关键词: 骨质疏松, 骨密度, CT平扫, 人工智能

Abstract: Background In recent years, the aging trend in China has gradually increased, and the incidence rate of osteoporosis has also gradually increased, becoming a major disease that plagues the physical and mental health of the elderly. Therefore, studying the artificial intelligence based thoracolumbar vertebral bone density measurement system and its calibration is of great significance. Objective This article establishes bone density measurement models for the thoracic and lumbar vertebrae using artificial intelligence technology based on CT plain scan images, and calibrates the bone density measurement of the thoracic vertebrae using the lumbar vertebrae measurement model to improve the accuracy of the bone density measurement model. Method Based on the exclusion criteria, 702 patients who underwent both chest CT and QCT examinations in the previous year from March 2022 to June 2023 were collected. Among them, 532 patients were randomly divided into a training set (80%) and a validation set (20%), with a random number seed set at 20. A model training set of 426 patients and a test set of 106 patients were obtained. Another 170 cases serve as external validation sets for the model. By collecting, analyzing, classifying, and post-processing patient related information and examination images, and using the scientific research platform system of relevant companies to segment and extract spinal features from the collected images, artificial intelligence technology is applied to chest and lumbar spine plain scan CT images. Osteoporosis classification models and bone density regression models are constructed using methods such as decision tree, logistic regression, random gradient descent, and random forest, At the same time, external validation was conducted on the established model. Based on the Bagging Decision Tree method, a diagnostic model for osteoporosis in the thoracic, lumbar, and thoracolumbar vertebrae was constructed. Sensitivity, specificity, accuracy, accuracy, and Area Under the Curve (AUC) were used to evaluate the diagnostic performance of the model. Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and determination coefficient R ² were used Evaluate the regression performance of the model with equal indicators; Compare the predicted BMD measurements of the developed model with the true values generated by QCT through Pearson correlation coefficient and Bland Altman plot analysis. Using the diagnostic results of QCT as a reference standard. Results The results showed that the AUC values of the classification model for osteoporosis in the thoracic and lumbar vertebrae were 0.948 and 0.968, respectively, and the MAE values of the bone density regression model were 10.534 and 9.449, respectively. After calibration, the AUC and MAE values of the thoracic vertebrae bone density measurement model increased to 0.967 and 10.511, respectively. The results of artificial intelligence based bone density measurements of the thoracic and lumbar vertebrae are highly correlated and consistent with those measured by QCT, providing great potential for the application and development of chest CT plain scan in the opportunistic screening of osteoporosis. Conclusion The results of artificial intelligence based bone density measurements of thoracic and lumbar vertebrae are highly correlated and consistent with those measured by QCT, which can effectively diagnose osteoporosis. The calibrated thoracic vertebral bone density measurement model has further improved its performance in diagnosis, providing great potential for the application and development of chest CT plain scan in the opportunistic screening of osteoporosis.

Key words: Osteoporosis, Bone density, CT plain scan, Artificial intelligence