Imaging of cartilage and bone: promises and pitfalls in clinical trials of osteoarthritis

summary

Imaging in clinical trials is used to evaluate subject eligibility, and/or efficacy of intervention, supporting decision making in drug development by ascertaining treatment effects on joint structure. This review focusses on imaging of bone and cartilage in clinical trials of (knee) osteoarthritis. We narratively review the full-text literature on imaging of bone and cartilage, adding primary experience in the implementation of imaging methods in clinical trials. Aims and constraints of applying imaging in clinical trials are outlined. The specific uses of semi-quantitative and quantitative imaging biomarkers of bone and cartilage in osteoarthritis trials are summarized, focusing on radiography and magnetic resonance imaging (MRI). Studies having compared both imaging methodologies directly and those having established a relationship between imaging biomarkers and clinical outcomes are highlighted. To make this review of practical use, recommendations are provided as to which imaging protocols are ideal for capturing specific aspects of bone and cartilage tissue, and pitfalls in their usage are highlighted. Further, the longitudinal sensitivity to change, of different imaging methods is reported for various patient strata. From these power calculations can be accomplished, provided the strength of the treatment effect is known. In conclusion, current imaging methodologies provide powerful tools for scoring and measuring morphological and compositional aspects of most articular tissues, capturing longitudinal change with reasonable to excellent sensitivity. When employed properly, imaging has tremendous potential for ascertaining treatment effects on various joint structures, potentially over shorter time scales than required for demonstrating effects on clinical outcomes.

Introduction

Clinical trials are designed to objectively test the effect of an intervention on the disease process. In osteoarthritis (OA), clinical outcomes (pain and function) are unable to elucidate whether the effect of intervention is purely symptomatic or modifying the disease process at a structural, joint tissue level. While molecular markers from serum or urine may identify processes of tissue formation and degradation, they represent all body tissue turn-over. Evaluating interventions at specific joints and tissues thus has to rely on imaging.

Imaging measures should be accurate, precise (reliable), specific, sensitive to longitudinal change, and acceptable to regulatory agencies. Current regulatory guidance for approval of disease modifying OA drugs (DMOADs) recommends that reduction of structural pathology in joint tissue should be accompanied by benefits in clinical outcomes. Studies relating imaging “biomarkers” to clinical outcomes therefore are of particular interest. All synovial joint tissues are known contribute to the development or progression of OA. However, articular cartilage warrants almost frictionless transmission of dynamic forces in joints, and when cartilage is lost and subchondral bone exposed, joint function declines. This review therefore focuses on imaging of cartilage and bone tissue in clinical trials.

The authors searched literature from Pubmed up to January 1st, 2014. Studies were prioritized for inclusion, when critically evaluating reliability, sensitivity to change, association with clinical outcomes, and efficacy in evaluating treatment effects for certain imaging methods, or when comparing different methodologies directly. Primary experience in implementing imaging methods was added, with the authors having a variety of backgrounds, including radiology, anatomy, rheumatology, medical image data analysis, work in a clinical research organization, or in the pharmaceutical industry. Recommendations are provided as to which imaging protocols are ideal for capturing specific aspects and imaging metrics of specific tissues (Table I), and pitfalls in their usage are outlined. Where available, sensitivity to change will be reported as “standardized response mean” (SRM) (Table II). Given the heterogeneity of study populations and observation periods, no attempt was made to derive summary measures across studies, or measures of consistency and risk bias. Rather, results were tabulated covering a wide range of conditions, to provide estimates of sensitivity to change for a variety of selection criteria and design parameters (Table II).

Table I.

MRI acquisition techniques supporting different imaging measures used in clinical trials

Imaging measure	Image acquisition required	Recommended resol.		Key validation papers	Other recommendations
		SlTh	In plane
Cartilage lesion scoring	IW F/TSE	3 mm	≤0.5 mm	Refs. 56,57	2–3 planes and TE < 45 ms required
BML scoring	IW F/TSE fs	3 mm	≤0.5 mm	Refs. 58,59	2–3 planes and fs required
Cartilage thickness/volume	FLASH/SPGR/FFE/WATSc fs/we	≤1.5 mm	≤0.35 mm	Refs. 60–62	TE < 10 ms
	DESS we	≤1.5 mm	≤0.35 mm	Refs. 65,66	With 0.7 mm every second slice sufficient
Cartilage composition
T2	MESE (6–7 echos)	3 mm	≤0.7 mm	Refs. 2,117,120	Eliminate shortest (10 ms) TE from fit
T2/diffusion	DESS	1.5 mm	≤0.35 mm	Refs. 118,119	Potentially simultaneous thickness analysis
T1rho	500 Hz spin lock	3 mm	≤0.7 mm	Refs. 120,122,123	Spin lock followed by SE or GRE readout
dGEMRIC	T1-mapping	3 mm	≤0.7 mm	Refs. 116,124,125	Requires iv Gd injection 90 min before
Sodium	PR or cones	4 mm	≤1.2 mm	Ref. 128	Requires specialized coil and high field
CEST	Multiple RF pulses	3 mm	≤0.7 mm	Refs. 129,130	Requires high field (3 or 7 T)
Ultra-short TE	PR or cones	3 mm	≤0.7 mm	Ref. 131	TE < 5 ms (available in 2D or 3D)

IW F/TSE fs = intermediate weighted fast or turbo spin echo sequence with fat suppression, SE = spin echo, GRE = gradient echo, T2 = measurement of the spin–spin relaxation time, T1rho = measurement of the spin-lattice relaxation time, iv = intravenous, Gd = gadolinium, SlTh = slice thickness, resol. = resolution, PR = projection-reconstruction.

Table II.

