Validation of quantitative magnetic resonance imaging-based apparent bone volume fraction in peri-articular tibial bone of cadaveric knees

Cadaveric specimen

Ten fresh cadaveric whole knee specimens were procured from five donors by the National Disease Research Interchange. Inclusion criteria were that the donor’s age was 51 to 80 years and neither knee had pathology with the exception of OA. Race and gender were not donor criteria. Each specimen was at least 16 cm in length and included at least 8 cm on either side of the tibiofemoral joint. Both knees were collected within 10 hours of death, wrapped in gauze, packed in wet ice or ice packs, and sealed for shipping. Within 36 hours of death, the knees underwent MR imaging. After the MR exam, the knees were repackaged with ice packs and shipped to a laboratory for preservation and preparation for microCT. The knees were placed in 4% paraformaldehyde in phosphate buffer within 58 hours of death and were immersion fixed in that solution for 1 week at 4°C, with three changes of fixative. The Institutional Review Board at Tufts Medical Center and Tufts University Health Sciences Campus declared that this study was not human subject research and therefore informed consent for this study was not collected. The National Disease Research Interchange requires that participating hospitals obtain informed consent for the bodies to be used for research purposes from the patient or family.

Magnetic resonance imaging

We obtained coronal-oblique 3-dimensional fast imaging with steady state precession (3D FISP) sequences (Figure 1) [17] on a Siemens Trio 3-Tesla MR system and a USA Instruments quadrature transmit-receive knee coil at one of the Osteoarthritis Initiative clinical sites (Memorial Hospital of Rhode Island). The double coronal-oblique orientation was used, with the posterior edge of the medial and lateral femoral condyles in the same slice. Thereafter, the head/foot orientation is aligned parallel to the femoral diaphysis. This orientation has provided reliable assessment of femoro-tibial cartilage loss [20, 21] and tibial subchondral trabecular bone [1, 2, 17] by optimizing the measurement plane to be perpendicular to weight-bearing knee cartilage and subchondral bone. The images were acquired in 10.5 minutes using 72 slices, 1 mm slice thickness, 0.23 mm × 0.23 mm in-plane spatial resolution, 12 cm field of view (FOV), 512 × 512 matrix (interpolated to 1024 × 1024), 4.92 ms echo time (TE) (fat-water in-phase), 20 ms recovery time (TR), 50° flip angle, 180 Hz/pixel readout bandwidth, and phase encode right/left. The chemical shift artifact is 2.4 pixels shifted superior, outside the femoral subchondral bone. We previously used this MR protocol among a convenience sample of the Osteoarthritis Initiative [1, 2]. Quality assessments of the MR images were performed for contrast, FOV placement, and absence of motion artifacts.

To determine if there were any bone marrow lesions in the tibia, contiguous sagittal intermediate-weighted turbo spin echo (IW-TSE), fat-suppressed MR images were obtained in the same session as the 3D FISP. Acquisition parameters were: 160 mm FOV, 3 mm slice thickness, 30 ms TE, 3200 ms TR, with an in-plane spatial resolution of 0.357 mm × 0.511 mm, 384 × 269 matrix (interpolated to 512 × 512), “strong” fat suppression (spectral spatial), echo train length 5, 40% phase oversampling, 248 Hz/pixel readout bandwidth, and phase encode anterior/posterior [20].

Magnetic resonance analyses

A single reader analyzed the coronal 3D FISP images using custom software. As previously described [1, 2] a single reader first determined a standardized cortical bone signal intensity threshold by placing 20 regions-of-interest (ROI; 0.69 mm × 0.69 mm) in the cortical bone along the subchondral plate of the medial and lateral femoral condyles (Additional file 1: Figure S1). We used a standardized cortical bone signal intensity as a conservative threshold. We hypothesized that this conservative threshold would cause MR-based apparent BV/TV to systematically underestimate microCT-based BV/TV but also minimize the influence of abnormal bone marrow signal on MR imaging. Additional file 1: Figure S2 is an example of an image after we applied this conservative threshold. To calculate the MR-based trabecular morphometry, a rectangular ROI was positioned on each of the 20 consecutive central MR images [1, 2, 17] in the proximal medial tibial epiphyseal subchondral bone, just distal to the cartilage (Figure 1). To improve the reliability of the trabecular morphometry metrics, the ROI included the subchondral cortical bone. The ROI had a constant height of 3.75 mm and width of 15.00 mm. Apparent BV/TV was calculated [22] for each image, thereafter the values from the 20 images were averaged. In brief, apparent BV/TV represents the ratio of the number of pixels with signal intensity at or below the signal intensity threshold defined by cortical bone divided by the total [17] number of pixels within the ROI.

