Radius fractures are among the most common fracture types;​ however, there is limited consensus on the standard of care. A better understanding of the fracture healing process could help to shape future treatment protocols and thus improve functional outcomes of patients. High-resolution peripheral quantitative computed tomography (HR-pQCT) allows monitoring and evaluation of the radius on the micro-structural level, which is crucial to our understanding of fracture healing. However, current radius fracture studies using HR-pQCT are limited by the lack of automated contouring routines, hence only including small number of patients due to the prohibitively time-consuming task of manually contouring HR-pQCT images.

In the present study, a new method to automatically contour images of distal radius fractures based on 3D morphological geodesic active contours (3D-GAC) is presented. Contours of 60 HR-pQCT images of fractured and conservatively treated radii spanning the healing process up to one year post-fracture are compared to the current gold standard, hand-drawn 2D contours, to assess the accuracy of the algorithm. Furthermore, robustness was established by applying the algorithm to HR-pQCT images of intact radii of 73 patients and comparing the resulting morphometric indices to the gold standard patient evaluation including a threshold- and dilation-based contouring approach. Reproducibility was evaluated using repeat scans of intact radii of 19 patients.

The new 3D-GAC approach offers contours within inter-operator variability for images of fractured distal radii (mean Dice score of 0.992 ± 0.005 versus median operator Dice score of 0.992 ± 0.006). The generated contours for images of intact radii yielded morphometric indices within the in vivo reproducibility limits compared to the current gold standard. Additionally, the 3D-GAC approach shows an improved robustness against failure (n = 5) when dealing with cortical interruptions, fracture fragments, etc. compared with the automatic, default manufacturer pipeline (n = 40). Using the 3D-GAC approach assures consistent results, while reducing the need for time-consuming hand-contouring.