Automated micro-CT morphometry of femoral biopsies from hip arthroplasties: adaptive local thresholding, volume of interest wrapping and removal of debris

Bone biopsies are an important biological tool for investigating bone microarchitecture, which can be non-destructively imaged in 3D via micro-computed tomography (micro-CT). Image thresholding and delineation of a region of interest (ROI) are prerequisites for quantifying bone parameters. Validated automatic protocols enable quantification of biopsies that contain trabecular and cortical bone. However, irregularly shaped trabecular bone biopsies with peripheral and internal debris have required manual ROI delineation, which is time-intensive and subject to inter and intra-observer variance. We hypothesise that an automated workflow will be a suitable alternative to overcome these issues and objectively determine bone microarchitecture in surgical biopsies, at higher throughput suitable for clinical studies. Hence, the aim of this study was to develop an objective, reproducible and automated workflow to analyse microarchitecture of trabecular bone biopsies. To accomplish this aim, we tested six different methods of ROI delineation: a whole biopsy ROI, and both manual (slow) and automatically delineated (fast) reduced ROIs to remove peripheral debris, each with (adaptive thresholding and a set of morphological operations to remove debris) and without (global thresholding) processing in a subset (n = 8) of intertrochanteric femoral biopsies obtained from patients undergoing hip arthroplasty. Number of objects, bone volume to tissue volume (BV/TV), trabecular separation (Tb.Sp), structure model index (SMI) and Euler number and trabecular pattern factor (Tb.Pf) were compared between the six workflows using Friedman's test and post-hoc pairwise comparisons with Bonferroni correction was performed. The two most reproducible techniques were tested for validation in a larger cohort of arthroplasty patients (n = 60) and results were compared with appropriate t-test. Subset analysis indicated that the manual and automated ROI with processing increased the ability to resolve real differences between these groups in parameters BV/TV, Tb.Sp and Euler number compared to with no processing and whole biopsy ROI approach. A validation cohort consisted of thirty osteoarthritis patients with a mean age 68.25 ± 8.64 and thirty neck of femur fracture with a mean age 82.4 ± 8.9. The manual technique failed to detect differences in BV/TV, SMI and Tb.Pf between the two patient groups (p > 0.05, for all) while the automated workflow demonstrated significant differences in these parameters between the OA and the NOF patients (p < 0.05). This is probably due to irregularity in the reference VOI volume introduced by manual ROI delineation reducing morphometric precision, compared to the automated method. In conclusion, our automated workflow performed better than customary practice; it represents a user-independent, high throughput technique to measure bone microarchitecture accurately in surgical biopsies.