SEARCH

SEARCH BY CITATION

Keywords:

  • T2;
  • cartilage;
  • osteoarthritis initiative (OAI);
  • texture analysis;
  • magnetic resonance imaging (MRI)

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

Cartilage magnetic resonance imaging T2 relaxation time is sensitive to hydration, collagen content, and tissue anisotropy, and a potential imaging-based biomarker for knee osteoarthritis. This longitudinal pilot study presents an improved cartilage flattening technique that facilitates texture analysis using gray-level co-occurrence matrices parallel and perpendicular to the cartilage layers, and the application of this technique to the knee cartilage of 13 subjects of the osteoarthritis initiative at baseline, 1-year follow-up, and 2-year follow-up. Cartilage flattening showed minimum distortion (∼ 0.5 ms) of mean T2 values between nonflattened and flattened T2 maps. Gray-level co-occurrence matrices texture analysis of flattened T2 maps detected a cartilage laminar organization at baseline, 1-year follow-up, and 2-year follow-up by yielding significant (P < 0.05) differences between texture parameters perpendicular and parallel to the cartilage layers. Tendencies showed higher contrast, dissimilarity, angular second moment, and energy perpendicular to the cartilage layers; and higher homogeneity, entropy, variance, and correlation parallel to them. Significant (P < 0.05) longitudinal texture changes were also detected reflecting subtle signs of a laminar disruption. Tendencies showed decreasing contrast, dissimilarity, and entropy; and increasing homogeneity, energy, and correlation. Results of this study warrant further investigation to complete the assessment of the usefulness of the presented methodology in the study of knee osteoarthritis. Magn Reson Med, 2010. © 2010 Wiley-Liss, Inc.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

The osteoarthritis initiative (OAI) is a public-private study sponsored by the National Institutes of Health aimed to better understand how to prevent and treat knee osteoarthritis (OA). This multicenter study, which involves ∼5000 research subjects, highlights the need to develop biochemical, genetic, and imaging-based biomarkers for development and progression of OA, which is the most common form of arthritis and the major cause of activity limitation and physical disability in older people (http://www.oai.ucsf.edu/).

Among the imaging-based biomarkers to be investigated in the OAI is cartilage magnetic resonance imaging (MRI) T2 relaxation time. This tissue property has demonstrated to be sensitive to hydration, collagen content, and tissue anisotropy (1). Knee cartilage studies involving cartilage T2 relaxation times have been conducted based on full-thickness mean values (2–5), Z-score maps (6, 7), laminar approaches (5, 8–11), and texture analysis using gray-level co-occurrence matrices (GLCM) (11, 12) all showing promising results. Recent cross-sectional (11, 12) and longitudinal (5, 12, 13) studies have suggested that assessing T2 spatial and laminar distribution could be more sensitive than full-thickness mean T2 values in detecting cartilage degeneration. A new technique recently proposed in (14) consists in flattening the cartilage relaxation time maps and then performing GLCM texture analysis on them, thus yielding an analysis of the laminar spatial distribution of relaxation time values with second-order texture measures. The advantages of this technique are that extracted texture information has correspondence to the natural laminar organization of cartilage, and that texture information based on pixels farther apart is possible at orientations parallel to the cartilage layers (14). However, the technique has not yet been applied to any study involving patients with knee OA.

The hypothesis of this pilot study is that the cartilage T2 laminar organization of patients with knee OA undergoes changes over a period of 2 years, and that these changes can be quantified using flattening of the cartilage relaxation time maps and then performing GLCM texture analysis (14). We also hypothesize that this technique might be more sensitive to change than full-thickness T2 values. Therefore, this study has three goals. First, to present an improvement of the technique presented in (14) concerning matching between points in the bone-cartilage interface and points in the articular surface, which is an essential step for cartilage flattening. Second, to longitudinally (baseline, 1-year follow-up, and 2-year follow-up) evaluate the improved technique in a small subset of patients of the OAI and to assess its potential in detecting a knee cartilage laminar organization. Third, to assess the potential of the technique in detecting longitudinal changes of the T2 laminar organization of knee cartilage.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

Subjects

Thirteen subjects from the Progression Cohort of the OAI were included in this study. Subjects of this cohort were required that both of the following conditions were present in at least one knee at baseline:

  • 1
    Radiographic signs of OA defined as definite osteophytes based on posteroanterior fixed flexed radiographs (OARSI atlas grade ≥1). Subjects with OARSI grade of 3 (severe narrowing), however, were excluded due to expected severe cartilage loss.
  • 2
    Presence of frequent knee symptoms defined as pain (Western Ontario and McMaster University (WOMAC) Osteoarthritis Index pain score), aching or stiffness on most days of a month in past year.

Subjects had a body mass index mean ±standard deviation of 30.1 ± 3.7, mean ±standard deviation age of 55.7 ± 10.6 years, and radiographic Kellgren-Lawrence (KL) scores (15) of 2 (n = 7) and 3 (n = 6). KL = 2 is defined as definite osteophytes without narrowing of joint space; while KL = 3 is defined as moderate multiple osteophytes, definite narrowing of joint space, some sclerosis, and possible deformity of bone contour.

