Longitudinal evaluation of T_1ρ and T₂ spatial distribution in osteoarthritic and healthy medial knee cartilage

SUMMARY

Objective

To investigate longitudinal changes in laminar and spatial distribution of knee articular cartilage magnetic resonance imaging (MRI) T_1ρ and T₂ relaxation times, in individuals with and without medial compartment cartilage defects.

Design

All subjects (at baseline n = 88, >18 years old) underwent 3-Tesla knee MRI at baseline and annually thereafter for 3 years. The MR studies were evaluated for presence of cartilage defects (modified Whole-Organ Magnetic Resonance Imaging Scoring – mWORMS), and quantitative T_1ρ and T₂ relaxation time maps. Subjects were segregated into those with (mWORMS ≥2) and without (mWORMS ≤1) cartilage lesions at the medial tibia (MT) or medial femur (MF) at each time point. Laminar (bone and articular layer) and spatial (gray level co-occurrence matrix – GLCM) distribution of the T_1ρ and T₂ relaxation time maps were calculated. Linear regression models (cross-sectional) and Generalized Estimating Equations (GEEs) (longitudinal) were used.

Results

Global T_1ρ, global T₂ and articular layer T₂ relaxation times at the MF, and global and articular layer T₂ relaxation times at the MT, were higher in subjects with cartilage lesions compared to those without lesions. At the MT global T_1ρ relaxation times were higher at each time point in subjects with lesions. MT T_1ρ and T₂ became progressively more heterogeneous than control compartments over the course of the study.

Conclusion

Spatial distribution of T_1ρ and T₂ relaxation time maps in medial knee OA using GLCM technique may be a sensitive indicator of cartilage deterioration, in addition to whole-compartment relaxation time data.

Introduction

Knee osteoarthritis (OA) most commonly affects the medial compartment¹ and degenerative cartilage lesions associated with knee OA have been reported more frequently at the medial compartment of the knee^2–4. Early degenerative changes in OA consist of reduction in the proteoglycan content and disruption of the collagen network⁵. T_1ρ and T₂ relaxation time mapping magnetic resonance imaging (MRI) techniques, among others, have been proposed for quantitative evaluation of early changes associated with OA in knee hyaline cartilage^6–10. An increase in T_1ρ and T₂ relaxation times indicates loss of proteoglycans and disruption of collagen matrix respectively^{7–9,11–13}. T₂ relaxation time has also been inversely correlated with proteoglycan concentration¹⁴, suggesting that this metric is sensitive to both collagen and proteoglycan concentration. Previous studies have demonstrated differences between superficial and deep layers of articular cartilage using laminar analyses, for mean T_1ρ¹⁰ and T₂¹⁵ relaxation times, possibly due to spatial differences in collagen orientation and content throughout the cartilage matrix. It has also been shown that individuals with greater number and severity of cartilage lesions in the medial femur (MF) have higher T_1ρ relaxation times at the MF⁴. However, longitudinal analysis of changes in T_1ρ and T₂ relaxation times for the superficial and deep layers of articular cartilage, and their association with medial knee cartilage defects, has not been performed.

Haralick et al.¹⁶ developed a method of texture analysis based on the gray level co-occurrence matrix (GLCM) that is used to evaluate spatial distribution of pixel intensities in an image along a corresponding angle or direction. Spatial analysis of T_1ρ and T₂ relaxation times in cartilage has been shown to provide supplementary information about specific patterns of degeneration when compared to standard metrics alone (compartment mean values and standard deviations)^17,18. Techniques to flatten regions of interest after image acquisition to more accurately classify tissues with well-defined layers have been proposed¹⁹. Carballido-Gamio et al.²⁰ reported significant increases in T_1ρ GLCM parameter reproducibility with flattened cartilage maps compared to non-flattened maps. Flattening of T_1ρ and T₂ cartilage maps allows for quantification of GLCM spatial heterogeneity both along (parallel to the bone–cartilage interface, corresponding to the A–P axis) and through (perpendicular to the bone–cartilage interface, corresponding to the S–I axis) the natural lamina present in articular cartilage. Longitudinal changes in knee articular cartilage GLCM parameters for both T_1ρ and T₂ relaxation times, using flattened cartilage maps, and their association with cartilage defects, have not been investigated to date.

The goals of this study were to (1) compare global, laminar (bone and articular layer), and flattened texture parameters of T_1ρ and T₂ relaxation times between medial knee compartments with and without cartilage lesions (cross-sectional), and (2) to compare the changes in global, laminar (bone and articular layer), and flattened texture parameters of T_1ρ and T₂ relaxation times in medial knee compartments with and without cartilage lesions over 3 years (longitudinal). We hypothesized that longitudinally, knee compartments with cartilage lesions will display elevated T_1ρ and T₂ relaxation times and will become increasingly more heterogeneous compared to compartments without cartilage lesions.

