Opportunistic osteoporosis screening using multiphase abdominal CT: reliable proximal femur BMD assessment with AI-assisted segmentation.
Authors
Affiliations (4)
Affiliations (4)
- Department of Radiology, Zhuhai Hospital, Guangdong Provincial Hospital of Chinese Medicine, Guangdong Province, Zhuhai, China.
- Medical Imaging Center, Qujing Central Hospital of Yunnan Province, Yunnan Province, Qujing, China.
- Canon Medical Systems China Co Ltd, Guangdong Province, Guangzhou, China.
- Department of MR, The First People's Hospital of Yunnan Province, The Affiliated Hospital of Kunming University of Science and Technology, Kunming, Yunnan, China.
Abstract
To evaluate the agreement and classification concordance of artificial intelligence (AI)-assisted proximal femur bone mineral density (BMD) assessment using multiphase contrast-enhanced abdominal computed tomography (CT), with unenhanced CT-derived quantitative computed tomography (QCT) serving as the reference standard. This retrospective study included 429 patients (mean age, 60.48 ± 12.30 years) who underwent a standardized five-phase abdominal CT protocol, comprising unenhanced, arterial, portal venous, late venous, and excretory phases. A commercially available AI-based QCT workstation was used for semi-automatic proximal femur segmentation and volumetric BMD (vBMD) and T-score calculation. Repeated-measures ANOVA and Bland-Altman analysis were performed to assess inter-phase differences and agreement. Classification concordance was evaluated using the unenhanced CT-derived QCT results as the reference standard. The study cohort comprised 378 osteoporotic patients and 51 non-osteoporotic patients, categorized according to unenhanced CT-derived T-score reference values. Results from repeated-measures ANOVA indicated significant phase-related variations in T-scores and vBMD (both <i>P</i> < 0.001). Subsequent <i>post-hoc</i> comparisons revealed that T-scores and vBMD in the arterial and portal venous phases were consistent with the unenhanced phase, whereas modest yet statistically significant deviations were observed in the late venous and excretory phases. Bland-Altman analyses showed minimal mean bias across arterial, portal venous, and excretory phases; however, the late venous phase exhibited broader limits of agreement (LoA). Classification concordance compared to the unenhanced reference was consistently high across all phases, ranging from 98.14 to 99.53%. Additionally, intraobserver reproducibility remained excellent, with ICC values between 0.964 and 0.996. AI-assisted proximal femur BMD assessment on contrast-enhanced abdominal CT showed high agreement and classification concordance with unenhanced CT-derived QCT. The arterial and portal venous phases were particularly suitable for opportunistic osteoporosis screening, although external validation against DXA or independent QCT remains necessary.