Sensitivity to change for different radiographic and MRI measures of progression, depending on specific inclusion criteria

Author	Study	Rx/MRI technique	n = X	Inclusion specifics	Measure	Time	Main finding (MC ± SD)	Sensitivity to change (SRM)
Radiography only (JSW change)
Brandt et al.46	Doxycycline (placebo group)	Semiflexed Rx (with fluoro)	180	KLG 2/3 [index knee]	mJSW	16 M	−0.24 ± 0.54 mm	−0.44
						30 M	−0.45 ± 0.70 mm	−0.64
				KLG 0/1 [contra-lateral knee]	mJSW	16 M	−0.23 ± 0.59 mm	−0.39
						30 M	−0.41 ± 0.67 mm	−0.61
Hellio et al.17	SD-6010 (placebo groups)	Mod. Lyon-Schuss	222	KLG 2*/3	mJSW	1 Y	−0.13 ± 0.36 mm	−0.36
			264		mJSW	2 Y	−0.22 ± 0.45 mm	−0.49
Oak et al.35	OAI	FFR	942	KLG 2–4	mJSW	4 Y	−0.54 ± 0.94 mm	−0.57
Reginster et al.47	SEKOIA (placebo groups)	FFR	472	KLG 2/3	mJSW	3 Y	−0.37 ± 0.59 mm	−0.63
MRI only (cartilage thickness or cartilage volume change)
Cicuttini et al.92	–	Cor MPR	117	WOMAC knee pain and osteophytes	MF	1.9 Y	0.15 ± 0.30 ml/year	−0.50
		GRASS			MT		0.10 ± 0.25 ml/year	−0.40
					LF		0.15 ± 0.22 ml/year	−0.68
					LT		0.12 ± 0.16 ml/year	−0.75
Raynauld et al.93	–	Sag FISP	32	KLG 2/3	MFTC	1 Y	4.2 ± 7.5 %	−0.56
						2 Y	7.6 ± 8.6 %	−0.88
Wluka et al.94	–	Sag GRASS	78	Pain + osteophytes	MT	4.5 Y	−63 ± 78 mm3/year	−0.81
					LT		−72 ± 73 mm3/year	−0.99
Eckstein et al.96	MAK	Cor FLASH	74	Neutral	MFTC	2.3 Y	−1.4 ± 2.8%/year	− 0.50
					LFTC		−1.0 ± 2.4%/year	− 0.42
			57	Varus	MFTC		−2.6 ± 4.2%/year	− 0.62
					LFTC		−0.7 ± 2.1%/year	− 0.34
			43	Valgus	MFTC		−0.4 ± 2.1%/year	− 0.19
					LFTC		−2.4 ± 3.9%/year	− 0.61
Eckstein et al.99	OAI	Cor FLASH	112	NEC	MFTC	1 Y	−0.002 ± 0.100 mm	−0.02
			310	KLG 2#	MFTC		−0.013 ± 0.100 mm	−0.13
			300	KLG 3	MFTC		−0.040 ± 0.129 mm	−0.31
			109	KLG 4	MFTC		−0.055 ± 0.145 mm	−0.38
Eckstein et al.102	OAI	Cor FLASH	255	§no pain	MFTC	1 Y	−0.016 ± 0.085 mm	−0.19
			117	§infreq. pain	MFTC		−0.021 ± 0.098 mm	−0.21
			111	§freq. pain	MFTC		−0.045 ± 0.148 mm	−0.30
Eckstein et al.7	OAI	Sag DESS	216	KLG 2#	MFTC	1 Y	−0.017 ± 0.142 mm	−0.12
					cMFTC	1 Y	−0.043 ± 0.205 mm	−0.21
						2 Y	−0.087 ± 0.249 mm	−0.35
						3 Y	−0.155 ± 0.337 mm	−0.46
						4 Y	−0.198 ± 0.388 mm	−0.51
				KLG 3	MFTC	1 Y	−0.086 ± 0.169 mm	−0.51
					cMFTC	1 Y	−0.163 ± 0.267 mm	−0.61
						2 Y	−0.256 ± 0.376 mm	−0.68
						3 Y	−0.334 ± 0.445 mm	−0.75
						4 Y	−0.498 ± 0.572 mm	−0.87
Pelletier et al.79	SEKOIA (placebo only)	FISP/SPGR	112	KLG 2/3	MT	3 Y	−7.2 ± 7.7%	−0.94
					MF		−8.5 ± 4.5%	−1.89
Wirth et al.101	OAI	Cor FLASH & Sag DESS	158	KLG 2#	MFTC	1 Y	−0.002 ± 0.087 mm	−0.02
			270	Med JSN1	MFTC		−0.029 ± 0.129 mm -	0.22
			280	Med JSN2	MFTC		−0.091 ± 0.171 mm	−0.53
			39	Med JSN3	MFTC		−0.092 ± 0.146 mm	−0.63
			158	KLG 2#	LFTC		−0.008 ± 0.084 mm	−0.09
			65	Lat JSN1	LFTC		−0.064 ± 0.132 mm	−0.48
			90	Lat JSN2	LFTC		−0.092 ± 0.193 mm	−0.48
			20	Lat JSN3	LFTC		−0.085 ± 0.158 mm	−0.53
Martel-Pelletier et al.107	OAI (Glu/CS) (groups without Glu/CS)	Sag DESS		KLG 0–4
			210	−NSAIDS	MFTC	1 Y	−1.9 ± 3.6%	−0.53
						2 Y	−3.0 ± 4.6%	−0.65
				−NSAIDS	LFTC	1 Y	−1.5 ± 3.8%	−0.39
						2 Y	−2.5 ± 4.5%	−0.56
			187	+ NSAIDS	MFTC	1 Y	−2.4 ± 4.7%	−0.51
						2 Y	−4.0 ± 6.1%	−0.66
				+ NSAIDS	LFTC	1 Y	−2.3 ± 4.1%	−0.56
						2 Y	−3.9 ± 5.0%	−0.78
Radiography (JSW) and MRI (cartilage thickness or volume change) in the same sample
Raynauld et al.95	–	Semiflexed Rx (with flouro) Sag FISP	107	Symptoms	mJSW	1 Y	Not reported
				Osteophytes		2 Y	−0.16 ± 0.49 mm	−0.33
					MFTC (vol.)	1 Y	−256 ± 211 mm3	−1.21
						2 Y	−405 ± 320 mm3	−1.27
Raynauld et al.112	Licofelone (Naproxen [control] group)	3D FISP/SPGR	<154	Pain +	mJSW	1 Y	−0.22 ± 0.47 mm	−0.47
				KLG 2/3	MFTC	1 Y	−299 ± 183 mm3	−1.64
					LFTC	1 Y	−244 ± 168 mm3	−1.45
					mJSW	2 Y	−0.38 ± 0.54 mm	−0.70
					MFTC	2 Y	−485 ± 237 mm3	−2.04
					LFTC	2 Y	−369 ± 161 mm3	−2.28
Le Graverand et al.148	A9001140	Lyon-Schuss	27	KLG 2	mJSW	2 Y (annualized)	−0.07 ± 0.22 mm	−0.32
		FFR			mJSW		+ 0.03 ± 0.64 mm	+ 0.11
		Cor FLASH			MFTC		−0.01 ± 0.03 mm	−0.03
					cMFTC		−0.01 ± 0.07 mm	−0.15
		Lyon-Schuss	28	KLG 3	mJSW	2 Y (annualized)	−0.22 ± 0.35 mm	−0.62
		FFR			mJSW		−0.07 ± 0.35 mm	−0.20
		Cor FLASH			MFTC		−0.05 ± 0.11 mm	−0.44
					cMFTC		−0.09 ± 0.19 mm	−0.48
Duryea et al.30	OAI	Sag DESS	116	KLG 2/3	MT	1 Y	+ 0.01 ± 0.11 mm3	+ 0.02
					LT		−0.02 ± 0.09 mm3	−0.26
					MF		−0.04 ± 0.11 mm3	−0.42
					LF		−0.01 ± 0.08 mm3	−0.09
		FFR			mJSW		−0.12 ± 0.64 mm	−0.18
					JSW (X = 0.25)		−0.21 ± 0.60 mm	−0.34
					JSW (X = 0.70)		−0.20 ± 0.73 mm	−0.27
Wirth et al.31	OAI	Cor FLASH	445	KLG 2–4	MFTC	1 Y	−1.1 ± 3.9%	−0.28
					cMFTC		−1.6 ± 5.0 %	−0.32
		FFR			mJSW		−2.3 ± 15.2 %	−0.15
					JSW (X = 0.250)		−2.2 ± 10.3 %	−0.22
		Sag DESS	522	KLG 2–4	MFTC	2 Y	−2.6 ± 6.1 %	−0.43
					cMFTC		−4.2 ± 8.2%	−0.51
		FFR			mJSW		−5.4 ± 17.5%	−0.31
					JSW (X = 0.225)		−5.3 ± 12.1%	−0.44
Cromer et al.147	LEGS	Semiflexed Rx	23	Chronic pain ≥ 1 mm mJSW	mJSW		−0.10 ± 0.22 mm	−0.46
		Sag FLASH			MT		−0.04 ± 0.07 mm	−0.56
					cMF		−0.09 ± 0.15 mm	−0.59
Cartilage composition measurement with MRI
McAlindon133	Placebo group	dGEMRIC index	13	Definite osteophytes, mJSW > 3 mm	MT	1 Y	2.4 ± 64.4	0.04
		Sag IR fast SPGR			cMF		−14.3 ± 52.1	−0.28
					pMF		−2.2 ± 55.4	−0.05
					LT		−17.3 ± 55.3	−0.31
					cLF		15.2 ± 43.7	0.35
					pLF		40.3 ± 88.3	0.46
		T2	11	Definite osteophytes, mJSW > 3 mm	MT		1.0 ms	Median, SD not reported
		Sag fast SE FS dual echo			cMF		−1.0 ms
					pMF		−2.0 ms
					LT		−1.0 ms
					cLF		0.0 ms
					pLF		0.0 ms
Anandacoomarasamy et al.132		Cartilage thickness	78	BMI > 30 kg/m2 (≈ 32% ACR OA)	MT	1 Y	−0.02 ± 0.1 mm	−0.20
		Sag FISP			cMF		−0.004 ± 0.2 mm	−0.02
					LT		−0.006 ± 0.1 mm	−0.06
					cLF		−0.07 ± 0.1 mm	−0.70
		dGEMRIC index (Sag T1Gd maps)	54	BMI > 30 kg/m2 (≈ 26% ACR OA)	MT		23 ± 126	0.18
					cMF		2 ± 123	0.02
					LT		23 ± 135	0.17
					cLF		23 ± 107	0.22

