Impact of coil design on the contrast-to-noise ratio, precision, and consistency of quantitative cartilage morphometry at 3 Tesla: A pilot study for the osteoarthritis initiative

Authors


Abstract

Phased-array (PA) coils generally provide higher signal-to-noise ratios (SNRs) than quadrature knee coils. In this pilot study for the Osteoarthritis Initiative (OAI) we compared these two types of coils in terms of contrast-to-noise ratio (CNR), precision, and consistency of quantitative femorotibial cartilage measurements. Test-retest measurements were acquired using coronal fast low-angle shot with water excitation (FLASHwe) and coronal multiplanar reconstruction (MPR) of sagittal double-echo steady state with water excitation (DESSwe) at 3T. The precision errors for cartilage volume and thickness were ≤2.6% for the quadrature coil and ≤2.3% for the PA coil with FLASHwe, and ≤2.3%/≤2.5% with DESSwe. The precision for aggregate medial and lateral cartilage measures was significantly higher than that for single plates, independently of coil and sequence. The PA coil measurements did not always fully agree with the quadrature coil measurements, and some differences were significant. The higher CNR of the PA coil did not translate directly into improved precision of cartilage measurement; however, summing up cartilage plates within the medial and lateral compartment reduced precision errors. Magn Reson Med 57:448–454, 2007. © 2007 Wiley-Liss, Inc.

Quantitative magnetic resonance imaging (qMRI) of articular cartilage can provide valuable insights into the structural status of joints and its changes in osteoarthritis (OA). qMRI of cartilage morphology therefore provides a powerful tool for epidemiological studies of OA and shows great promise for evaluating the treatment response of disease-modifying drugs (1–3). The Osteoarthritis Initiative (OAI), a program jointly sponsored by the National Institute of Health (NIH), the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS), and the pharmaceutical industry, is targeted at identifying the most reliable and sensitive biomarkers for evaluating the development and progression of symptomatic knee OA.

Validation studies have shown that at a field strength of 1.5T, MRI can provide accurate information on cartilage morphology (volume, thickness, and surface areas) if fat-suppressed or water-excited (we) spoiled gradient-recalled echo (SPGR or fast low-angle shot (FLASH)) sequences and state-of-the-art analysis tools are used (3). Recent studies have shown that the precision (reproducibility) errors of these measurements can be reduced by the use of a FLASHwe sequence at 3T, and that measurements at 3T are consistent with those previously validated at 1.5T (4). Also, it has been shown that in the femorotibial joint, double-echo steady state with water excitation (DESSwe) at 3T provides precision equivalent to that of FLASHwe, and that the measurements are highly correlated (5).

However, to date, quantitative measurements of cartilage morphology at 1.5T and 3T have been confined to quadrature transmit-receive (T/R) knee coils. Although phased-array (PA) coils (6) have been in clinical use for over a decade, they have not been used for qMRI of cartilage. Since PA coils generally result in improved signal-to-noise ratios (SNRs) and contrast-to-noise ratios (CNRs), there is potential for improved precision in qMRI of knee cartilage morphology. In this pilot study for the OAI we therefore tested the hypotheses that precision errors of cartilage morphometry can be effectively reduced using a quadrature transmit, eight-channel PA receive coil (InVivo Corp., Orlando, FL, USA) compared to a quadrature T/R coil (USA Instruments, Aurora, OH, USA), and that measurements obtained with the PA coil are consistent with those obtained with the previously validated quadrature knee coil (5). It is important to test the second hypothesis to determine whether one can potentially exchange a quadrature vs. a PA coil in the course of a longitudinal study to improve precision without introducing bias. Additionally, we tested the hypothesis that one can reduce precision errors by summing the measurements in the medial tibia (MT) and medial femur, and the lateral tibial (LT) and lateral femur, respectively. This approach could eliminate the variability that occurs when the tibial and femoral joint surfaces are segmented in the femorotibial contact zones. Because the segmentation variability may depend on the SNR/CNR and the specific cartilage contrast, we tested the latter hypothesis using data acquired with quadrature and PA coils, and with both FLASHwe and MPR DESSwe sequences.

