ORMIR_XCT: A Python package for high resolution peripheral quantitative computed tomography image processing

Archive of version 1.0.0 of the ORMIR_XCT Python package for the Journal of Open Source Software submission. 

Abstract:

The use of high-resolution peripheral quantitative computed tomography (HR-pQCT) allows for imaging

bones and joints at an isotropic voxel size of 61µm. However, processing images obtained from HR-pQCT

is mostly limited to the use of the scanner manufacturer’s scripting language, the Image Processing

Language (IPL). Moreover, the images generated by HR-pQCT scanners are saved in the scanner

manufacturer’s proprietary file format (AIM or ISQ) and limits the ability to perform more advanced image

processing outside of the IPL scripting language. Thus, the ORMIR_XCT package was developed to allow

for an open-source Python framework in which existing image processing workflows for HR-pQCT images

may be implemented, and more advanced image processing functions may be added in the future.

Preliminary workflows that have been developed for the ORMIR_XCT package include: 1) automatic

contouring of the periosteal surface, 2) joint space width (JSW) analysis, 3) bone mineral density (BMD)

calculation, and 4) segmentation of trabecular bone. A dataset of synthetic shapes of varying dimensions

(plates, hollow spheres and cylinders, and solid spheres and cylinders) and in vivo scans of the 2nd and 3rd

distal interphalangeal (DIP) and trapeziometacarpal (TMC) joints were used to compare IPL and

ORMIR_XCT workflows. Segmentation implementations were compared using DICE coefficients, Jaccard

indices, as well as mean and maximum Hausdorff distances. JSW and BMD calculations were compared

using Bland-Altman and regression plots. Results show excellent agreement between IPL and

ORMIR_XCT segmentations with BMD calculations. JSW results show very good agreement, but some

variations in thickness were found for shapes with thicknesses of one voxel, with the IPL implementation

overestimating thickness by one voxel. The ORMIR_XCT package currently implements four HR-pQCT

workflows with excellent agreement with IPL generated results. This Python package provides the

groundwork for expanding current HR-pQCT workflows in a reproducible, open-source format.