基于深度学习的机会性CT骨质疏松筛查及正常参考值的建立
本回顾性研究开发了卷积神经网络自动在胸腰椎松质骨放置三维感兴趣区,并建立校正方法以归一化不同CT协议和扫描仪型号的衰减值。分析了283499例患者共538946次CT检查,自动放置与放射科医生手动判读一致率超过99%,并基于世界卫生组织报告的骨质疏松患病率建立了诊断阈值和正常参考值。
要点
- 深度学习自动放置三维感兴趣区与放射科医生手动判读一致率超过99%。
- 80 kVp与120 kVp下松质骨衰减值相差23%,不同扫描仪型号间差异小于10%。
- 基于世界卫生组织报告的患病率建立了胸腰椎松质骨衰减值的骨质疏松诊断阈值。
核心结论
- 基于深度学习的自动机会性骨质疏松筛查可利用临床目的的CT扫描、在不同扫描仪和协议下识别骨密度低的患者。
研究设计与样本
- 回顾性研究,纳入283499例患者的538946次CT检查(平均年龄65岁±15[标准差],51.2%女性,55.5%白人),涉及43种扫描仪型号和6种管电压,参考标准为世界卫生组织报告的骨质疏松患病率。
关键影像表现/征象
- 自动三维感兴趣区放置经放射科医生手动复核验证1496个椎体,一致率超过99%。
- 80 kVp与120 kVp下衰减值相差23%,不同扫描仪型号间衰减值差异小于10%。
- 年轻女性(年龄<50岁)松质骨衰减值高于年轻男性(P<.001),衰减值随年龄下降,绝经后女性下降更陡。
- 50岁以上患者中男性松质骨衰减值高于女性(P<.001)。
- 松质骨衰减值黑人患者最高,亚裔患者次之,白人患者最低(P<.001)。
诊断标准与关键数值
- 基于世界卫生组织报告的骨质疏松患病率建立胸腰椎松质骨衰减值的骨质疏松诊断阈值。
- 采用统计方法校正不同管电压和扫描仪型号对衰减值的影响。
临床应用与阅片要点
- 可利用临床目的的CT扫描进行自动机会性骨质疏松筛查,无需额外检查。
- 判读时需注意管电压和扫描仪型号对松质骨衰减值的影响,80 kVp与120 kVp间差异达23%。
- 自动三维感兴趣区放置可用于胸腰椎松质骨衰减值量化,与手动判读一致率超过99%。
局限性
- 原文未报告局限性
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
Background Osteoporosis is underdiagnosed and undertreated, prompting the exploration of opportunistic screening using CT and artificial intelligence. Purpose To develop a reproducible convolutional neural network to automatically identify a three-dimensional (3D) region of interest (ROI) in trabecular bone, develop a correction method to normalize attenuation values across different CT protocols and scanner models, and establish thresholds for diagnosing osteoporosis in a large diverse population. Materials and Methods In this retrospective study, a deep learning-based method was developed to automatically quantify trabecular attenuation of the thoracic and lumbar spine on CT images with use of a 3D ROI. A statistical method was developed to adjust for different tube voltages and scanner models. Normative values and diagnostic thresholds for trabecular attenuation of the spine for osteoporosis were established based on the reported prevalence of osteoporosis by the World Health Organization. Differences between groups were assessed using the Student t test. Results A total of 538 946 CT examinations from 283 499 patients (mean age, 65 years ± 15 [SD]; 145 021 [51.2%] female; 157 457 [55.5%] White patients) were analyzed, representing 43 scanner models and six different tube voltages. The attenuation values at 80 kVp and 120 kVp differed by 23%, and different scanner models resulted in differences in values of less than 10%. The automated ROI placement of 1496 vertebrae was validated by manual radiologist review and demonstrated greater than 99% agreement. Trabecular attenuation was greater in young women (age <50 years) than in young men (P < .001) and decreased with age, with a steeper decline in postmenopausal women. In patients older than 50 years, trabecular attenuation was greater in male than in female patients (P < .001). Trabecular attenuation was highest in Black patients, followed by Asian patients, and lowest in White patients (P < .001). Conclusion Deep learning-based automated opportunistic osteoporosis screening can identify patients with low bone mineral density using CT scans obtained for clinical purposes with use of different scanners and protocols. © RSNA, 2025 Supplemental material is available for this article. See also the editorial by Feuerriegel and Sutter in this issue.
引用
- Westerhoff M, Gyftopoulos S, Dane B, et al. Deep Learning-based Opportunistic CT Osteoporosis Screening and the Establishment of Normative Values. Radiology. 2025. PMID: 41217284. doi:10.1148/radiol.250917
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。