Studies published on responsiveness of radiographic²² and MRI measures⁷⁵ prior to April 2009 have been systematically summarized in the two above articles. Therefore, only few studies published before 2009 were added to the above table as a reference, with the focus being on studies published 2009–2013. The pooled SRM for radiographic JSW change reported by Reichmann et al.²² was 0.33 (95% CI: 0.26, 0.41) (43 estimates). Estimates derived from studies with <1 Y and 1–2 Y follow-up had similar responsiveness (0.24 and 0.25 respectively). Estimates from studies with >2 Y follow-up had a pooled SRM of 0.57 (95% CI: 0.39, 0.75). The pooled SRM for cartilage thickness reported by Hunter et al.⁷⁵ for quantitative measures of cartilage morphometry (study durations 6 M to 2 Y) was −0.86 (95% CI: −1.26 to −0.46) for the medial femorotibial joint (31 estimates), −1.01 (95% CI: −2.04 to 0.02) for the lateral femorotibial joint (14 estimates), and −0.63 (95% CI: −0.90 to −0.37) for the patella (13 estimates). The pooled SRM for semi-quantitative measures of cartilage (increase in lesion scores) was 0.55 (95% CI: 0.47–0.64) for the medial femorotibial joint (three estimates), 0.37 (95% CI: 0.18–0.57) for lateral femorotibial joint (three estimates), and 0.29 (95% CI: 0.03–0.56) for the patella (two estimates). The pooled SRM for semi-quantitative measures of BMLs (six estimates) was 0.43 (95% CI: −0.17 to 1.03).

Abbreviations: MC = mean change; SD = standard deviation of change; SRM = standardized response mean = MC/SD. All changes in radiographic (Rx) medial JSW given in mm. MRI-based cartilage changes given in mm (or %) for change in cartilage thickness or in ml/mm³ for changes in cartilage volume (vol); JSW(X) = JSW measurement at a fixed location in the medial (X = 0.15–0.3) or lateral (X = 0.7–0.9) compartment; MT = medial tibia; MF = medial femur; cMF = central, weight-baring part of the medial femur; pMF = posterior part of the medial femur; MFTC = medial femorotibial compartment = MT + cMF; cMFTC = central part of the MFTC. LT = lateral tibia; LF = lateral femur; cLF = central, weight-baring part of the lateral femur; pLF = posterior part of the lateral femur; LFTC = lateral femorotibial compartment = LT + cLF; mod = modified; fluoro = fluoroscopy; 1 Y/2 Y/3 Y/4 Y = 1-/2-/3-/4-year follow-up; 16 M/30 M = 16-month/30-month follow-up; 48 W = 48-week follow-up; Cor = coronal; Sag = sagittal; MPR = multi-planar reconstruction; GRASS = T1-weighted fat-suppressed 3-dimensional gradient recall acquisition in the steady state; FFR = fixed flexion radiography; IR = inversion recovery; FISP = fast imaging with steady state precession; SE = spin echo; T2 = spin–spin relaxation time; T1Gd = gadolinium enhanced T1-weighted MRI; KLG 2* = knees with definite osteophytes and WITH radiographic JSN; ref = reference; KLG 2# = knees with definite osteophytes WITHOUT radiographic JSN; §mixture of KLG 2# and KLG 3 knees; ACR OA = criteria of the American College of Rheumatology for osteoarthritis; WOMAC = Western Ontario and McMaster Universities Arthritis Index; NEC = non-exposed controls.

