even more sensitive to early cartilage changes and it could be used to predict the development of OA before any radiographic signs.

2. MATERIALS AND METHODS

2.1. Study population

The proposed method was applied on a selected subset from a longitudinal cohort study, the OAI, enrolling 4796 subjects in total (https://nda.nih.gov/oai/study-details). 41 The OAI database is divided into three cohorts: a Progression cohort consisting of symptomatic knee OA patients with definite tibial‐femoral osteophyte (OARSI grades 1–3) at baseline and pain, aching or stiffness on most days of a month in the past year; an Incidence cohort including subjects with frequent knee symptoms without ROA but with two or more other eligibility risk factors; and a Control cohort with no symptoms, no ROA in either knee at baseline and no risk factors.

A total of 214 subjects were selected for our study from the OAI participants aged 65 years or younger with Kellgren‐Lawrence 42 (KL) grade <2 at baseline and DESS MR images of the right knee available at baseline (00m), 36‐month (36m), and 72‐month (72m) follow‐ups (Figure 1). They were classified in three groups based on the time of progression into ROA (KL ≥2). The control group (N = 65) included subjects selected from the OAI Control cohort with KL grade <2 at all time points (N = 65). Subjects in the slow progressor group were selected from the OAI Incidence and Progression cohorts diagnosed with ROA only at 72m (N = 71). Subjects in the fast progressor group were selected from OAI Incidence and Progression cohorts diagnosed with ROA already at 36m (N = 78).

Figure 1.

Figure 1

Open in a new tab

Flow chart of the subgroups selection based on the time of progression into radiologic osteoarthritis (Kellgren‐Lawrence ≥2), from the cohorts of the Osteoarthritis Initiative database for controls, fast progressors, and slow progressors

2.2. MRI acquisition and data preprocessing

MR images were acquired using 3T Siemens clinical MR systems (Siemens Healthcare) according to the OAI knee MRI protocol. 41 In total, 642 DESS images of right knees were segmented using the automatic deep learning‐based method that was previously trained and validated against manual segmentations. 43 As an output, the software produced separate segmentation masks for femoral, tibial, and patellar cartilage tissues, as well as menisci. Subsequently, the full cartilage masks were automatically segmented into the MOAKS‐based compartments via elastic registration to a multiatlas of 10 scan‐segmentation pairs. 44 For 3D texture analysis, full cartilage segmentations, and central medial and lateral compartments for femur and tibia are used. The average local thickness of cartilage was measured for the aforementioned compartments. 45 The mean thickness of full femoral and tibial cartilage tissues was determined from the compartmental measurements. Thickness data was used as a reference, to assess whether the textural feature changes were following the changes in the cartilage thickness.

2.3. 3D texture analysis

The proposed method for 3D texture analysis of cartilage DESS images was developed in‐house using Matlab (MathWorks Inc.). The software extracts textural features from a 3‐dimensional area defined by the DESS MR image and the segmentation mask. After anatomical normalization, the GLCM matrices for each configured direction are generated from neighboring pixels, which are layered on a plane parallel to the bone‐cartilage interface (BCI). These GLCMs are summed for the full cartilage thickness data (SUM). Additionally, the 1‐pixel thick layers, at 10% (L10), 50% (L50), and 90% (L90) of relative cartilage thickness from the BCI (Figure 2), have their own a GLCMs. Thickness in this context should not be confused with the average local cartilage thickness, because anatomical normalization requires interpolation and extrapolation of scattered data to grid‐aligned data, where pixel coordinates are rounded up to pixel level. More details about anatomical normalization and GLCM populating are provided in the Supporting Information B document.

Figure 2.

Figure 2

Open in a new tab

Sagittal view of 3D DESS MR image slice with femoral cartilage in blue. The textural features are extracted from one‐pixel thick 3D layers L10, L50, and L90. The SUM contains textural features from all cartilage data. The positions of the layers, thickness wise, are illustrated in the zoomed part of the cartilage, which is also showing collagen orientation. DESS, dual‐echo steady‐state; MR, magnetic resonance [Color figure can be viewed at wileyonlinelibrary.com]

For this study, 19 textural features 5 , 6 , 7 were extracted from the full femoral and tibial cartilages and the central medial and lateral compartments using our novel 3D texture analysis. Textural features were extracted for four different layers: for three one‐pixel thick 3D layers, L10, L50, and L90 (Figure 2), and 3D full cartilage thickness (SUM). Pixel intensities were divided into eight linear analysis bins using a gray level range of 0–300, where pixel intensities greater than 300 are placed in the eighth bin. Pixel offset value was set to 1, and the number of analysis directions (angles) to 8 (0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°).

