双源CT虚拟平扫技术联合人工智能在急性颅内出血诊断中的应用价值  

Application value of dual-source virtual non-contrast CT combined with AI in diagnosis of acute intracranial hemorrhage

作　　者：白珊玮 韦鑫 周治明[1] 艾光勇 张蓉[1] 陈金华 陈维娟[1] BAI Shanwei;WEI Xin;ZHOU Zhiming;AI Guangyong;ZHANG Rong;CHEN Jinhua;CHEN Weijuan(Department of Radiology,the Second Affiliated Hospital of Chongqing Medical University,Chongqing,400010,China)

机构地区：[1]重庆医科大学附属第二医院放射科,重庆400010

出　　处：《陆军军医大学学报》2023年第12期1320-1327,共8页Journal of Army Medical University

摘　　要：目的探讨双源CT虚拟平扫(virtual non-contrast,VNC)技术联合人工智能(artificial intelligence,AI)在急性颅内出血诊断中的应用价值。方法本研究为横断面研究,纳入2022年1-7月重庆医科大学附属第二医院门诊部收治的67例疑似颅内出血的头部平扫CT(conventional non-contrast CT,NCCT)和双能量CT血管成像(CT angiography,CTA)资料,其中男性32例,女性35例,年龄35~89(56.2±12.5)岁。分析VNCT与NCCT的平均CT值、噪声、图像信噪比(signal-to-noise ratio,SNR)、对比噪声比(contrast-to-noise ratio,CNR)和辐射剂量的差异。以2名放射学专家的共同诊断结果为参考标准,对比分析AI软件应用于VNCT和NCCT判断颅内出血的类型、中线结构偏移程度、出血最大层面的长径和短径与血肿体积的准确率。比较专家及AI得出以上定量评价指标所需耗时差异。结果客观评价方面,VNCT图像的平均CT值、SNR和CNR均明显低于NCCT(P均<0.001),噪声大于NCCT图像(P<0.001)。VNCT的辐射剂量比NCCT+CTA降低75.6%。定性分析方面,AI诊断颅内出血应用于NCCT和VNCT的准确率分别为94.0%和89.6%。VNCT+AI对IPH、IVH、SAH和SDH的准确率为:83.6%、86.0%、83.0%和60.0%,与NCCT+AI结果相比差异均无统计学意义。定量分析方面,VNCT+AI对中线偏移程度、最大面积长径和短径、血肿体积的测量结果与专家手动测量结果均一致性好(ICC=0.820、0.942、0.920和0.959,P均<0.001)。在AI软件诊断耗时方面,VNCT组显著低于NCCT组[(50.6±15.9)vs(58.5±24.9)s,P<0.001]。结论双源CT虚拟平扫技术辐射剂量低,检查流程简单、快捷,联合AI技术对急性颅内出血具有较高的诊断效能,且定量评价指标与放射学专家一致性好,具有潜在的临床应用价值。Objective To explore the clinical value of dual-source virtual non-contrast(VNC)CT combined with artificial intelligence(AI)in the diagnosis of acute intracranial hemorrhage.MethodsA cross-sectional study was conducted on the imaging data of non-contrast CT and dual-energy CT angiography(CTA)in 67 patients with suspected intracranial hemorrhage.They were 32 males and 35 females,at an average age of 56.2±12.5(35~89)years.The differences of average CT value,noise,signal to noise ratio(SNR),contrast noise ratio(CNR)and radiation dose between VNCT and NCCT were analyzed.Taken the common diagnosis results of two radiologists as the reference standard,the type of intracranial hemorrhage,the degree of midline structural deviation,the long and short diameters of the largest hemorrhage layer and the volume of hematoma were analyzed by AI software under VNCT and NCCT images.And the time taken by radiologists and AI to obtain the above quantitative evaluation indexes was compared.ResultsIn the objective evaluation,the average CT value,SNR and CNR were significantly lower(P<0.001),and the noise was higher in the VNCT images than the NCCT images(P<0.001).The radiation dose of VNCT was lower by 75.6%than that of NCCT+CTA(P<0.001).The qualitative analysis indicated that the accuracy of AI in diagnosing intracranial hemorrhage on NCCT and VNCT images was 94.0%and 89.6%respectively.The accuracy rates of VNCT+AI for IPH,IVH,SAH and SDH were 83.6%,86.0%,83.0%and 60.0%,respectively,which had no differences when compared with those of NCCT+AI(all P>0.05).In the quantitative analysis,VNCT+AI had good consistency with manual measurement results in terms of the above indicators(ICC=0.820,0.942,0.920,and 0.959,P<0.001).In terms of the time required for AI diagnosis,VNCT need significantly shorter time than NCCT(50.6±15.9 vs 58.5±24.9 s,P<0.001).ConclusionDual-source VNC technology has low radiation dose,simple and fast examination process,and high accuracy in the diagnosis of acute intracranial hemorrhage when combined with AI.And the

关 键 词：体层摄影术 X线计算机 虚拟平扫 颅内出血 人工智能 

分 类 号：R319[医药卫生—基础医学] R743.34R814.42