Clinical impact of adopting an AI-driven opportunistic bone health screening software using routine X-ray.

Abstract

Objectives: To evaluate the clinical impact of an artificial intelligence device, Rho, that opportunistically screens X-rays for low bone mineral density (BMD; DXA T-score < -1).

Materials and methods: Over 13 months, Rho analyzed X-rays from patients ≥ 50 years at a large independent health facility. Radiologists could opt to include Rho-flagged findings in their X-ray reports. DXAs that occurred within 6 months of an X-ray (from patients in the first 7 months of data collection) were categorized as being prompted by Rho ("Rho-generated") or arranged via usual standard-of-care practice ("pre-planned"), and their outcomes (diagnoses and 10-year fracture risk scores) were compared.

Results: Of 34,162 X-rays, Rho flagged 19,004 (56%) for low BMD, and radiologists included this information in 7726 (41%) reports. From the first 7 months of radiologists reporting Rho findings, initial and surveillance DXAs increased by factors of 1.8 and 1.4, respectively. Of 299 Rho-generated DXAs, 193 had low bone mass (- 2.5 < T-score < - 1) and 65 had osteoporosis. Rho-generated DXAs vs. pre-planned DXAs identified a greater proportion of patients with low BMD (87% vs. 69%; p < 0.001) and similar proportions of patients with elevated fracture risk (34% vs. 40%). The diagnostic yield was particularly marked in patients undergoing their first-ever DXA (84% vs. 60%; p < 0.001), and in males ≥ 65 years (83% vs. 45%; p < 0.001).

Conclusion: Incorporating Rho in radiology workflow nearly doubled initial bone health assessments and prompted additional surveillance DXA evaluations. Rho-generated DXAs captured higher rates of true low BMD than standard-of-care practices, particularly in older men.