The study protocol, amendments, and informed consent documentation including analysis plans were reviewed and approved by the local institutional review boards. Data used in the preparation of this article were obtained from the OAI database, which is available for public access at http://www.oai.ucsf.edu/. Baseline clinical dataset 0.2.2 was used in this study.

Magnetic Resonance Imaging

Magnetic resonance images of the right knee were obtained at 3 Tesla (Siemens Magnetom Trio, Erlangen, Germany) using quadrature transmit-receive knee coils (USA Instruments, Aurora, OH). The right knee was chosen since this side included sequences for quantification of both cartilage morphology and T2 relaxation times. Sagittal three-dimensional (3D) double echo in steady state (DESS) images with water excitation (WE) were used for cartilage segmentation. Scanning parameters consisted of repetition time (TR) of 16.3 ms, echo time (TE) of 4.7 ms, bandwidth (BW) of 185 Hz/pixel, in-plane spatial resolution of 0.365 mm × 0.456 mm (0.365 mm × 0.365 mm after reconstruction), and slice thickness of 0.7 mm. Sagittal two-dimensional (2D) multiecho spin-echo (MESE) images were used for T2 quantification. Scanning parameters consisted of TR of 2700 ms, seven TEs (10, 20, 30, 40, 50, 60, and 70 ms), BW of 250 Hz/pixel, in-plane spatial resolution of 0.313 mm × 0.446 mm (0.313 mm × 0.313 mm after reconstruction), slice thickness of 3.0 mm, and 0.5 mm gap. Please refer to the report of Peterfy et al. (16) for further details of image acquisition parameters. Images used in this study are available for public access at http://www.oai.ucsf.edu/. The specific image datasets used in this study were 0.B.1, 1.B.1, and 3.B.1.

Image Analysis

Cartilage was segmented into six distinct compartments using a semiautomatic technique based on Bezier splines and edge detection (17). The cartilage compartments were: medial femoral condyle (MFC), lateral femoral condyle (LFC), trochlea (TRO), patella (PAT), medial tibia (MT), and lateral tibia (LT).

T2 maps at baseline, 1-year follow-up, and 2-year follow-up were created on a pixel-by-pixel basis assuming mono-exponential signal decay and by excluding the first echo (TE1 = 10 ms) using in-house developed software based on the Levenberg-Marquardt algorithm (5). The first echo was excluded since studies have suggested that by excluding it from a multiecho Carr-Purcell-Meiboom-Grill (CPMG) sequence minimizes error from stimulated echoes in calculated T2 values for cartilage (18, 19). The expression to be fitted was of the following form:

  • equation image(1)

In Eq. 1, S0 is the signal intensity when TE = 0 ms and S is the signal intensity in a T2-weighted image with a certain TE.

To correct for patient motion between DESS and T2 mapping scans, DESS images were 3D rigidly registered to the second echoes. Image registration was accomplished using in-house developed software written in MATLAB (The Mathworks, Natick, MA) based on normalized mutual information (20). Cartilage contours represented by Bezier splines in the DESS-image domain were converted to binary regions of interest (ROIs) representing knee cartilage. The calculated 3D registration matrix was then applied to the binary ROIs, and the aligned ROIs, now in the T2-map domain, were converted back to splines and automatically labeled as bone-cartilage interface and articular surface.

Cartilage was flattened as previously described in (14). The technique relies on nonlinear deformation of the cartilage shape. The geodesic length (λ) of the bone-cartilage interface is computed and points along this curve are uniformly sampled. Each sampled point in the bone-cartilage interface is then matched to a point in the articular surface and the corresponding thicknesses (t) at those locations are calculated. For each sampled point in the bone-cartilage interface and its matched articular point a target position is defined. The target positions of the bone-cartilage interface lay on a horizontal line (vertical for the PAT) with a length equal to the geodesic length previously computed. The target positions of the corresponding articular points are below this line for the femoral compartments (MFC, LFC, and TRO), above this line for the tibial compartments (MT and LT), and to the right of this line for the patellar compartment. The target positions of the articular points are at vertical distances (horizontal for the PAT) equal to the local thicknesses previously computed. Wendland's radial basis functions with local support (21) as suggested by Fornefett et al. (22) are then used to warp the cartilage maps to bring the original cartilage points to their corresponding target positions with the inherent constraints of preserving the geodesic length of the bone-cartilage interface and cartilage thickness. Image warping is applied with backward mapping and spline interpolation. Figure 1 visually demonstrates the cartilage flattening of a LFC compartment. In this figure the preservation of the length of the bone-cartilage interface (λ) and cartilage thicknesses (t) is easier to understand.

thumbnail image

Figure 1. Schematic representation of the cartilage flattening technique for the lateral femoral condyle. At the bottom of this figure, preservation of the length (λ) of the bone-cartilage interface, and preservation of 2D cartilage thickness (t) can be appreciated on the flattened map.