Materials and methods

Subjects

Patients with OA and control subjects without OA were recruited from UCSF orthopedic surgeons and the communityas part of a natural evolution study on knee OA. The data in this study include ongoing analyses from these previouslycollected data. The inclusion criteria for OA patients were frequent clinical symptoms of OA (including pain, stiffness and dysfunction) and demonstration of typical signs of OA in radiographs [Kellgren–Lawrence (KL)grade>0]²¹. The controlshad no history of diagnosed OA, clinical OA symptoms, previous knee injuries, or signs of OA on radiographs. Standard standing antero-posterior radiographs of the knee were obtained in all subjects at baseline to determine the KL grade and OA severity²². At baseline, the 88 subjects (41 men, 47 women) that participated in this study had a mean age of 50.1 ± 14 years and a mean BMI of 26.1 ± 4.6 kg/m².

MRI

All subjects underwent MR imaging of the knee at baseline, and at 1 year intervals for 3 more years. MR data were acquired on a 3 T Signa HDx MR (GE Healthcare, Piscataway, NJ) scanner with a dedicated 8-channel phased array knee coil. Clinical scoring of cartilage lesions was performed on a sagittal T₂ fast-spin echo (FSE) sequence (repetition time (TR)/echo time (TE) = 4300/51 ms, field of view (FOV) = 6–8 cm, matrix = 512 × 256, slice thickness (ST) = 1 mm, echo train length = 9, bandwidth (BW) = 31.25 kHz, NEX = 2, acquisition time = 4 min). A fat-saturated 3D spoiled gradient-echo (SPGR) sequence (TR/TE = 15/6.7 ms, flip angle = 12, FOV = 6–8 cm, matrix = 512 × 512, ST = 1 mm, BW = 31.25 kHz, number of excitations (NEX) = 1, acquisition time = 8 min 30 s) was acquired for the purposes of cartilage segmentation. Cartilage T_1ρ and T₂ maps were generated using 3D T_1ρ mapping techniques²⁰ based on a gradient echo sequence (TR/TE = 9.3/3.7 ms, FOV = 6–8 cm, matrix = 256 × 128, ST = 2 mm, BW = 31.25 kHz, views per segment = 64, Trec = 1.5 s, spin-lock time (TSL) = 0, 10, 40, 80 ms, spin-lock frequency (FSL) = 500 Hz, acquisition time = 13 min)²³. T₂-weighted images were acquired using sagittal 3D T₂ mapping (TR = 3700 ms, TE = 4.1, 14.5, 25, 45.9 ms, FOV = 6–8 cm, matrix = 256 × 128, ST = 2 mm, BW = 31.25 kHz, views per segment = 64, time of recovery (Trec) = 1.5 s, acquisition time = 13 min). Parallel imaging was used on all imaging sequences utilizing Array Spatial Sensitivity Encoding Technique (ASSET) with an acceleration factor of 2. Fig. 1 displays representative T_1ρ relaxation time color overlays of baseline and year 2 time points for both groups.

Fig. 1.

Representative sagittal SPGR images with T_1ρ relaxation times superimposed on articular cartilage as a color overlay of a healthy control at (A) baseline and (B) at the 2-year follow-up. OA patient at (C) baseline and (D) at the 2-year follow-up. Qualitative OA spatial heterogeneity increases are visible near the anterior portion of the MF/MT. Color scale (right) measured in milliseconds.

Clinical grading

UCSF modified Whole-Organ Magnetic Resonance Imaging Score (mWORMS)²⁴ was used to assess cartilage morphology at each time point, on a sagittal intermediate-weighted FSE fat-saturated image (Fig. 2) by board certified radiologists (TML with 20 and LN with 4 years of experience with musculoskeletal MRI). The radiologists were blinded to subject information and performed separate readings, with a consensus in case of disagreement. Cartilage was graded as follows: 0: normal signal and thickness; 1: normal thickness and elevated signal; 2: partial-thickness focal defect less than 1 cm in width; 2.5: full-thickness focal defect less than 1 cm in width; 3: multiple areas of partial-thickness focal defects mixed with areas of normal thickness or a grade 2 defect wider than 1 cm but less than 75% of the region; 4: diffuse partial thickness loss (≥75% of region); 5: multiple areas of full-thickness cartilage loss less than 1 cm or a full-thickness lesion greater than 1 m but less than 75% of the region; 6: diffuse full-thickness cartilage loss. Subjects were stratified into those with cartilage lesions (mWORMS ≥2) and those without cartilage lesions (mWORMS ≤1) at each time point.

Fig. 2.

Sagittal T₂-weighted FSE images displaying (A) a MF osteoarthritic partial-thickness lesion (arrow) associated with underlying bone marrow edema mWORMS grade 2 (0.7 mm) and (B) a healthy control with intact cartilage.

