Abstract

Background

This study aimed to evaluate the performance of DeepXray™ Spina, a software that estimates bone mineral density (BMD) and T-scores from frontal lumbar spine X-ray (FLS-X), in predicting osteoporosis.

Methodology

Patients from a Japanese cohort who underwent both FLS-X and dual-energy X-ray absorptiometry (DXA) using Hologic systems within 30 days at Tohoku University Hospital (May 2014-April 2024) were included. BMD was estimated from FLS-X using DeepXray™ Spina, which was developed using dataset from a Taiwanese Cohort. BMD assessed by DXA (observed BMD) and BMD estimated from FLS-X by DeepXray™ Spina (estimated BMD) were compared using Pearson’s correlation coefficient (PCC) and normalized root mean square error (NRMSE). T-scores were converted to osteoporosis classifications as normal, osteopenia, or osteoporosis following the World Health Organization criteria. Classification performance was evaluated by accuracy, sensitivity, specificity, Cohen’s kappa, and quadratic-weighted Cohen’s kappa.

Results

The correlation between estimated and observed BMD was strong, with a PCC of 0.901 and an NRMSE of 0.070. For osteoporosis classification, the accuracy, sensitivity, specificity, and Cohen’s kappa were as follows: 0.902, 1.000, 0.842, and 0.803 for normal; 0.854, 0.729, 0.924, and 0.673 for osteopenia; 0.951, 0.810, 1.000, and 0.863 for osteoporosis. The quadratic-weighted Cohen’s kappa was 0.884.

Conclusion

This study evaluated the performance of Deep Xray™ Spina in predicting osteoporosis from FLS-X. The software is a practical and reliable tool for predicting osteoporosis, with high performance and robustness. Journal of Clinical Densitometry