Download figure to PowerPoint

In (14), matching of bone-cartilage and articular surface points was accomplished with normal vectors to the bone-cartilage interface ending at the articular surface. The length of these vectors represented the 2D local cartilage thicknesses as represented in Figs. 2–5a for the different cartilage compartments investigated in this study. In this work, the matching was accomplished based on the Laplace's equation (23). The bone-cartilage interface was set to a fixed potential, and the articular surface to a higher potential. The Laplace's equation was numerically solved in two dimensions with Dirichlet boundary conditions in an iterative way:

  • equation image(2)
thumbnail image

Figure 2. Matching of bone-cartilage interface and articular cartilage points in the lateral femoral condyle and lateral tibia. a: Normal vectors to the bone-cartilage interface end at the articular surface. b: Streamlines resulting from a tangent field generated by numerically solving the Laplace's equation. Layers are color encoded to show the transition from the bone cartilage interface (dark) to the articular surface (bright). Vectors and streamlines have been sampled along the bone cartilage interface to facilitate their visualization. Black arrows indicate regions were normal vectors either crossed or were assigned to the same point.

Download figure to PowerPoint

thumbnail image

Figure 3. Matching of bone-cartilage interface and articular cartilage points in the patella. a: Normal vectors to the bone-cartilage interface end at the articular surface. b: Streamlines resulting from a tangent field generated by numerically solving the Laplace's equation. Layers are color encoded to show the transition from the bone cartilage interface (dark) to the articular surface (bright). Vectors and streamlines have been sampled along the bone cartilage interface to facilitate their visualization. Black arrow indicates a problematic matching.

Download figure to PowerPoint

thumbnail image

Figure 4. Matching of bone-cartilage interface and articular cartilage points in the trochlea. a: Normal vectors to the bone-cartilage interface end at the articular surface. b: Streamlines resulting from a tangent field generated by numerically solving the Laplace's equation. Layers are color encoded to show the transition from the bone cartilage interface (dark) to the articular surface (bright). Vectors and streamlines have been sampled along the bone cartilage interface to facilitate their visualization. Black arrows indicate regions were normal vectors either crossed or were assigned to the same point.

Download figure to PowerPoint

thumbnail image

Figure 5. Matching of bone-cartilage interface and articular cartilage points in the medial femoral condyle and medial tibia. a: Normal vectors to the bone-cartilage interface end at the articular surface. b: Streamlines resulting from a tangent field generated by numerically solving the Laplace's equation. Layers are color encoded to show the transition from the bone cartilage interface (dark) to the articular surface (bright). Vectors and streamlines have been sampled along the bone cartilage interface to facilitate their visualization. Black arrows indicate regions were normal vectors either crossed or were assigned to the same point.

Download figure to PowerPoint

In Eq. 2, ψi(x,y) is the value of the potential at position (x,y) during the ith iteration. Then normalized gradients of ψ were computed to produce a tangent vector field that was used to lead each point in the bone-cartilage interface to a point in the articular surface. The length of the streamlines was taken as the 2D local cartilage thicknesses. This matching strategy yields a better laminar partition since it provides 1-to-1 matching (23) between bone-cartilage interface and articular points, as well as streamlines that do not cross between each other, situations that are not guaranteed by the normal vector technique or by using a minimum Euclidean distance approach. In addition, by solving the Laplace's equation the streamline segments truly represent a laminar transition from the bone-cartilage interface to the articular surface. Figures 2–5b show examples of this point matching strategy for the different cartilage compartments investigated in this study. The corresponding streamlines in these figures are overlaid on the potential maps obtained by solving Eq. 2. The mean 2D cartilage thickness based on the lengths of streamlines was also calculated for each compartment. It should be noted that these values should not be confused with the actual cartilage thickness, which can be more accurately computed in 3D using the DESS-WE images. The mean streamline lengths were calculated to further help in the interpretation of GLCM texture analysis and they are summarized in Table 1.

Table 1. Mean 2D Cartilage Thickness Based On Streamline Lengths for Each Compartment and Time Point Expressed in Pixels
CompartmentBaseline1-year follow-up2-year follow-up
  1. MFC, medial femoral condyle; LFC, lateral femoral condyle; TRO, trochlea; PAT, patella; MT, medial tibia; LT, lateral tibia.

MFC6.8977.1406.790
LFC7.2097.3137.012
TRO8.1338.3028.266
PAT7.3637.3017.337
MT5.9005.9565.968
LT6.1606.6526.056

Texture analysis using GLCM (24) was applied to the flattened cartilage T2 maps. GLCM estimate the joint probability of a pixel intensity i and pixel intensity j (P(i,j)) for a given pixel spacing and direction. The pixel spacing is also known as offset and it is equivalent to the distance between the pixel-of-interest i and its neighbor j. Joint probabilities are calculated by constructing second-order histograms based on quantized images with N distinct gray levels, from which texture information can be extracted by computing statistical features. Since GLCM parameters consider the relationship between groups of two pixels instead of isolated properties such as mean and variance, they are classified as second-order texture measures. In this work, nine statistical features at five different spacings, and two different orientations were investigated. Pixels pairs with a noncartilage pixel neighbor were excluded from the GLCM. GLCM parameters included three from the contrast group: contrast, dissimilarity, and homogeneity; three from the orderliness group: angular second moment (ASM), energy, and entropy; and three from the stats group: mean, variance, and correlation. The orientations were parallel and perpendicular to the cartilage layers. The equation for GLCM entropy is shown below as just one example of a GLCM measure:

  • equation image(3)

Statistical Analysis

Statistical analysis was performed in MATLAB (The Mathworks, Natick, MA) and was done individually for each cartilage compartment as well as by pooling the compartments together using mean values. Results were considered significant if P values were less than 0.05.