2.4. Statistics

Statistical analysis was performed using Matlab. Differences between groups at the same time points were analyzed using the Kruskal–Wallis test and longitudinal changes within the groups using Friedman's test. All p values were adjusted with Bonferroni correction for multiple comparisons for both textural features and thickness. The Bonferroni correction was used in triplets for each textural feature: at each time point and in each layer for the group comparisons, and in each layer within each group for the longitudinal comparisons. In addition, the Benjamini‐Hochberg method with a 10% false discovery rate was applied to textural feature p values of each group comparison, in each time point in each layer, and p values of the longitudinal comparisons between each time point in each layer. The correlation between body mass index (BMI) and texture features was assessed using Spearman's correlation.

3. RESULTS

3.1. Full cartilage textural features

All textural features, from femoral and tibial cartilage, showed significant longitudinal changes across all groups and layers. Most, but not all, longitudinal changes were observed in progressor groups. Several features, like contrast (Figure 3) and entropy (Figure 4), increased over time, in both cartilages; on the other hand, features like energy (Figure 5) and homogeneity (Figure 6) decreased over time. Other features such as correlation (Figure 7), did not show a monotonic trend and showed less longitudinal changes. In a number of features, including contrast, energy, entropy, and homogeneity, the larger differences between time points were found in the tibia. Differences between groups were seen mostly at baseline and 72 months.

Figure 3.

Figure 3

Open in a new tab

Full cartilage femoral and tibial box plots of textural feature contrast for controls (ctrl), slow progressors, and fast progressors at baseline (00m), 36‐month (36m), and 72‐month (72m) follow‐ups. L10, L50, and L90 indicate results of one‐pixel thick 3D layers at 10%, 50%, and 90% of relative cartilage thickness from the bone‐cartilage interface. The SUM contains results from full cartilage thickness. Horizontal lines inside the boxplots indicate median values, boxplot edges indicate interquartile range, whiskers indicate 1.5 times the interquartile range, and diamonds indicate the outliers (***p < 0.001, **p < 0.01, *p < 0.05) [Color figure can be viewed at wileyonlinelibrary.com]

Figure 4.

Figure 4

Open in a new tab

Femoral and tibial box plots of textural feature entropy for controls (ctrl), slow progressors, and fast progressors at baseline (00m), 36‐month (36m), and 72‐month (72m) follow‐ups. L10, L50, and L90 indicate results of one‐pixel thick 3D layers at 10%, 50%, and 90% of relative cartilage thickness from the bone‐cartilage interface. The SUM contains results from full cartilage thickness. Horizontal lines inside the boxplots indicate median values, boxplot edges indicate interquartile range, whiskers indicate 1.5 times the interquartile range, and diamonds indicate the outliers (***p < 0.001, **p < 0.01, *p < 0.05) [Color figure can be viewed at wileyonlinelibrary.com]

Figure 5.

Figure 5

Open in a new tab

Femoral and tibial box plots of textural feature energy for controls (ctrl), slow progressors, and fast progressors at baseline (00m), 36‐month (36m), and 72‐month (72m) follow‐ups. L10, L50, and L90 indicate results of one‐pixel thick 3D layers at 10%, 50%, and 90% of relative cartilage thickness from the bone‐cartilage interface. The SUM contains results from full cartilage thickness. Horizontal lines inside the boxplots indicate median values, boxplot edges indicate interquartile range, whiskers indicate 1.5 times the interquartile range, and diamonds indicate the outliers (***p < 0.001, **p < 0.01, *p < 0.05) [Color figure can be viewed at wileyonlinelibrary.com]

Figure 6.

Figure 6

Open in a new tab

Femoral and tibial box plots of textural feature homogeneity for controls (ctrl), slow progressors, and fast progressors at baseline (00m), 36‐month (36m), and 72‐month (72m) follow‐ups. L10, L50, and L90 indicate results of one‐pixel thick 3D layers at 10%, 50%, and 90% of relative cartilage thickness from the bone‐cartilage interface. The SUM contains results from full cartilage thickness. Horizontal lines inside the boxplots indicate median values, boxplot edges indicate interquartile range, whiskers indicate 1.5 times the interquartile range, and diamonds indicate the outliers (***p < 0.001, **p < 0.01, *p < 0.05) [Color figure can be viewed at wileyonlinelibrary.com]

Figure 7.

Figure 7

Open in a new tab

Femoral and tibial box plots of textural feature correlation for controls (ctrl), slow progressors, and fast progressors at baseline (00m), 36‐month (36m), and 72‐month (72m) follow‐ups. L10, L50, and L90 indicate results of one‐pixel thick 3D layers at 10%, 50%, and 90% of relative cartilage thickness from the bone‐cartilage interface. The SUM contains results from full cartilage thickness. Horizontal lines inside the boxplots indicate median values, boxplot edges indicate interquartile range, whiskers indicate 1.5 times the interquartile range, and diamonds indicate the outliers (***p < 0.001, **p < 0.01, *p < 0.05) [Color figure can be viewed at wileyonlinelibrary.com]