Image processing

Cartilage compartments were segmented on multiple slices semi-automatically in high resolution SPGR images using the in-house software developed with Matlab (Mathworks, Natick, MA, USA) based on edge detection and Bezier splines²⁵. The cartilage compartments analyzed for this study included the MF and medial tibia (MT). T_1ρ and T₂ maps were reconstructed by fitting T_1ρ- and T₂-weighted images pixel-by-pixel to the equations below using in-house developed software:

$$S\left(\mathrm{TSL}\right)\infty S_0\exp \left(\frac{\mathrm{TSL}}{T{1}_{\rho }}\right)$$ (1)

$$S\left(\mathrm{TE}\right)\infty S_0\exp \left(\frac{\mathrm{TE}}{T_2}\right)$$ (2)

Post-processing of T_1ρ and T₂ maps for this study was identical to that of previous studies from our group which used the same dataset^26,27. MF and MT ROIs were further partitioned into two equal layers: bone (closer to the subchondral bone) and articular (closer to articular surface) lamina automatically using in-house developed software²⁵.

Cartilage T_1ρ and T₂ maps were flattened before quantification of the GLCM contrast, entropy, and variance parameters in the horizontal (corresponding to the A–P axis) and vertical (corresponding to the S–I axis) directions, for the regions of interest²⁰. Flattening was achieved using a Bezier spline, non-linear warping technique setting the bone–cartilage interface spline as the reference for warped flattening. Relaxation times were analyzed at a one pixel offset. Elevated contrast indicates a greater number of adjacent pixels of differing values. Entropy is a measure of pixel orderliness with elevated entropy indicating a more uniform histogram (i.e., equal numbers of each pixel value). Variance is a measure in reference to how much pixel values vary from the compartment mean. Equations (3)–(5) denote three representative GLCM measurements¹⁶.

$$\text{Entropy}=\sum _{i=1}^N\sum _{j=1}^{N}P(i,j)(-\ln \left[P(i,j)\right])$$ (3)

$$\text{Variance}=\sum _{i,j=0}^{N=1}{P}_{i,j}{\left(i-{\mu }_{i,j}\right)}^2$$ (4)

where ${\mu }_{i,j}={\#}_{i,j=0}^{N^{″}1}i\left(P_{i,j}\right)$

$$\text{Contrast}=\sum _{i=1}^N\sum _{j=1}^{N}P(i,j){(i-j)}^2$$ (5)

P indicates the probability of pixel values i and j co-occur in an image and N indicates the total number of pixel co-occurrences in each region of interest. A pixel offset of one pixel was chosen based on the fact that approximately three to four pixels span the cartilage thickness. Methods of using these specific representative measurements from each GLCM group have been widely applied in the study of T_1ρ and T₂ mapping of auricular cartilage^18,28–30.

Statistical analysis

Independent two-tail Student's t tests were carried out to compare differences in subject age and BMI for compartments in the presence and absence of cartilage lesions at baseline. Similarly, chi-square tests were employed to calculate gender differences between the two groups. For cross-sectional statistics, a linear regression model was fit to each outcome, adjusting for age, gender and BMI. To evaluate whether lesion and control groups changed differentially over time, we utilized Generalized Estimating Equations (GEEs) to accommodate the repeated measures. All analyses were conducted in SAS 9.3 (SAS Institute, Cary, NC).

Results

Subject characteristics

Age, BMI and gender distribution at each time point for both groups are presented in Table I. Subjects with lesions tended to be older and heavier. Overall, there were 27 subjects with lesions in both MF and MT compartments, eight subjects with a lesion in the MF but not in the MT compartment, 0 subject with a lesion in the MT but not in the MF compartment, and 53 subjects without a lesion in either MF or MT compartments.

Table I.

Age, BMI, and gender distribution for the groups. P values from independent samples t-tests for age and BMI, and from chi-square tests for gender distribution

	Baseline (n = 88)		1 Year (n = 60)		2 Year (n = 38)		3 Year (n = 27)
	Control (n = 53)	Lesion (n = 35)	Control (n = 37)	Lesion (n = 23)	Control (n = 28)	Lesion (n = 10)	Control (n = 15)	Lesion (n = 12)
MF
Age (years)	43.9 (12.3)	59.5 (11)	45.3 (12.4)	55 (10.7)	45.3 (12)	57.5 (7)	47.8 (13.6)	51.3 (10.4)
P-value	<0.0001		0.002		0.001		0.461
BMI (kg/m2)	25.1 (4.6)	27.5 (4.4)	24.8 (4.4)	26.9 (4.1)	24 (3.1)	26.4 (6.3)	22.6 (2.7)	24.4 (3.6)
P-value	0.021		0.074		0.279		0.173
Gender (F:M)	26:27	21:14	16:21	12:11	10:18	4:6	8:7	6:6
P-value	0.385		0.500		0.810		0.863

	Control (n = 61)	Lesion (n = 27)	Control (n = 43)	Lesion (n = 17)	Control (n = 29)	Lesion (n = 9)	Control (n = 20)	Lesion (n = 7)
MT
Age (years)	44.4 (11.9)	61.9 (9.9)	46 (11.8)	56.5 (11.5)	46.1 (11.9)	56.2 (9.5)	49.2 (12.5)	49.9 (12.1)
P-value	<0.0001		0.004		0.018		0.898
BMI (kg/m2)	25.3 (4.5)	27.8 (4.8)	24.9 (4.1)	27.5 (4.5)	24.1 (3.1)	26.2 (6.6)	23.3 (2.9)	23.7 (4.3)
P-value	0.022		0.050		0.396		0.830
Gender (F:M)	30:30	16:11	20:23	8:9	11:18	3:6	12:8	2:5
P-value	0.422		0.970		0.802		0.148

The bold indicates significance at P < 0.05.

