摘要： 目的 基于人工智能系统（AI）肺结节系统结合能谱CT比较不同单能量肺结节的诊断效能，筛选不同性质结节的最佳单能量。方法 收集28例能谱CT图像，AI肺结节系统结合70 keV，共标注231个肺结节作为 “ 金标准”，结节性质包括实性、亚实性（部分实性、纯磨玻璃）和钙化3类。分析40、60、80和100 keV 4组图像结节，对比 “ 金标准”，计算灵敏度、阳性预测值及假阳性率。结果 4组单能量检测肺结节的灵敏度及阳性预测值差异均有统计学意义（P均<0.05），80 keV灵敏度、60 keV阳性预测值较高。单能量组内，不同性质结节灵敏度差异有统计学意义（P均<0.05），除40 keV外其他能量组内阳性预测值差异有统计学意义（P均<0.05）。实性结节中不同能量组间灵敏度及阳性预测值差异有统计学意义（P均<0.001），80 keV以上组灵敏度显著高于40、60 keV，60 keV以上组阳性预测值显著高于40 keV。亚实性结节中不同能量组间灵敏度及阳性预测值差异无统计学意义（P>0.05）。钙化结节中4组灵敏度均为100%，阳性预测值差异无统计学意义（P=0.843）。结论 AI肺结节系统与80 keV以上组结合具有较高的肺结节检出灵敏度。不同单能量中80 keV以上组检测肺实性结节效能最优。

人工智能肺结节筛查系统结合能谱CT比较不同性质肺结节的检测效能

钟宇, 周姝, 张立波, 刘文源, 杨本强, 邹明宇

北部战区总医院放射诊断科, 沈阳 110016

收稿日期:2022-05-19出版日期:2023-07-30发布日期:2023-07-08
通讯作者:邹明宇E-mail:mingyu_zou@msn.com
作者简介:钟宇(1985-),女,主治医师,硕士.
基金资助:辽宁省自然科学基金（2019-ZD-1056）


关键词: 人工智能, 肺结节筛查, 能谱CT, 体层摄影术
Abstract: Objective To compare the diagnostic efficacy of different single energy levels for detecting pulmonary nodules using artificial intelligence（AI）combined with energy spectrum CT，and to identify the best single energy for different types of nodules. Methods Twenty-eight cases of energy spectrum CT images were collected，and 231 pulmonary nodules were labeled as the “gold standard” using an AI pulmonary nodule system combined with 70 keV. The nodules were classified as solid，sub solid（partial solid，pure ground glass），or calcified. We analyzed the four groups of images of nodules obtained at 40，60，80，and 100 keV，and compared them to the “gold standard”. We then calculated the sensitivity，positive predictive value，and false positive rate. Results There were statistically significant differences in the sensitivity and positive predictive value of single energy detection of pulmonary nodules among the four groups（P<0.05）. The sensitivity of 80 keV and the positive predictive value of 60 keV were higher than those of the other energies. Within the single energy group，there was a statistically significant difference in the sensitivity of nodules with different properties（P<0.05），and there was a statistically significant difference in the positive predictive value in other energy groups except for 40 keV（P<0.05）. For solid nodules，there was a statistically significant difference in the sensitivity and positive predictive value between different energy groups（P<0.001）. The sensitivity of energy groups above 80 keV was significantly higher than those of 40 and 60 keV groups； moreover， the positive predictive value of energy groups above 60 keV was significantly higher than those of the 40 keV group. For sub solid nodules， there was no statistically significant difference in the sensitivity and positive predictive value between the different energy groups（P>0.05）. For calcified nodules，the sensitivity of all four groups was 100%，and the difference in positive predictive value was not statistically significant（P=0.843）. Conclusion The combination of AI pulmonary nodule system with energies of above 80 keV has a high sensitivity for detecting pulmonary nodules. Different single energy levels with 80 keV and above are optimum for detecting solid pulmonary nodules.
Key words: artificial intelligence, pulmonary nodule screening, energy spectrum CT, tomography
PDF全文下载地址:

https://journal.cmu.edu.cn/CN/article/downloadArticleFile.do?attachType=PDF&id=3247