Primary aims and constraints of clinical trials

Medical imaging is used for screening, diagnosis/prognosis, evaluating the natural history of disease, or monitoring therapy. Imaging requirements for safety evaluations will not be considered here, given space limitations. In efficacy evaluation, imaging is used for monitoring natural history (placebo) vs therapy. The metrics to be satisfied by imaging parameters for clinical trials have been described¹. A key metric is “precision” (also “reproducibility” or “reliability”), which refers to the degree to which repeated (test–retest) measurements show the same result under unchanged conditions. Currently only radiographs are accepted for DMOAD Phase III trials by regulatory agencies, but there is new guidance that may provide a pathway for the utilization of magnetic resonance imaging (MRI) (http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM230597.pdf), whereas in cartilage repair MRI already has been accepted as an endpoint.

The decision of which imaging technique to use in a clinical trial depends on the mechanism of DMOAD action. Radiographic assessment of femorotibial “joint space width (JSW) or narrowing” (JSN) represents a composite measure, because more than one tissue is present between femoral and tibial subchondral bone. MRI is more specific to a variety of articular tissues, but careful consideration must be made as to which joint region and tissue is to be evaluated, and which specific scoring or measuring methods are to be applied. For MRI of the knee, magnets with ≥1.5 T and dedicated knee coils are highly preferable. Subtle differences have been reported between quadrature and phased array coils^2,3, the latter providing greater signal. Signal homogeneity, high contrast-to-noise, minimization of technical artifacts, and full anatomical coverage of structures of interest are crucial. Patient positioning, image orientation, and spatial resolution must be matched to the specific goals and measurement method. Compromises have to be achieved regarding acquisition time, patient comfort, cost, image quality, resolution, and number/types of imaging protocols obtained supporting various imaging measures (Table I). Monthly phantom measurements and quality assurance procedures in the Osteoarthritis Initiative (OAI) have revealed that geometric MRI measures were consistent between different sites, and that geometric measures and MRI relaxation times (T2) were stable over now up to 8 years follow-up⁴. Whichever imaging and scoring/measurement method used, central review of the images by expert readers is highly preferable⁵. These have to agree on the specific approach before evaluating images, to ensure consistent application of scoring or measurement technology. Further, intra- (and inter-) reader variability should be tested within clinical trials, to ensure consistency and precision under specific conditions. In cross-sectional studies, the inter-observer error must be substantially lower than the between-subject variability to ensure robust differentiation of study participants. In longitudinal studies, baseline and follow-images should always be evaluated by the same reader, because intra-observer errors are generally smaller than inter-observer errors⁶. Preferably, also, the intra-observer error should be small in relation to the “effect” that is being measured to avoid large sample sizes⁷.

Use of radiography in OA clinical trials

Radiographic acquisition techniques and quality control

Radiographs can be used for establishing disease severity during screening to determine patient eligibility, and for evaluating disease progression with and without DMOAD treatment. While historically diagnosis and disease severity have been established on conventional extended radiographs (bilateral weight-bearing antero-posterior view of both knees in full extension) differences in grading have been observed between radiographic protocols⁸. Therefore use of the same protocol for evaluating eligibility and progression is recommended for clinical trials. Accurate and precise evaluation of disease progression requires acquisition of a reproducible image of radiographic JSW, and adherence to standards of knee positioning⁹. Several protocols have been developed to address the characteristics of knee joint anatomy. Common to all is a standard for both weight-bearing and knee flexion; weight-bearing acquisition ensures similar cartilage compression/hydration and joint positioning between examinations. Loaded joints exhibit significantly smaller and more medially located minimum medial joint space width (mJSW) than unloaded joints¹⁰. Knee flexion provides contact between the tibia and the central/posterior aspect of the femoral condyle where OA-related cartilage damage was reported to be prominent¹¹. The first standardized positioning protocol was the semiflexed antero-posterior view: fluoroscopy was used to guide knee flexion and rotation and to achieve reproducible and parallel alignment of the medial tibial plateau relative to a horizontal X-ray beam¹². This alignment warrants a more reproducible measurement of JSW compared to extended views^12,13. Subsequently, the posterior–anterior Lyon-Schuss view was developed, using a co-planar alignment of the hip, patella, and great toe to achieve greater degrees of fixed knee flexion in repeat examinations¹⁴. However, due to increased radiation exposure and decreased fluoroscopic use, alternative protocols were developed to facilitate use in clinical trials, including fixed flexion¹³ and metatarsophalangeal protocols¹⁵. Positioning devices⁹ permit standardization of knee flexion, foot rotation and the inclusion of markers in the field of view permit correction for magnification caused by beam divergence and film focus distance. A single knee (rather than both) should be imaged at a time, to ensure minimal beam divergence. All standardized radiographic protocols offer greater precision than the conventional extended knee radiograph. However, among the rare head-to-head comparisons of alternative positioning protocols, greater sensitivity to change has been reported for fluoroscopy-assisted than for non-fluoroscopic protocols^13,16. Since parallel radio-anatomic alignment of the medial tibial plateau with the X-ray beam significantly improves sensitivity to change, multiple radiographic acquisitions per knee have been used alternatively, to ensure optimal acquisition without fluoroscopy¹⁷. A maximum 1.5 mm inter-margin distance (IMD), i.e., the projected difference between the anterior and posterior rims of the medial tibial plateau, was identified to be key quality control parameter¹⁸. However, some studies have even recommended deviations to be kept below 1.0 mm^12,19. Detailed methodology for acquisition and quality control of the modified Lyon-Schuss was recently reported from a large DMOAD trial^17,20.

Semi-quantitative grading of radiographs

OA severity can be graded semi-quantitatively, the Kellgren and Lawrence grade (KLG)²¹ being the oldest and most commonly used system. It permits demographic characterization and stratification of study participants²², and can be used to define eligibility for clinical trials. Radiographic OA is defined as KLG ≥ 2 (definite osteophytes). However, limitations are such that KLG 2 may include knees with and without JSN¹⁷, and that KLG 3 includes various degree of JSN²³; hence each KLG encompasses a wide spectrum of joint pathology (Fig. 1). JSN is not caused by cartilage loss alone, but also by meniscal extrusion, independently of cartilage status²⁴. Several modifications of KLG system exist²⁵, some of which proposed to improve sensitivity to change²⁶. Modifications of existing systems²⁵ should be clearly reported to be able to compare disease burden in populations or drug efficacy across disease severities. The Osteoarthritis Research Society International (OARSI) atlas²⁷ grades osteophytes and JSN separately in medial and lateral femorotibial compartments (Fig. 1). Using a between-knee comparison with MRI in subjects with unilateral JSN, the amount of cartilage loss per JSN grade was estimated, and was found to vary substantially between subjects²³. JSN grades were found to be more strongly related to pain than osteophytes²⁸. Disease progression can be defined as an increase in KLG, osteophyte, or JSN grade, with greater inter-observer variability for KLG and osteophytes than for JSN⁵. Given these limitations, there is an ongoing debate regarding the suitability of radiographic measures to define eligibility criteria and for outcome assessment of clinical trials²⁹.

