Texture Analysis of Cartilage Repair Tissue Maturation: Comparison of Two Cartilage Repair Methods and Correlation with MOCART 2.0

Abstract

Objective

The objective of this study was to assess the maturation of matrix-associated autologous chondrocyte transplantation (MACT) grafts up to 2 years after the surgery using gray-level co-occurrence matrix (GLCM) texture analysis of quantitative T₂ maps, compare the results with the microfracturing technique (MFX) control group, and relate these results to the morphological MOCART 2.0 score.

Design

A subcohort of 37 patients from prospective, multi-center study underwent examination on a 3T MR scanner, including a T₂ mapping sequence at 3, 12, and 24 months after surgery. Changes between the time-points in the mean T₂ values and 20 GLCM features extracted from T₂ maps were assessed in repair tissue, tissue adjacent to the repair site, and the reference cartilage for both procedures.

Results

Significant correlations were found between the MOCART 2.0 and GLCM features for both surgical procedures. There were no significant differences between MACT and MFX. We identified significant intra-group changes in T₂ and autocorrelation (3M-12M: P = 0.002; 3M-24M: P = 0.004), dissimilarity (3M-24M: P = 0.01), homogeneity (3M-24M: P = 0.013), and correlation (3M-24M: P = 0.036), sum average (3M-12M: P = 0.001; 3M-24M: P = 0.002), and information measure (3M-24M: P < 0.001) in the MACT repair tissue. MACT models revealed differences in GLCM between all combinations of ROI types at almost all time-points. In the case of MFX, the significant differences were mainly between repair and reference tissue at 12 months.

Conclusion

Texture analysis provides a useful extension to T₂ mapping. Texture features are correlated to the morphological outcome and reveal differences in the process of maturation between MACT and MFX.

Keywords

Articular Cartilage Tissue Magnetic resonance imaging Diagnostics Knee Joint involved Cartilage repair Repair Radiomics

Introduction

The articular cartilage, a highly specialized tissue that facilitates joint motion, has a limited capacity for self-healing because it is an avascular and aneural tissue.^1-3 Defects resulting from trauma can result in debilitating joint conditions, such as early osteoarthritis (OA).⁴ For young adults, joint replacement is not an option, as prostheses can wear out early after approximately 10 years.⁵

In cases where conservative treatment is not effective and lesions are small, minimally invasive techniques based on the stimulation of bone marrow cells, such as microfracturing (MFX) or cartilage restoration techniques, such as matrix-associated autologous chondrocyte transplantation (MACT), are used.^6-10 Previous research has indicated that the outcome of cartilage repair surgery can vary from hyaline-like tissue to mixed fibrous-hyaline tissue or fibrous tissue with poorer bio-mechanical qualities, which are typically seen after MFX,^11-13 but problems or suboptimal results can arise from both MFX and MACT.^14,15

To avoid invasive procedures during the monitoring of patients after cartilage repair surgery, the Magnetic Resonance Observation of Cartilage Repair Tissue (MOCART)¹⁶ score and MOCART 2.0¹⁷ score have been introduced. These scores evaluate the quality of cartilage repair by examining factors such as surface of the repair tissue, subchondral bone, defect filling, repair tissue signal intensity, and integration with the surrounding cartilage. However, the gray-level co-occurrence matrix (GLCM)¹⁸ of collagen-specific T₂ maps is currently being explored as a quantitative tool with which to assess cartilage repair tissue maturation.

The GLCM identifies the co-occurrence of signal intensities at a designated offset within an image. The matrix then allows the extraction of various quantitative features. These GLCM features (second-order image statistics) reveal linear and quadratic relationships among pixel pairs. As degradation of cartilage matrix leads to increased water mobility and subsequent T₂ elevation,¹⁹ differences in T₂ distributions are often visible on T₂ maps. GLCM offers new insights into cartilage structure when combined with T₂ values of cartilage regions. GLCM has already been explored in recent years in the context of detecting early osteoarthritic (OA) changes,^19-27,28 or in cartilage maturation assessment.²⁹

Because GLCM features are handcrafted, calculation parameters must be based on expertise and the problem at hand. The displacement distance (d) determines the distance between neighboring pixels,¹⁸ and quantization into eight, 16, or 32 gray levels simplifies the calculation. It has been shown that 16 gray levels are usually adequate.³⁰ Considering the nonhomogeneous cartilage structure, the angle of GLCM calculation is crucial. This was addressed in an earlier study.³¹ To address nonrectangular ROIs, studies have implemented ROI flattening for GLCM reproducibility.^25,32

The objective of this study was to assess MACT graft maturation up to 2 years after the surgery using GLCM texture analysis of quantitative T₂ maps, compare the results to the MFX control group, and relate these results to the morphological MOCART 2.0 score.

Methods

Study Design and Patient Baseline Characteristics

Ethical approvals for this study were obtained separately for each individual participating and each participant gave written, informed consent. This article covers a prospective, multi-center, randomized, controlled, open-label (blinded MRI reading), phase III study comparing the efficacy and safety of MACT using NOVOCART 3D plus (TETEC—Tissue Engineering Technologies AG, Reutlingen, Germany) versus MFX in patients with focal cartilage defects of the knee. The clinical trial was registered under EudraCT number 2011-005798-22.

Patients, between 18 and 65 years of age, or minors at least 14 years old with a closed epiphyseal growth plate, were randomly assigned to either the MACT or the MFX group in a 2:1 ratio. Eligible participants included both males and females with localized articular cartilage defects in the femoral condyle or trochlea of the knee, graded as III or IV according to the International Cartilage Repair Society (ICRS) classification. Enrollment criteria specified a maximum of two defects, with the total defect size restricted to a maximum of 6 cm² and a minimum of 2 cm². In case of a pediatric patient (14-17 years old), the confirmation of closure of epiphyseal growth plate of the index knee by x-ray or MRI was required. The 2:1 randomization ratio was specified in the original clinical trial registration.

The overall patient cohort (n = 262) comprised 189 males (72.1%) and 73 females (27.9%). The average age of the patients was 39.9 ± 10.6 years. The majority of participants (233 patients, 88.9%) had a solitary lesion, while 29 patients (11.1%) presented with two lesions. The majority of cartilage lesions in both treatment groups originated from trauma (77.4% in the MFX group and 78.3% in the MACT group). Both the total lesion size (MFX group: 3.6 ± 1.3 cm²; MACT group: 3.8 ± 1.4 cm²) and the mean size of the larger lesion (MFX group: 3.5 ± 1.2 cm²; MACT group: 3.6 ± 1.2 cm²) were comparable between the two treatment groups. The MRI substudy comprised 110 patients, (MACT: 75 patients; MFX: 35 patients).

MRI Examination

Patients underwent MRI examinations at 3 months (3M), 12 months (12M), and 24 months (24M) after surgery. Only 27 MACT and 10 MFX patients had T₂ maps available for all three time-points. The 3T MRI protocol consisted of morphological imaging protocols (three-dimensional proton density-weighted spin-echo sequence, and two-dimensional proton-density, T₁- and T₂-weighted fast-spin echo sequences) and sagittal T₂ mapping multi-echo spin echo sequences (slice thickness: 3 mm; slice spacing: 3.3 mm; repetition time: 2000 ms; echo times (ms): 12.5, 25, 37.5, 50, 62.5, 75, 87.5, 100; averages: 1; flip angle: 90° and 180°; acquisition matrix: 320 × 256; field of view: 16 cm × 16 cm; total acquisition time: 10 minutes 36 seconds). T₂ mapping was performed off-line in MATLAB R2019a (MathWorks, Natick, MA) using mono-exponential, two-parametric least-squares decay fitting.

Evaluation of MRI Images

Morphological images were evaluated by an expert radiologist (S.T.) using the MOCART 2.0¹⁷ score for semi-quantitative assessment of repair tissue. This score regions-of-interest (ROI) were also defined by the same radiologist on two or three consecutive slices on T₂ mapping sequence (usually on third echo) images using JiveX (Visus, Bochum, Germany). The number of slices varied according to the extent of the repair tissue. For each slice, repair cartilage, cartilage adjacent to the repair cartilage, and a reference cartilage were defined (see Fig. 1 ). An adjacent ROI was always positioned next to the repair ROI. Reference ROI was placed at least 10 mm from the edge of the repair tissue. ROIs were positioned in the same area on the cartilage at all time-points. ROIs were then transferred onto corresponding slices of T₂ maps for processing using MATLAB R2019.a (MathWorks, Natick, MA), as illustrated in Fig. 1 . Mean T₂ was calculated and averaged through the slices, ROI-wise, resulting in three T₂ values per patient per time-point (one for repair cartilage, one for adjacent cartilage, and one for reference cartilage).