The first analysis consisted in the evaluation of the cartilage flattening technique. In addition to the visual assessment of the point matching and preservation of patterns of T2 values, mean T2 values before and after flattening were compare using paired t tests at baseline, 1-year follow-up, and 2-year-follow-up.The magnitude of the T2 differences was also assessed with three different metrics:

  • 1
    Difference of mean values:
    • equation image(4)
  • 2
    Absolute difference of mean values
    • equation image(5)
  • 3
    Percent discrepancy
    • equation image(6)

The Pearson product-moment correlation coefficient was also computed and correlation plots with least-squares regression lines were inspected to measure the level of agreement between flattened and nonflattened T2 values.

The second analysis consisted of the longitudinal quantification of full-thickness T2 cartilage changes: baseline vs. 1-year follow-up, baseline vs. 2-year follow-up, and 1-year follow-up vs. 2-year follow-up. This analysis was done using repeated measures analysis of variance followed by Bonferroni corrections (25). The purpose of this analysis was to demonstrate the need of additional techniques for longitudinal cartilage T2 quantification in knee OA.

The third analysis consisted of comparing GLCM texture parameters that were parallel to the cartilage layers to those that were perpendicular to the cartilage layers. These comparisons were done individually at baseline, 1-year follow-up, and 2-year follow-up based on paired t tests. Comparisons were made to investigate the potential of GLCM texture parameters to detect a laminar organization on flattened T2 cartilage maps.

The fourth and last analysis quantified the longitudinal changes (baseline vs. 1-year follow-up, baseline vs. 2-year follow-up, and 1-year follow-up vs. 2-year follow-up) of texture parameters using repeated-measures analysis of variance followed by Bonferroni corrections (25). Changes were investigated in both orientations: parallel and perpendicular to the cartilage layers. The purpose of this analysis was to assess the potential of GLCM texture parameters to quantify cartilage degeneration on flattened T2 maps.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

Cartilage Flattening

Table 2 summarizes the flattening results at the compartmental level, while Table 3 summarizes the results at the knee level. T2 maps of the different cartilage compartments (n = 234) were successfully flattened using the new technique showing small mean T2 value changes when compared to nonflattened maps.

Table 2. Comparison of Mean T2 Values Between Nonflattened and Flattened Cartilage Maps for Each Compartment and Time Point (Mean ± Standard Deviation)
Time pointCompartment
MFCLFCTROPATMTLT
  1. MFC = Medial Femoral Condyle; LFC = Lateral Femoral Condyle; TRO = Trochlea; PAT = Patella; MT = Medial Tibia; LT = Lateral Tibia.

  2. P values were obtained by comparing mean T2 values before and after flattening using paired t tests.

Difference (ms) = T2-nonflattenedT2-flattened
Baseline−0.19 ± 0.62−0.23 ± 0.54−0.37 ± 0.82−0.21 ± 0.88−0.08 ± 0.63−0.06 ± 0.70
 P < 0.003P < 0.001P < 0.001P < 0.003  
1-Year−0.02 ± 0.50−0.18 ± 0.63−0.49 ± 0.99−0.25 ± 0.99−0.23 ± 0.93−0.13 ± 0.71
  P < 0.004P < 0.001P < 0.002P < 0.012 
2-Year−0.09 ± 0.72−0.26 ± 0.91−0.48 ± 0.87−0.25 ± 0.93−0.09 ± 0.74−0.13 ± 0.90
  P < 0.003P < 0.001P < 0.001  
Absolute difference (ms) = |T2-nonflattenedT2-flattened|
Baseline0.46 ± 0.460.46 ± 0.360.65 ± 0.620.56 ± 0.710.48 ± 0.420.49 ± 0.51
 P < 0.003P < 0.001P < 0.001P < 0.003  
1-Year0.35 ± 0.360.46 ± 0.470.69 ± 0.870.69 ± 0.760.58 ± 0.760.48 ± 0.55
  P < 0.004P < 0.001P < 0.002P < 0.012 
2-Year0.46 ± 0.560.55 ± 0.770.62 ± 0.760.65 ± 0.710.54 ± 0.510.57 ± 0.71
  P < 0.003P < 0.001P < 0.001  
Percent discrepancy (%) = 100 * (T2-nonflattenedT2-flattened)/T2-non-flattened
Baseline−0.36 ± 1.16−0.46 ± 1.10−0.73 ± 1.59−0.48 ± 2.03−0.19 ± 1.42−0.20 ± 1.69
 P < 0.003P < 0.001P < 0.001P < 0.003  
1-Year−0.03 ± 0.98−0.32 ± 1.18−0.98 ± 1.99−0.54 ± 2.03−0.57 ± 2.34−0.32 ± 1.57
  P < 0.004P < 0.001P < 0.002P < 0.012 
2-Year−0.22 ± 1.65−0.53 ± 1.77−0.94 ± 1.60−0.54 ± 1.89−0.24 ± 1.67−0.36 ± 2.05
  P < 0.003P < 0.001P < 0.001  
Table 3. Global Comparisons of Mean T2 Values Between Nonflattened and Flattened Cartilage Maps for Each Time Point (Mean ± Standard Deviation)
 Baseline1-year follow-up2-year follow-up
  1. P values were obtained by comparing mean T2 values before and after flattening using paired t tests.

