AI肺结节筛查系统结合能谱CT单能图像用于不同位置肺结节检测的效能研究

Study on the Efficacy of AI Pulmonary Nodule Screening System Combined with Energy Spectrum CT Single-Energy Images for Nodules at Different Locations

本研究结合人工智能(Artificial Intelligence,AI)肺结节筛查系统与能谱CT单能图像技术,筛选适合AI系统获得最优肺结节检出率的最佳单能图像;同时评估AI系统对能谱CT单能图像不同位置结节检出的准确性。利用AI肺结节筛查系统分别对30例胸部能谱CT不同单能量图像进行读片筛查,与“金标准”进行比对,计算敏感度、假阳性率和阳性预测值。AI肺结节筛查系统在70 keV图像中具有最高的结节检测敏感性。不同单能量图像中,肺内带、肺外带、气管血管树旁和胸膜下四个位置的肺结节检出敏感性无统计学差异;对不同位置肺结节,70 keV及以上组单能图像的检测敏感度显著高于低keV组。AI肺结节筛查系统结合能谱CT中70 keV单能图像能够获得较高检测敏感性;不同单能量图像中,结节位置对人工智能系统结节检出无显著影响。

To explore the optimal single-energy images for AI pulmonary nodule screening system to achieve the best nodules detection rate.In addition,the precision of AI system in detecting pulmonary nodules at different positions in the energy spectrum CT images was studied.AI system was utilized to screen pulmonary nodules on different single-energy images of 30 chest energy spectrum CT sequences.The detected nodules and their locations were compared with the“gold standard”,the number of true positive nodules,false positive nodules,and false negative nodules detected in different single-energy images by AI system were counted and recorded as well.AI system displayed the highest sensitivity for nodule detection on 70 keV single-energy images.In each images of different single-energy,no statistical difference was found in sensitivity of AI system in detecting pulmonary nodules at four different locations(P>0.05),including intrapulmonary,extrapulmonary,para-bronchovascular,and subpleural nodules.In terms of pulmonary nodules at different locations,the detection sensitivity was significantly higher for 70 and above keV singleenergy images comparing with the low keV images.70 keV single energy image is the optimal choice for AI pulmonary nodule screening system to obtain higher detection sensitivity.Positions of nodules exhibited no significant effect on AI systems in detecting nodules efficiently in each of the single-energy images.