At baseline, significant differences between controls and progressor groups were found in several textural features and layers. Most of the differences were found between controls and slow progressors in the femur L90, while differences between controls and fast progressors were found often in the tibia L10. Correlation, cluster prominence, information measure of correlation 2, sum of squares, and sum variance, showed significant differences between controls and both progressors in all layers in femur and tibia. In several features, most differences occurred either in the femur (energy, entropy, sum entropy, maximum probability, and autocorrelation) or in the tibia (contrast, cluster shade, difference variance). Other features differed between controls and either fast or slow progressors in different tissues and layers. For example, difference entropy, dissimilarity, homogeneity, inverse difference, and sum average had significances between controls and slow progressors, but not between controls and fast progressors. On the other hand, cluster shade in the femur had significances between controls and fast progressors but not between controls and slow progressors. Finally, information measure of correlation 1 had significances between controls and slow progressors only in the femur (all layers) but between controls and fast progressors in the tibia (all layers); while entropy and sum of entropy in the tibia differentiated controls from fast progressors in L10 and controls from slow progressors in L90. At 36 months, almost all the significant differences were observed between fast progressors and controls, mostly in the femur in L10 and L50. The only feature showing differences at 36 months, between controls and slow progressors, occurring only in the tibia (all layers except L90), was information measure of correlation 1. At 72 months, significant differences between controls and progressors were found for all textural features and most of the layers, with the majority of the significances seen between controls and fast progressors. Finally, significant differences between slow and fast progressors were observed only at 72 months in the tibia SUM. Descriptions of full cartilage results per textural feature, medians, interquartile ranges, and group comparison p values of each textural feature for all groups, layers, and time points are presented in the Supporting Information C document (Tables SC1–SC6).

3.2. Textural features of central medial and central lateral compartments

All textural features, from femoral and tibial cartilage, showed significant longitudinal changes across all groups, layers, and compartments. Most of the significant longitudinal changes were observed in progressor groups. The control group showed significant longitudinal changes mostly between baseline and 36 months. On the other hand, both progressors showed significant changes also between 36 and 72 months, the majority of them observed in the fast group. Textural features for the SUM behaved similarly as in full cartilage data: for example, contrast and entropy statistically significantly increased over time, energy and homogeneity statistically significantly decreased, and correlation showed less longitudinal significant changes compared to other features. The majority of the longitudinal significant changes in the control group in the tibia were found on the lateral side, otherwise significant longitudinal changes were spread more evenly between lateral and medial compartments.

All the baseline significances in the femur were observed between controls and slow progressors, and the majority of them on the lateral side. However, most baseline significances in the tibia, like cluster prominence, cluster shade, correlation, information measure of correlation 1 and 2, sum of squares, and sum variance, were observed between controls and fast progressors, on the medial side. At 36 months textural features like contrast, difference entropy, dissimilarity, entropy, homogeneity, and inverse difference showed significances between controls and both progressors in the lateral tibia. At 72 months, most of the significant differences in the femur were observed on the medial side, but in the tibia, most of the significances were on the lateral side. Finally, at 72 months, the majority of textural features showed significances between controls and progressors, but only energy, entropy, maximum probability, sum entropy, sum of squares, and sum variance showed significances between slow and fast progressor in the tibia on the central lateral compartment. Descriptions of compartmental SUM results per textural feature, medians, interquartile ranges, and group comparison p values of each textural feature for all groups, layers, and time points are presented in the Supplemental C document (Tables SC7–SC18).

3.3. Textural features versus BMI and cartilage thickness

Correlation coefficients between texture features and BMI were mostly nonsignificant or very weak‐to‐weak. Cartilage thickness, conversely to texture analysis, was significantly different at baseline only between controls and slow progressors in femur. Significant differences in cartilage thickness between fast progressors and ctrl were observed only at 36 months in femur. For cartilage thickness results and transversal and longitudinal p‐values, see the Supplemental A document (Tables SA1 and SA2, Figures SA1–SA3).

4. DISCUSSION

In the present study, we compared 3D textural features of 3D DESS MR images between a control group and two progressor groups with different rates of progression into ROA. The fast progressor group developed ROA within 3 years, while the slow progressor group developed ROA between three and 6 years. Nineteen textural features were extracted from full cartilage, and medial and lateral central compartments, in the femur and tibia. Textural features were extracted from four different layers. Three of them are one‐pixel thick 3D layers at a certain depth of cartilage thickness (L10, L50, and L90) and one is the 3D full cartilage thickness (SUM). 3D texture analysis was able to detect longitudinal changes in both progressor groups, but also in controls, suggesting worsening of cartilage health due to OA or aging, in all layers. The earliest changes from baseline were observed already at 36 months in all the three groups and in all layers in both femur and tibia and no particular layer seemed to be more frequently affected than other. Differences between controls and progressors were slightly more frequent in femur than in tibia. At baseline, several textural features showed significant differences between control and progressor groups, enabling the possibility of early diagnosis and prediction of OA development. Compared to controls, early changes at baseline in both progressor groups were seen in all layers. Those changes were most frequently found in superficial femur in slow progressors, which is consistent with the current understanding that early OA changes often start in the most superficial layer of cartilage. 46 , 47 In fast progressors, changes were also most frequently seen in superficial femur as well as in deep tibia, which might be indicating a possible role of changes occurring in deep layer of tibial cartilage in accelerating the OA onset. Changes in textural features did not follow changes in cartilage thickness, which suggests that cartilage thickness is not a confounding factor. Extracting textural features from femoral and tibial central lateral and central medial compartments did not increase the sensitivity of 3D texture analysis and resulted in a reduced number of significant differences compared to full cartilage analysis. Findings from later time points at 36 and 72 months indicate that 3D texture analysis has the potential to monitor cartilage changes associated with OA onset and progression.