MF

Mean values (95% confidence intervals (CI), estimated model differences) for T_1ρ and T₂ global, laminar, and GLCM texture data for MF are shown in Table II. For the global T_1ρ relaxation times, the subjects with lesions displayed higher T_1ρ at year 1 and 2 (P < 0.05) but not at baseline and year 3. For laminar T_1ρ the subjects with lesions had higher articular layer T_1ρ at year 1 (P = 0.015) and higher deep layer T_1ρ at year 3 (P = 0.001). For the GLCM measures at baseline, the subjects with lesion had higher contrast, entropy, and variance in both directions (P < 0.05). At year 1, the subjects with lesions had higher vertical contrast (P = 0.03) as well as higher entropy and variance in both directions (P < 0.05). At year 2, the subjects with lesions had higher horizontal entropy (P = 0.02), higher contrast and variance in both directions (P < 0.05). At year 3, there were no differences between the groups for any of the GLCM measures. Longitudinal change in global mean T_1ρ relaxation time between the two groups approached a significant difference (P = 0.056) (Table IV). The lesion group global mean displayed increasingly longer relaxation time until year 2, experiencing the largest drop-off from year 2 to year 3 (Fig. 3). Meanwhile, the control cartilage group experienced a slight yet consistent decrease in global mean T_1ρ relaxation time (roughly 2 ms throughout the course of the study) (Fig. 3).

Table II.

MF mean values (95% CI, estimated differences) of each variable cross-sectionally (linear regression models adjusting for age, gender, BMI)

