基于AI技术光谱CT非增强模式椎体松质骨无水碘图测量值与骨密度对比研究

A Comparative Study of Iodine-No-Water Map Observed Value and Bone Density of Vertebral Cancellous Bone in Non-Enhanced Spectral CT Mode Based on AI

摘要: 目的：基于人工智能(Artificial intelligence, AI)技术提取光谱CT非增强模式无水碘图椎体松质骨测量值与骨密度测定的相关性对比研究，为骨密度测量提供新的方法和思路。方法：回顾性纳入317例接受了光谱CT腰椎平扫检查的健康体检者，分别使用AI骨密度测量系统和双能X线骨密度仪(dual x-ray absorptiometry, DXA)测量受检者第一腰椎到第四腰椎(L1~L4)椎体的骨密度(bone mineral density, BMD)值。采用线性回归方法分析非增强模式无水碘图测量值与上述两种方法测定骨密度值的相关性。以DXA结果为参考标准将受检者分为正常(T-score ≥ −1), 骨质减少症(−2.5 < T-score < −1)和骨质疏松症(T-score < −2.5)，将AI骨密度测量系统预测的骨质情况以及非增强无水碘图测量值预测的骨质情况分别与DXA生成的真实骨质情况进行比较。绘制受试者工作特征(receiver operating characteristic, ROC)曲线，计算受试者工作特征分析的曲线下面积(area under the curve, AUC)，利用达到约登指数的截断值评估AI系统以及非增强模式无水碘图椎体松质骨测量值诊断受检者骨质情况的准确性、敏感性和特异性。结果：非增强模式无水碘图测量值与AI系统测量L1~L4椎体的骨密度线性回归拟合优度R2为0.91~0.93；非增强模式无水碘图测量值与DXA测量L1~L4椎体的骨密度之间的相关性有所降低，相关系数R²为0.57~0.65。以DXA为诊断标准，AI系统与非增强模式无水碘图测量值在评估检测骨质疏松症上，二者均有较好的性能。结论：光谱CT非增强模式腰椎松质骨无水碘图测量值同椎体BMD值密切相关，非增强模式腰椎松质骨无水碘图测量值能很好体现椎体骨密度状态，有望成为椎体骨矿含量高低变化的敏感指标。

Abstract: Objective: To study the correlation between vertebral cancellous bone measurement and bone mineral density measurement based on the extraction of non-enhanced spectral CT Iodine-no- wa-ter Map observed value with artificial intelligence (AI), and to provide a new method and idea for bone mineral density measurement. Method: A total of 317 healthy subjects who underwent spec-tral CT lumbar scan were retrospectively enrolled using AI bone densitometry system and dual x-ray absorptiometry bone mineral density (BMD) was measured from the first to the fourth lumbar vertebrae (L1~L4). Linear regression method was used to analyze the correlation between the non-enhanced mode anhydrous iodiogram measurements and the BMD values measured by the above two methods. The patients were classified as normal (T-score ≥ −1), osteopenia (−2.5 < T-score < −1) and osteoporosis (T-score < −2.5) using DXA results as the reference standard. The bone condition predicted by the AI bone densitometry system and the non-enhanced Iodine-no- water Map observed value measurements were compared with the actual bone condition generated by DXA. The receiver operating characteristic (ROC) curve was drawn, and the area under the curve (AUC) was calculated. Truncation values reaching the Yoden index were used to evaluate the accu-racy, sensitivity, and specificity of the AI system and the non-enhanced mode Iodine-no-water Map observed value vertebral cancellous bone measurements in diagnosing patients with bone condi-tions. Results: The linear regression goodness of fit R2 between the measured values of anhydrous iodiogram in non-enhanced mode and the bone mineral density of L1~L4 vertebral body measured by AI system was 0.91~0.93. The correlation between non-enhanced anhydrous iodiogram meas-urements and DXA measurements of bone mineral density in the L1~L4 vertebral body was re-duced, with a correlation coefficient of 0.77~0.65 in R². Using DXA as diagnostic criteria, both AI system and non-enhanced mode anhydrous iodiogram showed good performance in evaluating and detecting osteoporosis. Conclusions: The measurement values of lumbar cancellous bone Io-dine-no-water Map observed value in non-enhanced CT mode are closely related to vertebral BMD values. The measurement values of lumbar cancellous bone Iodine-no-water Map observed value in non-enhanced CT mode can well reflect the status of vertebral bone mineral density, which is ex-pected to be a sensitive index for the change of vertebral bone mineral content.