人工智能在基于颅脑CT灌注数据血管后处理的应用

Application of artificial intelligence in vascular reconstruction based on cerebral CT perfusion data

目的探讨人工智能(AI)在基于颅脑CT灌注数据重组CTA的图像后处理中的应用价值。方法回顾性分析2020年1至7月河北省人民医院疑似脑血管疾病的100例患者的临床及影像资料。按检查方案的不同将分为A、B 2组,每组50例。A组行颅脑CT灌注检查(选取动静脉CT值差值最大期的前5个时相生成的时间最大密度投影数据集为A1亚组,差值最大期的原始薄层图像为A2亚组),B组行单时相CTA检查,分别进行人工和AI后处理。测量3组图像CT值、噪声(SD)、信噪比(SNR)、对比噪声比(CNR)客观评价指标并进行主观评分,记录并统计人工和AI后处理的血管分割合格率与后处理时间。采用单因素方差分析比较图像的客观评价指标,Kruskal-Wallis H检验比较主观评分差异。结果3组横断面原始图像的主观评分、客观评价指标差异均有统计学意义(P均<0.05),其中A1组动脉强化程度及小动脉细节显示评分、脑动脉CT值、SNR、CNR均高于A2亚组和B组(P均<0.05)。100例患者共1100条血管,A1亚组[98.4%(541/550)]、B组[98.7%(543/550)]的AI血管重组合格率均高于人工重组[82.9%(456/550)、87.1%(479/550),χ^(2)值分别为77.392、56.521,P<0.001],A2亚组的[78.4%(431/550)]AI血管重组合格率低于人工重组[85.6%(471/550),χ^(2)=9.855,P=0.002]。AI处理完成时间较人工分别缩短56.30%、49.63%、50.81%。结论与人工后处理相比,AI在基于颅脑CT灌注降噪数据的CTA后处理的图像质量与工作效率上具有一定优势,结合人工质控,在脑血管病影像后处理方面值得推广应用。

Objective To explore the application value of artificial intelligence(AI)in image post-processing of reconstructed CTA based on CT cerebral perfusion(CTP).Methods Clinical and radiological data of 100 patients suspected of cerebrovascular diseases in Hebei General Hospital from January to July 2020 were retrospectively selected.All patients were divided into A and B group on average according to the different examination schemes.Cerebral CTP examination was performed in group A(the temporal maximum intensity projective data set generated by the first 5 time phases in the maximum period of the difference between arteriovenous CT values selected as subgroup A1,and the corresponding original thin-layer images selected as subgroup A2),single phase CTA examination was performed in group B,manual and AI image post-processing were performed respectively.Subjective scoring of the image data was performed,and the objective bid evaluation indexes such as CT value,noise(SD),signal-to-noise ratio(SNR),contrast to noise ratio(CNR)were measured,the qualified rate of artificial and AI vascular segmentation was counted,and post-processing time were recorded.The objective evaluation indexes were compared between three groups using one-way ANOVA,and the Kruskal-Wallis H test was used to compare the difference of subjective scores.Results Statistically significant differences were observed in subjective score and objective evaluation index of original images among group A1,group A2 and group B(all P<0.05).Among them,arterial enhancement,arteriolar detail display score,cerebral artery CT value,SNR and CNR in group A1 were higher than those in group A2 and group B(all P<0.05).In a total of 100 patients with 1100 blood vessels,the qualified rates of AI vascular segmentation in group A1[98.4%(541/550)]and group B[98.7%(543/550)]were higher than those of manual[82.9%(456/550),87.1%(479/550),χ^(2)=77.392,56.521,P<0.001],but the qualified rate of AI vascular segmentation of group A2[78.4%(431/550)]was lower than that of manual[85.6%(471/550),χ^(2)=9.855,P=0.002].The completion time of AI post-processing were reduced by 56.30%,49.63%,50.81%,respectively than those with manual.Conclusion Compared with manual image post-processing,AI has certain advantages in image quality and work efficiency of reconstructed CTA post-processing based on CTP de-noising dataset,and it is worth popularizing and applying in the image post-processing of cerebrovascular disease,combined with artificial quality control.