Figure 1.

(A) Regions of interest (ROIs) selection. Each tissue type is delineated by ROI bounds and arrows. Red: repair tissue, yellow: adjacent tissue, green: reference tissue. (B) ROI analysis pipeline.

GLCM Analysis

Only ROIs with more than 15 pixels were analyzed with GLCM. Using an in-house-written script in MATLAB, ROIs were rotated (built-in MATLAB functions “regionprops” and “imrotate”), flattened and quantized into 16 gray levels and consecutive GLCM analysis was computed with an offset of 0° (direction parallel to the cartilage surface) and a step of length 1 (considering a pixel and its immediate neighbor), as illustrated in Fig. 1 . Using the GLCM_features1 function from the MATLAB Repository,³³ 20 quantitative features were extracted (autocorrelation, cluster prominence, cluster shade, contrast, correlation, difference entropy, difference variance, dissimilarity, energy, entropy, homogeneity, information measure, information measure of correlation 2, inverse difference moment normalized, inverse difference normalized INN, maximum probability, sum average, sum entropy, sum of squares, sum variance). For each patient at each time-point, GLCM features values were averaged through the slices, ROI-wise, resulting in nine values for every feature per patient (repair cartilage at 3M, 12M, and 24M; adjacent cartilage at 3M, 12M, and 24M; reference cartilage at 3M, 12M, and 24M). For visualization purposes, features heatmaps of repair tissue were calculated for one MACT and one MFX patient at each time-point. For rotated and flattened ROI, autocorrelation, homogeneity, and dissimilarity were calculated based on surrounding pixels selected by a 5 × 5-pixel sliding window with a step of length 1. These features were calculated in eight directions (0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°) and averaged, yielding three maps per patient per time-point. Map calculation and visualization was performed in Python 3.9.18.

Statistical Analysis

All statistical analyses were performed using R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria) in RStudio version 2022.07.2 (Rstudio, PBC). The Shapiro-Wilk test was used to assess the normality of the examined variables. Based on normality, Student’s t test or the Wilcoxon signed-rank test was used to determine the significance of differences in mean T₂ values and GLCM features between the procedures, as well as all combinations of time-points for MACT and MFX separately. Differences are described as mean difference ± standard error of the difference between means. The Spearman correlation coefficient was used to examine the relationship between GLCM features and the MOCART 2.0 score. Based on simple comparisons between time-points, mean T₂, autocorrelation, dissimilarity, homogeneity, correlation, sum average, and information measure were further analyzed. (See Appendix A, Supplemental Digital Content 1, which provides mathematical formulas and explanations of selected features.)

Linear mixed-effects models with a random slope and a random intercept with Kenward-Roger degrees of freedom correction were fitted using the function lmer from the R package lme4 version 1.1-35.1.³⁴ In all models, we considered procedure, time-point and ROI type as fixed effects and we allowed individual intercepts for each patient with the slopes of the variable ROI type varying by patient. Final models were fitted with Restricted Maximum Likelihood (REML). We present the results as estimated differences with 95% confidence intervals. P-values lower than 0.05 were considered statistically significant.

Results

Correlation of GLCM Features with MOCART 2.0

Correlation analysis showed weak to strong correlation of GLCM features and the MOCART 2.0 score at different time-points. In the case of MACT, significant correlations were found between the MOCART 2.0 and autocorrelation (3M, r = −0.34), contrast (12M, r = −0.49; 24M, r = −0.58), correlation (12M, r = 0.35; 24M, r = 0.48), dissimilarity (12M, r = −0.47; 24M, r = −0.53), homogeneity (12M, r = 0.52; 24M, r = 0.44), sum of squares (3M, r = −0.36), difference variance (12M, r = −0.49; 24M, r = −0.58), difference entropy (12M, r = −0.43; 24M, r = −0.55), difference variance (12M, r = −0.49; 24M, r = −0.58), information measure (24M, r = −0.42), inverse difference normalized INN (12M, r = 0.51; 24M, r = 0.50), and inverse difference moment normalized (12M, r = 0.49; 24M, r = 0.58).

For MFX, the MOCART 2.0 score was significantly correlated with mean T₂ (12M, r = −0.82), correlation (3M, r = 0.64), cluster shade (3M, r = 0.68; 24M, r = −0.61), sum average (3M, r = −0.68), sum entropy (12M, r = −0.69), and information measure of correlation 2 (3M, r = 0.69). All correlations are listed in Table 1 .

Table 1.

Correlations between MOCART 2.0 and Gray-Level Co-Occurrence Matrix (GLCM) Features.

	MACT			MFX
	3M	12M	24M	3M	12M	24M
T₂	–0.18	0.04	–0.09	–0.19	–0.82*	–0.15
Autocorrelation	–0.34*	–0.14	0.12	–0.56	0.18	–0.31
Contrast	–0.04	–0.49*	–0.58*	–0.56	–0.37	–0.20
Correlation	–0.05	0.35*	0.48*	0.64*	–0.50	–0.37
Cluster prominence	–0.06	–0.05	–0.10	0.41	–0.37	0.00
Cluster shade	0.15	0.13	–0.23	0.68*	–0.61*	0.06
Dissimilarity	–0.01	–0.47*	–0.53*	–0.47	–0.13	–0.10
Energy	–0.07	0.05	0.25	0.15	0.25	0.26
Entropy	0.01	–0.09	–0.19	–0.06	–0.34	–0.19
Homogeneity	–0.04	0.52*	0.44*	0.47	0.03	–0.03
Maximum probability	–0.11	–0.05	0.24	0.15	–0.08	0.05
Sum of squares	–0.36*	–0.14	0.12	–0.56	0.06	–0.31
Sum average	–0.31	–0.13	0.08	–0.68*	0.18	–0.20
Sum variance	–0.32	–0.17	0.11	–0.56	0.13	–0.31
Sum entropy	–0.07	0.16	0.06	0.30	–0.69*	–0.46
Difference variance	–0.04	–0.49*	–0.58*	–0.56	–0.37	–0.20
Difference entropy	–0.03	–0.43*	–0.55*	–0.44	–0.28	–0.20
Information measure	0.04	–0.24	–0.42*	–0.56	0.39	0.28
Information measure of correlation 2	–0.08	0.27	0.31	0.69*	–0.50	–0.38
Inverse difference normalized INN	–0.01	0.51*	0.50*	0.47	0.13	0.03
Inverse difference moment normalized	0.02	0.49*	0.58*	0.56	0.37	0.20

MACT = matrix-associated autologous chondrocyte transplantation; MFX = microfracturing; 3M = 3 months follow-up; 12M = 12 months follow-up; 24M = 24 months follow-up.

Statistically significant correlations are marked with an asterisk. P-Values < 0.05 are Marked in Bold.

Comparison of MACT and MFX

For a descriptive statistics of all parameters, see Appendix B, Supplemental Digital Content 1.

Based on unpaired tests, we found no significant differences in the mean T₂ or GLCM features between MACT and MFX across all time points for all regions of interest (ROIs), except for cluster prominence at 24 months in adjacent tissue (P = 0.034). When examining the changes in these parameters separately for each procedure, it was observed that, for MFX, only the mean T₂ showed significant differences, specifically between 3M and 12M (P < 0.001) and between 3M and 24M (P = 0.004). For MACT, mean T₂ values also changed significantly between 3M and 12M (P < 0.001), 3M and 24M (P < 0.001), and between 12M and 24M (P = 0.034). Furthermore, GLCM features changed significantly for MACT, namely, autocorrelation, correlation, homogeneity, information measure, information measure of correlation 2, sum average, sum of squares, and sum variance. All significant differences are listed in Table 2 . The variables chosen for further analysis via linear mixed-effects models are visualized in Fig. 2.

Table 2.

Significant Differences in Gray-Level Co-Occurrence Matrix (GLCM) Features between Time-Points.

