JACR has published our manuscript, prepared in collaboration with our research partners at CaMos (Canadian Multicentre Osteoporosis Study), that reports the performance of Rho in three independent data sets. Rho takes as inputs a conventional x-ray (of the lumbar spine, thoracic spine, chest, pelvis, hand or knee), age, and sex, and outputs a Rho Score of 1-10. The paper reports the performance of the Rho Score at distinguishing patients with and without low BMD (at FN or L1-4 as measured by DXA), assessed using receiver-operating characteristic (ROC) analyses in patients from community imaging centres (n=3,729, 83% female), the Canadian Multicentre Osteoporosis Study (n=1,780, 71% female) and the Osteoarthritis Initiative (n=591, 50% female). The areas under the ROC curves were 0.89 (0.87–0.90), 0.87 (0.85-0.88) and 0.82 (0.79–0.85), respectively, and subset analyses showed similar results for each sex, body part and race.
The journal also printed a commentary by Drs. Varney and Kapoor, which stated that "overall accuracy for the three datasets was excellent and consistent" and "Rho shows strong potential for widespread use given the vast number of radiographs conducted on a routine basis". The authors of the commentary concluded that there "remains a need for validation in a prospective and multi-institutional screening population to confirm Rho’s real-world utility in the opportunistic detection of low BMD". Stay tuned for a CAN Health study at Sunnybrook Hospital which evaluated Rho prospectively in over 1000 patients!
Bilbily A, Syme CA, Adachi JD, Berger C, Morin SN, Goltzman D, Cicero MD. Opportunistic screening of low bone mineral density from standard x-rays. J Am Coll Radiol. 2023 Oct 5:S1546-1440(23)00715-9. doi: 10.1016/j.jacr.2023.07.024. Epub ahead of print. PMID: 37805012.
Varney ET, Kapoor N. Utility of Machine Learning Algorithms for Opportunistic Bone Density Screening Using Conventional Radiographs. J Am Coll Radiol. 2023 Sep 16:S1546-1440(23)00716-0. doi: 10.1016/j.jacr.2023.08.043. Epub ahead of print. PMID: 37722464.