Difference (ms) = T2-nonflattenedT2-flattened−0.19 ± 0.73−0.22 ± 0.84−0.21 ± 0.86
 P < 0.001P < 0.001P < 0.001
Absolute difference (ms) = |T2-nonflattenedT2-flattened|0.52 ± 0.550.55 ± 0.670.57 ± 0.68
 P < 0.001P < 0.001P < 0.001
Percent discrepancy (%) = 100 * (T2-nonflattenedT2-flattened)/T2-nonflattened−0.40 ± 1.58−0.46 ± 1.78−0.47 ± 1.80
 P < 0.001P < 0.001P < 0.001

All correlations at the compartmental and knee level yielded values larger than 0.98. Figures 6–8 show scatter plots of mean flattened and nonflattened T2 values at the knee level for the three time points. In these plots the solid lines represent the least-squares regression lines, which slopes (m) and intercepts (b) were very close to 1 and 0 ms, respectively.

thumbnail image

Figure 6. Scatter plot of mean T2 values of flattened and nonflattened cartilage maps with least-squares regression line for all regions of interest at baseline.

Download figure to PowerPoint

thumbnail image

Figure 7. Scatter plot of mean T2 values of flattened and nonflattened cartilage maps with least-squares regression line for all regions of interest at 1-year follow-up.

Download figure to PowerPoint

thumbnail image

Figure 8. Scatter plot of mean T2 values of flattened and nonflattened cartilage maps with least-squares regression line for all regions of interest at 2-year follow-up.

Download figure to PowerPoint

A global difference of −0.21 ±ms, absolute difference of 0.55 ±0.64 ms, and percent discrepancy of −0.44% ±1.72% was observed, with a global correlation of 0.994 (m = 0.99 and b = 0.25 ms) between mean nonflattened and flattened cartilage T2 values.

Figures 2–5 demonstrate the improvement of the Laplace's equation approach over the normal vector approach for the matching of points between the bone-cartilage interface and articular surface at allcartilage compartments; while Figs. 9–11 show side-to-side examples of nonflattened and flattened cartilage T2 maps using the new technique, where preservation of the laminar cartilage T2 patterns can be appreciated in the flattened maps of all compartments.

thumbnail image

Figure 9. Comparison of nonflattened and flattened cartilage T2 maps in the lateral femoral condyle.

Download figure to PowerPoint

thumbnail image

Figure 10. Comparison of nonflattened and flattened cartilage T2 maps in the trochlea.

Download figure to PowerPoint

thumbnail image

Figure 11. Comparison of nonflattened and flattened cartilage T2 maps in the medial tibia.

Download figure to PowerPoint

Full-Thickness T2 Changes

Individual cartilage compartments showed no significant longitudinal changes of full-thickness mean T2 values. In this study, as previously mentioned above, all compartments were also pooled, however the global longitudinal analysis showed no significant full-thickness T2 changes either.

Laminar Organization

Comparisons of GLCM texture parameters that were parallel to the cartilage layers to those that were perpendicular to the cartilage layers demonstrated a T2 laminar organization at baseline, 1-year follow-up, and 2-year follow-up by at least one GLCM texture parameter of each category (contrast, orderliness, and stats) in all compartments, individually and as a whole.

Because of the large amount of data quantitative results of the whole knee will be presented but tendencies of individual compartments only will be mentioned. Table 4 summarizes the GLCM texture laminar analysis of flattened cartilage T2 maps of the whole knee.

Table 4. Global Comparison of GLCM Texture Measures that are Parallel to those that are Perpendicular to the Cartilage T2 Layers
Texture parameterOffsetOrientation with higher values
Baseline1-year2-yearBaseline1-year2-year
  1. Offsets that were not significant were omitted; n-m signifies all offsets from n to m.

  2. Values were significantly different with P < 0.05.

  3. ⟂, signifies perpendicular; =, signifies parallel; ASM, angular second moment.

ContrastContrast1–51–51–5
GroupDissimilarity1–51–51–5
 Homogeneity1–51–51–5===
OrderlinessASM1, 2–52–52–5=, ⟂
GroupEnergy1, 2–51, 2–51, 2–5=, ⟂=, ⟂=, ⟂
 Entropy2–51, 2–52–5=⟂, ==
StatsMean111===
GroupVariance1–3, 51–3, 51–3, 5=, ⟂=, ⟂=, ⟂
 Correlation1–51–51–5===

The most consistent results were those of the GLCM contrast group showing significantly higher contrast and dissimilarity in the perpendicular direction, and significantly higher homogeneity parallel to the cartilage layers at all offsets and compartments, individually and as a whole (except contrast and dissimilarity in the PAT at baseline with 1-pixel offset).

Texture parameters of the orderliness group also showed a consistent tendency at baseline, 1-year follow-up, and 2-year follow-up. At 1-pixel offset higher ASM and energy parallel to the cartilage layers, and higher entropy perpendicular to them was observed (except in some instances in the PAT at 1-pixel offset). The opposite pattern was obtained at higher pixel offsets.

