Robust microCT image preprocessing workflow for quantitative morphometric analysis (QMA) of the knee in small and medium animal models

Pholpat Durongbhan, PhD Candidate
Dr. Kathryn S. Stok

Integrative Cartilage Research Group at the University of Melbourne
Department of Biomedical Engineering, The University of Melbourne, Victoria 3010 Australia

​Impact: Modern medical imaging serves as a powerful tool for understanding disease progression, performing accurate diagnosis, and facilitating development of appropriate treatments. Computed tomography (CT) imaging provides 3-dimensional structural information in a non-invasive and real-time manner, making it a key modality in musculoskeletal preclinical and clinical research. Automating the image processing workflow, as described here, will allow for a rapid and reproducible quantitative analysis of joints. Fast and reliable measurements can then be used to track and assess disease progression in osteoarthritic joints.

Introduction: Recent studies [1], [2] have proposed a suite of 17 quantitative morphometric analysis measures (QMA) describing structures of the joint to assess it as a single organ and have demonstrated its reproducibility and sensitivity in assessing joint health in preclinical rat and rabbit models using micro-computed tomography (microCT). However, the accessibility and use of quantitative measurements of joint morphometry have been limited by its high sensitivity to tibial alignment and appropriate volume of interest (VOI) selection of joint compartments; often a challenging and time-consuming manual task.

Objective: The objective of this work is to develop a novel automatic, efficient, and model-invariant image preprocessing pipeline that allows for highly reproducible 3D quantitative morphometric analysis (QMA) of the joint.

Methods: Two modules working as a pipeline were developed to tackle to problems of tibial alignment and volume of interest (VOI) selection of joint compartments. Joint alignment is achieved by representing the tibia’s basic form using lower degree spherical harmonic basis functions (SPHARM) [3] and performing alignment analytically using principal component analysis. The second module subdivides the joint into appropriate lateral and medial VOIs with a novel application of a watershedding approach based on persistence homology [4]. Multiple repeated microCT scans of small (rat) and medium (rabbit) animal knees were processed using the pipeline to demonstrate its model invariance. Existing 3D joint QMA was performed to evaluate the pipeline’s ability to generate reproducible measurements. A summary of the workflow is shown in Figure 1.

Results: Typical results of the alignment module can be seen in Figure 2. Measurements for the joint centre of mass, cartilage contact area under virtual loading, joint space width, and joint space volume showed excellent reproducibility (all intraclass correlation coefficients > 0.75) with less than 9.5% root-mean-squared error compared to manual processing results from previous studies [1], [2]. Compared to earlier manual work, this workflow has reduced time (average time < 4 minutes per sample) and technical requirements (fully automated, no recalibration between models) needed to preprocess joint images for 3D joint QMA.

Conclusions: The software provides an automated and efficient preprocessing solution that allows highly reproducible 3D joint QMA for two animal models. This work can increase the accessibility of 3D joint QMA measurements, which shows potential as a platform to quantify disease-based morphometric features for joint research from microCT scans using multiple preclinical animal models.
Please find further details of this work at: https://doi.org/10.1038/s41598-021-04542-8.

Figure 1: Overview of the pipeline as well as the relevant joint QMA used to evaluate each process. 3D microCT masks of cartilage (left), femur (central), and tibia (right) of a typical rat knee is used to highlight each process’s input and result.

Figure 2: Typical images of a segmented rat knee joint (a) before alignment and (b) after processing through the automated alignment software.