Overall, the current study supports several findings from earlier 2D texture analysis studies using qMRI T2 and T1ρ. However, a comparison of our findings to earlier studies, which were using only one or two slices of knee cartilage, is not so straightforward. In the case of a 2D slice, the number of pixels is in hundreds, and in our case of full femoral cartilage, we are on a scale of over one hundred thousand pixels. Besides the differences in the number of pixels, many of the earlier 2D studies are not following the curvature of the cartilage, which means the texture analysis direction is not on a plane parallel, as in our case, or perpendicular to the bone cartilage interface. Offsets and the number of analyzing directions also vary. With 3D DESS MR images, there is also a significant difference in image contrast, compared to T2 and T1ρ.

Blumenkrantz et al. 10 reported a decrease in energy and an increase in entropy of cartilage T2 and T1ρ for OA patients. Our analysis of DESS images shows similarity only at a 72 months time point in L90 in the femur, where both progressors are significantly different from control. At 72 months in the tibia, energy and entropy in SUM can distinguish slow and fast groups, but L90 shows significance only between control and fast. Longitudinally, but only in a 9‐month time frame, Blumenkrantz et al. 10 reported that entropy of cartilage T2 significantly decreased in OA patients, but controls showed no significant changes in energy or entropy. However, we found out that over time, energy values are decreasing, and entropy values are increasing, in both progressor and control groups. We observed the biggest longitudinal changes between energy and entropy in the femur in the control group in L90, where entropy has significant change only between baseline and 36 months, and energy has significant change only between 36 and 72 months.

Joseph et al. 13 used subjects from OAI incidence and control cohorts to extract textural features of T2, at baseline. In the incidence group, entropy, contrast, and variance were increased differentiating it from control. In our study textural features can differentiate controls from progressors, especially variance (sum of squares) has strong significances between control and both progressors at baseline. The least amount of significance at baseline can be seen in the contrast in the femur.

Baum et al. 16 conducted a longitudinal study, using baseline and 36 months time points from OAI data. They reported a normal control group and subjects with OA risk factors stratified into three groups based on BMI. At baseline, subjects with risk factors for OA had no ROA (KL grade <2). Baum et al. 16  reported constantly elevated T2 entropy in overweight and obese subjects over 36 months. In their study on average over all compartments, entropy is showing a significant difference between control and two highest BMI value groups at 36 months, but not at baseline. Our results, at baseline, are showing significant differences between control and progressors in both femoral and tibial layers. However, at later time points, there are no significant differences until 72 months. Over time, Baum et al. 16 reported an increase in contrast and variance and a decrease in entropy. Our longitudinal findings agree with the contrast and variance findings of Baum et al. 16 but do not agree with the entropy, which is increasing over time.

Kretzschmar et al. 48 investigated compositional changes of knee cartilage at the site of newly appearing cartilage lesions and the surrounding cartilage. Their method requires a specific ROI in a specific slice, to be able to detect the difference. They used the longitudinal OAI data and reported that local cartilage ROIs had higher T2‐values compared to the surrounding cartilage, 4 years before lesion onset. They also reported of the 57 new cartilage lesions studied, most occurred in the medial femoral condyle (30%), and the lateral tibia (25%). Even though our study is not focused on finding lesions, similarities between the distribution of significances between medial and lateral compartments can be seen. In our textural features, at 72 months in the femur, the majority of significances were found in the central medial compartment. However, at 72 months in the tibia, the majority of the significances, and only significances between slow and fast progressors, were observed on the central lateral compartment.

Peuna et al. 21 studied compartmental two‐dimensional T2 textural features of symptomatic and asymptomatic subjects. Their method follows the cartilage surface and executes textural analysis parallel to the BCI. They reported significant differences between groups in femoral and tibial cartilage in medial and lateral central compartments. Our findings, in contrast, correlation, energy, entropy, homogeneity, and variance agree with results from Peuna et al. 21 in terms of the direction of change of parameter values in progressor groups.

