American Heart Technologies, Torrance, California.
American Heart Technologies, Torrance, California.
J Am Coll Radiol. 2024 Apr;21(4):624-632. doi: 10.1016/j.jacr.2023.05.006. Epub 2023 Jun 17.
Previously we reported a manual method of measuring thoracic vertebral bone mineral density (BMD) using quantitative CT in noncontrast cardiac CT scans used for coronary artery calcium (CAC) scoring. In this report, we present validation studies of an artificial intelligence-based automated BMD measurement (AutoBMD) that recently received FDA approval as an opportunistic add-on to CAC scans.
A deep learning model was trained to detect vertebral bodies. Subsequently, signal processing techniques were developed to detect intervertebral discs and the trabecular components of the vertebral body. The model was trained using 132 CAC scans comprising 7,649 slices. To validate AutoBMD, we used 5,785 cases of manual BMD measurements previously reported from CAC scans in the Multi-Ethnic Study of Atherosclerosis.
Mean ± SD for AutoBMD and manual BMD were 166.1 ± 47.9 mg/cc and 163.1 ± 46 mg/cc, respectively (P = .006). Multi-Ethnic Study of Atherosclerosis cases were 47.5% male and 52.5% female, with age 62.2 ± 10.3. A strong correlation was found between AutoBMD and manual measurements (R = 0.85, P < .0001). Accuracy, sensitivity, specificity, positive predictive value and negative predictive value for AutoBMD-based detection of osteoporosis were 99.6%, 96.7%, 97.7%, 99.7% and 99.8%, respectively. AutoBMD averaged 15 seconds per report versus 5.5 min for manual measurements (P < .0001).
AutoBMD is an FDA-approved, artificial intelligence-enabled opportunistic tool that reports BMD with Z-scores and T-scores and accurately detects osteoporosis and osteopenia in CAC scans, demonstrating results comparable to manual measurements. No extra cost of scanning and no extra radiation to patients, plus the high prevalence of asymptomatic osteoporosis, make AutoBMD a promising candidate to enhance patient care.
此前，我们报道了一种使用非对比心脏 CT 扫描进行冠状动脉钙 (CAC) 评分的定量 CT 手动测量胸腰椎骨密度 (BMD) 的方法。在本报告中，我们介绍了一种基于人工智能的自动 BMD 测量（AutoBMD）的验证研究，该方法最近获得了 FDA 的批准，可作为 CAC 扫描的一种机会性附加检查。
一个深度学习模型被训练来检测椎体。随后，开发了信号处理技术来检测椎间盘和椎体的小梁成分。该模型使用包含 7649 个切片的 132 个 CAC 扫描进行训练。为了验证 AutoBMD，我们使用了之前在动脉粥样硬化多民族研究中从 CAC 扫描报告的 5785 例手动 BMD 测量值。
AutoBMD 和手动 BMD 的平均值±标准差分别为 166.1±47.9mg/cc 和 163.1±46mg/cc（P=0.006）。动脉粥样硬化多民族研究的病例中，男性占 47.5%，女性占 52.5%，年龄为 62.2±10.3。AutoBMD 与手动测量值之间存在很强的相关性（R=0.85，P<0.0001）。AutoBMD 基于骨质疏松症的检测的准确性、敏感性、特异性、阳性预测值和阴性预测值分别为 99.6%、96.7%、97.7%、99.7%和 99.8%。AutoBMD 每份报告平均耗时 15 秒，而手动测量则耗时 5.5 分钟（P<0.0001）。
AutoBMD 是一种获得 FDA 批准的人工智能支持的机会性工具，可报告 Z 评分和 T 评分的 BMD，并在 CAC 扫描中准确检测骨质疏松症和低骨量，结果与手动测量相当。这种方法不会增加扫描的成本和患者的辐射，再加上无症状骨质疏松症的高患病率，使得 AutoBMD 成为增强患者护理的有前途的候选方案。