ROI	Procedure	Parameter	Contrast	Mean difference	SE	P-value
Repair tissue	MACT	Autocorrelation	3M vs. 12M	25.11	9.29	0.001
Repair tissue	MACT	Autocorrelation	3M vs. 24M	23.25	9.93	0.014
Repair tissue	MACT	Correlation	12M vs. 24M	0.05	0.03	0.034
Repair tissue	MACT	Homogeneity	12M vs. 24M	0.03	0.02	0.049
Repair tissue	MACT	Information measure	3M vs. 24M	–0.08	0.03	0.016
Repair tissue	MACT	Information measure of correlation 2	3M vs. 24M	0.04	0.02	0.031
Repair tissue	MACT	Information measure of correlation 2	12M vs.24M	0.03	0.02	0.031
Repair tissue	MACT	Sum average	3M vs. 12M	2.99	1.03	0.001
Repair tissue	MACT	Sum average	3M vs. 24M	2.91	1.14	0.012
Repair tissue	MACT	Sum of squares	3M vs. 12M	24.9	9.25	0.001
Repair tissue	MACT	Sum of squares	3M vs. 24M	22.98	9.88	0.013
Repair tissue	MACT	Sum variance	3M vs. 12M	88.85	35.41	0.001
Repair tissue	MACT	Sum variance	3M vs. 24M	80.36	36.82	0.015
Repair tissue	MACT	Mean T₂	3M vs. 12M	16.98	4.94	0.000
Repair tissue	MACT	Mean T₂	3M vs. 24M	22.27	5.18	0.000
Repair tissue	MACT	Mean T₂	12M vs. 24M	5.28	4.26	0.034
Repair tissue	MFX	Mean T₂	3M vs. 12M	17.58	4.76	0.012
Repair tissue	MFX	Mean T₂	3M vs. 24M	26.2	6.36	0.012
Adjacent issue	MACT	Mean T₂	3M vs. 24M	4.63	4.54	0.042
Reference tissue	MACT	Energy	3M vs. 12M	0.02	0.01	0.016
Reference tissue	MACT	Information measure	3M vs. 12M	-0.05	0.03	0.045
Reference tissue	MACT	Information measure	12M vs. 24M	0.04	0.03	0.045
Reference tissue	MACT	Maximum probability	3M vs. 12M	0.03	0.01	0.04

MACT = matrix-associated autologous chondrocyte transplantation; MFX = microfracturing; 3M = 3 months follow-up; 12M = 12 months follow-up; 24M = 24 months follow-up.

Figure 2.

Mean values of the T₂ and the selected gray-level co-occurrence matrix (GLCM) features at 3, 12, and 24 months after surgery.

Linear Mixed-Effects Models

Models did not show significant differences between the procedures (P > 0.05). However, when analyzing MACT and MFX separately, we found a significant decrease in the T₂ of repair tissue in the MACT group (3Mvs12M: P < 0.001; 3Mvs24M: P < 0.001) and in the MFX group (3Mvs12M: P = 0.001; 3Mvs24M: P < 0.001; 12Mvs24M: P = 0.034).

Moreover, there was a significant change in autocorrelation (3Mvs12M: P = 0.002; 3Mvs24M: P = 0.004), dissimilarity (3Mvs24M: P = 0.01), homogeneity (3Mvs24M: P = 0.013), correlation (3Mvs24M: P = 0.036), sum average (3Mvs12M: P = 0.001; 3Mvs24M: P = 0.002) and information measure (3Mvs24M: P < 0.001) in the MACT repair tissue. The only significant change in the texture of MFX was observed in the repair tissue, as measured by the information measure (3M vs 24M: P < 0.016). All estimated differences in time are provided in Table 3 .

Table 3.

Differences between Time-Points in All Regions of Interest (ROIs) Estimated by the Models. In the Case of Logarithmic Variable Transformation During Modeling, the Difference Is Expressed as a Ratio in the Original Scale.

Fixed variable: Tissue type	Variable contrasts*	T₂ (ms)(ratio)	95% CI			p-value	Autocorrelation(difference)	95% CI			p-value	Dissimilarity(ratio)	95% CI			p-value
MACT
Repair tissue
	3M vs. 12M	1.26	1.14	–	1.39	<0.001	25.1	8.04	–	42.16	0.002	0.93	0.80	–	1.08	0.501
	3M vs. 24M	1.39	1.25	–	1.54	<0.001	23.25	6.18	–	40.31	0.004	0.83	0.71	–	0.96	0.010
	12M vs. 24M	1.1	1	–	1.22	0.063	–1.86	–18.92	–	15.21	0.964	0.89	0.77	–	1.03	0.162
Adjacent tissue
	3M vs. 12M	1.06	0.96	–	1.18	0.344	5.07	–11.99	–	22.13	0.763	0.93	0.80	–	1.08	0.511
	3M vs. 24M	1.07	0.97	–	1.19	0.231	4.6	–12.47	–	21.66	0.801	0.93	0.80	–	1.09	0.537
	12M vs. 24M	1.01	0.91	–	1.12	0.968	–0.47	–17.54	–	16.59	0.998	1.00	0.86	–	1.16	0.999
Reference tissue
	3M vs. 12M	1.06	0.96	–	1.18	0.334	–4.46	–21.52	–	12.6	0.811	0.95	0.82	–	1.11	0.711
	3M vs. 24M	1.06	0.95	–	1.17	0.412	2.56	–14.5	–	19.62	0.933	0.92	0.80	–	1.07	0.434
	12M vs. 24M	0.99	0.9	–	1.1	0.989	7.02	–10.04	–	24.08	0.596	0.97	0.84	–	1.13	0.896
MFX
Repair tissue
	3M vs. 12M	1.28	1.08	–	1.51	0.002	6.74	–21.29	–	34.78	0.837	0.99	0.78	–	1.27	0.999
	3M vs. 24M	1.53	1.29	–	1.81	<0.001	0.44	–27.6	–	28.47	0.999	0.91	0.71	–	1.16	0.622
	12M vs. 24M	1.2	1.01	–	1.41	0.034	–6.3	–34.34	–	21.73	0.856	0.91	0.71	–	1.17	0.654
Adjacent tissue
	3M vs. 12M	0.96	0.82	–	1.14	0.865	4.27	–23.76	–	32.31	0.931	0.95	0.74	–	1.22	0.887
	3M vs. 24M	1.1	0.93	–	1.31	0.342	3.79	–24.25	–	31.82	0.945	1.05	0.82	–	1.34	0.884
	12M vs. 24M	1.15	0.97	–	1.36	0.137	–0.48	–28.52	–	27.55	0.999	1.10	0.86	–	1.41	0.616
Reference tissue
	3M vs. 12M	0.99	0.84	–	1.17	0.995	11.24	–16.79	–	39.28	0.611	0.89	0.69	–	1.14	0.499
	3M vs. 24M	1.14	0.96	–	1.35	0.168	1.27	–26.76	–	29.31	0.994	0.99	0.77	–	1.27	0.996
	12M vs. 24M	1.15	0.97	–	1.36	0.138	–9.97	–38	–	18.07	0.679	1.12	0.87	–	1.43	0.550
		Homogeneity(difference)	95% CI			p-value	Correlation(difference)	95% CI			p-value	Sum average(difference)	95% CI			p-value
MACT
Repair tissue
	3M vs. 12M	0.02	–0.02	–	0.05	0.545	0.01	–0.08	–	0.11	0.919	2.99	1.01	–	4.96	0.001
	3M vs. 24M	0.04	0.01	–	0.08	0.013	0.10	0.01	–	0.19	0.019	2.91	0.94	–	4.89	0.002
	12M vs. 24M	0.03	–0.01	–	0.06	0.168	0.09	0.00	–	0.18	0.053	–0.07	–2.05	–	1.90	0.996
Adjacent tissue
	3M vs. 12M	0.02	–0.02	–	0.05	0.5	0.04	–0.05	–	0.13	0.545	0.72	–1.26	–	2.69	0.669
	3M vs. 24M	0.01	–0.03	–	0.05	0.755	–0.01	–0.10	–	0.08	0.925	0.85	–1.13	–	2.82	0.570
	12M vs. 24M	–0.01	–0.04	–	0.03	0.912	–0.05	–0.15	–	0.04	0.328	0.13	–1.84	–	2.11	0.986
Reference tissue
	3M vs. 12M	0.02	–0.02	–	0.05	0.54	0.04	–0.05	–	0.13	0.549	–0.39	–2.37	–	1.58	0.885
	3M vs. 24M	0.02	–0.01	–	0.06	0.352	–0.01	–0.10	–	0.08	0.937	0.36	–1.62	–	2.33	0.903
	12M vs. 24M	0	–0.03	–	0.04	0.944	–0.05	–0.14	–	0.04	0.349	0.75	–1.22	–	2.73	0.640
MFX
Repair tissue
	3M vs. 12M	0.01	–0.05	–	0.06	0.974	0.08	–0.07	–	0.23	0.401	0.84	–2.41	–	4.08	0.816
	3M vs. 24M	0.02	–0.03	–	0.08	0.584	0.13	–0.01	–	0.28	0.085	–0.23	–3.47	–	3.02	0.985
	12M vs. 24M	0.02	–0.04	–	0.08	0.722	0.05	–0.10	–	0.20	0.676	–1.06	–4.31	–	2.18	0.720
Adjacent tissue
	3M vs. 12M	0.02	–0.04	–	0.07	0.819	–0.09	–0.24	–	0.06	0.336	0.59	–2.66	–	3.83	0.905
	3M vs. 24M	0	–0.06	–	0.06	0.987	–0.05	–0.20	–	0.10	0.675	0.45	–2.80	–	3.69	0.944
	12M vs. 24M	–0.02	–0.08	–	0.04	0.729	0.04	–0.11	–	0.18	0.838	–0.14	–3.39	–	3.10	0.994
Reference tissue
	3M vs. 12M	0.02	–0.04	–	0.08	0.723	–0.01	–0.16	–	0.14	0.990	1.07	–2.17	–	4.32	0.716
	3M vs. 24M	0	–0.06	–	0.06	0.994	0.03	–0.12	–	0.17	0.913	0.07	–3.17	–	3.32	0.998
	12M vs. 24M	–0.02	–0.08	–	0.04	0.657	0.03	–0.11	–	0.18	0.853	–1.00	–4.25	–	2.25	0.748
		Information measure(difference)	95% CI			p-value
MACT
Repair tissue
	3M vs. 12M	–0.03	–0.08	–	0.01	0.237
	3M vs. 24M	–0.08	–0.13	–	–0.03	<0.001
	12M vs. 24M	–0.05	–0.09	–	0.00	0.051
Adjacent tissue
	3M vs. 12M	–0.03	–0.07	–	0.02	0.402
	3M vs. 24M	0.01	–0.04	–	0.05	0.944
	12M vs. 24M	0.03	–0.01	–	0.08	0.243
Reference tissue
	3M vs. 12M	–0.05	–0.10	–	0.00	0.037
	3M vs. 24M	–0.01	–0.05	–	0.04	0.943
	12M vs. 24M	0.04	0.00	–	0.09	0.081
MFX
Repair tissue
	3M vs. 12M	–0.05	–0.13	–	0.02	0.249
	3M vs. 24M	–0.09	–0.17	–	–0.01	0.016
	12M vs. 24M	–0.04	–0.12	–	0.04	0.465
Adjacent tissue
	3M vs. 12M	0.01	–0.07	–	0.09	0.933
	3M vs. 24M	0.01	–0.06	–	0.09	0.904
	12M vs. 24M	0.00	–0.07	–	0.08	0.997
Reference tissue
	3M vs. 12M	–0.01	–0.08	–	0.07	0.985
	3M vs. 24M	0.03	–0.05	–	0.10	0.671
	12M vs. 24M	0.03	–0.04	–	0.11	0.566