MATERIALS AND METHODS

Study Participants and MRI

Ten adult subjects (three men and seven women; five healthy and five with a clinical diagnosis of OA) underwent a test-retest examination of either their left (N = 4) or right (N = 4) knee, and two subjects underwent examination of both their right and left knees. In total, 12 knees were examined (six with clinical OA and six without OA). The participants were selected according to the OAI study design and documentation, and recruited from two clinical centers (Ohio State University and Memorial Hospital of Rhode Island). The symptomatic knee OA patients had to have experienced knee pain, aching, or stiffness on the majority of days within a month in the past 12 months, and to have knee OA as diagnosed by a physician. Participants without knee OA had only infrequent knee pain, aching, or stiffness over the prior year, if at all. The exclusion criteria included rheumatoid or inflammatory arthritis, orthopedic hardware in the knee, pregnancy, or other contraindications for MRI. Seven of the 10 subjects in this pilot study also participated in the OAI and underwent nonfluoroscopically guided fixed-flexion posterior-anterior (P/A) radiography (7) on both knees. The other three subjects were asymptomatic and had no prior diagnosis of OA. Based on the OAI knee radiographs, five subjects had Kellgren-Lawrence grade (KLG) 1, one subject had KLG 2, and one subject had KLG 3. The mean age of the participants was 52.2 years (range = 45–73 years) and the mean body mass index (BMI) 28.2 kg/m2 (range = 21.8–34.6 kg/m2). The study was performed in compliance with principles originating in or derived from the Declaration of Helsinki (Revised Edinburgh, 2000) and with IRB informed-consent regulations and International Congress of Harmonization Good Clinical Practices guidelines. The study protocol, amendments, and informed-consent documentation were reviewed and approved by the local institutional review boards.

Images were acquired on a 3T MRI scanner (Siemens Magnetom Trio, Erlangen, Germany) using both a quadrature T/R knee coil (USA Instruments Inc.) and a quadrature transmit, eight-channel PA receive coil (InVivo Corp). The acquisition sequences included a double oblique coronal 3D FLASH sequence with water excitation (corFLASHwe) with a slice thickness of 1.5 mm, an in-plane resolution of 0.31 mm × 0.31 mm, and an acquisition time of 8 min 30 s. This protocol has been validated at 1.5T (3, 8–10) and was recently cross-calibrated between 1.5T and 3T (4). The other acquisition was a sagittal 3D DESSwe (sagDESSwe) sequence that was prescribed orthogonal to the corFLASHwe with a slice thickness of 0.7 mm, and 0.37 mm × 0.46 mm in-plane resolution (interpolated to an isotropic in-plane resolution of 0.37 mm), and had an acquisition time of 10 min 23 s (5). The specific parameters for the corFLASHwe were TR = 20 ms, TE = 7.6 ms, FA = 12°, 80 slices, FOV = 160 mm; matrix = 512 × 512, bandwidth = 130 Hz/pixel, 0% phase oversampling, 0% slice oversampling, 100% phase resolution, 75% slice resolution, one average, elliptical filter on, asymmetric echo off, right/left phase encoding, fast gradient, and fast RF options. The orientation of the images was similar to previous recommendations (4, 5, 11), with the posterior ends of the medial and lateral femoral condyle being ideally located in the same (coronal) slice and not more than two slices apart. In contrast to the study by Glaser et al. (11), slices were oriented parallel to the femoral shaft, and not perpendicular to the tibial plateau. The pulse sequence parameters for the sagDESSwe were TR = 16.3 ms, TE = 4.7 ms, FA = 25°, 160 slices, FOV = 140 mm; matrix = 384 × 307; bandwidth = 185 Hz/pixel, 0% phase oversampling, 10% slice oversampling, 80% phase resolution, 100% slice resolution, one average, elliptical filter on, asymmetric echo off, A/P phase encoding, fast gradient, and fast RF options. A multiplanar reformat (MPR) of the sagDESSwe (corMPR DESSwe) was performed in the double oblique coronal plane, identically oriented as the corFLASHwe. The slice thickness was 1.5 mm, the in-plane matrix 384 × 269, and the in-plane resolution 0.37 mm × 0.7 mm (interpolated to an isotropic in-plane resolution of 0.37 mm).