Carballido‐Gamio et al. 11 reported that the subdivision of lateral and medial femoral compartments into weight‐bearing and nonweight‐bearing regions did not improve discrimination between healthy controls and subjects with mild OA, in two‐dimensional textural features of T2 and T1ρ. Our findings agree with this. In both, femur and tibia, when we compare significances between full cartilage data and medial and lateral compartments, it can be observed that full cartilage data provides more significances between groups at baseline. This could suggest that early cartilage changes, seen by texture analysis, might not be solely on weight‐bearing compartments of the cartilage.

The current study has several limitations. First, the studied groups were not BMI matched, but on the other hand, BMI did not show a strong correlation with textural features between groups. Second, different knee radiofrequency coils were used during the OAI study. Most of the data used in this study was acquired with similar quadrature transmit‐receive coils (same brand and design), which provided comparable signal‐to‐noise ratio. However, part of the knee MRIs of the control group at 72‐month follow‐up visit were performed using an eight‐channel phased‐array coil, with higher performance in terms of signal‐to‐noise ratio compared to the other coils. This might explain, at least in part, the large variance of the textural features observed in the control group at the last time point. After excluding those subjects from the analysis, significant differences were still observed between control and progressor groups, although the results changed for individual textural features at 72 months, see the Supporting Information D document (Figures SD1–SD7). Third, analysis parameters of texture analysis (e.g., gray‐level range, number of bins, and quantization method) can affect the results and need to be thoroughly optimized in the future for optimal sensitivity and robustness. Also, a dataset acquired at isotropic resolution would be preferred for less distorted coordinate transformations.

In conclusion, our novel method is sensitive to early and late cartilage degenerative changes in knee articular cartilage. It was able to distinguish controls and different progressor groups before, and after, any radiographic signs of OA. Moreover, 3D texture analysis of DESS images was able to detect longitudinal changes in controls and progressors, so it is sensitive and capable of distinguishing cartilage changes due to OA and aging. Thus, it could be used as a possible marker for predicting OA onset and progression in clinical trials of new disease‐modifying OA drugs. The method does not have any special hardware requirements, it works with a standard MR morphological sequence that can be run on any clinical MRI scanner and does not require acquisition of a separate quantitative image dataset. Therefore, it can be implemented as an adjunct methodology to any ongoing clinical trial in OA.

CONFLICT OF INTERESTS

Ali Mobasheri is Senior Advisor to the World Health Organization Collaborating Center for Public Health Aspects of Musculoskeletal Health and Aging and “Collaborateur Scientifique de l'Université de Liège” at the Université de Liège in Belgium. He has consulted for Genacol and Sterifarma, companies that produce and market collagen supplements. Ali Mobasheri has also consulted for Sanofi (Brazil), Pfizer Consumer Health, GSK Consumer Health, and Aché (Aché Laboratórios Farmacêuticos), companies that have ongoing R&D activities in joint health supplements. The other authors declare that they have no conflict of interests.

AUTHOR CONTRIBUTIONS

Ari Väärälä: 3D texture analysis method implementation, data analysis, and manuscript drafting. Victor Casula: Design of the study, interpretation of the results, supervising, and manuscript editing. Arttu Peuna: 3D texture analysis idea and method implementation support. Egor Panfilov: Cartilage segmentations using the automated DL segmentation tool, thickness calculations, and manuscript editing. Ali Mobasheri: interpretation of the results and manuscript editing. Marianne Haapea: Statistics support and manuscript editing. Eveliina Lammentausta: 3D texture analysis idea and method implementation support. Miika T. Nieminen: Design of the study, interpretation of results, supervising, and manuscript editing.

Supporting information

Supporting information.

Click here for additional data file. (34.7MB, docx)

Supporting information.

Click here for additional data file. (334.3KB, docx)

Supporting information.

Click here for additional data file. (312.6KB, docx)

Supporting information.

Click here for additional data file. (288.2KB, docx)

Supporting information.

Click here for additional data file. (697.6KB, docx)

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the financial support of the Jane and Aatos Erkko Foundation and strategic funding of the University of Oulu (Infotech Oulu). The funding sources had no role in the study design, collection, analysis, interpretation of data, in writing of the manuscript, or in a decision to submit the manuscript for publication.

Väärälä A, Casula V, Peuna A, et al. Predicting osteoarthritis onset and progression with 3D texture analysis of cartilage MRI DESS: 6‐year data from osteoarthritis initiative. J Orthop Res. 2022;40:2597‐2608. 10.1002/jor.25293

Ari Väärälä and Victor Casula contributed equally to this manuscript.