		T1ρ Global (ms)	Estimated difference	95% CI	P-value	T1ρ Articular layer (ms)	Estimated difference	95% CI	P-value	T1ρ Bone layer (ms)	Estimated difference	95% CI	P-value
Baseline	Control	41.93	0.811	–1.926, 3.5	0.56	47.16	1.082	–1.94, 4.1	0.48	36.43	0.387	–2.7, 3.4	0.8
	Lesion	45.55				51.62				39.31
1 Year	Control	41.48	3	0.553, 5.5	0.017	47.37	3.4	0.69, 6.2	0.015	35.26	2.5	–0.61, 5.5	0.11
	Lesion	45.88				52.55				38.99
2 Year	Control	41.48	3.9	0.664, 7.2	0.02	47.40	3.4	–0.25, 7.0	0.067	35.28	3.8	–0.59, 8.2	0.088
	Lesion	47.77				54.20				41.01
3 Year	Control	40.28	2.3	–0.274, 4.9	0.077	47.37	1.321	–2.69, 5.3	0.5	32.82	4.60	2.1, 7.2	0.001
	Lesion	43.60				50.04				37.24
		T1ρ Contrast-horizontal	Estimated difference	95% CI	P-value	T1ρ Contrast-vertical	Estimated difference	95% CI	P-value
Baseline	Control	10.62	23	8.6, 37	0.002	81.84	41	8.4, 73	0.014
	Lesion	35.32				148.90
1 Year	Control	9.43	16	–3.76, 36	0.11	81.23	36	4, 69	0.028
	Lesion	30.09				137.36
2 Year	Control	6.22	22	4.8, 40	0.015	85.65	81	29, 134	0.0034
	Lesion	33.88				202.60
3 Year	Control	9.51	10.2	–7.38, 28	0.24	95.41	–5.678	–37.66, 26	0.72
	Lesion	22.37				100.51
		T1ρ Entropy-horizontal	Estimated difference	95% CI	P-value	T1ρ Entropy-vertical	Estimated difference	95% CI	P-value
Baseline	Control	5.26	0.211	0.039, 0.382	0.017	5.55	0.225	0.07, 0.34	0.004
	Lesion	5.56				5.84
1 Year	Control	5.21	0.258	0.102, 0.414	0.0016	5.61	0.251	0.11, 0.40	0.001
	Lesion	5.54				5.89
2 Year	Control	5.13	0.271	0.057, 0.484	0.015	5.64	0.13	–0.10, 0.36	0.25
	Lesion	5.52				5.83
3 Year	Control	5.28	0.047	–0.183, 0.277	0.68	5.74	0.031	–0.22, 0.28	0.8
	Lesion	5.41				5.84
		T1ρ Variance-horizontal	Estimated difference	95% CI	P-value	T1ρ Variance-vertical	Estimated difference	95% CI	P-value
Baseline	Control	72.35	85	55, 116	<0.0001	66.05	77	48, 106	<0.0001
	Lesion	185.54				167.22
1 Year	Control	75.92	73	43, 103	<0.0001	68.04	57	32, 82	<0.0001
	Lesion	167.96				141.72
2 Year	Control	80.14	127	67, 187	0.0001	71.43	85	39, 130	0.0006
	Lesion	232.83				183.11
3 Year	Control	97.18	32	–43.08, 107	0.39	85.56	24	–31.89, 80	0.38
	Lesion	148.24				125.27
		T2 Global mean (ms)	Estimated difference	95% CI	P-value	T2 Articular layer (ms)	Estimated difference	95% CI	P-value	T2 Bone layer (ms)	Estimated difference	95% CI	P-value
Baseline	Control	31.82	1.349	–0.929, 3.6	0.24	33.60	1.753	–0.699, 4.2	0.16	29.93	1.003	–0.97, 3.2	0.42
	Lesion	34.88				37.38				32.33
1 Year	Control	30.58	3.8	1.781, 1.551	0.0004	32.37	3.7	1.551, 5.8	0.001	28.67	4	1.52, 6.4	0.002
	Lesion	35.39				37.37				33.32
2 Year	Control	31.18	2.6	0.162, 5.1	0.038	32.68	2.4	–0.048, 4.9	0.054	29.60	2.9	–0.30, 6.1	0.074
	Lesion	35.30				37.39				33.16
3 Year	Control	29.64	4.1	1.74, 6.5	0.0016	32.07	3.7	0.985, 6.4	0.0098	27.05		2.3, 6.8	0.0004
	Lesion	34.41				36.89				31.80	4.60
		T2 Contrast-horizontal	Estimated difference	95% CI	P-value	T2 Contrast-vertical	Estimated difference	95% CI	P-value
Baseline	Control	10.66	11.9	–2.284, 26	0.099	59.12	63	28, 99	0.0007
	Lesion	25.26				139.79
1 Year	Control	4.08	8.4	3.5, 13.3	0.0012	53.76	42	19.1, 65	0.0006
	Lesion	13.60				100.70
2 Year	Control	3.35	9.9	3.5, 16.3	0.0035	48.85	47	8.4, 86	0.019
	Lesion	13.92				113.73
3 Year	Control	3.16	2.2	0.299, 4	0.025	45.14	16.2	1.878, 31	0.028
	Lesion	5.70				66.65
		T2 Entropy-horiizontal	Estimated difference	95% CI	P-value	T2 Entropy-vertical	Estimated difference	95% CI	P-value
Baseline	Control	5.09	0.094	–0.05, 0.24	0.19	5.48	0.056	–0.11, 0.22	0.49
	Lesion	5.32				5.61
1 Year	Control	4.90	0.25	0.09, 0.41	0.003	5.45	0.134	–0.01, 0.28	0.064
	Lesion	5.27				5.66
2 Year	Control	4.84	0.166	–0.05, 0.38	0.13	5.48	–0.087	–0.26, 0.09	0.31
	Lesion	5.13				5.48
3 Year	Control	4.86	0.107	–0.07, 0.29	0.23	5.47	0.083	–0.10, 0.28	0.36
	Lesion	5.07				5.66
		T2 Variance-horizontal	Estimated difference	95% CI	P-value	T2 Variance-vertical	Estimated difference	95% CI	P-value
Baseline	Control	73.12	78	39, 117	0.0001	67.09	50	22, 78	0.0006
	Lesion	186.63				146.76
1 Year	Control	68.57	61	30, 92	0.0002	61.54	47	22, 73	0.0005
	Lesion	139.81				116.40
2 Year	Control	63.52	101	32, 169	0.0056	57.00	59	11.9, 107	0.016
	Lesion	180.03				136.00
3 Year	Control	64.19	18.1	–6.19, 42	0.14	56.86	12.3	–5.936, 31	0.18
	Lesion	92.75				78.23

The bold indicates significance at P < 0.05.

Table IV.

Longitudinal interactions for variables approaching or displaying significantly divergent interactions using GEE models. Data adjusted for age, gender, BMI

Variable	95% CI	Estimated difference	P-value
MF T1ρ global mean (ms)	–0.019, 1.596	0.788	0.056
MF T2 global mean (ms)	0.03, 1.576	0.803	0.042
MF T2 articular layer mean (ms)	0.027, 1.6	0.813	0.043
MT T2 global mean (ms)	–0.073, 2.706	1.317	0.063
MT T2 articular layer mean (ms)	0.474, 3.482	1.978	0.0099
MT T2 horizontal entropy	0.002, 0.117	0.059	0.043
MT T1ρ vertical entropy	0.017, 0.212	0.114	0.021
MT T1ρ horizontal entropy	0.058, 0.21	0.134	0.0006

The bold indicates significance at P < 0.05.

Fig. 3.