MACT = matrix-associated autologous chondrocyte transplantation; MFX = microfracturing; 3M = 3 months follow-up; 12M = 12 months follow-up; 24M = 24 months follow-up. P-Values < 0.05 are Marked in Bold.

Mean T₂ differed significantly between repair and adjacent tissue and repair and reference tissue in both MACT and MFX (all P < 0.001). In MACT, models revealed differences in GLCM between all combinations of ROI types at almost all time-points (see Table 4 ). In the case of MFX, the significant differences were mainly between repair and reference tissue at 12 months. For all comparisons, please refer to Table 4 .

Table 4.

Differences Between Regions of Interest (ROIs) at All Time-Points Estimated by the Models. In Case of Logarithmic Variable Transformation During Modeling, the Difference Is Expressed as a Ratio in the Original Scale.

Fixed variable: Tissue type	Variable contrasts*	T₂ (ms)(ratio)	95% CI			p-value	Autocorrelation(difference)	95% CI			p-value	Dissimilarity(ratio)	95% CI			p-value
MACT
3M
	repair vs. adjacent	1.32	1.16	–	1.51	<0.001	–4.55	–26.56	–	17.45	0.875	0.78	0.64	–	0.94	0.005
	repair vs. reference	1.33	1.16	–	1.53	<0.001	–29.96	–50.34	–	–9.58	0.002	0.64	0.52	–	0.78	<0.001
	adjacent vs. reference	1.01	0.9	–	1.12	0.991	–25.4	–44.41	–	–6.39	0.005	0.82	0.70	–	0.96	0.008
12M
	repair vs. adjacent	1.12	0.98	–	1.28	0.122	–24.59	–46.59	–	–2.58	0.025	0.78	0.64	–	0.94	0.005
	repair vs. reference	1.13	0.98	–	1.29	0.107	–59.52	–79.9	–	–39.14	<0.001	0.65	0.53	–	0.80	<0.001
	adjacent vs. reference	1.01	0.9	–	1.13	0.988	–34.94	–53.95	–	–15.93	<0.001	0.84	0.71	–	0.98	0.021
24M
	repair vs. adjacent	1.02	0.9	–	1.17	0.906	–23.21	–45.21	–	–1.2	0.036	0.88	0.72	–	1.06	0.280
	repair vs. reference	1.01	0.88	–	1.16	0.969	–50.65	–71.03	–	–30.27	<0.001	0.71	0.58	–	0.87	<0.001
	adjacent vs. reference	0.99	0.89	–	1.11	0.976	–27.44	–46.45	–	–8.43	0.002	0.81	0.69	–	0.95	0.005
MFX
3M
	repair vs. adjacent	1.44	1.16	–	1.8	<0.001	–28.71	–64.87	–	7.44	0.147	0.82	0.60	–	1.12	0.382
	repair vs. reference	1.46	1.17	–	1.83	<0.001	–44.26	–77.75	–	–10.77	0.006	0.76	0.54	–	1.06	0.145
	adjacent vs. reference	1.01	0.85	–	1.22	0.981	–15.55	–46.78	–	15.69	0.467	0.92	0.71	–	1.19	1.000
12M
	repair vs. adjacent	1.09	0.87	–	1.35	0.646	–31.19	–67.34	–	4.97	0.105	0.79	0.58	–	1.07	0.191
	repair vs. reference	1.13	0.9	–	1.42	0.382	–39.76	–73.25	–	–6.27	0.016	0.68	0.48	–	0.95	0.018
	adjacent vs. reference	1.05	0.87	–	1.25	0.835	–8.57	–39.81	–	22.66	0.792	0.86	0.66	–	1.12	0.488
24M
	repair vs. adjacent	1.04	0.84	–	1.3	0.902	–25.36	–61.52	–	10.79	0.222	0.95	0.70	–	1.30	1.000
	repair vs. reference	1.09	0.87	–	1.36	0.653	–43.42	–76.91	–	–9.93	0.007	0.83	0.59	–	1.16	0.524
	adjacent vs. reference	1.04	0.87	–	1.25	0.837	–18.06	–49.3	–	13.18	0.359	0.87	0.67	–	1.13	0.585
		Homogeneity(difference)	95% CI			p-value	Correlation(difference)	95% CI			p-value	Sum average(difference)	95% CI			p-value
MACT
3M
	repair vs. adjacent	0.07	0.03	–	0.12	<0.001	0.22	0.09	–	0.35	<0.001	–0.82	–3.40	–	1.76	0.732
	repair vs. reference	0.12	0.07	–	0.17	<0.001	0.41	0.25	–	0.56	<0.001	–3.66	–6.08	–	–1.24	0.001
	adjacent vs. reference	0.05	0.01	–	0.09	0.012	0.19	0.09	–	0.29	<0.001	–2.84	–5.05	–	–0.64	0.008
12M
	repair vs. adjacent	0.07	0.03	–	0.12	<0.001	0.24	0.11	–	0.37	<0.001	–3.09	–5.67	–	–0.51	0.015
	repair vs. reference	0.12	0.07	–	0.17	<0.001	0.43	0.28	–	0.58	<0.001	–7.04	–9.46	–	–4.62	<0.001
	adjacent vs. reference	0.05	0.01	–	0.08	0.014	0.19	0.09	–	0.29	<0.001	–3.95	–6.16	–	–1.75	0.000
24M
	repair vs. adjacent	0.04	–0.01	–	0.09	0.101	0.10	–0.03	–	0.23	0.183	–2.88	–5.46	–	–0.30	0.025
	repair vs. reference	0.1	0.05	–	0.15	<0.001	0.29	0.14	–	0.44	<0.001	–6.21	–8.63	–	–3.79	0.000
	adjacent vs. reference	0.06	0.02	–	0.1	0.002	0.19	0.09	–	0.29	<0.001	–3.33	–5.53	–	–1.12	0.001
MFX
3M
	repair vs. adjacent	0.06	–0.02	–	0.13	0.178	0.34	0.13	–	0.56	0.001	–3.58	–7.82	–	0.66	0.115
	repair vs. reference	0.08	0	–	0.17	0.057	0.42	0.17	–	0.67	<0.001	–5.06	–9.03	–	–1.08	0.009
	adjacent vs. reference	0.03	–0.04	–	0.09	0.591	0.08	–0.09	–	0.25	0.497	–1.48	–5.10	–	2.15	0.599
12M
	repair vs. adjacent	0.07	–0.01	–	0.14	0.097	0.17	–0.04	–	0.39	0.137	–3.83	–8.07	–	0.41	0.085
	repair vs. reference	0.1	0.01	–	0.18	0.022	0.33	0.08	–	0.58	0.006	–4.82	–8.80	–	–0.84	0.013
	adjacent vs. reference	0.03	–0.03	–	0.09	0.494	0.16	–0.01	–	0.33	0.064	–0.99	–4.61	–	2.63	0.794
24M
	repair vs. adjacent	0.03	–0.05	–	0.11	0.644	0.16	–0.06	–	0.37	0.198	–2.91	–7.15	–	1.33	0.237
	repair vs. reference	0.06	–0.03	–	0.14	0.266	0.31	0.06	–	0.56	0.011	–4.76	–8.73	–	–0.78	0.015
	adjacent vs. reference	0.03	–0.04	–	0.09	0.561	0.16	–0.01	–	0.33	0.068	–1.85	–5.47	–	1.77	0.449
		Information measure(difference)	95% CI			p-value
MACT
3M
	repair vs. adjacent	–0.07	–0.13	–	0.00	0.037
	repair vs. reference	–0.14	–0.22	–	–0.06	<0.001
	adjacent vs. reference	–0.07	–0.13	–	–0.02	0.008
12M
	repair vs. adjacent	–0.06	–0.12	–	0.00	0.067
	repair vs. reference	–0.16	–0.23	–	–0.08	<0.001
	adjacent vs. reference	–0.10	–0.15	–	–0.04	<0.001
24M
	repair vs. adjacent	0.02	–0.05	–	0.08	0.790
	repair vs. reference	–0.07	–0.15	–	0.01	0.089
	adjacent vs. reference	–0.09	–0.14	–	–0.03	0.002
MFX
3M
	repair vs. adjacent	0.02	–0.22	–	–0.01	0.023
	repair vs. reference	–0.07	–0.28	–	–0.03	0.011
	adjacent vs. reference	–0.09	–0.13	–	0.06	0.600
12M
	repair vs. adjacent	–0.05	–0.16	–	0.05	0.429
	repair vs. reference	–0.11	–0.24	–	0.02	0.098
	adjacent vs. reference	–0.06	–0.15	–	0.04	0.348
24M
	repair vs. adjacent	–0.01	–0.12	–	0.09	0.944
	repair vs. reference	–0.04	–0.16	–	0.09	0.742
	adjacent vs. reference	–0.02	–0.12	–	0.07	0.811