REFERENCES

  • 1. Friedrich KM, Reiter G, Kaiser B, et al. High‐resolution cartilage imaging of the knee at 3T: Basic evaluation of modern isotropic 3‐D MR‐sequences. Eur J Radiol. 2011;78(3):398‐405. 10.1016/j.ejrad.2010.01.008 [DOI] [PubMed] [Google Scholar]
  • 2. Wang YXJ, Griffith JF, Ahuja AT. Non‐invasive MRI assessment of the articular cartilage in clinical studies and experimental settings. World J Radiol. 2010;2(1):44‐54. 10.4329/wjr.v2.i1.44 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Nieminen MT, Rieppo J, Töyräs J, et al. T2 relaxation reveals spatial collagen architecture in articular cartilage: a comparative quantitative MRI and polarized light microscopic study. Magn Reson Med. 2001;46(3):487‐493. 10.1002/mrm.1218 [DOI] [PubMed] [Google Scholar]
  • 4. Li X, Majumdar S. Quantitative MRI of articular cartilage and its clinical applications. J Magn Reson Imaging. 2013;38(5):991‐1008. 10.1002/jmri.24313 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Haralick RM, Shanmugam K, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern. 1973;SMC‐3(6):610‐621. 10.1109/TSMC.1973.4309314 [DOI] [Google Scholar]
  • 6. Soh LK, Tsatsoulis C. Texture analysis of SAR sea ice imagery using gray level co‐occurrence matrices. IEEE Transactions on Geoscience and Remote Sensing. 1999;37(2):780‐795. 10.1109/36.752194 [DOI] [Google Scholar]
  • 7. Clausi DA. An analysis of co‐occurrence texture statistics as a function of grey level quantization. Can. J. Remote Sens. 2002;28(1):45‐62. 10.5589/m02-004 [DOI] [Google Scholar]
  • 8. Castellano G, Bonilha L, Li LM, Cendes F. Texture analysis of medical images. Clin Radiol. 2004;59(12):1061‐1069. 10.1016/j.crad.2004.07.008 [DOI] [PubMed] [Google Scholar]
  • 9. Carballido‐Gamio J, Link TM, Majumdar S. New techniques for cartilage magnetic resonance imaging relaxation time analysis: Texture analysis of flattened cartilage and localized intra‐ and inter‐subject comparisons. Magn Reson Med. 2008;59(6):1472‐1477. 10.1002/mrm.21553 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Blumenkrantz G, Stahl R, Carballido‐Gamio J, et al. The feasibility of characterizing the spatial distribution of cartilage T2 using texture analysis. Osteoarthritis Cartilage. 2008;16(5):584‐590. 10.1016/j.joca.2007.10.019 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Carballido‐Gamio J, Stahl R, Blumenkrantz G, Romero A, Majumdar S, Link TM. Spatial analysis of magnetic resonance and relaxation times improves classification between subjects with and without osteoarthritis. Med Phys. 2009;36(9):4059‐4067. 10.1118/1.3187228 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Carballido‐Gamio J, Joseph GB, Lynch JA, Link TM, Majumdar S. Longitudinal analysis of MRI T2 knee cartilage laminar organization in a subset of patients from the osteoarthritis initiative: a texture approach. Magn Reson Med. 2011;65(4):1184‐1194. 10.1002/mrm.22693 [DOI] [PubMed] [Google Scholar]
  • 13. Joseph GB, Baum T, Carballido‐Gamio J, et al. Texture analysis of cartilage T2 maps: individuals with risk factors for OA have higher and more heterogeneous knee cartilage MR T2 compared to normal controls ‐ data from the osteoarthritis initiative. Arthritis Res Ther. 2011;13(5):R153. 10.1186/ar3469 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Joseph GB, Baum T, Alizai H, et al. Baseline mean and heterogeneity of MR cartilage T2 are associated with morphologic degeneration of cartilage, meniscus, and bone marrow over 3years – data from the Osteoarthritis Initiative. Osteoarthritis Cartilage. 2012;20(7):727‐735. 10.1016/j.joca.2012.04.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. de Oliveira MS, Betting LE, Mory SB, Cendes F, Castellano G. Texture analysis of magnetic resonance images of patients with juvenile myoclonic epilepsy. Epilepsy & Behavior. 2013;27(1):22‐28. 10.1016/j.yebeh.2012.12.009 [DOI] [PubMed] [Google Scholar]