Intra-rater (measurement-remeasurement; at least 48 hours apart) reliability was excellent with an intraclass-correlation coefficient (3,1 model) of 0.99 (n = 10) [23]. Furthermore, the smallest and largest paired differences (measurement-remeasurement) in apparent BV/TV were -0.002 and 0.007. The root mean square coefficient of variation (RMS %CV) for the test-retest measurements was 21.4%.

Micro-computed tomography

The tibial plateau of each knee specimen was divided into medial and lateral compartments using bone saws (Mar-Med Inc., Cleveland, OH, USA). The center of each medial plateau was identified. We then measured 10 mm into the anterior and posterior planes, each, from this center point, and marked the bones with permanent marker at those locations; next we measured 15 mm into the medial planes, each, from this center point, again marking those points before cutting the medial tibia into 14 mm (medial-lateral) × 20 mm (anterior-posterior) × 6 mm in height (with articular cartilage still intact) rectangular pieces using a bone saw (Dremel 4000, Robert Bosch Tools Corporation, Racine WI, USA). These osteochondral specimens were scanned in air with a high resolution, ex vivo, cone-beam microCT scanner (Skyscan 1172, 12 megapixel camera model; Microphotonics, Allentown, PA, USA): X-ray source voltage of 59 kV, current of 167 μA, source spot size of 300 nm, aluminum 0.5 mm filter, a rotation step of 0.40°, frame averaging of 4, a ring artifact correction of 10, a beam hardening correction of 40%, 1335 slices, and isotropic voxel resolution of 9 μm. The average scan duration for each osteochondral specimen was 36 minutes. The images were reconstructed into 3D images using cone-beam reconstruction software (Skyscan NRecon, Aartselaar, Belgium) based on the Feldkamp algorithm, a process that yielded 9 μm thick sections in the axial plane. Note, calibration of the Skyscan is performed twice monthly; background corrections were performed before each scan.

Micro-computed tomography analyses

Structural indices were calculated using the Skyscan CT Analyzer software (CTAn; Aartselaar, Belgium). Trabecular morphometric traits were computed from binarized images using direct 3D techniques that do not rely on prior assumptions from the underlying structures. The volume of interest for trabecular microarchitectural variables was based on the MR ROI, and on the positioning and selection of the osteochondral specimens that underwent the microCT exam. The osteochondral specimens were extracted by identifying the center weight-bearing zone of each medial plateau. We then measured 10 mm into the anterior and posterior planes, each, from this center point – this strategy replicated the 20 consecutive central MR images. Next, we measured 15 mm medial and lateral from this center point. In addition, several study team members (JBD, TEM, GHL, and MFB) reviewed the final microCT volumes to ensure that they corresponded to the MR ROI. The volume of interest was 17 mm (anterior-posterior), 13.729 mm (medial-lateral), and 3.939 mm (vertical). Thresholding or “segmentation” was performed using simple global methods. The binary grayscale range of the Skyscan instrument is from 0 (air, black) to 255 (most solid structure, white), and is indicative of the resorptive properties of the structure scanned, in this case bone. Thus, we used an upper threshold of 255, which captures the densest bone. We also chose a lower threshold of 80 using the grayscale histogram feature of the software, which showed a clear dip in detection of bone versus non-bone structures. We have also used this lower threshold in a number of other publications examining bone structures [24]. The density range of the system is regularly calibrated against "phantoms" of known bone mineral density content and thus Houndsfield units, in which the lower grayscale density of air (0) is equal to -1000 HU, and the highest density of 255 is equal to 9200 Houndsfield units. We computed BV/TV using a marching-cubes algorithm.

Statistical analyses

The primary outcome measures were MR-based and microCT-based trabecular BV/TV in the proximal medial tibia. We evaluated validity by calculating the association between MR-based and microCT-based trabecular morphometry with Spearman rank correlation coefficients as well as agreement between measures with Bland-Altman analyses. Based on a priori power computations, a sample size of 10 knees was expected to provide adequate power to determine criterion validity (r > 0.80, power > 0.80, alpha < 0.05). This power calculation is supported by prior research in other anatomical locations, which detected correlation coefficients r > 0.78 using as few as five cadaveric specimens [4]. For these analyses, each knee was considered an independent measurement; however, we also conducted secondary analyses specific to left and right knees.