With respect to GLCM texture parameters of the stats group, mean did not show a tendency in all compartments, however all compartments showed a tendency towards higher variance and correlation parallel to the cartilage layers at baseline, 1-year follow-up, and 2-year follow-up.

Longitudinal Laminar Changes

No significant longitudinal changes of full-thickness mean T2 values were detected, and a laminar organization was identified at the three time points (baseline, 1-year follow-up, and 2-year follow-up) in all compartments, both individually and as a whole. However, longitudinal GLCM texture analysis detected significant changes. Tables 5 and 6 summarize the longitudinal analysis of GLCM texture parameters of flattened cartilage T2 maps at the compartmental and whole knee level, respectively.

Table 5. Compartmental Longitudinal Changes of GLCM Texture Measures of Flattened Cartilage T2 Maps for All Orientations and Offsets
Texture parameterBaseline vs. 1-yearBaseline vs. 2-year1-year vs. 2-year
OffsetOrientationΔOffsetOrientationΔOffsetOrientationΔ
  1. Offsets and directions that were not significant were omitted; n-m signifies all offsets from n to m.

  2. Values were significantly different with P < 0.05.

  3. ⟂, signifies perpendicular; =, signifies parallel; [DOWNWARDS ARROW], signifies longitudinal decrease; [UPWARDS ARROW], signifies longitudinal increase; MFC, medial femoral condyle; LFC, lateral femoral condyle; TRO, trochlea; MT, medial tibia; LT, lateral tibia; ASM, angular second moment.

MFC
ASM      3–5=[UPWARDS ARROW]
Energy      2[UPWARDS ARROW]
       3–5=[UPWARDS ARROW]
Entropy      1–2= and ⟂[DOWNWARDS ARROW]
       3–5=[DOWNWARDS ARROW]
LFC
ASM2–5[UPWARDS ARROW]      
Energy5[UPWARDS ARROW]      
Correlation1=[UPWARDS ARROW]      
TRO
Homogeneity1= and ⟂[UPWARDS ARROW]      
Correlation1= and ⟂[UPWARDS ARROW]      
 2=[UPWARDS ARROW]      
MT
Contrast1–3= and ⟂[DOWNWARDS ARROW]1–3= and ⟂[DOWNWARDS ARROW]   
 4–5=[DOWNWARDS ARROW]4–5=[DOWNWARDS ARROW]   
Dissimilarity1–3= and ⟂[DOWNWARDS ARROW]1–4= and ⟂[DOWNWARDS ARROW]   
 4–5=[DOWNWARDS ARROW]5=[DOWNWARDS ARROW]   
Homogeneity2,4=[UPWARDS ARROW]2[UPWARDS ARROW]   
    4=[UPWARDS ARROW]   
Variance4[DOWNWARDS ARROW]1–5= and ⟂[DOWNWARDS ARROW]   
Correlation1–3=[UPWARDS ARROW]      
LT
Contrast3[UPWARDS ARROW]      
 4–5=[UPWARDS ARROW]      
Energy      2, 5=[UPWARDS ARROW]
       3= and ⟂[UPWARDS ARROW]
Entropy      1–3, 5= and ⟂[DOWNWARDS ARROW]
       4=[DOWNWARDS ARROW]
Variance1–4= and ⟂[UPWARDS ARROW]      
 5=[UPWARDS ARROW]      
Table 6. Global Longitudinal Changes of GLCM Texture Measures of Flattened Cartilage T2 Maps for All Orientations and Offsets
Texture parameterBaseline vs. 1-yearBaseline vs. 2-year1-year vs. 2-year
OffsetOrientationΔOffsetOrientationΔOffsetOrientationΔ
  1. Offsets and orientations that were not significant were omitted; n-m signifies all offsets from n to m.

  2. Values were significantly different with P < 0.05.

  3. ⟂, signifies perpendicular; =, signifies parallel; [DOWNWARDS ARROW], signifies longitudinal decrease; [UPWARDS ARROW], signifies longitudinal increase; ASM, angular second moment.

Contrast1–3=[DOWNWARDS ARROW]1= and ⟂[DOWNWARDS ARROW]   
Dissimilarity1–3=[DOWNWARDS ARROW]1= and ⟂[DOWNWARDS ARROW]   
Homogeneity1= and ⟂[UPWARDS ARROW]2[UPWARDS ARROW]4[UPWARDS ARROW]
 2–3, 5=, =[UPWARDS ARROW]      
ASM         
Energy   2–3[UPWARDS ARROW]4[UPWARDS ARROW]
    5=[UPWARDS ARROW]   
Entropy   1–2= and ⟂[DOWNWARDS ARROW]1–4= and ⟂[DOWNWARDS ARROW]
    3–5=[DOWNWARDS ARROW]5=[DOWNWARDS ARROW]
Mean         
Variance         
Correlation1–2= and ⟂[UPWARDS ARROW]1= and ⟂[UPWARDS ARROW]   
 3–5=[UPWARDS ARROW]2=[UPWARDS ARROW]   

The MT showed early significant changes which become permanent. This was demonstrated by significant changes of GLCM texture parameters from baseline to 1-year follow-up and from baseline to 2-year follow-up. Tendencies indicated a decrease in contrast and dissimilarity, an increase in homogeneity, and a decrease in variance. Most changes were observed parallel and perpendicular to the cartilage layers.