  • 16. Baum T, Joseph GB, Nardo L, et al. Correlation of magnetic resonance imaging–based knee cartilage T2 measurements and focal knee lesions with body mass index: thirty‐six–month followup data from a longitudinal, observational multicenter study. Arthritis Care & Research. 2013;65(1):23‐33. 10.1002/acr.21741 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Urish KL, Keffalas MG, Durkin JR, Miller DJ, Chu CR, Mosher TJ. T2 texture index of cartilage can predict early symptomatic OA progression: data from the osteoarthritis initiative. Osteoarthritis Cartilage. 2013;21(10):1550‐1557. 10.1016/j.joca.2013.06.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Schooler J, Kumar D, Nardo L, et al. Longitudinal evaluation of T1ρ and T2 spatial distribution in osteoarthritic and healthy medial knee cartilage. Osteoarthritis Cartilage. 2014;22(1):51‐62. 10.1016/j.joca.2013.10.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Depeursinge A, Foncubierta‐Rodriguez A, De Ville V, Müller H. Three‐dimensional solid texture analysis in biomedical imaging: review and opportunities. Med Image Anal. 2014;18(1):176‐196. 10.1016/j.media.2013.10.005 [DOI] [PubMed] [Google Scholar]
  • 20. Williams A, Winalski CS, Chu CR. Early articular cartilage MRI T2 changes after anterior cruciate ligament reconstruction correlate with later changes in T2 and cartilage thickness. J Orthop Res. 2017;35(3):699‐706. 10.1002/jor.23358 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Peuna A, Hekkala J, Haapea M, et al. Variable angle gray level co‐occurrence matrix analysis of T2 relaxation time maps reveals degenerative changes of cartilage in knee osteoarthritis: Oulu knee osteoarthritis study. J Magn Reson Imaging. 2018;47(5):1316‐1327. 10.1002/jmri.25881 [DOI] [PubMed] [Google Scholar]
  • 22. Li X, Pai A, Blumenkrantz G, et al. Spatial distribution and relationship of T1ρ and T2 relaxation times in knee cartilage with osteoarthritis. Magn Reson Med. 2009;61(6):1310‐1318. 10.1002/mrm.21877 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Chen X, Wei X, Zhang Z, Yang R, Zhu Y, Jiang X. Differentiation of true‐progression from pseudoprogression in glioblastoma treated with radiation therapy and concomitant temozolomide by GLCM texture analysis of conventional MRI. Clin Imaging. 2015;39(5):775‐780. 10.1016/j.clinimag.2015.04.003 [DOI] [PubMed] [Google Scholar]
  • 24. Barros I, Peuna A, Casula V, Haapea M, Lammentausta E, Nieminen M Grey‐level Co‐Occurrence Matrix Texture Analysis of osteoarthritic knee cartilage on T2, Adiabatic T1ρ, Adiabatic T2ρ and Dual‐Echo Steady‐State Magnetic Resonance Imaging contrasts. In: Proc. Intl. Soc. Magn. Reson. Med 26(1381); 2018. [Google Scholar]
  • 25. Thakkar RS, Flammang AJ, Chhabra A, Padua A, Carrino JA. 3T MR imaging of cartilage using 3‐D dual echo steady state (DESS). MAGNETOM Flash. 2011;3:33‐36. [Google Scholar]
  • 26. Hardy PA, Recht MP, Piraino D, Thomasson D. Optimization of a dual echo in the steady state (DESS) free‐precession sequence for imaging cartilage. J Magn Reson Imaging. 1996;6(2):329‐335. 10.1002/jmri.1880060212 [DOI] [PubMed] [Google Scholar]
  • 27. Ruehm S, Zanetti M, Romero J, Hodler J. MRI of patellar articular cartilage: evaluation of an optimized gradient‐echo sequence (3‐D‐DESS). J Magn Reson Imaging. 1998;8(6):1246‐1251. 10.1002/jmri.1880080611 [DOI] [PubMed] [Google Scholar]
  • 28. Welsch GH, Scheffler K, Mamisch TC, et al. Rapid estimation of cartilage T2 based on double echo at steady state (DESS) with 3 Tesla. Magn Reson Med. 2009;62(2):544‐549. 10.1002/mrm.22036 [DOI] [PubMed] [Google Scholar]
  • 29. Panfilov E, Tiulpin A, Klein S, Nieminen MT, Saarakkala S Improving Robustness of Deep Learning Based Knee MRI Segmentation: Mixup and Adversarial Domain Adaptation. In: 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW). IEEE; 2019:450‐459. 10.1109/ICCVW.2019.00057 [DOI]
  • 30. Zaworski C, Cheah J, Koff MF, et al. MRI‐based texture analysis of trabecular bone for opportunistic screening of skeletal fragility. J Clin Endocrinol Metab. 2021;106(8):2233‐2241. 10.1210/clinem/dgab342 [DOI] [PubMed] [Google Scholar]
  • 31. Bay‐Jensen AC, Thudium CS, Mobasheri A. Development and use of biochemical markers in osteoarthritis: current update. Curr Opin Rheumatol. 2018;30:121‐128. 10.1097/BOR.0000000000000467 [DOI] [PubMed] [Google Scholar]
  • 32. Hunter DJ, Le Graverand MPH, Eckstein F. Radiologic markers of osteoarthritis progression. Curr Opin Rheumatol. 2009;21:110‐117. 10.1097/BOR.0b013e3283235add [DOI] [PubMed] [Google Scholar]
  • 33. Leyland KM, Hart DJ, Javaid MK, et al. The natural history of radiographic knee osteoarthritis: A fourteen‐year population‐based cohort study. Arthritis & Rheumatism. 2012;64(7):2243‐2251. 10.1002/art.34415 [DOI] [PubMed] [Google Scholar]
  • 34. Thorstensson CA, Andersson MLE, Jonsson H, Saxne T, Petersson IF. Natural course of knee osteoarthritis in middle‐aged subjects with knee pain: 12‐year follow‐up using clinical and radiographic criteria. Ann Rheum Dis. 2009;68(12):1890‐1893. 10.1136/ard.2008.095158 [DOI] [PubMed] [Google Scholar]