Fig. 1.

Radiography [(D)–(I) are from the OAI]. (A) Leg positioning and direction of X-ray tube, perpendicular to the medial tibial plateau. (B) Layout of the patient positioning device and bucky. (C) Properly acquired X-ray: the tibial spines are centered within the femoral condyle notch. Adequate spacing between the tibia and fibula indicates that there is no internal or external rotation. The knee joint space is centered from top-to-bottom and left-to-right on the film, and at least five beads above and below knee joint space. The outer margins of the femur and tibia are clearly visible. The exposure is optimized for visualizing bone details. Both rows of calibration markers positioning unit are visible. (D) IMD of second radiograph (first was 100) 8 mm still showing sub-optimal IMD. (E) Automated measurement of the minimum (JSW) between the femoral and tibial margins. (F) Coordinate system for fixed location JSW in medial (X = 0.15–0.3) and lateral (X = 0.7–0.9) compartment. (G) X-ray of a knee classified as KLG2, definite osteophyte and OARSI medial JSN grade0. (H) X-ray of a knee classified as KLG2, definite osteophyte and OARSI medial JSN grade1. (I) X-ray of a knee classified as KLG3, definite osteophyte and OARSI medial JSN grade2.

Quantitative measurement of radiographic JSW

Measurement of JSW by manual and semi-automated methods has been reviewed previously⁹ and a meta-analysis of its reliability and sensitivity to change has been presented²². Today, JSW can be measured almost fully automatically using computer-delineation of femoral and tibial margins³⁰. As stated above, accurate measurement depends on the alignment of the medial tibial plateau with the X-ray beam¹⁹. JSW may be defined as the minimum distance between the opposing bones of the joint (mJSW) or as the distance at specific locations with respect to defined anatomical landmarks³⁰. mJSW was found to have greater longitudinal sensitivity to change (Table II) in knees without definite radiographic OA (KLG ≤ 1), but location-specific JSW measures were more responsive in knees with OA (KLG ≥ 2)^30,31. Being a continuous outcome, JSW loss generally is reported as the mean [and standard deviation (SD)] difference between baseline and follow-up. However, efforts have been made to define cut-offs that separate “progression” from normal measurement variability³². The disadvantage of this approach is a potential loss of information, but its strength lies in the reduction of measurement noise and in the easier (statistical) handling of dichotomized outcomes of progressors vs non-progressors³². Albeit various thresholds of JSW loss have been suggested, no consensus currently exists as to which one accurately defines structural progression³². Evidence exists that JSW loss is associated with the risk of subsequent knee replacement^33,34, an important and economically relevant clinical outcome. Also, JSW loss was associated with concurrent worsening of pain, function and quality of life³⁵. Further, quantitative JSW has been used in a large number of studies that have explored risk factors for knee OA progression³⁶; however, methodological challenges (e.g., collider bias) have been described that need to be considered in their interpretation³⁷. Yet, varus/valgus mal-alignment of the knee has been consistently shown to be a strong predictor of JSW loss and functional decline³⁸, including the medial and lateral compartment³⁹. However, the effect of mal-alignment may be modified by variability in body weight⁴⁰ and baseline KLG⁴¹. Knees with incident radiographic disease (KLG ≥ 2) were at higher risk of subsequent progression than those that were in a stable radiographic state⁴². Being an accepted structural outcome for Phase III clinical trials, JSW loss has been used in a series of DMOAD intervention studies^20,43–47. Clinical trials aiming at measuring JSW loss in the medial compartment often have included knees with at least some medial JSN, to ensure sufficient progression in the placebo group¹⁷. However, with some DMOADs, treatment effects may only be demonstrable relatively early in the disease, with some trials having shown efficacy in reducing JSW loss in knees with mild but not in those with advanced radiographic OA^20,43,46. Along the same lines, in some trials effects were seen over 1, but lost over 2 years^20,44,45. Also, clinical trials often have excluded severely mal-aligned knees, because extensive biomechanical challenge may negate potential benefits of DMOAD intervention⁴⁸. Those with valgus may develop lateral OA, potentially associated with medial pseudo-widening of medial JSW measurements²⁰.

Radiographic analysis techniques and alignment measures

Radiographs can be used further to measure the femorotibial knee alignment (varus/valgus). The most widely accepted measure is the hip–knee–ankle angle (HKA; i.e., mechanical axis), requiring full-length lower-limb radiographs⁴⁹. The femur-tibia angle (FTA; anatomical axis) can be determined from conventional knee radiographs, but has a defined offset (approx. 4°) and limited correlation with the HKA^50,51. Yet, FTA was used in a recent clinical trial²⁰ and was reported to be equally predictive of JSW loss as HKA⁵¹. Measures of trabecular texture, such as fractal signature analysis (FSA) may provide estimates of bone density and microstructural change associated with OA and its therapy⁵². FSA was found predictive of structural progression by radiography and MRI⁵³.

Use of MRI in clinical trials

MRI acquisition techniques

The OARSI/Outcome Measures in Rheumatoid Arthritis Clinical Trials (OMERACT) consensus review summarized the MRI sequences that should be used for each OA feature of OA⁵⁴. A protocol using a minimum number of sequences and permitting whole-organ assessment of the majority of articular features should comprise “fluid-sensitive” (proton density-, intermediate- or T2-weighted) fast or turbo spin echo sequences^55–59 (Fig. 2). These must be acquired in ≥2 orthogonal planes to avoid that partial volume effects at the joint margins of sagittal images are misinterpreted as bone marrow lesion (BML), if not confirmed in the coronal plane; further volume estimates from one plane are unreliable. Imaging cartilage with heavy T2 weighting prevents differentiation of the deep cartilage from the subchondral plate, and therefore echo times (TEs) <45 ms (proton density- or intermediated-weighted sequences) are recommended. Non-fat suppressed pulse sequences were shown to be precise and to agree with arthroscopic evaluation, provided that adequate spatial resolution, short inter-echo spacing, and a wide receiver bandwidth were utilized⁵⁶. However, fat suppression (fs) is required if BMLs are to be assessed from the same images. Some investigators have used gradient-echo type sequences [e.g., double echo steady state (DESS, Fig. 2), fast low-angle shot (FLASH), or spoiled gradient recalled acquisition in steady state (SPGR)] for assessing BMLs and focal cartilage defects, but several studies have shown that gradient-echo sequences are insensitive to these features and are not suited for such purpose^57–59.