The subjects were positioned on the patient table feet first and supine, with the lower end of the patella located at coil isocenter. The coils were positioned directly on the patient table, 60 mm right/left off the magnet isocenter. For the quadrature coil, the knee was slightly flexed, with a product coil cushion placed beneath the knee, and the heel positioned directly on the patient table. A strap was placed across the top of the quadrature coil and the contralateral knee to secure the position. The knee angle on the InVivo PA coil was fixed at about 10°. With both coils the foot was secured in a vertical position.

Each subject underwent eight MR acquisitions. On one day, a test-retest examination was performed with the corFLASHwe and sagDESSwe, respectively, using one of the two coils. Between the MR exams the participant was removed from the magnet and allowed to walk for about 10 min. On another day, within 1 month of the first MR exam, the same four acquisitions were repeated using the other coil.

All MR images were reviewed for quality and immediately reacquired if the scans were unacceptable (due to inappropriate orientation, incomplete anatomical coverage, motion artifact, etc.). Only a sagittal (and not a coronal) DESSwe was acquired because sagittal DESS allows analysis of the entire knee, including the femoropatellar joint and posterior femoral cartilage. However, a previous study showed that cartilage measurements from a corMPR DESSwe displayed similar precision and consistent findings in comparison with sagDESSwe and corFLASHwe (5). In this study we analyzed only the corMPR DESSwe because of the better comparability with the orientation of the corFLASHwe, and because the 1.5-mm sections used in corMPR DESSwe require less segmentation time than 0.7-mm sections.

Measurements of tissue signal levels were made mid-joint (A/P), near the slice where the trochlea diverges into the femoral condyles (Fig. 1). Four cartilage and four bone marrow signal intensity (SI) measurements were made in the middle of each condyle (two on the medial and lateral femur, and two on MT and LT) for the corMPR DESSwe and corFLASHwe acquired in both coils. Additionally, two measurements of noise, subcutaneous fat, and meniscal SI levels were made on the medial and lateral sides of the knee, respectively. The signal levels from the four acquisitions were measured without a reference (except for slice location) because this was important for detection of fluid in the joint.

Figure 1.

Comparative images of corMPR DESSwe (left) and corFLASHwe (right) acquired with the quadrature coil (top) and the PA coil (bottom).

Cartilage Morphometry and Statistical Analysis

The 96 image series (12 × 4 corFLASHwe, and 12 × 4 corMPR DESSwe) were anonymized, and the image analysis center was blinded to subject identification. However, we grouped the two repeat measurements (with subject repositioning in between exams) together to perform quantitative analysis in paired fashion (4, 12). One experienced reader (M.K.), who was formally trained in image segmentation, processed all of the images and analyzed the following cartilage plates: MT, LT, central medial femur (cMF), and central lateral femur (cLF) (4, 12). Segmentation involved manual tracing of the total area of the subchondral bone (tAB) and the area of the cartilaginous joint surface (AC) on a slice-by-slice basis, with osteophyte cartilage excluded (4, 5). The MT and LT were segmented throughout all slices that displayed cartilage, except for slices that had substantial partial-volume averaging at the margins of the cartilage plates. Analysis of the femur was confined to the portion of the condyles that is displayed without excessive partial-volume effect in coronal images (cMF and cLF) (4, 5). The femoral coronal landmarks started anteriorly with the first slice showing an interruption of the subchondral bone (divergence of the trochlea into the femoral condyles), and ended posteriorly with the last slice showing the circular structure of the posterior femoral condyles (bone centrally surrounded by cartilage). The slice located 60% between the anterior and posterior landmarks was the most posterior slice to be included in the cMF and cLF. These slices were automatically identified by the software (see below) after the anterior and posterior landmarks were marked interactively, and the software permitted segmentation entries for cMF and cLF only in these slices and not in others. The 60% criterion was selected based on previous findings that in the anterior 60% of the slices, partial-volume averaging was small enough to support segmentation, whereas more posteriorly partial-volume contributions became too large for clear identification of tAB and AC (5). The same regions were identified in the corMPR DESSwe. We matched the number of slices in the paired data sets by importing the ROI after interactively identifying the anterior landmark in the partner data set.