  • 35. Spector TD, Dacre JE, Harris PA, Huskisson EC. Radiological progression of osteoarthritis: an 11 year follow up study of the knee. Ann Rheum Dis. 1992;51(10):1107‐1110. 10.1136/ard.51.10.1107 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Lachance L, Sowers MF, Jamadar D, Hochberg M. The natural history of emergent osteoarthritis of the knee in women. Osteoarthritis Cartilage. 2002;10(11):849‐854. 10.1053/joca.2002.0840 [DOI] [PubMed] [Google Scholar]
  • 37. Cooper C, Snow S, McAlindon TE, et al. Risk factors for the incidence and progression of radiographic knee osteoarthritis. Arthritis & Rheumatism. 2000;43(5):995‐1000. 10.1002/1529-0131(200005)43:5<995::AID-ANR6>3.0.CO;2-1 [DOI] [PubMed] [Google Scholar]
  • 38. Schouten JSAG, van den Ouweland FA, Valkenburg HA. A 12 year follow up study in the general population on prognostic factors of cartilage loss in osteoarthritis of the knee. Ann Rheum Dis. 1992;51(8):932‐937. 10.1136/ard.51.8.932 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Burki V, Paternotte S, Dougados M, Ayral X. Rapidly destructive tibiofemoral knee osteoarthritis: clinicoradiological presentation and outcome after global medical treatment including non‐arthroscopic joint lavage plus corticosteroid injection. A single center retrolective study. Joint Bone Spine. 2014;81(2):142‐148. 10.1016/j.jbspin.2013.05.009 [DOI] [PubMed] [Google Scholar]
  • 40. Driban JB, Harkey MS, Barbe MF, et al. Risk factors and the natural history of accelerated knee osteoarthritis: a narrative review. BMC Musculoskelet Disord. 2020;21(1):332. 10.1186/s12891-020-03367-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Peterfy CG, Schneider E, Nevitt M. The osteoarthritis initiative: report on the design rationale for the magnetic resonance imaging protocol for the knee. Osteoarthritis Cartilage. 2008;16(12):1433‐1441. 10.1016/j.joca.2008.06.016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Kohn MD, Sassoon AA, Fernando ND. Classifications in brief: Kellgren‐Lawrence classification of osteoarthritis. Clin Orthop Relat Res. 2016;474(8):1886‐1893. 10.1007/s11999-016-4732-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Panfilov E, Tiulpin A, Nieminen MT, Saarakkala S, Casula V. Deep learning‐based segmentation of knee MRI for fully automatic subregional morphological assessment of cartilage tissues: data from the osteoarthritis initiative. J Orthop Res. Published online July 29 2021. 10.1002/jor.25150 [DOI] [PubMed] [Google Scholar]
  • 44. Hunter DJ, Guermazi A, Lo GH, et al. Evolution of semi‐quantitative whole joint assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score). Osteoarthritis Cartilage. 2011;19(8):990‐1002. 10.1016/j.joca.2011.05.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Dougherty R, Kunzelmann KH. Computing local thickness of 3‐D structures with image. J. Microscopy and Microanalysis. 2007;13(S02):1678‐1679. 10.1017/S1431927607074430 [DOI] [Google Scholar]
  • 46. Guilak F, Ratcliffe A, Lane N, Rosenwasser MP, Mow VC. Mechanical and biochemical changes in the superficial zone of articular cartilage in canine experimental osteoarthritis. J Orthop Res. 1994;12(4):474‐484. 10.1002/jor.1100120404 [DOI] [PubMed] [Google Scholar]
  • 47. Pratta MA, Su JL, Leesnitzer MA, et al. Development and characterization of a highly specific and sensitive sandwich ELISA for detection of aggrecanase‐generated aggrecan fragments. Osteoarthritis Cartilage. 2006;14(7):702‐713. 10.1016/j.joca.2006.01.012 [DOI] [PubMed] [Google Scholar]
  • 48. Kretzschmar M, Nevitt MC, Schwaiger BJ, Joseph GB, McCulloch CE, Link TM. Spatial distribution and temporal progression of T2 relaxation time values in knee cartilage prior to the onset of cartilage lesions – data from the Osteoarthritis Initiative (OAI). Osteoarthritis Cartilage. 2019;27:737‐745. 10.1016/j.joca.2018.10.016 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting information.

Click here for additional data file. (34.7MB, docx)

Supporting information.

Click here for additional data file. (334.3KB, docx)

Supporting information.

Click here for additional data file. (312.6KB, docx)

Supporting information.

Click here for additional data file. (288.2KB, docx)

Supporting information.

Click here for additional data file. (697.6KB, docx)

Articles from Journal of Orthopaedic Research are provided here courtesy of Wiley

RESOURCES