We found a significant effect of time-point 24M in all models: T₂ (P < 0.001); autocorrelation (P = 0.002); dissimilarity (P = 0.003); homogeneity (P = 0.005); correlation (P = 0.007); sum average (P = 0.001); and information measure (P < 0.001). The effect of reference tissue was also significant in all models: T₂ (P < 0.001); autocorrelation (P = 0.001); dissimilarity (P < 0.001); homogeneity (P < 0.001); correlation (P < 0.001); sum average (P = 0.001); and information measure (P < 0.001). All significant effects and interactions are listed in Table 5 .

Table 5.

Summary of Linear Mixed Effects Models of T₂ and Selected Gray-Level Co-Occurrence Matrix (GLCM) Features in Repair, Adjacent, and Reference Tissue at 3, 12, and 24 Months after Surgery. To Achieve Adequate Model Fit, It Was Necessary to Apply a Natural Logarithm Transformation to Both T₂ and Dissimilarity Variables and a Cubic Transformation to the Correlation Variable.

Predictors	Log (mean T₂)			Autocorrelation			log(Dissimilarity)			Homogeneity			Correlation³			Sum average			Information measure
Predictors	Estimates	CI	P	Estimates	CI	p	Estimates	CI	p	Estimates	CI	p	Estimates	CI	p	Estimates	CI	P	Estimates	CI	p
(Intercept)	4.35	4.26 to 4.44	<0.001	96.79	84.14 to 109.45	<0.001	−0.4	−0.56 to −0.24	<0.001	0.73	0.69 to 0.76	<0.001	0.69	0.62 to 0.77	<0.001	18.28	16.82 to 19.74	<0.001	−0.5	−0.54 to −0.46	<0.001
Time-point 12M	−0.23	−0.31 to −0.14	<0.001	−25.1	−39.35 to −10.85	0.001	0.07	−0.05 to 0.20	0.263	−0.02	−0.05 to 0.01	0.294	−0.01	−0.09 to 0.06	0.696	−2.99	−4.64 to −1.34	<0.001	0.03	−0.01 to 0.07	0.106
Time-point 24M	−0.33	−0.41 to −0.24	<0.001	−23.25	−37.50 to −9.00	0.002	0.19	0.06 to 0.31	0.003	−0.04	−0.07 to −0.01	0.005	−0.1	−0.18 to −0.03	0.007	−2.91	−4.56 to −1.26	0.001	0.08	0.04 to 0.12	<0.001
Adjacent tissue	−0.28	−0.39 to −0.17	<0.001	4.55	−13.79 to 22.90	0.623	0.25	0.10 to 0.41	0.002	−0.07	−0.11 to −0.03	<0.001	−0.22	−0.32 to −0.11	<0.001	0.82	−1.33 to 2.97	0.453	0.07	0.01 to 0.12	0.014
Reference tissue	−0.29	−0.40 to −0.17	<0.001	29.96	12.96 to 46.96	0.001	0.45	0.28 to 0.62	<0.001	−0.12	−0.16 to −0.08	<0.001	−0.41	−0.53 to −0.28	<0.001	3.66	1.64 to 5.68	0.001	0.14	0.08 to 0.20	<0.001
Procedure MFX	0.06	−0.11 to 0.24	0.478	−15.73	−40.07 to 8.60	0.203	0.06	−0.25 to 0.36	0.715	−0.02	−0.08 to 0.05	0.634	0.03	−0.12 to 0.18	0.694	−1.61	−4.42 to 1.19	0.257	0.01	−0.07 to 0.09	0.823
Time-point 12M * Adjacent tissue	0.17	0.05 to 0.29	0.006	20.03	−0.12 to 40.18	0.051	0	−0.18 to 0.18	0.991	0	−0.04 to 0.04	0.959	−0.03	−0.13 to 0.08	0.64	2.27	−0.06 to 4.60	0.056	−0.01	−0.06 to 0.05	0.813
Time-point 24M * Adjacent tissue	0.26	0.14 to 0.38	<0.001	18.65	−1.50 to 38.80	0.069	−0.12	−0.30 to 0.06	0.183	0.03	−0.01 to 0.08	0.13	0.12	0.01 to 0.23	0.029	2.07	−0.27 to 4.40	0.082	−0.08	−0.14 to −0.03	0.003
Time-point 12M * Reference tissue	0.17	0.05 to 0.29	0.007	29.56	9.41 to 49.72	0.004	−0.02	−0.20 to 0.16	0.813	0	−0.04 to 0.04	0.996	−0.03	−0.13 to 0.08	0.644	3.38	1.05 to 5.71	0.005	0.02	−0.04 to 0.07	0.546
Time-point 24M * Reference tissue	0.27	0.15 to 0.39	<0.001	20.69	0.54 to 40.84	0.044	−0.11	−0.29 to 0.07	0.225	0.02	−0.02 to 0.07	0.296	0.12	0.01 to 0.22	0.031	2.55	0.22 to 4.89	0.032	−0.07	−0.13 to −0.02	0.011
Time-point 12M * MFX	−0.02	−0.18 to 0.15	0.835	18.36	−9.05 to 45.77	0.188	−0.07	−0.31 to 0.17	0.59	0.01	−0.05 to 0.07	0.719	−0.07	−0.21 to 0.08	0.369	2.15	−1.02 to 5.32	0.183	0.02	−0.06 to 0.09	0.604
Time-point 24M * MFX	−0.1	−0.26 to 0.07	0.242	22.81	−4.60 to 50.22	0.102	−0.09	−0.33 to 0.15	0.459	0.02	−0.04 to 0.08	0.52	−0.03	−0.18 to 0.12	0.685	3.14	−0.03 to 6.31	0.052	0.01	−0.06 to 0.09	0.759
Adjacent tissue * MFX	−0.09	−0.30 to 0.13	0.431	24.16	−11.13 to 59.45	0.177	−0.06	−0.35 to 0.24	0.714	0.02	−0.06 to 0.09	0.674	−0.13	−0.34 to 0.08	0.228	2.76	−1.37 to 6.90	0.188	0.05	−0.05 to 0.15	0.326
Reference tissue * MFX	−0.09	−0.31 to 0.13	0.401	14.3	−18.40 to 47.00	0.388	−0.17	−0.50 to 0.15	0.295	0.04	−0.05 to 0.12	0.385	−0.02	−0.26 to 0.23	0.89	1.4	−2.48 to 5.28	0.476	0.02	−0.11 to 0.14	0.796
Time-point 12M * Adjacent tissue * MFX	0.11	−0.12 to 0.35	0.333	−17.56	−56.32 to 21.20	0.373	0.04	−0.30 to 0.38	0.797	−0.01	−0.09 to 0.07	0.838	0.2	−0.01 to 0.40	0.062	−2.02	−6.51 to 2.46	0.375	−0.06	−0.16 to 0.05	0.292
Time-point 24M * Adjacent tissue * MFX	0.07	−0.16 to 0.30	0.558	−22	−60.76 to 16.76	0.264	−0.03	−0.37 to 0.31	0.879	0	−0.09 to 0.08	0.922	0.07	−0.14 to 0.27	0.51	−2.74	−7.23 to 1.75	0.23	−0.02	−0.13 to 0.09	0.722
Time-point 12M * Reference tissue * MFX	0.09	−0.15 to 0.32	0.468	−34.06	−72.83 to 4.70	0.085	0.13	−0.21 to 0.47	0.441	−0.01	−0.10 to 0.07	0.743	0.11	−0.09 to 0.32	0.272	−3.62	−8.11 to 0.87	0.113	−0.06	−0.17 to 0.04	0.24
Time-point 24M * Reference tissue * MFX	0.02	−0.21 to 0.26	0.839	−21.52	−60.29 to 17.24	0.275	0.02	−0.32 to 0.36	0.904	0	−0.08 to 0.09	0.907	−0.01	−0.21 to 0.20	0.934	−2.85	−7.34 to 1.63	0.211	−0.05	−0.15 to 0.06	0.395
Random Effects
σ²	0.03			705.33			0.05			0			0.02			9.45			0.01
τ00 Patient	0.03			392.07			0.11			0			0.02			5.15			0.01
τ11 Patient—Adjacent tissue	0.03			893.32			0.06			0			0.04			12.73			0.01
τ11 Patient—Reference tissue	0.04			576.9			0.08			0.01			0.07			9.07			0.02
ρ01	–0.39			–0.64			–0.16			–0.12			–0.05			–0.63			–0.58
	–0.74			–0.49			–0.36			–0.37			–0.39			–0.59			–0.78
ICC	0.54			0.4			0.7			0.67			0.7			0.4			0.55
N Patient	37			37			37			37			37			37			37
Observations	333			333			333			333			333			333			333
Marginal R² / Conditional R²	0.198 / 0.630			0.251 / 0.553			0.129 / 0.741			0.173 / 0.729			0.267 / 0.783			0.264 / 0.559			0.205 / 0.645

