Supporting data for ＂Development of a novel bone void detection algorithm based on phantom-less quantitative computed tomography technique and its application in osteoporosis and osteoporotic fracture screening＂

Background: Early screening and intervention for osteoporosis is the key to prevent osteoporotic vertebral compression fracture (VCF). Phantom-less quantitative computed tomography (PL-QCT) technique provides opportunistic screening of osteoporosis but its accuracy is influenced by ROI (region of interest) selection of reference soft tissues. An automatic ROI selection might improve the performance of PL-QCT. Moreover, novel algorithms based on routine computed tomography (CT) images could provide alternative tools for bone quality assessment.

Objectives: This thesis aimed to (1) validate the accuracy of an automatic PL-QCT system and extend its application to endplates and subchondral bones; (2) develop a novel bone void detection algorithm and validate its accuracy; (3) explore the bone void distribution in thoracic and lumbar vertebrae; (4) assess the reliability and feasibility of bone void in screening osteoporotic VCF.

Methods and results: (1) The automatic PL-QCT system was compared with a phantom-based QCT (PB-QCT) system via a retrospective cohort and a prospective multi-center cohort. Bone mineral density (BMD) measured by two systems highly correlated, and the bias was nearly negligible (-0.9984±16.52mg/cm3). QCT identified higher osteoporosis proportion than dual energy X-ray absorptiometry (DEXA) and screened osteoporosis with good sensitivity and specificity. Furthermore, PL-QCT was used to detect the BMD distribution of endplates and subchodral bones. The lower endplate and subchondral bone demonstrated higher BMD than the upper ones, respectively. The outer edge exhibited higher BMD than central area in the endplates. The posterior regions presented higher BMD than anterior regions in the subchondral bones. (2)Based on PL-QCT technique, a novel bone void detection algorithm was developed. A bone void was defined as area with a BMD <40mg/cm3, and a volume >16.5mm3. Artificial bone voids were made in the vertebral body of pig lumbar specimen and detected by our algorithm and third-party software, respectively. The results showed a high consistency and an acceptable bias in void volume between two groups. (3)The multi-center QCT data was used to explore the bone void distribution in thoracolumbar spine. The lumbar vertebrae displayed a higher bone void detection rate and a larger normalized void volume than the thoracic vertebrae. Additionally, void volume correlated positively with age and increased rapidly after age 55. Bone void could distinguish normal, osteopenia, and osteoporosis with high sensitivity and specificity. (4) Retrospective age and gender-matched case and control cohorts were used to investigate the discriminatory ability of bone void for VCF. Results showed that VCF group exhibited lower BMD and T-score, but larger bone void indicators relative to control group. The fracture severity positively correlated with estimated bone void indicator. Moreover, bone void showed superior discriminatory ability compared to BMD and T-score, offering higher sensitivity and specificity.

Conclusions: The automatic PL-QCT system was a reliable technique and could be applied on vertebral endplates and subchonral bones. This study further developed a novel bone void detection algorithm and validated its reliability for bone quality assessment. The bone void could be a novel intuitive indicator to screening osteoporosis and discriminate osteoporotic fracture, outperforming DEXA-derived BMD.