Global mean T_1ρ and T₂ relaxation times (A and B) and mean articular layer T₂ relaxation times (C) in the MF. Single asterisk indicates P < 0.05, double asterisk indicates P < 0.01, and cross indicates P = 0.07–0.051 (approaching significance). Longitudinal significance between the groups is denoted above the horizontal bracket.

For MF global T₂, the subjects with lesions had higher relaxation times at years 1, 2 and 3 (P < 0.05) (Table II). For laminar T₂, the subjects with lesions had higher articular and deep layer T₂ relaxation times at years 1 and 3 (P < 0.05). For T₂ GLCM measures at baseline, the subjects with lesions had higher vertical contrast (P = 0.0007), and higher variance in both directions (P < 0.05) (Table II). At year 1, the subjects with lesions had higher contrast and variance in both directions (P < 0.05) and higher horizontal entropy (P = 0.003). At year 2, the subjects with lesions had higher contrast and variance in both directions (P < 0.05). At year 3, the subjects with lesions had higher contrast in both directions (P < 0.05). Global T₂ relaxation time displayed significant longitudinal changes between lesion and control cartilage groups (P = 0.042) (Table IV). Lesion group global T₂ relaxation time remained relatively constant throughout the study, fluctuating less than 1 ms from baseline to year 3, while control compartment global mean T₂ relaxation time longitudinally decreased more than 2 ms (Fig. 3). Articular layer T₂ relaxation time for lesion and control compartment groups also showed significantly different longitudinal changes (P = 0.043). Similarly to global mean T₂, lesion group articular T₂ fluctuated very little throughout the course of the study (less than 0.5 ms) while the control group decreased roughly 1.5 ms throughout all time points (Fig. 3) (Table IV).

MT

Mean values (95% CI, estimated model differences) for T_1ρ and T₂ global, laminar, and GLCM texture data for MT are shown in Table III. For global and laminar T_1ρ relaxation times, the subjects with MT lesions had higher values for all parameters at all time points (P < 0.05). For T_1ρ GLCM measures, at baseline and years 1 and 2, the subjects with lesions had higher contrast and variance in both directions (P < 0.05). Horizontal entropy was higher at years 1, 2 and 3, and vertical entropy was higher at years 2 and 3 (P < 0.05) (Table III). Subjects with lesions in the MT compartment also showed an increase in horizontal entropy (P = 0.021) and vertical entropy (P = 0.0006) over time compared to subjects without lesions (Table IV) (Fig. 4).

Table III.

MT mean values (95% CI, estimated differences) of each variable cross-sectionally (linear regression models adjusting for age, gender, BMI)