Discussion

Inventing noninvasive, clinically relevant approaches for the assessment of cartilage repair tissue remains a persistent challenge. GLCM analysis was previously applied to monitoring of OA,^19-27,28 but it is essential to recognize that the maturation of repair tissue is a different, yet similar process compared with cartilage degradation. As such, findings from OA studies are not directly transferable to our current investigation. Up to date published studies compared GLCM features of MACT and MFX at 24M after surgery³⁵ and the interconnection between repair and adjacent tissue, suggesting that cartilage tissue adjacent to the repair site changes along with the cartilage implant.³⁶ Therefore, we included both adjacent and reference tissue in this analysis. Although the T₂ value is linked to the cartilage matrix structure, measured voxels are too large to distinguish fiber orientation precisely, T₂ maps offer rough information about cartilage composition. Consequently, GLCM should be viewed not as an ultra-structural assessment, but as a metric between micro- and macroscopic.

This is the first study to analyze the correlation of textural features with the MOCART 2.0 score. Both surgical procedures show significant correlations with the MOCART 2.0 score at 3M and 12M. However, only textural features of MACT showed significant correlations with the MOCART score at 24M. Specifically, these were contrast (r = −0.58), correlation (r = 0.48), dissimilarity (r = −0.53), homogeneity (r = 0.44), difference variance (r = −0.58), difference entropy (= −0.55), difference variance (r = −0.58), information measure (24M, r = −0.42), inverse difference normalized INN (r = 0.50), and inverse difference moment normalized (r = 0.58). This suggests that GLCM texture of the repair tissue is tied to the morphological outcome and this relationship is more prominent in the MACT group mainly at the later follow-ups. As morphological change is preceded by micro-structural alteration³⁷ and reflected by T₂ values to some extent, a connection between textural features and the MOCART 2.0 score is logical. Experiments on cadaveric knees,³⁸ demonstrated a correlation between the features autocorrelation, contrast, and entropy extracted from T₂ maps with the histologically assessed stage of cartilage degeneration (Mankin score). Also, correlation between collagen fiber parallelism and feature autocorrelation was discovered. Schagemann et al.³⁹ showed mild but significant correlation between the 3D MOCART score category defect overall and histological ICRS II overall assessment in a large-animal model at 3T. Goebel et al.⁴⁰ found correlations between “defect fill” and “total points” of the 2D and 3D MOCART scores and the corresponding categories of the Sellers and Wakitani histological scores at 9.4T in an animal model. To date, we are not aware of any study investigating a direct relationship between MACT graft texture and histology; however, our results indicate a possible link.

Unpaired tests and linear mixed-effects models found no significant differences between MACT and MFX regarding T₂ and texture, except for cluster prominence at 24M in adjacent tissue (P = 0.034), which was, potentially, a fluke among the multiple parameters tested. Previous studies demonstrated the ability of T₂ mapping to differentiate between repair types;^29,41 for example, Welsch et al.²⁹ found significantly lower T₂ values in MFX (47.9 ± 9.8 ms) compared with T₂ values in MACT (53.6 ± 11.9 ms), which they linked to the fibrocartilaginous-like structure of MFX and the hyaline-like structure of MACT. In our cohort, changes in bulk T₂ might not have been present at 24 months but might have developed later. A previous analysis,³⁵ specifically focusing on femoral cartilage at the 24-month mark with a sample size of 57 patients, found significant differences in texture between MACT and MFX. These differences were observed in autocorrelation (P < 0.001), sum of squares (P < 0.001), sum average (P = 0.01), sum variance (P < 0.001), and sum entropy (P = 0.05). However, in the current analysis, a smaller subset of only 34 patients with measurements available at all three time-points was considered. Consequently, we were unable to demonstrate the differences between MACT and MFX in this subset.

However, we did identify intra-group changes over the various time-points, as visible in Fig. 3 . Bulk T₂ decreased significantly only in the repair tissue of the MACT group (3M vs 12M, 3M vs 24M) and the MFX group (3M vs 12M, 3M vs 24M, 12M vs 24M). Repair tissue texture developed significantly over time almost exclusively in the MACT group, MFX repair tissue texture remained the same during the 24-month period, except for information measure between 3 and 24 months (P = 0.016). This is visible in Fig. 3 , where the overall texture of MFX did not change, although the overall T₂ values decreased, particularly the hydration of the superficial layer. The change of MACT texture is clearly visible. The Wilcoxon signed-rank test revealed significant differences in autocorrelation, correlation, homogeneity, information measure, information measure of correlation 2, sum average, sum of squares, and sum variance. These features are measures of gray-level homogeneity or correlation of gray levels. The features that are linked to the presence of edges or disorderliness (e.g., contrast or dissimilarity) were not significantly different between time-points. Nevertheless, our focus on analyzing dissimilarity stems from its correlation with the MOCART score and its correlation with other features related to disorderliness. Models indicated a reduction in gray-level homogeneity and a decline in correlations between gray levels in the repair tissue over time, for example a decrease in autocorrelation (3M-12M: Δ = 25.1, P = 0.002; 3M-24M: Δ = 23.25, P = 0.004) and homogeneity (3M-24M: Δ = 0.04, P = 0.004). However, disorderliness increased, particularly dissimilarity (3M/24M: ratio = 0.83, P = 0.01). For detailed formulas and an explanation of features, please refer to the Appendix. Initially, when the repair tissue was highly hydrated, differences between the highest and lowest T₂ voxel values were large, with similar T₂ values falling into one quantization bin, which resulted in larger areas of uniform gray levels. When water content decreased and a finer texture with a smaller variance of T₂ values was revealed, quantization resulted in greater local disparity in gray levels ( Fig. 4 ). Therefore, homogeneity decreased and dissimilarity increased. Heatmaps ( Fig. 4 ) depict local feature variations calculated with different parameters than the overall ROI feature values. Nonetheless, they illustrate how T₂ variations translate into shades of gray and subsequent feature values.

