不同CT扫描条件对胸部模体实性结节人工智能检出效率及辐射剂量的影响

目的探讨不同CT扫描条件下人工智能(AI)系统对胸部模体内实性结节检出效率与辐射剂量的影响。方法于仿真胸部拟人模体内各肺叶和肺段均匀放置不同CT值和直径的60颗不同形态的仿真结节。应用GE Revolution evo CT对胸部模体进行扫描, 通过调节管电压80、100、120和140 kV, 噪声指数(NI 10～40, 间隔2), 其他参数固定, 采集64组不同参数图像。在AI软件上记录仿真结节检出情况并计算检出率与误检率, 不同形态结节分别计算;记录每次扫描平均容积CT剂量指数(CTDIvol)、剂量长度乘积(DLP)。结果不同管电压对类球形结节和不规则结节的检出率、误检率差异均无统计学意义(P>0.05);不同噪声指数对类球形结节和不规则结节的检出率、误检率差异均存在统计学意义(F=10.57、17.77、9.33, P<0.001)。不同管电压对CTDIvol、DLP差异无统计学意义(P>0.05), 不同噪声指数对CTDIvol、DLP差异具有统计学意义(F=59.87、60.92, P<0.001)。结节的检出率与噪声指数、CTDIvol、DLP...

The influence of different CT scanning protocols on AI detection efficiency and radiation dose of solid nodules in chest phantom

Objective To investigate the radiation dose and detection efficiency of artificial intelligence (AI) system for solid nodules in chest phantom with different scanning protocols.Methods A total of 60 simulated nodules with different CT values and diameters were uniformly placed in each lung lobe and lung segment of the anthropomorphic chest phantom. GE Revolution evo CT was used to scan the chest phantom. 64 groups of images with different scanning parameters were collected at the tube voltage of 80, 100, 120, 140 kV, different noise indexes (NI 10-40 with interval 2), and other fixed parameters. The detection result of simulated nodules were recorded on AI software, and the detection rate and false detection rate were calculated, respectively, for different shapes of nodules. The mean volume CT dose index (CTDI_vol) and dose length product (DLP) of each scan were recorded.Results There were no statistically significant differences in the detection rate and false detection rate of spherical nodules and irregular nodules at different tube voltages(P > 0.05), but there were and statistically significant with different noise indices (F=10.57, 17.77, 9.33, P < 0.001). Different tube voltages had no statistical significance for CTDI_vol and DLP (P > 0.05), while different noise indices had statistical significance for CTDI_vol and DLP (F=59.87, 60.92, P < 0.001). The detection rates of nodules were moderately or weakly correlated with noise indices, CTDI_vol and DLP (r=0.43, 0.56, -0.58, -0.78, P < 0.05), but no correlation with tube voltage (P > 0.05).Conclusions Scanning protocol has an impact on AI detection efficiency of pulmonary nodules. Reasonable scanning parameters should be selected according to different image quality requirements in clinical practice.