Proprietary software (Chondrometrics GmbH, Ainring, Germany) was used to calculate the cartilage volume (VC), tAB, cartilage volume divided by tAB (VCtAB), AC, part of the tAB covered with cartilage (cAB), mean cartilage thickness (ThCcAB), and mean thickness when counting all denuded areas as 0-mm cartilage thickness (ThCtAB) (4, 5, 13) from the segmented regions. To determine whether summed measurements of the medial femorotibial compartment (MFTC) were more precise than the individual measurements of MT and cMF across coils and acquisition protocols, we added values in the MT and cMF. The same was done in the lateral femorotibial compartment (LFTC) with LT and cLF.

We determined the precision for paired analysis by computing the root-mean-square (RMS) coefficient of variation (CV%) for all cartilage plates (14). To quantify differences between the coils and MR sequence contrast across cartilage plates, we computed the average precision error across MT, cMF, LT, and cLF for each sequence and coil combination. The individual log (CV%) values between coils were compared by means of a paired Student's t-test. We evaluated the consistency between the two coils by computing the Pearson correlation coefficients, the random (%) pairwise differences, and the systematic (%) pairwise differences, with the mean of the two quadrature coil measurements being in the denominator. Systematic differences between coils were tested for statistical significance using a paired t-test. To evaluate whether the precision of aggregate values in MFTC and LFTC were better than those averaged in the single cartilage plates, we compared the individual log(CV%) values with the average values of MT, cMF and LT, cLF, respectively, using a paired t-test.

RESULTS

SNRs and CNRs

Images acquired with the PA coil showed higher SNR and CNR of cartilage vs. surrounding tissues compared to the quadrature coil (Fig 1). The CNR increases were greater with corFLASHwe than with corMPR DESSwe (Fig. 2). With corFLASHwe, low-SI joint fluid was detected in only two of eight knees. An effusion was clearly visible on eight of eight images with corMPR DESSwe.

Figure 2.

Bar graphs showing the CNR of the cartilage vs. surrounding joint tissues for the quadrature and PA coils for both the corFLASHwe and corMPR DESSwe sequences.

Cartilage Morphometry

The average precision errors for area measurements (tAB, cAB, and AC) across all four cartilage plates were ≤1.0% for corFLASHwe on the quadrature coil, ≤1.1% for corFLASHwe on the PA coil, ≤1.3% for corMPR DESSwe on the quadrature coil, and ≤1.3% for corMPR DESSwe on the PA coil (Tables 1 and 2). For measures of area, there were no significant differences in precision errors between the coils, except for tAB of cMF with the corFLASHwe (P < 0.05) (Table 1).

Table 1. Coefficients of Variation (RMS CV) for Repeated Measurements of Quantitative Knee Cartilage Morphology for Test-Retest Acquisitions of the Coronal FLASHwe Sequence Using the Previously Validated Quadrature (Quad) Coil and the Phased Array (PA) Coil
 tABcABACVCVCtABThCtABThCcAB
  • All values are shown in %.

  • a

    Aggregate values of MFTC are derived by first adding individual values in MT and cMF, and aggregate values of LFTC by first adding individual values in LT and cLF, and then by computing the precision errors for the aggregate values.

  • b

    Average precision obtained by computing the average of the four RMS CV values observed in MT, cMF, LT, and cLF.

  • *

    P < 0.05 for comparing the log(CV) values of the PA with the quadrature coil (paired t-test).

  • MT = medial tibia, cMF = central medial femur, MFTC = medial femorotibial compartment, LT = lateral tibia, cLF =central lateral femur, LFTC = lateral femorotibial compartment, AVG = average precision errors in MT, cMF, LT, and cLF, tAB = total area of the subchondral bone, cAB = cartilage covered area of the subchondral bone area, VC = cartilage volume, VCtAB = volume of cartilage normalized to subchondral bone area, ThCtAB = mean cartilage thickness over total subchondral bone area, including denuded areas with 0 mm cartilage thickness, ThCcAB = mean cartilage thickness over cartilage covered subchondral bone area.

