Seven knee specimens from 3 males and 4 females, aged 80.6 ± 14 years (mean ± standard deviation) were obtained from the Institut d’Anatomie Paris (France). The collection of these human tissue specimens was conducted according to the relevant protocols established by the Human Ethics Committee from the Institute of Medical Research. No additional information was available regarding cause of death, previous illnesses, or any medical treatments of these subjects except for an absence of hepatitis and human immunodeficiency virus. The protocol was approved by the French Ministry of Higher Education and Research (CODECOH number DC-2019-3422). After soft tissue removal, the knee specimens were stored at -20 °C.
Data acquisition and image reconstruction
The knee specimens were imaged using a clinical PCSCT prototype system (Philips Healthcare, Amsterdam, Netherland) installed at CERMEP, Lyon. This is a modified clinical system that is equipped with a conventional X-ray tube that can be set to tube voltages from 80 to 120 kVp, and tube currents between 10 mA and 500 mA; it was set at 120 kVp and 100 mA in the present study. The tube filtration absorbs low-energy X-rays, so the final spectrum ranges from 30 keV to 120 keV. The system is based on PCDs of 2-mm-thick cadmium zinc telluride, with a pixel pitch of 270 × 270 μm2 at the isocenter, and coupled with application-specific integrated circuits (ChromAIX2, Philips Research Europe, Aachen, Germany) that operate in single-photon-counting mode with energy discrimination . The acquisition time was less than 5 min. The PCDs allow up to five consecutive energy thresholds between 30 keV and 120 keV, which were set in the present study at 30, 51, 62, 72, and 81 keV. The acquisition field of view was 500 mm in-plane, with a z-coverage of 17.5 mm in the scanner isocenter. Axial scans were performed over 360° with 2,400 projections per rotation, on a grid of 64 × 1,848 pixels.
Fifty stacks of eight slices were acquired to cover the entire knee specimens (height, 10 cm). After data acquisition, the projections in the different energy bins were decomposed on a Compton/photoelectric basis using the maximum-likelihood method . Then, the decomposed Compton/photoelectric sinograms (Radon transform) were reconstructed using filtered back-projection. The whole reconstructed images were made of 640 × 640 × 400 voxels, with a voxel size of 250 × 250 × 250 μm3 with a reconstructed scan field of view about 160 × 160 × 100 mm3. Seven virtual monoenergetic images from 40 keV to 110 keV were then computed from the linear combination of the reconstructed Compton/photoelectric images and expressed in Hounsfield units (HU) units. In addition, the conventional HU image obtained by combining all of the bins together was computed.
For comparison, we used a standard HR-pQCT imaging protocol (Scanco Wangen-Brüttisellen, Switzerland) with voxel size of 82 μm for all specimens, and synchrotron radiation monochromatic CT at 55 KeV (European Synchrotron Radiation Facility, Grenoble beamline ID 17) with a voxel size at 45 μm for 5 specimens. The Kellgren-Lawrence classification  was performed on the HR-pQCT images.
First, we investigated the multienergy feature of PCSCT. For this purpose, we analysed the monoenergetic images and compared them to the conventional HU images to select an optimal monoenergetic image for cartilage assessment. This required selection of the energy that led to the best characterisation of the cartilage in terms of noise and contrast-to-noise ratio (CNR). The noise was computed as the standard deviation (SD) for a manually selected circular region of interest with radius of 4 pixels in a homogeneous part within the cartilage. The CNR was computed as the difference for the grey levels along a line inside the cartilage and a line inside the joint space outside the meniscus divided by the SD previously computed. The monoenergetic approach allows to retrieve the information that links the X-ray attenuation to the type of material crossed and the energy. We analysed the attenuation variation of cartilage, of both trabecular and cortical bone and soft tissue with energy for all knees.
The second part of the image analysis consisted of the qualitative assessment of OA from the monoenergetic images at the selected energy. For this, the feasibility of visual assessment of cartilage defects, cartilage calcification, bone cysts, and bone osteophytes was investigated by an expert radiologist who had been in practice for > 20 years (J.B.P.).
The third part concerned the extraction of quantitative parameters from the monoenergetic images at the selected energy. We analysed bone cysts observed in the monoenergetic images for the two severe OA knee specimens. To this aim, we performed the segmentation of the subchondral bone of the femur and patella in the femoro-patellar compartments to a total depth of 10 mm from the subchondral surface. The bone cysts were segmented using the commercially available Avizo 9.0 software (FEI Visualization Sciences Group, Burlington MA, Avizo v.9.0) with the semiautomatic region growing tool (Magic Wand). Here, a seed point was first defined by the operator, and all of the connected voxels with grey levels in a given tolerance range were selected. Any object with a volume < 20 voxels (0.31 mm3) was considered as noise, and was removed from the final calculations. The bone cysts in the medial and lateral compartments were analysed separately. After segmentation, the following parameters were measured: number of cysts/mm3, total cyst volume (mm3), and maximum cyst volume (mm3). Our quantitative analysis was exploratory, so not supported by statistical analysis.