At the whole knee level, early changes were in general consistent with respect to the MT showing a decrease in contrast and dissimilarity mainly parallel to the cartilage layers, and an increase in homogeneity mainly perpendicular to the cartilage layers. Early changes were also demonstrated as an increase in correlation in both orientations. Late changes were observed as manifested by changes of GLCM texture parameters from baseline to 2-year follow-up and from 1-year to 2-year follow-up. Tendencies indicated an increase in energy mainly perpendicular to the cartilage layers, and a decrease in entropy parallel and perpendicular to the cartilage layers. Homogeneity showed a continuous increase in the perpendicular direction.

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES

In this pilot study the potential of a new T2 cartilage quantification technique for knee OA has been demonstrated. The presented technique flattens cartilage T2 relaxation time maps to analyze the laminar organization with second-order texture features. Cartilage flattening was validated visually and quantitatively. Texture parameters detected a T2 cartilage laminar organization in a subset of knee OA patients of the OAI at three different time points (baseline, 1-year follow-up, and 2-year follow-up), as well as longitudinal laminar T2 cartilage changes, which were not demonstrated by longitudinal full-thickness T2 analysis.

Although the in vivo MRI detection of a laminar organization in knee cartilage T2 maps was reported many years ago (8) as well as its relationship with age (26, 27) and exercise (27, 28), the number of publications addressing this topic has recently increased (5, 9–11, 29). Special attention has been given to knee cartilage repair tissue (9, 10, 29), cross-sectional (7, 11) and longitudinal knee OA studies (5), and initial steps on hip cartilage have also been reported (30, 31). Furthermore, laminar organization in T2 knee cartilage maps has motivated T laminar studies in the OA field as well (3, 7, 31). Currently, there are three main laminar analysis techniques that have been reported in the literature: (a) those based on line profiles (8, 14, 26), (b) those based on individual mean T2 values of inner and outer layers (5, 9–11, 29), and (c) those based on composite measures of mean T2 values of inner and outer layers (5). All these strategies have shown potential for being more sensitive to T2 cartilage changes than full-thickness mean values. However, the need for techniques for early detection of cartilage degeneration and its longitudinal quantification based on T2 values still prevails, and motivated this study.

A key component of the proposed T2 cartilage quantification technique is flattening of cartilage ROIs, where point matching between the bone-cartilage interface and the articular surface plays a central role. Two desired characteristics of a point matching technique are one-to-one matching and no crossing of the lines connecting corresponding points. Since point matching based on normal vectors does not guarantee the above characteristics, we replaced it by a point matching based on the Laplace's equation, which clearly performed better as demonstrated in Figs. 2–5. Flattened maps using this new approach were qualitatively and quantitatively validated demonstrating preservation of T2 cartilage patterns (Figs. 9–11) and mean values (Tables 2 and 3). Although in some instances the mean T2 values of flattened cartilage maps were significantly different than the mean T2 values of nonflattened maps, the differences were very small and could be considered quite good in terms of T2 reproducibility of knee cartilage studies reported in the literature (32). Furthermore the very high correlations (∼ 0.99) between mean T2 values of flattened and mean T2 values of nonflattened cartilage maps, and least-squares regression lines with slopes very close to 1 and intercepts very close to 0 ms, indicate the very high level of agreement between flattened and nonflattened cartilage T2 maps.

Longitudinal analysis of full-thickness T2 values did not show significant cartilage changes which further motivated the other two experiments. First, the ability of second-order texture measures to detect a T2 cartilage laminar organization. Second, the potential of second-order texture measures to detect longitudinal changes of laminar T2values.

Second-order texture parameters of the three categories (contrast, orderliness, and stats) based on GLCM detected a cartilage T2 laminar organization at the three time points (baseline, 1-year follow-up, and 2-year follow-up). Trends were as expected. Higher contrast and dissimilarity perpendicular to the cartilage layers, and higher homogeneity parallel to them at all offsets and compartments, indicate that is more likely to find pairs of pixels with very different T2 values perpendicular to the cartilage layers, and pairs of pixels with very similar T2values parallel to them. These results are perhaps easy to understand since in vivo (8, 18, 26) and ex vivo studies (3) have reported a smooth increase of T2 knee cartilage values from the bone-cartilage interface towards the articular surface.

High values of GLCM ASM and energy indicate that certain co-occurrences of T2 values are dominant, while higher values of GLCM entropy indicate that the distribution of co-occurrences of T2 values is flatter. Numbers of the orderliness features showed one pattern for pixels that wereimmediate neighbors and another pattern for higher pixel spacings. The pattern at higher offsets suggested a laminar organization since higher energy and ASM was observed perpendicular to the cartilage layers, while higher entropy was observed parallel to them.

GLCM results of the stats category indicate that there is a higher dispersion of T2 co-occurrences around the GLCM mean (pixel offset <5) and a higher linear dependency between neighboring pixels parallel to the cartilage layers. GLCM mean did not show a clear tendency since this parameter measures the degree of pixel co-occurrences involving high relaxation times, which occur at both orientations. This measure may be of better value in knee-OA cross-sectional studies (11) since it has been shown that knee-OA patients have longer mean T2 cartilage values than normal controls (3, 4, 11, 33).