Quad Coil FLASHwe
 MT0.90.90.92.11.72.02.0
 cMF0.71.01.13.43.53.33.3
 MFTCa0.70.70.51.11.61.31.4
 LT1.11.21.21.71.21.21.1
 cLF0.80.80.93.12.92.72.7
 LFTCa0.90.90.91.51.51.41.4
 AVGb0.91.01.02.62.32.32.3
PA Coil FLASHwe
 MT1.01.00.82.32.02.02.0
 cMF1.2*1.11.13.13.02.62.6
 MFTCa0.90.90.82.22.12.02.0
 LT1.11.21.31.61.41.51.4
 cLF0.90.91.12.1*1.81.71.7
 LFTCa0.91.01.11.61.41.41.4
 AVGb1.11.01.12.32.02.02.0
Table 2. Coefficients of Variation (RMS CV) for Repeated Measurements of Quantitative Knee Cartilage Morphology Parameters for Test-Retest Acquisitions of the Coronal MPR DESSwe Sequence Using the Previously Validated Quadrature (Quad) Coil and the Phased Array (PA) coil*
 tABcABACVCVCtABThCtABThCcAB
  • *

    All values are shown in %.

  • a

    Aggregate values of MFTC are derived by first adding individual values in MT and cMF, and aggregate values of LFTC by first adding individual values in LT and cLF, and then by computing the precision errors for the aggregate values.

  • b

    Average precision obtained by computing the average of the four RMS CV values observed in MT, cMF, LT, and cLF.

  • *P < 0.05 for comparing the log(CV) values of the PA with the quadrature coil (paired t-test).

  • MT = medial tibia, cMF = central medial femur, MFTC = medial femorotibial compartment, LT = lateral tibia, cLF =central lateral femur, LFTC = lateral femorotibial compartment, AVG = average precision errors in MT, cMF, LT, and cLF, tAB = total area of the subchondral bone, cAB = cartilage covered area of the subchondral bone area, VC = cartilage volume, VCtAB = volume of cartilage normalized to subchondral bone area, ThCtAB = mean cartilage thickness over total subchondral bone area, including denuded areas with 0 mm cartilage thickness, ThCcAB = mean cartilage thickness over cartilage covered subchondral bone area.

Quad Coil DESSwe
 MT1.31.31.21.82.01.91.9
 cMF1.31.31.42.62.32.12.0
 MFTCa1.11.11.10.81.21.11.2
 LT1.41.41.52.21.81.81.8
 cLF1.21.21.12.72.42.42.4
 LFTCa1.21.21.31.91.21.11.1
 AVGb1.31.31.32.32.12.02.0
PA Coil DESSwe
 MT1.71.81.92.71.71.61.6
 cMF0.71.31.12.22.32.22.0
 MFTCa1.21.61.62.11.71.51.1
 LT1.11.21.02.11.92.02.0
 cLF0.90.91.02.82.92.92.9
 LFTCa1.01.00.81.92.02.02.1
 AVGb1.11.31.32.52.22.22.1

The average precision errors for cartilage volume and thickness values (VC, VCtAB, ThCtAB, and ThCcAB) across all four cartilage plates were ≤2.6% for corFLASHwe on the quadrature coil and ≤2.3% on the PA coil, and they were ≤2.3% for corMPR DESSwe on the quadrature coil and ≤2.5% on the PA coil (Tables 1 and 2). Measures of volume and thickness tended to have slightly higher precision errors with the PA coil, but there were no significant differences in log (CV%) values except for VC of cLF with corFLASHwe (P < 0.05) (Table 1).

Precision errors in MFTC and LFTC were generally smaller than averaged precision errors in MT/cMF and LT/cLF, respectively, and the trend was consistent across sequences and coils (Tables 1 and 2).

Correlation coefficients for FLASHwe (Table 3) measurements made on the two coils were relatively high (≥0.94). No significant differences between coils were found in most variables, except for measures of area (tAB, cAB, and AC) in cMF and cLF, and ThCcAB in cMF and cLF. With the DESSwe image contrast, correlations between coils were somewhat smaller (r ≥ 0.90; Table 4) and a significant overestimation (up to 5.4%) was observed in most PA coil measures of MT. There were also significant overestimations of cAB (2.5%), AC (2.8%), VC (3.5%) in LT, and of tAB (2.2%) in cMF (Table 4). With FLASHwe image contrast, random differences between the two coils of up to 6.6% ± 9.9% were observed (VC of cMF), and with DESSwe contrast differences of up to 5.4% ± 2.7% (VC of MT) were noted.