Fig. 2.

Semi-quantitative scoring of cartilage and BMLs. The images show cartilage damage using 3 T MRIs from the Osteoarthritis Initiative. (A) Sagittal DESS sequence showing discrete retro-patellar cartilage damage (white arrow) and small cystic subchondral BML (arrowhead). (B) Corresponding sagittal intermediate-weighted turbo spin echo fat-suppressed (IW-TSE fs) image showing the true extent of focal cartilage defect due to an arthrographic effect of joint fluid (black arrow) and also depicting a large ill-defined subchondral BML (arrowheads) in addition to the small cystic lesion depicted in (A). (C) Sagittal IW TSE fs MRI of a different knee showing subtle focal intra-chondral hyper-intensity (arrow). It remains unclear if this finding represents intra-cartilaginous signal change of unknown significance or a true surface defect. Furthermore depth of the lesion cannot be unequivocally determined. (D) Coronal intermediate-weighted MRI without fs showing that the lesion in (C) represents a full thickness focal defect that reaches the subchondral plate (arrow).

The imaging sequences required for quantitative measurement of cartilage morphology (i.e., volume, thickness) differ from those used for semi-quantitative grading in that spatial resolution is more critical, and the cartilage surface and bone interface need to be delineated accurately (Fig. 3). Adequate contrast and adequate spatial resolution (preferably ≤1.5 mm slice thickness and ≤0.35 mm in-plane) can be obtained at ≥1.5 T field strength, using T1-weighted FLASH (Siemens), SPGR (General Electric), or fast field echo (FFE) or water selective cartilage scan (WATSc) (Philips) images with fs⁶ (Table I). Shorter acquisition times can be obtained by using selective water excitation (we) (Fig. 3) and these protocols have been validated⁶⁰ and were shown to be stable over time in multicenter trials⁶¹. Comparisons have been made between coils³, field strengths⁶², magnets of different vendors⁶³, and magnets and implementations by the same vendor⁶⁴. These comparisons generally revealed small systematic offsets, suggesting that baseline and follow-images must always be obtained on the same system with identical image parameters; however test–retest errors were in a similar range between protocols. Ideally, slices should be orientated perpendicular to the region of interest; in the knee, axial slices are optimal for imaging the patella, coronal slices for the (weight-bearing) femorotibial joint; sagittal slices cover all knee joint surfaces simultaneously but should be sufficiently thin to avoid partial volume effects. Recently, DESS imaging was introduced and cross-calibrated with FLASH for cartilage thickness measurements⁶⁵, but is only available from Siemens scanners (Table I; Fig. 3). The 12-month longitudinal sensitivity to change in central medial femorotibial joint cartilage thickness was reported to be −0.34 (SRM) for coronal FLASH, −0.37 for coronal multiplanar reconstructed DESS, and −0.36 for sagittal DESS. Using every second 0.7 mm slice of the sagittal DESS yielded similar SRMs as using every slice, and longitudinal correlations of r ≥ 0.79 were reported between the three protocols⁶⁶. For multicenter trials including magnets from different vendors, FLASH, SPGR, FFE/WATSc are therefore recommended, preferably with water excitation. Caution has to be applied when acquiring these sequences and delayed gadolinium enhanced MRI of cartilage (dGEMRIC) in the same session, because lower sensitivity to change in cartilage thickness has been reported with SPGR in the presence of Gd-DTPA⁶⁷.

Fig. 3.

Quantitative analysis of articular cartilage from MRI. (A) Sagittal FLASH with water excitation showing different regions of interest in the knee:P = Patella, TrF = femoral trochlea, LT = lateral tibia, cLF = central (weight-bearing) lateral femur, pLF = posterior (weight-bearing) lateral femur (LT + cLF = lateral femorotibial compartment = LFTC). (B) Coronal FLASH with water excitation. The segmentation of the total area of subchondral bone (tAB) is shown in dark green, the area of the cartilage surface (AC) in magenta. The AC of the MT only partially covers the tAB, separating it into a cartilage-covered part (cAB) and a denuded part (dAB). (C) 3D reconstruction of the knee showing cartilage thickness distribution in the tibia in colors, the TrF cartilage in gray, and the femoral bone using grid lines. (D) Sagittal DESS showing segmentation of the medial femur (MF) and tibia (MT). (E) Coronal multiplanar reconstruction of the DESS showing segmentation of the MT and central (weight-bearing) part of the medial femur (cMF); MT + cMF = medial femorotibial compartment = MFTC. (F) Axial multiplanar reconstruction of the DESS without segmentation.

Semi-quantitative grading systems for cartilage and bone

The Whole Organ Magnetic Resonance Imaging Score (WORMS) is a widely cited semi-quantitative scoring system for knee OA⁵⁴. It uses a complex subregional division of the knee rather than lesion-oriented approach to scoring, especially for cartilage and BMLs. The Boston–Leeds Osteoarthritis Knee Score (BLOKS) uses a lesion-oriented approach for BMLs⁶⁸. Recent studies demonstrated relative strengths and weaknesses of these scoring systems^69,70; to overcome the latter, the MRI Osteoarthritis Knee Score (MOAKS)⁵⁹ added subregional assessment, refined the scoring, and removed redundancies for cartilage and BMLs. While MOAKS is new and requires further validation, it has been recently used in the OAI⁷¹ and in another multicenter trial⁷² that examines whether arthroscopic partial meniscectomy results in better functional outcomes than non-operative therapy. Relative lack of sensitivity to change is a potential weakness of semi-quantitative vs quantitative approaches. Therefore, scoring “within-grade” changes between time points is now common practice and involves better longitudinal sensitivity, although requires unblinding to acquisition order⁷³. Within-grade changes are not part of published scoring systems, but are clinically relevant, as within-grade changes were shown to be associated with known OA risk factors and outcomes⁷³. Scoring within-grade change is particularly important in clinical trials, since full-grade change may not occur within relatively short follow-up of <1 year. Yet, a recent systematic review reported semi-quantitative scoring of cartilage and BMLs to be adequately responsive, and the pooled SRMs broadly consistent with those for quantitative cartilage measurement (0.55 for medial tibiofemoral cartilage and 0.43 for BMLs)⁷⁴. Another systematic review provided evidence that knee OA pain is cross-sectionally and longitudinally associated with BMLs⁷⁵. Semi-quantitative MRI outcomes (i.e., changes in cartilage damage and BMLs) have been used in several clinical trials^76–78, and recently strontium ranelate was reported to reduce the progression of BMLs in the medial femorotibial compartment⁷⁹.

