基于多层螺旋CT图像的机器深度学习技术测量仿真胸部体模肺结节体积及长径的效果

Effect of multi-slice spiral CT images-based deep learning-based computer-aided diagnosis system on measuring pulmonary nodule volume and diameter in a chest phantom

目的观察在多层螺旋CT图像中机器深度学习技术测量仿真胸部体模肺结节体积及长径的效果。方法取4种直径(5 mm、8 mm、10 mm、12 mm)的模拟肺结节共12个,每种直径结节都有3种CT密度值(100 HU、-630 HU、-800 HU)。将12枚模拟肺结节随机放置在仿真胸部体模,使用256排螺旋CT进行扫描,重建层厚分别为0.625 mm、1.25 mm、2.5 mm。使用机器深度学习软件检测所有的模拟肺结节,记录结节数目、长径、体积,计算结节长径、体积测量的绝对错误率(APE)。结果在不同图像层厚下,肺结节的长径和体积APE差异均无统计学意义(均P>0.05);不同直径肺结节的体积APE差异亦无统计学意义(P>0.05)。结论对于直径≥4 mm的肺结节,体积或是机器深度学习软件测量肺结节时更可靠的指标。重建层厚对机器深度学习技术测量肺结节长径及容积的结果影响较小,故采用人机协同阅片工作模式时,可适当加大图像层厚以减少影像医师阅片数量,缓解临床工作压力。

Objective To observe the effect of multi-slice spiral CT images-based deep learning-based computer-aided diagnosis(DL-CAD)system on measuring pulmonary nodule volume and diameter in a chest phantom.Methods Totally 12 artificial pulmonary nodules with four different diameters(5 mm,8 mm,10 mm and 12 mm)were selected,and the pulmonary nodules with each diameter consisted of three nodules with various CT values(100 HU,-630 HU,and-800 HU).The 12 artificial pulmonary nodules were randomly placed in a chest phantom,then scanning using a 256-detector row spiral CT was performed,and the images were reconstructed with different thicknesses(including 0.625 mm,1.25 mm,and 2.5 mm).A DL-CAD system was used to measure all artificial pulmonary nodules,and the number,diameter and volume of nodules were recorded,then the absolute percentage error(APE)for nodule diameter or volume measurement was calculated.Results There was no statistically significant difference in diameter APE or volume APE of pulmonary nodules among different slice thicknesses(all P>0.05);no statistically significant difference was found in volume APE among pulmonary nodules with different diameters(all P>0.05).Conclusion For pulmonary nodules with diameter≥4 mm,volume may be an relatively reliable indicator when measuring pulmonary nodules by DL-CAD.Slice thickness of reconstruction exerts a minor effect on the measurement results of pulmonary nodule diameter and volume by DL-CAD,thus,slice thickness can be increased properly to reduce frequency of imaging diagnosis and relieving clinical stress for radiologists using the man-machine coordination imaging diagnosis pattern.