As China's aging population continues to grow, the incidence of osteoporosis has been steadily increasing, posing a significant health challenge for the elderly population. Furthermore, the high cost of diagnosing and treating osteoporosis highlights the importance of early diagnosis as a key strategy to reduce both patient suffering and healthcare expenses.

The objective of this study is to develop a chest and abdominal bone mineral density (BMD) measurement model using conventional chest and abdominal CT scans, with deep neural networks and machine learning algorithms. The abdominal BMD model is subsequently employed to calibrate the chest BMD measurements, with the goal of enabling automated BMD measurement and the diagnosis of osteoporosis.

This retrospective study collected 702 patients from Suining Central Hospital in Sichuan Province who underwent both chest CT scans and quantitative CT (QCT) examinations during the period from March 2022 to June 2023 (spanning approximately one year) as research subjects. Among them, 532 patients were randomly divided into a training set (426 cases, 80%) and a validation set (106 cases, 20%). An additional 170 patients were included in the internal testing set. This study used the diagnostic results of QCT as the reference standard and employs machine learning methods such as logistic regression, stochastic gradient descent, and random forest to construct osteoporosis classification models and bone density regression models for the chest and abdomen, the established model was also tested internally. The performance of the classification models was evaluated using sensitivity, specificity, accuracy, precision, and area under the receiver operating characteristic curve (AUC), while regression model performance was assessed using mean absolute error (MAE), root mean square error (RMSE), and R-squared.

The results showed that the AUC values for the osteoporosis classification models in the validation set were 0.948 for the chest model and 0.968 for the abdominal model. The mean absolute errors of the BMD regression models were 10.534 and 9.449, respectively. In the internal testing set, the AUC values for the classification models were 0.905 and 0.926, and the MAE for the regression models were 9.255 and 7.924, respectively. After calibration, the AUC and MAE of the chest BMD measurement model in the validation set improved to 0.967 and 10.511, respectively.

The AI-based chest and abdominal BMD measurements demonstrate a high correlation and consistency with QCT measurements, effectively diagnosing osteoporosis. The calibrated chest BMD measurement model further enhances diagnostic performance and offers significant potential for the application of chest CT scans in opportunistic osteoporosis screening.

近年来我国老龄化趋势逐渐增高，骨质疏松的发病率也逐渐攀升，成为困扰老年人身心健康的主要疾病，而且，国内外诊断及治疗骨质疏松的成本较高，因此骨质疏松的早期诊断成了降低患者疾病痛苦及治疗成本的关键。

基于常规胸腹部CT平扫图像，通过深度神经网络和机器学习算法建立胸部和腹部骨密度测定模型，并通过腹部测定模型对胸部骨密度测定结果进行校准，实现自动化的骨密度测量和骨质疏松诊断。

回顾性收集四川省遂宁市中心医院2022年3月—2023年6月既行胸部CT检查又完成定量CT（QCT）检查的702例患者为研究对象，其中532例按照随机分组的方式分为训练集（426例，80%）和验证集（106例，20%）。另外170例作为模型的内部测试集。本文以QCT的诊断结果作为参考标准，使用逻辑回归、随机梯度下降及随机森林等机器学习方法构建胸部和腹部的骨质疏松分类模型和骨密度回归模型，同时对建立的模型进行了内部测试，并采用灵敏度、特异度、准确率、精确率及受试者工作特征曲线下面积（AUC）等指标评估模型分类性能，采用平均绝对误差、均方根误差及决定系数等指标评估模型回归性能。

胸部和腹部的骨质疏松分类模型验证集AUC值分别为0.948和0.968，骨密度回归模型平均绝对误差分别为10.534和9.449；在内部测试集中分类模型AUC值分别为0.905和0.926，回归模型平均绝对误差分别为9.255和7.924；校准后的胸部骨密度测定模型验证集AUC和平均绝对误差分别提高至0.967和10.511。

基于人工智能的胸部和腰部骨密度测定结果与QCT测定的骨密度具有高度相关性及一致性，可有效诊断骨质疏松症。经校准后的胸部骨密度测定模型也进一步提高了模型在诊断中的性能，为胸部CT平扫在骨质疏松症的机会性筛查中的应用发展提供了巨大潜力。