Quantitative measurement of cartilage loss

Quantitative measurement of cartilage requires segmentation of the tissue in joint (compartments) of interest. The precision and inter-observer reliability for different imaging protocols, orientations, and segmentation approaches have been well documented^6,65,74. Various quantitative metrics can then be extracted (Fig. 3), for which a nomenclature was proposed by a panel of experts (Fig. 3)⁸⁰. Face-to-face comparisons of the precision and sensitivity to change of these metrics have been presented^81,82, and a core set of measures was identified, providing independent information on cartilage status and loss⁸³. This encompassed the total and denuded area of subchondral bone (tAB; dAB), and cartilage thickness (ThC)⁸³ (Fig. 3). Further, technology has been developed to extract the above metrics for defined subregions^71,81,84. As for JSW loss (see above), some studies have dichotomized structural progressors from non-progressors using variable thresholds of cartilage loss, such as precision⁸⁵, smallest detectable change (SDC)⁷, or (subregional) change in healthy subjects⁸⁶. With regard to correlation with clinical outcomes, dABs (Fig. 3), have been associated with concurrent⁸⁷ and incident knee pain⁸⁸. Knees with larger rates of femorotibial cartilage loss were more likely to subsequently receive knee replacement (KR) than those with lower rates^89,90. A recent case–control study confirmed substantially larger rates of (central medial tibial) cartilage loss in the year prior to KR than in non-replaced controls, independent of radiographic status⁹¹. The sensitivity to change of cartilage volume and thickness loss has been reported in a wide range of studies, covering different cartilage measures, cohorts and lengths of observation^7,92–95 (Table II). Predictors of cartilage loss that may be useful for enriching clinical trials with fast progressors have been reported; these have included mal-alignment^85,96, advanced radiographic OA status (i.e., greater baseline KLG or JSN grade, or smaller JSW)^97–101; frequent pain^93,100,102, and high bodyweight/body mass index (BMI)^95,97. Molecular markers were less successful in identifying progressors^98,103. Meniscus extrusion^97,104 also was identified to predict progression, but requires MRI for recruitment purposes and was suggested to not increase sensitivity to change⁸¹. Further, the sensitivity was not markedly improved when measurements were confined to specific joint regions^71,81, albeit central (femorotibial) subregions displayed somewhat greater rates and sensitivity to change than peripheral ones^7,99 (Table II). In peripheral subregions, in contrast, cartilage thickening was observed, particularly in knees with early radiographic OA⁸⁶. In contrast, a non-region specific (ordered value) system of subregion change was identified to considerably improve the differentiation of cartilage loss between knees with and without JSN¹⁰⁵. Knees with osteophytes but without JSN displayed rates of change almost indistinguishable from those in healthy knees (Table II)⁹⁹. Quantitative measures of cartilage morphology were used as outcomes of observational trials on therapeutic effects, including non-steroidal anti-inflammatory drugs (NSAIDs)^106,107, bisphosphonates¹⁰⁸, or glucosamine and chondroitin sulfate¹⁰⁷. They also were used in interventional trials, evaluating effects of physical exercise¹⁰⁹ and joint distraction¹¹⁰, and potential DMOADs such as celecoxib¹¹¹, licofelone¹¹², chondroitin sulfate⁷⁷, vitamin D supplementation¹¹³, strontium ranelate⁷⁹, and sprifermin (fibroblast growth factor 18)¹¹⁴. When drug effects on knee cartilage loss were seen, these were often greater in the lateral than medial femorotibial joint^77,112,114, potentially because the lateral compartment showed less/earlier disease. Quantitative methods were reported to be superior to semi-quantitative ones in assessing the association between risk factors and cartilage loss⁸⁵, and structure modification by drug treatment¹¹⁵.

Quantitative measurement of cartilage composition

Different MRI parameters have been proposed as quantitative markers of cartilage composition and mechanical properties (Fig. 4), with similar ability to predict compressive moduli at 1.5 T and 9.4 T¹¹⁶. One is the spin–spin (T2) relaxation time derived from a multiple echo spin echo (MESE) acquisition. Methodological work suggests that the image with the shortest TE should be eliminated from T2 computations¹¹⁷, and that T2 offsets exist between coils². It is possible to derive T2 and diffusion values from DESS^118,119, creating the potential to measure cartilage composition and thickness from the same images. T2 is sensitive to collagen matrix structure and organization in intact cartilage, and has reasonable reproducibility in multicenter studies¹²⁰ (intraclass correlation coefficient (ICC) 0.61–0.98; root mean square coefficient of variation (RMS CV) 4–14%). Satisfactory precision was also reported after splitting the cartilage layer arbitrarily into two or three horizontal layers, regions between which collagen orientation varies². Although these regions do not perfectly match the superficial, middle and deep cartilage layers as defined histologically, precise matching is challenging given limited in-plane resolution. T2 is considered more useful in morphologically intact cartilage undergoing compositional change and may have limited sensitivity to change in patients with advanced cartilage defects¹²¹.

Fig. 4.

Compositional cartilage imaging. Compositional maps for a healthy volunteer are shown as overlays on gray scale morphologic images. (A) Spin–spin relaxation time (T2) map with higher values (red) showing increased T2 (ms). (B) Relaxation time in a rotating field (T1rho) map with higher values (red) showing increased T1rho (ms). (C) Sodium MRI map with heat scale showing signal intensity that reflects cartilage sodium concentration. T2 is sensitive to collagen matrix composition and organizational changes. T1rho and sodium measures have been shown to be sensitive to changes in glycosaminoglycan content. These measures may be useful in detection and characterization of early disease before tissue loss occurs.

