Investigating the Repeatability and Reproducibility of MRI-Based Porosity Assessment of Mandibular Bone
Abstract
Purpose
Radiation-induced toxicities have become more prevalent with increasing incidence of human papillomavirus-associated oropharyngeal cancer. Osteoradionecrosis (ORN) is a severe toxicity defined by decay of bone, commonly the mandible. Presently, ORN is subjectively diagnosed with clinical symptoms and confirmed on CT. Our group has implemented two MRI sequences, "black bone" (BB) and ultrashort echo time (UTE), for bony visualization used in routine follow-up imaging. We aim to use these sequences in tandem to generate a functional porosity index of bone. This preliminary study assesses the repeatability and reproducibility of said porosity index (PI) in a cohort of healthy volunteers and patients.
Methods
Three IRB-consented healthy volunteers and five IRB-consented patients were imaged with BB and UTE on 1.5T or 3T MR-simulation scanners. Volunteer repeatability measurements were performed by scanning twice in one session with repositioning in between. Reproducibility measurements were performed in volunteers and patients with an average of two weeks separation. Patients received both scans prior to treatment to ensure no changes due to radiation. PI was defined as the ratio of single-timepoint UTE-to-BB. Pre-processing included: reslicing UTE (ITK-SNAP v4.2.2); division to generate PI, normalization via the spinal cord at C1/2 level, mandible segmentation (ImageJ v1.54p). Absolute coefficient of variation (CV), repeatability and reproducibility coefficients (RC, RDC), and intraclass correlation coefficient (ICC) assessed variability.
Results
Volunteers had lower repeatability ([0.07, 0.10]) and reproducibility ([0.02, 0.10]) CVs than patients ([0.05, 0.58]). Volunteers had lower RDC ([0.17, 0.53]) than patients ([1.07, 21.18]). Additional results for volunteers include RC ([0.52, 0.73]) and ICC ([0.05, 0.43]). All metrics are measured in intensity value.
Conclusion
Volunteers maintained lower variability overall, plausibly due to their homogeneity, low tissue variance, and fewer artifacts. Future work will consist of increasing cohort size and a secondary analysis to assess the accuracy of PI measurements based on prior studies.