Evaluating an Artificial Intelligence software for opportunistic low bone mineral density and osteoporosis screening:...

Area of Science:

• Radiology
• Artificial Intelligence
• Osteoporosis Screening

Background:

• Osteoporosis poses a significant fracture risk, yet access to gold-standard DXA scans is limited, particularly in rural areas.
• Existing diagnostic methods for osteoporosis are not universally accessible, creating a gap in early detection and intervention.

Purpose of the Study:

• To independently validate the performance of Rho, an artificial intelligence (AI) software, in identifying patients at risk for low bone mineral density (BMD) and osteoporosis using radiographs.
• To evaluate Rho's effectiveness as an opportunistic screening tool, especially in populations with limited access to DXA scans.

Main Methods:

• A retrospective study involving 4878 patients (mean age 70 ± 10, 80% female) with DXA scans acquired within one year of a radiograph.
• The area under the curve (AUC) was calculated to assess Rho's performance in predicting low BMD (T-Score < -1) and osteoporosis (T-Score ≤ -2.5).
• Subgroup analyses were conducted based on radiograph location, sex, and rural versus urban populations.

Main Results:

• Rho demonstrated high accuracy in predicting low BMD (AUC = 0.840) and osteoporosis (AUC = 0.815).
• The AI tool showed strong and comparable performance in both rural and urban populations for both low BMD and osteoporosis detection.
• Rho performed effectively in both men and women, although optimal thresholds varied by sex.

Conclusions:

• Rho software is highly effective in identifying individuals at risk for low BMD and osteoporosis using standard radiographs.
• The findings support Rho's utility as an opportunistic screening tool, potentially bridging the gap in communities with limited DXA access.
• AI-powered radiographic analysis offers a promising solution to improve osteoporosis screening accessibility and early detection.