T1rho relaxation time has been shown to be sensitive to glycosaminoglycan content¹²², but has greater precision error than T2 (ICC 0.20–0.93; RMS CV 4–19%)¹²⁰. Also, there is disagreement on the specificity of T1rho to cartilage glycosaminoglycan¹²³. As a marker of cartilage glycosaminoglycan, dGEMRIC¹²⁴ has good reproducibility (ICC 0.87–0.95)¹²⁵, and longitudinal change was found inversely associated with that in cartilage thickness¹²⁶. However, dGEMRIC requires intravenous contrast and a 90-min delay before scanning, making study logistics challenging⁶⁷. Other imaging markers of cartilage glycosaminoglycan are Computed Tomography (CT) after injection of a charged contrast agent into the knee joint¹²⁷, sodium MRI¹²⁸ and chemical exchange saturation transfer (CEST)¹²⁹, but precision data on these techniques are lacking. Sodium MRI requires specialized coils, and sodium and CEST have greater accuracy at very high field strength¹³⁰, limiting their applicability. Ultra-short TE imaging has been applied to image the calcified cartilage layer¹³¹. In small clinical trials, dGEMRIC has been shown to be sensitive to weight-loss¹³² and, in contrast to T2, has indicated potential treatment effects of oral collagen hydrolysate on cartilage composition¹³³.

Quantitative measures of bone

Recent advances in MRI-based analysis of bone have been made, such as semi-(automated) analysis of BMLs^134–136, total bone area and shape^137–140, trabecular bone structure^134,141,142, and periarticular bone structure^136,143. Using a DESS MRI sequences, bone was mapped using >100,000 points per knee¹⁴⁰. By applying a triangulation method, certain bone shape vectors identifying knees with greater risk of incident knee OA 1 year later compared with non-incident controls¹⁴⁰. Further, change in bone area displayed greater sensitivity than change in cartilage thickness and radiographic JSW for assessing structural progression¹³⁹. These methods have not yet been applied in clinical trials, but may be of particular interest when anti-resorptive effects on bone are studied^108,144.

Dual Energy X-ray absorptiometry (DXA) demonstrated greater bone mineral density (BMD) in knees with greater radiographic OA severity and specifically in areas with BMLs¹⁴³. The latter finding was confirmed by quantitative CT, and the medial/lateral BMD ratio was significantly greater in bones with medial BMLs¹⁴⁵. CT can further provide very accurate measures of the density and structure of subchondral bone in vivo¹⁴⁶, but has not been applied in clinical trials thus far.

Direct comparison between radiography and MRI

Meta-analyses suggested a greater sensitivity to change of MRI-based cartilage loss⁷⁴ than change in radiographic JSW²², but were performed on different cohorts. Face-to-face comparison generally reported greater sensitivity to change of MRI-based cartilage measures than of radiographic JSW^30,93,95,147, but one study found a somewhat greater SRM for fluoroscopy-based Lyon-Schuss minimum JSW than for cartilage thickness loss in the medial compartment¹⁴⁸. The SRM for JSW loss in fixed flexion radiography, in contrast, was shown to be substantially less than for MRI in the same study¹⁴⁸ (Table II). The authors proposed that the high responsiveness of Lyon-Schuss may be owed to the fluoroscopic guidance providing a small IMD, and to radiography being performed under weight-bearing conditions, with diseased cartilage being potentially more compressible than healthy one. Fixed location measures of JSW have been suggested to partially compensate limitations of fixed flexion acquisition, with SRMs found similar to those of MRI-based cartilage loss³⁰. Yet, in a larger sample, greater sensitivity of MRI-based measures was reported compared with fixed location JSW over 12 and 24 months³¹ (Table II). A significant relationship was reported between side differences in radiographic minimum and fixed location JSW vs percent tibial plateau coverage by the meniscus (r² ≈ 25%) and central medial femoral cartilage thickness (r² ≈ 50%)²⁴. Other meniscus measures and cartilage regions showed weaker associations. Including only knees with reasonable alignment between the tibial plateau and the X-ray beam improved the correlation with femoral cartilage thickness (r² ≈ 65%), but not with meniscus coverage, suggesting that JSW provides a better representation of (central) femoral cartilage thickness when satisfactory radiographic positioning is achieved²⁴. Initial studies found only weak correlations between longitudinal cartilage loss in MRI and JSW loss in radiography^93,149,150. However, stronger correlations were observed when minimum (or central fixed location) JSW change was compared with cartilage loss in the central aspect of the medial femorotibial compartment^31,79,97. MRI was found superior to radiography in differentiating rates of cartilage loss in knees with advanced vs early radiographic OA¹⁰⁵, and a very small study reported superiority of MRI in predicting knee replacement relative to JSW change from extended radiographs¹⁴⁹. Also, interventional studies indicated that drug effects may be more efficiently detected using MRI-based cartilage loss than radiography^107,112

Recommendations and summary

General

• Imaging biomarkers used in clinical trials should be precise and take into account the specific mechanism of DMOAD action thereby lying on the therapeutic pathway.

• Tissues and joint regions of interest should be imaged with full anatomical coverage and adequate image orientation.

• All hardware and image settings must be kept constant between baseline and follow-up, to avoid systematic offsets biasing assessment of longitudinal change.

• Central review/measurement of images by expert readers warrants consistency and sensitivity to change.

• Power calculations should take into account the potential magnitude of DMOAD effects as well as reported sensitivity to change for various selection criteria and design parameters.

Radiography

• Standards of knee positioning (weight-bearing and fixed flexion) are important.

• Each of both knees should be imaged separately to ensure minimal X-ray beam divergence.

• The X-ray beam must be aligned with the medial tibial plateau; the IMD should be ≤1.0 mm.

• Several modifications of radiographic classification systems exist, these should therefore be clearly defined and documented.

• In fixed flexion radiographs of OA knees, central fixed location JSW measurements may be more responsive than that of minimum JSW.

MRI

• Imaging should be performed at ≥1.5 T, using dedicated (knee) coils.

• Fluid-sensitive MR images in 2–3 orthogonal planes are suited for assessing focal cartilage defects, and fat-suppressed images for BMLs.

• Scoring “within-grade” changes of semi-quantitative MRI scores between time points involves better longitudinal sensitivity, but requires unblinding to acquisition order.

• Fat-suppressed or water excitation T1-weighted gradient-echo or DESS images with ≤1.5 mm a slice thickness of and ≤0.35 mm in-plane resolution are suited for cartilage thickness analysis.

• T2 should be derived from multiple (6–7) echoes. Superficial and deep cartilage layers should be assessed separately, because collagen orientation varies. However, T2 may have limited responsiveness in patients with advanced cartilage defects.

• Other techniques for measuring cartilage composition and bone shape/structure are available and should be further evaluated in clinical trials.

When the above recommendations are taken into account, current imaging methodologies provide powerful tools for scoring and measuring morphological and compositional aspects of articular tissues, capturing longitudinal change with reasonable to excellent sensitivity. Under these conditions, imaging has tremendous potential for ascertaining treatment effects on various joint structures, potentially over shorter time scales than required for demonstrating effects on clinical outcomes.