Figure 3.

Example of cartilage repair healing. Examined tissue is delineated by white arrows. Matrix-associated autologous chondrocyte transplantation (MACT) change in T₂ distribution is clearly visible between the 3M and 12M mark. Microfracturing (MFX) T₂ distribution remains approximately the same, but the overall T₂ values decrease.

Figure 4.

Visualization of quantized and rotated regions of interest (ROIs) and calculated features heatmaps. Value of each voxel is calculated based on surrounding pixels selected by a 5 × 5-pixel sliding window with a step of length 1. Features values for each voxel were calculated in eight directions (0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°) and averaged. These visualizations represent local features variation, and the visual change might not correspond to the overall ROI values analyzed.

Similar trends in texture feature changes between 12 and 24 months were estimated in a prior study that examined the MACT treatment of larger lesions (5.5 ± 1.9 cm²) using different MACT product. More specifically, autocorrelation (-5.14, CI [-10.84, 0.56], P = 0.077), homogeneity (-0.02, CI [-0.04, -0.01], P = 0.001), and cube of correlation (-0.06, CI [-0.09, -0.03], P < 0.001).³⁶ In reference to the three-month follow-up of this cohort, we estimated the difference in autocorrelation (-25.10, CI [-39.35, -10.85], P = 0.001) at 12M and (-23.25, CI [-37.50, -9.00], P = 0.002) at 24M. For homogeneity, differences were estimated to be (-0.07, CI [-0.11, -0.03], P < 0.001) at 12M and (-0.12, CI [-0.16, -0.08], P < 0.001) at 24M. Last, for the cube of correlation, the estimation was (-0.01, CI [-0.09, 0.06], P = 0.70) at 12M and (-0.10, CI [-0.18, -0.03], P = 0.007) at 24M. This clearly shows the robustness of the method across different cohorts and different MACT products.

In our analysis, we found that with regard to T₂, repair tissue differs from adjacent and reference cartilage only at 3 months in both types of procedures. However, in the case of MACT, texturally, the ROI types differed significantly over the course of the study, but we would like to draw attention to the nonsignificant differences between repair and adjacent tissue. At 3M, autocorrelation (P = 0.875) and sum average (P = 0.732), and at 24M (P = 0.280), homogeneity (P = 0.101), correlation (P = 0.183), and information measure (P = 0.790) show no significant differences either. This demonstrates that texturally adjacent tissue is close to repair tissue, confirming the results of the aforementioned study of different MACT product.³⁶ It is worth noting that, at 24M, information measure was significantly different between adjacent and reference tissue (P = 0.002), suggesting adjacent tissue can be between repair and reference tissue texture-wise. When looking at MFX, there were significant differences between repair and reference tissue, particularly at the 12-month mark, and for autocorrelation and sum average at all three time-points. However, there were almost no significant differences between repair and adjacent tissue, as well as adjacent and reference tissue pairings. Exceptions include repair and adjacent tissue at 3 months in terms of correlation and information measure (P = 0.001 and P = 0.023, respectively). This shows that MACT affects the whole cartilage and the maturation process is probably not finished at 24M when compared with MFX.

The effect of reference tissue in the linear mixed-effects models has a similar direction and roughly equal magnitude when compared with a previous study. Namely, autocorrelation (ours: 29.6, P = 0.001; previous:³⁶ 49.11, P < 0.001), homogeneity (ours: -0.12, P < 0.001; previous:³⁶ −0.07, P = 0.002), and cube of correlation (ours: -0.41, P < 0.001; previous:³⁶ −0.17, P = 0.015), once again pointing to the method’s robustness.

Gradual changes in tissue texture can be captured noninvasively and conveniently, thanks to the availability of T₂ mapping protocols on clinical scanners and GLCM libraries. However, the maturation of MACT grafts into hyaline-like tissue is a gradual process, spanning up to five years.⁴² In our study, there is apparent development in MACT structure after two years not only in terms of the reduction of increased water content and swelling compared with MFX, but also in terms of T₂ redistribution (change in texture), implying tissue reorganization. However, we were unable to distinguish whether these changes in features reflect hyaline-like cartilage or fibrocartilage because no histological assessment of biopsies was performed, as this would have put a significant burden on the patient. Although a link between texture and histology was previously shown in the case of OA,³⁸ more studies of cartilage repair will be needed to validate correlations between GLCM features and the histological characteristics of repair tissue.

Our study has several limitations. The number of cases is relatively low (27 MACT and 10 MFX patients). Therefore, some differences, although important, might not be significant in this study, but the direction of the change is comparable to other cartilage repair trials. Texture analysis was conducted on three consecutive slices rather than utilizing a 3D approach due to the nonzero slice distance in multi-echo spin-echo T₂ mapping. However, utilization of the same calculation pipeline ensured comparability between studies. Last, the absence of histological evaluation prevented us from establishing a direct connection between textural features and cartilage maturation, highlighting the need for such analysis in future studies.

Conclusion

In conclusion, GLCM analysis of cartilage repair tissue has proven to be robust across multiple trials and MACT products. Even though there was no significant difference in parameters between MACT and MFX, significant tissue development over time was uncovered by the GLCM analysis in the case of MACT only, allowing noninvasive monitoring of graft maturation. Last, GLCM features have the potential to become the parameter that links microscopic structure with morphological appearance.

Supplemental Material

sj-docx-1-car-10.1177_19476035241313047 – Supplemental material for Texture Analysis of Cartilage Repair Tissue Maturation: Comparison of Two Cartilage Repair Methods and Correlation with MOCART 2.0

Supplemental material, sj-docx-1-car-10.1177_19476035241313047 for Texture Analysis of Cartilage Repair Tissue Maturation: Comparison of Two Cartilage Repair Methods and Correlation with MOCART 2.0 by Veronika Janacova, Pavol Szomolanyi, Diana Sitarcikova, Alexandra Kirner, Siegfried Trattnig and Vladimir Juras in CARTILAGE

Footnotes

Acknowledgments and Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The financial support by the Austrian Federal Ministry for Digital and Economic Affairs, the National Foundation for Research, Technology and Development and the Christian Doppler Research Association is gratefully acknowledged. This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/KLI917 and 10.55776/PIN5555423 and grant no. APVV-21-0299 of the Slovak Research and Development Agency. The funding bodies had no role in the design of the study or collection, analysis, and interpretation of the data or manuscript preparation and submission.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: A.K. is an employee of TETEC Tissue Engineering Technologies. S.T., MD, is a Research and Development editor of the Cartilage journal. Remaining authors declare that they have no competing interests.

Ethics Approval

Ethical approvals for this study were obtained separately for each individual participating and each participant gave written, informed consent. The clinical trial was registered under EudraCT number 2011-005798-22.

Informed Consent

Written informed consent was obtained from all subjects before the study.

ORCID iDs

Veronika Janacova

Pavol Szomolanyi

Diana Sitarcikova

Alexandra Kirner

Vladimir Juras

Supplementary material for this article is available on the Cartilage website at .

References

Bhosale

Richardson

. Articular cartilage: structure, injuries and review of management. Br Med Bull. 2008;87:77-95.

Sophia Fox

Bedi

Rodeo

. The basic science of articular cartilage. Sports Health. 2009;1(6):461-8.

Randsborg

Brinchmann

Løken

Hanvold

Aae

Årøen

. Focal cartilage defects in the knee: a randomized controlled trial comparing autologous chondrocyte implantation with arthroscopic debridement. BMC Musculos Dis. 2016;17(1):117.

Sellards

Nho

Cole

. Chondral injuries. Curr Opin Rheumatol. 2002;14(2):134-41.

Bentley

Minas

. Treating joint damage in young people. BMJ. 2000;320(7249):1585-8.

Basad

Ishaque

Bachmann

Stürz

Steinmeyer

. Matrix-induced autologous chondrocyte implantation versus microfracture in the treatment of cartilage defects of the knee: a 2-year randomised study. Knee Surg Sports Traumatol Arthrosc. 2010;18(4):519-27.