		T1ρ Global (ms)	Estimated difference	95% CI	P-value	T1ρ Articular layer (ms)	Estimated difference	95% CI	P-value	T1ρ Bone layer (ms)	Estimated difference	95% CI	P-value
Baseline	Control	34.57	5.5	2.5, 8.6	0.0005	41.02	4.5	1.056, 8	0.011	28.04	5.9	2.2, 9.6	0.0024
	Lesion	40.66				47.28				33.75
1 Year	Control	34.52	8	4.2, 11.9	0.0001	41.13	10.1	5.4, 14.9	<0.0001	27.68	5.8	1.579, 10	0.0079
	Lesion	42.40				51.55				33.47
2 Year	Control	35.38	6.6	2.7, 10.6	0.0017	42.30	6.3	1.864, 10.8	0.0069	28.33	5.1	0.4, 9.8	0.034
	Lesion	41.46				49.23				32.21
3 Year	Control	33.58	6.2	2.6, 9.7	0.0018	40.33	8.3	2.3, 14.4	0.0091	26.71	5.8	1.55, 10	0.0099
	Lesion	38.11				47.38				30.12
		T1ρ Contrast-horizontal	Estimated difference	95% CI	P-value	T1ρ Contrast-vertical	Estimated difference	95% CI	P-value
Baseline	Control	13.14	28	15, 41	<0.0001	106.24	118	63, 173	<0.0001
	Lesion	42.96				260.12
1 Year	Control	13.68	55	18.8, 91	0.0035	94.21	183	112, 253	<0.0001
	Lesion	64.90				307.70
2 Year	Control	8.59	50	19.4, 81	0.0023	106.12	110	33, 187	0.0064
	Lesion	57.93				256.48
3 Year	Control	13.83	–2.187	–10.35, 6	0.58	98.72	–7.996	–40.63, 25	0.62
	Lesion	10.87				97.71
		T1ρ Entropy-horizontal	Estimated difference	95% CI	P-value	T1ρ Entropy-vertical	Estimated difference	95% CI	P-value
Baseline	Control	5.09	–0.045	–0.279, 0.188	0.7	5.21	–0.07	–0.28, 0.14	0.51
	Lesion	5.08				5.14
1 Year	Control	5.02	0.248	0.101, 0.395	0.0014	5.23	0.058	–0.125, 0.24	0.53
	Lesion	5.27				5.26
2 Year	Control	4.97	0.355	0.188, 0.522	0.0001	5.22	0.187	–0.006, 0.38	0.05
	Lesion	5.34				5.37
3 Year	Control	5.11	0.267	0.024, 0.51	0.033	5.27	0.524	0.207, 0.841	0.0024
	Lesion	5.39				5.82
		T1ρ Variance-horizontal	Estimated difference	95% CI	P-value	T1ρ Variance-vertical	Estimated difference	95% CI	P-value
Baseline	Control	97.44	124	70, 179	<0.0001	84.71	89	44, 134	0.0002
	Lesion	262.71				212.65
1 Year	Control	95.14	201	138, 264	<0.0001	87.72	159	104, 215	<0.0001
	Lesion	315.53				257.53
2 Year	Control	100.03	133	79, 187	<0.0001	86.38	120	71, 169	<0.0001
	Lesion	258.85				226.33
3 Year	Control	103.72	30	–3.777, 64	0.079	95.90	24	–13.87, 62	0.2
	Lesion	139.75				122.61
		T2 Global mean (ms)	Estimated difference	95% CI	P-value	T2 Articular layer (ms)	Estimated difference	95% CI	P-value	T2 Bone layer (ms)	Estimated difference	95% CI	P-value
Baseline	Control	28.09	3.7	1.216, 6.3	0.0042	31.15	4.1	1.301, 6.9	0.0045	24.97	3.6	0.017, 7.2	0.049
	Lesion	33.97				37.40				30.76
1 Year	Control	25.96	6.6	4, 9.2	<0.0001	29.30	7.9	5.4, 10.4	<0.0001	22.56	5.5	1.765, 9.3	0.0047
	Lesion	32.60				37.77				27.74
2 Year	Control	24.89	7.9	4.8, 11	<0.0001	28.52	9.3	5.9, 12.6	<0.0001	21.17	6.8	3.2, 10.5	0.0006
	Lesion	33.22				38.86				27.88
3 Year	Control	26.50	7.3	3.8, 10.8	0.0003	29.56	6.9	2.2, 11.6	0.0056	23.28	8	3.7, 12.3	0.0009
	Lesion	32.80				36.53					29.30
		T2 Contrast-horizontal	Estimated difference	95% CI	P-value	T2 Contrast-vertical	Estimated difference	95% CI	P-value
Baseline	Control	15.86	23	6.6, 40	0.0066	115.95	123	52, 194	0.001
	Lesion	41.70				262.36
1 Year	Control	11.06	14.2	8, 20	<0.0001	80.42	173	106, 240	<0.0001
	Lesion	25.00				285.26
2 Year	Control	6.96	11.6	2.1, 21	0.019	73.21	127	68, 186	0.0001
	Lesion	23.57				241.23
3 Year	Control	7.05	12.7	3.8, 22	0.0073	108.58	68	–33.1, 169	0.18
	Lesion	21.21				172.17
		T2 Entropy-horizontal	Estimated difference	95% CI	P-value	T2 Entropy-vertical	Estimated difference	95% CI	P-value
Baseline	Control	4.94	0.146	–0.033, 0.326	0.11	5.18	–0.005	–0.23, 0.22	0.97
	Lesion	5.14				5.14
1 Year	Control	4.88	0.313	0.135, 0.491	0.0009	5.15	0.154	–0.05, 0.35	0.13
	Lesion	5.18				5.31
2 Year	Control	4.77	0.372	0.162, 0.582	0.0011	5.19	0.242	–0.02, 0.51	0.071
	Lesion	5.15				5.38
3 Year	Control	4.82	0.25	0.051, 0.45	0.017	5.17	0.226	–0.02, 0.47	0.068
	Lesion	5.12				5.46
		T2 Variance-horizontal	Estimated difference	95% CI	P-value	T2 Variance-vertical	Estimated difference	95% CI	P-value
Baseline	Control	108.51	137	77, 197	<0.0001	96.96	106	54, 158	0.0001
	Lesion	287.60	198			238.24
1 Year	Control	82.89		143, 254	<0.0001	73.23	152	108, 196	<0.0001
	Lesion	304.81				240.65
2 Year	Control	75.31	118	66, 171	<0.0001	64.60	96	54, 139	<0.0001
	Lesion	229.27				190.26
3 Year	Control	89.83	110	43, 177	0.0026	77.43	92	37, 147	0.0022
	Lesion	205.07				172.35

The bold indicates significance at P < 0.05.

Fig. 4.

Mean T_1ρ entropy (A and B) in the MT. Single asterisk indicates P < 0.05, double asterisk indicates P < 0.01. Longitudinal significance between the groups is denoted above the horizontal bracket.

