Sample acquisition and imaging
This was a retrospective analysis of a CT database of images from approximately 30 total shoulder arthroplasty patients. This database included both preoperative clinical-CT scans and postoperative micro-CT scans of the humeral head osteotomies acquired at surgery. These CT scans were assessed for the presence of cysts, and we excluded scans that had cysts with undefined boundaries, including cysts that went into the articular surface and cysts that bled into each other. We found that 10 patients had mild to moderate cystic presence with subchondral bone cysts that were suitable for measurement. Thus, the inclusion criteria was osteoarthritis with measurable cysts in the region of the humeral head osteotomy.
The ten selected humeral head osteotomies, excised at the head to neck junction, were collected from a subset of patients with primary end-stage osteoarthritis who had previously undergone total shoulder arthroplasty (mean age: 65 ± 12 years old; 7 males; 3 females). Each humeral head osteotomy was scanned using a micro-CT scanner (Nikon XT H 225 ST, Nikon Metrology, Cambridge, ON, Can; 20 µm isotropic voxels, 95 kV, 80 µA, 3141 projections, 1000 ms exposure). Corresponding preoperative clinical-CT scans (GE Discovery CT750 HD, Milwaukee, WI, USA; pixel dimension 0.527–0.645 mm, slice thickness 1.25 mm, 120 kVp, 200 mA) for each patient were gathered retrospectively from patient charts. The micro-CT osteotomy was registered to the respective clinical-CT scan by iterative closest point registration of the bone [9]. This study was approved by the Health Sciences Research Ethics Board at Western University (HSREB# 113,023).
Image analysis
Cysts were defined as an elliptical, spherical or irregularly shaped volume > 1.25 mm in diameter of lower grayscale (lower Hounsfield unit [HU]) surrounded by an area of higher grayscale (higher HU) [4, 6, 12]. Individual cysts were identified on the micro-CT scan for measurement in this study (Fig. 1). If multiple cysts were identified in a subject, each were treated and analyzed independently as separate cases for comparison between segmentation methods; two subjects had multiple cysts, providing a total of 13 cysts to be evaluated (n = 13). Identified cysts were manually measured in micro-CT using free hand segmentation aided by automated edge detection, similar to a previously introduced gold standard segmentation technique [10] (Mimics v.20.0, Materialise, Leuven, BE). These micro-CT volume measurements served as the reference standard to which the clinical-CT segmentation techniques were compared.
Corresponding cyst volumes were measured in clinical-CT using four different techniques: Qualitative, Edge Detection, Region Growing, and Thresholding (Fig. 2). Cysts were segmented using the built-in image processing tools in commercially available software (Mimics v.20.0). These techniques were selected based on their prevalence in image analysis software and reported use in previously published studies [5, 14, 15]. All clinical-CT segmentation measurements were completed by two independent, blinded graders (JM & AP). A fellowship-trained shoulder surgeon completed the measurements (JM), as well as a researcher (AP) trained by senior researchers with guidance from a radiographic atlas [1] and previous literature evaluating cysts using CT scans. Volume data were collected following all measurements to reduce bias.
Qualitative measurements consisted of slice-by-slice freehand segmentation of voxels qualitatively identified to be part of the cyst by the grader based on their relative intensity. Edge Detection worked by semi-automated determination of the cyst boundary using object contours automatically computed using gradient magnitude and manually placed points on multiple planes of the scan. Region Growing worked by extracting all connected voxels from a manually placed seed point guided by cyst-specific thresholds defining the midpoint HU value of the boundary between cyst interior and surrounding bone, similar to the half maximum height threshold technique [13]. Thresholding is a technique where every voxel below a specified intensity value is selected. The mean cyst-specific midpoint intensity value calculated for Region Growing measurements was used as a universal threshold value for all Thresholding measurements. Segmentation was completed primarily on the axial and coronal planes. Detailed descriptions of the segmentation methods used are provided as an additional supporting file (Additional file 1: Table S1).
Statistical analysis
Measurement techniques were assessed on two parameters: accuracy to micro-CT standard and reproducibility, with cyst volumes measured in micro-CT used as the standard for comparison. Statistical analysis was performed in SPSS 26.0 (IBM Corp., Chicago, IL). The CT database from which this retrospective analysis was derived was a representative sample of osteoarthritic shoulder arthroplasty patients at our surgical centre, and the final number of 10 patients included in this study represents the incidence of mild to moderate cystic OA within the original group.
Descriptive statistics (mean, standard deviation [SD], and range) of cyst volumes measured in clinical-CT are reported. Accuracy of each technique relative to the micro-CT standard volume measurements was assessed using linear regression and Bland–Altman analysis, as volume measurements are continuous variables. Values of p < 0.05 were considered statistically significant. Root mean square error (RMSE) from the micro-CT standard was also calculated to assess accuracy. The reproducibility of each technique was assessed by calculating intraclass correlation coefficient (ICC) for the inter-observer measurements based on a single rating, consistency, 2-way mixed-effects model. The 95% CI of the ICC estimates were used, with values > 0.90 indicating excellent reliability, values between 0.75 and 0.9 indicating good reliability, and values between 0.5 and 0.75 indicating moderate reliability [8].