Table 3. Random and Systematic Differences (With Significance Levels) and Correlation Coefficients for Quantitative Knee Cartilage Morphology Parameters from Coronal FLASHwe With the Phased (PA) Array Coil vs. the Previously Validated Quadrature Coil
 tABcABACVCVCtABThCtABThCcAB
  • Systematic differences given for PA vs. quadrature coil with quadrature coil used as reference:

  • **

    P < 0.05;

  • **

    P < 0.01; (paired t-test).

MT
 Random pairwise difference (%)1.7 ± 1.32.1 ± 1.92.1 ± 1.93.2 ± 6.23.0 ± 5.22.9 ± 5.53.0 ± 5.4
 Systematic pairwise difference (%)+0.1+0.4+0.2+2.4+2.3+2.6+2.3
 Correlations0.990.990.990.990.950.950.94
LT
 Random pairwise difference (%)2.5 ± 2.42.4 ± 2.32.5 ± 2.13.0 ± 2.43.5 ± 2.13.5 ± 2.23.1 ± 1.8
 Systematic pairwise difference (%)+0.2+0.3+0.3−0.2−0.4−0.4−0.6
 Correlations0.980.990.980.990.960.970.97
cMF
 Random pairwise difference (%)4.1 ± 3.35.8 ± 8.65.5 ± 8.76.6 ± 9.96.3 ± 6.75.8 ± 6.54.1 ± 2.6
 Systematic pairwise difference (%)+3.8**+5.5**+5.2**+2.6−1.3−1.2−2.6**
 Correlations0.970.940.950.980.970.970.98
cLF
 Random pairwise difference (%)3.3 ± 3.33.7 ± 3.73.7 ± 3.55.9 ± 3.03.6 ± 2.83.6 ± 2.63.7 ± 2.5
 Systematic pairwise difference (%)+3.1**+3.5**+3.2**+0.9−2.2−2.2−2.6**
 Correlations0.980.980.980.970.940.940.95
Table 4. Random and Systematic Differences (With Significance Levels) and Correlation Coefficients for Quantitative Knee Cartilage Morphology Parameters From Coronal MPR DESSwe With the Phased Array (PA) Coil vs. the Previously Validated Quadrature Coil
 tABcABACVCVCtABThCtABThCcAB
  • Systematic differences given for PA vs. quadrature coil with the quadrature coil used as reference:

  • *

    P < 0.05

  • **

    P < 0.01

  • ***

    P < 0.0001; (paired t-test).

MT
 Random pairwise difference (%)2.7 ± 1.92.7 ± 2.02.9 ± 2.65.4 ± 2.73.9 ± 2.33.4 ± 2.23.1 ± 2.6
 Systematic pairwise difference (%)+1.4+2.1*+2.7**+5.4***+3.9***+3.4***+2.9**
 Correlations0.990.990.991.000.990.990.98
LT
 Random pairwise difference (%)3.4 ± 2.33.2 ± 2.13.8 ± 2.84.0 ± 3.92.9 ± 2.53.2 ± 2.02.9 ± 1.8
 Systematic pairwise difference (%)+2.0+2.5*+2.8*+3.5*+1.5+1.6+1.2
 Correlations0.980.990.980.990.960.970.97
cMF
 Random pairwise difference (%)2.9 ± 3.02.7 ± 3.12.5 ± 2.34.3 ± 3.43.7 ± 2.83.1 ± 2.73.1 ± 2.6
 Systematic pairwise difference (%)+2.2*+1.8+1.7+1.7−0.4−0.5−0.1
 Correlations0.970.980.990.990.980.980.98
cLF
 Random pairwise difference (%)2.1 ± 2.01.8 ± 1.72.3 ± 1.74.3 ± 3.04.7 ± 3.04.1 ± 2.73.9 ± 2.4
 Systematic pairwise difference (%)+1.3+1.0+1.0+1.0−0.2−0.20.0
 Correlations0.990.990.990.980.900.920.94