Dekker

Aman

DePhillipo

Dickens

Anz

LaPrade

. Chondral lesions of the knee: an evidence-based approach. J Bone Joint Surg Am. 2021;103(7):629-45.

Kreuz

Steinwachs

Erggelet

Krause

Konrad

Uhl

, et al. Results after microfracture of full-thickness chondral defects in different compartments in the knee. Osteoarthr Cartilage. 2006;14(11):1119-25.

Seo

Kim

Jung

. Management of focal chondral lesion in the knee joint. Knee Surg Relat Res. 2011;23(4):185-96.

10.

Jarecki

Waśko

Widuchowski

Tomczyk-Warunek

Wójciak

Sowa

, et al. Knee cartilage lesion management: current trends in clinical practice. J Clin Med. 2023;12(20):6434.

11.

Song

Park

. Microfracture for cartilage repair in the knee: current concepts and limitations of systematic reviews. Ann Transl Med. 2019;7(Suppl 3):S108.

12.

Orth

Gao

Madry

. Microfracture for cartilage repair in the knee: a systematic review of the contemporary literature. Knee Surg Sports Traumatol Arthrosc. 2020;28(3):670-706.

13.

Erggelet

Vavken

. Microfracture for the treatment of cartilage defects in the knee joint: a golden standard. J Clin Orthop Trauma. 2016;7(3):145-52.

14.

Niemeyer

Pestka

Kreuz

Erggelet

Schmal

Suedkamp

, et al. Characteristic complications after autologous chondrocyte implantation for cartilage defects of the knee joint. Am J Sports Med. 2008;36(11):2091-9.

15.

Harris

Siston

Brophy

Lattermann

Carey

Flanigan

. Failures, re-operations, and complications after autologous chondrocyte implantation: a systematic review. Osteoarthritis Cartilage. 2011;19(7):779-91.

16.

Marlovits

Singer

Zeller

Mandl

Haller

Trattnig

. Magnetic resonance observation of cartilage repair tissue (MOCART) for the evaluation of autologous chondrocyte transplantation: determination of interobserver variability and correlation to clinical outcome after 2 years. Eur J Radiol. 2006;57(1):16-23.

17.

Schreiner

Raudner

Marlovits

Bohndorf

Weber

Zalaudek

, et al. The MOCART (magnetic resonance observation of cartilage repair tissue) 2.0 knee score and Atlas. Cartilage. 2021;13(1_Suppl):571S-87.

18.

Haralick

Shanmugam

Dinstein

. Textural features for image classification. IEEE Trans Syst Man Cybernet. 1973; SMC-3(6):610-21.

19.

Joseph

Baum

Carballido-Gamio

Nardo

Virayavanich

Alizai

, 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.

20.

Heilmeier

Wamba

Joseph

Darakananda

Callan

Neumann

, et al. Baseline knee joint effusion and medial femoral bone marrow edema, in addition to MRI-based T2 relaxation time and texture measurements of knee cartilage, can help predict incident total knee arthroplasty 4–7 years later: data from the Osteoarthritis Initiative. Skeletal Radiol. 2019;48(1):89-101.

21.

Blumenkrantz

Stahl

Carballido-Gamio

Zhao

Munoz

, et al. The feasibility of characterizing the spatial distribution of cartilage T2 using texture analysis. Osteoarthr Cartilage. 2008;16:584-90.

22.

Zeitoune

Erbes

Casco

Adur

. Improvement of co-occurrence matrix calculation and collagen fibers orientation estimation. 12th International Symposium on Medical Information Processing and Analysis. 2017 January 26; Tandil, Argentina.

23.

Carballido-Gamio

Blumenkrantz

Lynch

Link

Majumdar

. Longitudinal analysis of MRI T2 knee cartilage laminar organization in a subset of patients from the osteoarthritis initiative. Magn Reson Med. 2010;63(2):465-72.

24.

Carballido-Gamio

Joseph

Lynch

Link

Majumdar

. 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):1184-94.

25.

Schooler

Kumar

Nardo

McCulloch

Link

, et al. Longitudinal evaluation of T1ρ and T2 spatial distribution in osteoarthritic and healthy medial knee cartilage. Osteoarthr Cartilage. 2014;22(1):51-62.

26.

Peuna

Thevenot

Saarakkala

Nieminen

Lammentausta

. Machine learning classification on texture analyzed T2 maps of osteoarthritic cartilage: Oulu Knee Osteoarthritis Study. Osteoarthr Cartilage. 2021;29(6):859-69.

27.

Carballido-Gamio

Stahl

Blumenkrantz

Romero

Majumdar

Link

. Spatial analysis of magnetic resonance and relaxation times improves classification between subjects with and without osteoarthritis. Med Phys. 2009;36(9 Part 1):4059-67.

28.

Peuna

Hekkala

Haapea

Podlipská

Guermazi

Saarakkala

, et al. Variable angle gray level co-occurrence matrix analysis of T 2 relaxation time maps reveals degenerative changes of cartilage in knee osteoarthritis: Oulu Knee Osteoarthritis Study. J Magn Reson Imaging. 2018;47(5):1316-27.

29.

Welsch

Trattnig

Domayer

Marlovits

White

Mamisch

. Multimodal approach in the use of clinical scoring, morphological MRI and biochemical T2-mapping and diffusion-weighted imaging in their ability to assess differences between cartilage repair tissue after microfracture therapy and matrix-associated autologous chondrocyte transplantation: a pilot study. Osteoarthritis Cartilage. 2009;17(9):1219-27.

30.

Mohanaiah

Sathyanarayana

GuruKumar

. Image texture feature extraction using GLCM approach. 2013 [Accessed January 16, 2024]. https://www.semanticscholar.org/paper/Image-Texture-Feature-Extraction-Using-GLCM-Mohanaiah-Sathyanarayana/7628d73c73b8ae20c6cf866ce0865a1fb64612a3.

31.

Janáčová

Juras

Szomolanyi

Trattnig

. Optimization of knee cartilage texture analysis of quantitative MRI T2 maps. Osteoarthritis and Cartilage. 2021;29S329-30.

32.

Carballido-Gamio

Link

Majumdar

. 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):1472-7.

33.

Uppuluri

. GLCM_Features4.m: vectorized version of GLCM_Features1.m [with code changes]. Accessed January 17, 2024. https://www.mathworks.com/matlabcentral/fileexchange/22354-glcm_features4-m-vectorized-version-of-glcm_features1-m-with-code-changes.

34.

Bates

Mächler

Bolker

Walker

. Fitting linear mixed-effects models using lme4. J Stat Soft. 2015;67:1-48.

35.

Juras

Szomolanyi

Janáčová

Kirner

Angele

Trattnig

. Differentiation of cartilage repair techniques using texture analysis from T2 maps. Cartilage. 2021;13(1 Suppl):718S-28S.

36.

Janacova

Szomolanyi

Kirner

Trattnig

Juras

. Adjacent cartilage tissue structure after successful transplantation: a quantitative MRI study using T2 mapping and texture analysis. Eur Radiol. 2022;32(12):8364-75.

37.

Goldring

. Chondrogenesis, chondrocyte differentiation, and articular cartilage metabolism in health and osteoarthritis. Ther Adv Musculoskelet Dis. 2012;4(4):269-85.

38.

Juras

Toegel

Hager

Schreiner

Janacova

Heule

, et al. Histological validation of the textural features from quantitative MRI for determination of cartilage degeneration. Osteoarthr Cartilage. 2023;31S40-2.

39.

Schagemann

Galle

Gille

Frydrychowicz

Welsch

Salzmann

, et al. Correlation of the histological ICRS II score and the 3D MOCART score for the analysis of aged osteochondral regenerates in a large animal model. Cartilage. 2022;13(1):1072254.

40.

Goebel

Zurakowski

Müller

Pape

Cucchiarini

Madry

. 2D and 3D MOCART scoring systems assessed by 9.4 T high-field MRI correlate with elementary and complex histological scoring systems in a translational model of osteochondral repair. Osteoarthr Cartilage. 2014;22(10):1386-95.

41.

Domayer

Apprich

Stelzeneder

Hirschfeld

Sokolowski

Kronnerwetter

, et al. Cartilage repair of the ankle: first results of T2 mapping at 7.0 T after microfracture and matrix associated autologous cartilage transplantation. Osteoarthr Cartilage. 2012;20(8):829-36.

42.

Paatela

Vasara

Nurmi

Kautiainen

Jurvelin

Kiviranta

. Biomechanical changes of repair tissue after autologous chondrocyte implantation at long-term follow-up. Cartilage. 2021;13(1 Suppl);1085S-91S.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.06 MB