The fact that some tendencies in Table 4 are not fulfilled at 5-pixel offsets (GLCM variance) may be due to mean 2D cartilage thicknesses (Table 1) that indicate that some compartments such as the medial and lateral tibia might not be suitable for high offsets at orientations perpendicular to the cartilage layers. In fact, only the medial (1-year follow-up and 2-year follow-up) and lateral tibia (baseline) showed this discrepancy at 5-pixel offset. Small discrepancies were also seen at 1-pixel offset (orderliness measures in the PAT) for unknown reasons. Because of the small number of subjects involved in this study as well as the heterogeneity of the pathology in question discrepancies were actually expected.

Longitudinal changes of second-order texture parameters at the compartmental level (Table 5) showed the most interesting results of this study. The MT showed early significant changes from baseline to 1-year follow-up which were still observed at 2-year follow-up. Changes consisted in decreases in contrast and dissimilarity perpendicular to the cartilage layers at small pixel offsets (≤3), decreases of the same parameters parallel to the cartilage layers at all offsets, and increases in homogeneity parallel to the cartilage layers at 4-pixel offsets. These findings are in agreement with another recent pilot OAI study (5) where it was suggested that cartilage degeneration might cause T2 cartilage changes that lead to T2 cartilage layers with mean values that are more similar between them. Furthermore, in that study a laminar change was also seen in the MT, although it was not detected until 2-year follow-up. The decrease in variance perpendicular to the cartilage layers at 4-pixel offset was relatively unexpected since at this pixel offset variance did not detect a laminar organization. In the MT variance was higher parallel to the cartilage layers at 1- and 2-pixel offsets (baseline, 1-year follow-up, and 2-year follow-up) and lower perpendicular to the cartilage layers at 5-pixel offsets (1-year follow-up and 2-year follow-up).

Longitudinal changes of second-order texture measures at the knee level (Table 6) also showed interesting results. At this level three different patterns were observed: (1) continuous changes, (2) early changes that persisted at 2-year follow-up, and (3) late changes.

Homogeneity showed the most interesting finding at the knee level with a continuous increase from baseline to 1-year follow-up and from 1-year follow-up to 2-year follow-up perpendicular to the cartilage layers. Even more interesting was the fact that these increases were seen gradually at higher pixel offsets: 1, 2, and 4. These changes again suggest the idea of T2 cartilage layers becoming more homogeneous as knee OA progresses.

Early changes from baseline to 1-year follow-up that persisted at 2-year follow-up were seen in contrast and stats texture features. From the contrast group a decrease in contrast and dissimilarity parallel to the cartilage layers and consistently at 1-pixel offset was observed. From the stats group increases in correlation perpendicular to the cartilage layers at 1-pixel offset, and parallel to the cartilage layers at 1- and 2-pixel offset was observed. Although the linear dependency along the layers seem to remain strong, early changes across the layers were also seen, which together with the changes of the texture features of the contrast category indicate laminar disruption.

Late global changes were observed with orderliness measures. Consistent results were seen for entropy with decreases parallel to the cartilage layers at all pixel offsets, and decreases perpendicular to the cartilage layers at 1- and 2-pixel offsets. Simultaneous changes were observed for energy with increases perpendicular to the layers at gradually larger offsets. The decreases in entropy along the layers are in agreement with the idea of disruption of the cartilage layers. In a previous study by Blumenkrantz et al. (12) changes between baseline and 9-month follow-up were also seen as decreases in GLCM entropy. The authors speculated that possible reasons were cartilage swelling in the early stages of OA, or short-term changes in disease progression, demonstrating the heterogeneity of the pathology. Although a direct comparison to the study of Blumenkrantz et al. (12) is inadequate due to differences in orientation of texture measures and follow-up times, the trends are similar.The increases in energy perpendicular to the layers are more difficult to comprehend, however they were not observed at all pixel offsets, in fact at 2-year follow-up they were only seen at 4-pixel offset, i.e. mainly between the deep radial and superficial layer.

The main limitation of this study is the small number of subjects, which implies that the interpretations given above are not definite, requiring studies with larger patient samples to be confirmed. Such studies are currently in progress. This study presents three more limitations. First short T2 cartilage components, which are important for normal cartilage homeostasis, might not be fully represented since the authors dropped the first echo for T2 fitting. Second, at the time this work was done, no phantom studies were available to the public validating the T2 sequences, which would have provided material to choose between T2 fittings using all echoes and those discarding the first one. The third additional limitation was the lack of images to test reproducibility of full-thickness T2 values and GLCM parameters.

In conclusion, in this pilot study we have presented and validated an improved cartilage flattening technique that facilitates T2 laminar analysis with GLCM texture parameters. Second-order texture measures detected both a laminar organization and longitudinal laminar changes in patients with knee OA. Longitudinal differences were seen as continuous changes, early changes that persisted, and late consistent changes. The most interesting finding was the early detection of reasonable changes of texture parameters in the MT which persisted until 2-year follow-up. Texture changes in general indicated subtle signs of a possible longitudinal cartilage laminar disruption not detected by longitudinal analysis of full-thickness mean T2 values. Based on the current results, future longitudinal and cross-sectional studies are warranted to further validate the presented texture laminar analysis technique.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. REFERENCES