Global, bone and articular layer T₂ relaxation times were higher in subjects with lesions in the MT compartment at each time points (Table III). Subjects with lesions had greater contrast in the horizontal direction at each time point, and greater contrast in the vertical direction at each time point except year 3 (Table III). Subjects with lesions also had higher T₂ variance in both directions at each time point compared to subjects without lesions. Horizontal MT T₂ entropy in compartments with lesions was higher at year 1 (P = 0.001), year 2 (P = 0.001), and year 3 (P = 0.017) but was not significantly different at baseline. As observed in MF, articular layer T₂ relaxation time in MT showed significantly different longitudinal trends between the lesion and control compartment groups (P = 0.01) caused by increases in articular layer T₂ for the lesion group and decreases in the control group (Table IV) (Fig. 5). Similar longitudinal trends approaching significance were observed for global T₂ relaxation time (P = 0.06) although for this variable control compartment T₂ decreased while lesion T₂ remained relatively constant. Additionally, T₂ horizontal entropy of the two groups changed differently with time. T₂ entropy in compartments with lesions increased slightly, then decreased slightly from year 2 to year 3, while control compartments experienced a longitudinal decrease (P = 0.043) (Fig. 5) (Table IV).

Fig. 5.

Global mean and articular T₂ relaxation times (A and B) and mean T₂ entropy in the MT. Single asterisk indicates P < 0.05, double asterisk indicates P < 0.01, and cross indicates P = 0.07–0.051 (approaching significance). Longitudinal significance between the groups is denoted above the horizontal bracket.

Discussion

In this study we investigated longitudinal changes in global, laminar and flattened texture parameters of articular cartilage T_1ρ and T₂ relaxation times in medial knee compartments with and without cartilage lesions. It is established that the prevalence of cartilage lesions due to OA is greater in the medial knee joint^31,32. In the MF, baseline cross-sectional T_1ρ global mean values were not significantly different between the two groups, but the lesion group T_1ρ was significantly more heterogeneous. This trend is consistent with the other reports^29,33,34 of higher spatial variation of T₂ values in people with knee OA compared to controls, which predicts clinical deterioration over the long term. Additionally, there was no significant difference in global mean MF T_1ρ relaxation times or GLCM texture measurements between the two groups at the year 3 time point, suggesting prolonged cartilage degeneration may reduce the capacity of the tissue to bind to motion-restricted water molecules.

Longitudinally, we discovered that lesion group MT T_1ρ and T₂ relaxation times became progressively more heterogeneous than healthy control compartments, as measured by GLCM entropy. Longitudinal changes in MT T_1ρ GLCM entropy were significantly different between the groups in both the horizontal and vertical directions. MT T_1ρ entropy progressively increased in the lesion group and remained constant in the control group. Qazi et al. studied heterogeneity of T₁-weighted images of OA and control patients using entropy calculated from histogram signal intensities. They described increases in entropy as a widening bandwidth of pixel signal intensity values and a reduction of the more dominant pixel values seen in homogenous histograms. Our results suggest that over time MT cartilage with lesions will develop a progressively more diverse array of T_1ρ values when compared with control compartments. The longitudinal significance of this relationship in both the horizontal and vertical directions supplement previous studies displaying increasing entropy in T_1ρ values in OA cartilage compared to controls¹⁸, and show the utility of using this metric to supplement global mean T_1ρ values. MT T₂ horizontal entropy in control cartilage became increasingly homogeneous over time while entropy in the lesion group remained higher (significantly higher at years 1, 2 and 3). This relationship displayed significant longitudinal differences in voxel heterogeneity between groups. These results are consistent with previous longitudinal studies that displayed elevated medial knee OA mean T₂ values along with increased entropy^30,35.

This study has several limitations. Firstly, the study focused on investigating the relationship between medial knee cartilage lesions and quantitative MR parameters of cartilage composition. Hence, the findings are not generalizable to the whole knee and pertain to individuals with cartilage lesions in the medial compartment, which are more common than lesions in the lateral compartment. Future studies would need to be done to investigate these relationships for lateral knee cartilage lesions. Secondly, there was a significant reduction in follow-up data collection due to late enrollment and subject attrition that may have limited the power to investigate differences at the year 2 and 3 time points, especially in the lesion group MT (n = 7 year 3). However, even with the limited sample size, we observed a large number of significant differences between the groups.

In summary, T_1ρ and T₂ MRI provide some promising methods by which the classification of biochemical changes in medial knee joint OA is possible. MF T_1ρ and T₂ global mean values were not significantly different at baseline, but GLCM contrast and variance were significantly higher in the lesion group indicating that GLCM calculations may provide a heightened level of sensitivity which may be undetectable via global mean analysis alone. MT T_1ρ and T₂ entropy displayed progressive, longitudinal increases in the lesion group. Thus the longitudinal evolution of cartilage T_1ρ and T₂, and the heterogeneity of these measures may be different at different stages of OA, and are strongly dependent on compartment and cartilage layer. The results presented here underscore the potential of using flattened T_1ρ and T₂ cartilage GLCM calculations along with laminar analysis to provide a more detailed characterization of longitudinal biochemical and structural changes in medial osteoarthritic knee articular cartilage.