DISCUSSION

As expected, we found that the SNR and CNR of cartilage and other tissues were improved with the PA coil. However, we cannot explain why the gain in SNR was stronger for the corFLASHwe than for the corMPR DESSwe, and why it was stronger for cartilage and muscle (cor FLASHwe). Phantom measurements (data not shown) were not able to clarify these observations. Contrary to expectations, however, the reproducibility of cartilage morphology measurements was not improved with the PA coil for either corFLASHwe (which displayed considerable improvements in SNR and CNR) or corMPR DESSwe (which displayed relatively little improvement in SNR and CNR). We therefore assume that the higher CNR did not directly translate into improved delineation of the bone interface (tAB) and cartilage surface (AC). This may be because SNR/CNR measurements are typically performed at the center of the tissues, whereas segmentation is performed on the edges, where sharp contrast and unambiguous SI characteristics are required to attribute the voxels to either cartilage or noncartilage tissue. Our results therefore indicate that these local SI gradients, which are required for segmentation, are not sufficiently improved by the PA coil to reduce precision errors of quantitative cartilage measurements in the knee using FLASHwe or DESSwe. Nevertheless, because precision errors were similar overall between the two coils, these data demonstrate that PA coils can be used for longitudinal qMRI studies on cartilage morphology.

Significant systematic differences in the absolute values of some measures were observed in some (but not all) plates. These differences are likely due to the higher SNR provided by the PA coil, which can potentially result in an overestimation of the cartilage due to a more extensive inclusion of surface voxels compared to the quadrature coil. For this reason we recommend that the coil should not be exchanged between baseline and follow-up acquisitions in a longitudinal study, because even small systematic deviations or bias can affect longitudinal measurements given that the annual changes in knee OA measured by qMRI are relatively small (15–19).

The limitations of this study include the relatively small number of participants, the limited number of repeated measurements performed (only a test and retest), and the low grade of OA present in the participants. Because all measurements were taken in the same subjects, however, paired statistical tests were applicable and reduced the impact of intersubject variability. Another limitation is that the femoropatellar and posterior femoral cartilages were not measured. These regions cannot be accurately delineated with a corFLASHwe because of the large partial-volume effects involved. The acquisition of sagittal FLASHwe images, however, would have required additional (fourfold greater) imaging time. The use of the PA coil only moderately improved the SNR and CNR of the DESSwe sequence; therefore, our analysis focused on the central (weight-bearing) femorotibial compartment since we could compare the data from that compartment with those from the corFLASHwe.

Precision errors were found to be significantly improved for aggregate cartilage morphometry values in the medial and lateral femorotibial compartments compared to the single cartilage plates. It is interesting and important that this finding was consistent across the two coils and the two sequences. Raynauld et al. (19) previously reported precision errors for aggregate values in the tibia and femur at 1.5T, but did not compare these with the reproducibility of the contacting cartilage plates. We show that by eliminating the variability of the surface definition of the tibial and femoral cartilage in the contact zone of the two plates, which is one of the most difficult tasks of segmentation, one can significantly improve the precision. In turn, there is a loss of information from the contacting cartilage plates. Because a longitudinal change in knee cartilage in OA may not be highly correlated between the tibia and femur (20), we suggest that measurements should be computed initially for the contacting cartilage plates before the aggregate values for the compartment are synthesized.

In conclusion, this study shows that the improved SNR/CNR observed for cartilage imaging using a PA coil does not directly translate into improved (but similar) precision (reproducibility) of quantitative cartilage measurements in the knee using 3D FLASH or 3D DESS image contrast compared to a quadrature knee coil. Since significant differences in absolute values between the two coils were observed in some plates, changing the coil over the course of a longitudinal study is problematic. However, one can significantly reduce precision error by using aggregate values for the tibial and femoral compartments. The latter finding was consistent across acquisitions with differences in SNR/CNR and image contrast.

Acknowledgements

This pilot study was conducted as part of the OAI, in collaboration with OAI investigators and consultants. The manuscript was reviewed by the OAI Publications Committee for scientific content and data interpretation. We are grateful to the Ohio State University team (particularly Kim Toussant) and the Center for Primary Care and Prevention team of Memorial Hospital of Rhode Island for recruiting the study subjects, and to Larry Martin RTR(MR) and Lynn Fanella RTR(MR) for acquiring the MR images.

Ancillary