【摘要】 目的 探讨颅内动脉瘤(intracranial aneurysm, IAN)CT血管成像(CT angiography, CTA)检查中人工智能(artificial intelligence, AI)技术的诊断价值。方法 回顾性分析45例临床疑似IAN患者的CTA和数字减影血管造影(digital subtraction angiography, DSA)影像资料。以DSA结果为“金标准”,以动脉瘤个数为单位计算AI辅助下CTA诊断动脉瘤的灵敏度、特异度和准确度。并采取Kappa检验，对AI辅助下CTA或DSA检测到的IAN位置和长短径的一致性进行统计学分析。结果 45例患者中经DSA确诊IAN 33例，共48个动脉瘤。以DSA诊断动脉瘤个数为“金标准”,AI诊断<3 mm动脉瘤、3～5 mm动脉瘤、>5 mm动脉瘤的灵敏度、特异度、准确度分别为53.33%、57.58%和56.25%,76.19%、96.30%和87.50%,83.33%、97.22%和93.75%。AI自动识别动脉瘤的位置与DSA人工识别一致性较好(κ=0.788)。AI自动测量<3 mm瘤体长径及短径测量结果与DSA人工测量结果的一致性差[组内相关系数(intraclass correlation coefficient, ICC)=0.364、0.315];AI自动测量3～5 mm瘤体长径及短径和>5 mm瘤体长径及短径的测量结果与DSA人工测量结果的一致性为中等和较好(ICC=0.713、0.676和0.813、0.798)。结论 AI辅助下CTA对直径>5 mm的IAN的诊断灵敏度、特异度较高，并对动脉瘤的定位和长短径的分析测量均比较准确。而对于较小的IAN,特别是直径<3 mm动脉瘤，AI获得的CTA诊断结果仍需要人工校正，以减少漏诊和误判。更多还原

【Abstract】 Objective To investigate the diagnostic value of artificial intelligence(AI) technology in CT angiography(CTA) examination of intracranial aneurysm(IAN). Methods CTA and digital subtraction angiography(DSA) images of 45 patients with clinically suspected IAN were retrospectively analyzed. The sensitivity, specificity and accuracy of AI-assisted CTA for the diagnosis of aneurysms were calculated by using DSA results as the "gold standard" and the number of aneurysms as the unit. Kappa test was taken to statistically analyze the consistency of IAN location and long and short diameter detected by AI-assisted CTA or DSA. Results IAN was confirmed by DSA in 33 of 45 patients with a total of 48 aneurysms. Using the number of aneurysms diagnosed by DSA as the "gold standard", the sensitivity, specificity and accuracy of AI in diagnosing <3 mm aneurysms, 3-5 mm aneurysms and >5 mm aneurysms were 53.33%, 57.58% and 56.25%, 76.19%, 96.30% and 87.50%, 83.33%, 97.22% and 93.75% respectively.The AI automatically identified the location of the aneurysms in good consistancy with the DSA manual identification(κ=0.788). AI automatic measurement <3 mm tumor diameter had poor consistancy with DSA manual measurement results [intraclass correlation coefficient(ICC)=0.364,0.315]; the AI automatic measurements of 3-5 mm tumor′s length and short diameter >5 mm tumor′s length and short diameter were in moderate and good consistancy with the DSA manual measurements(ICC=0.713, 0.676 and 0.813, 0.798). Conclusion AI-assisted CTA has higher diagnostic sensitivity and specificity for IAN which diameters >5 mm, which is more accurate for both localization of aneurysms and analytical measurement of long and short diameters. Compared for smaller IAN, especially for aneurysms <3 mm in diameter, the CTA diagnostic results obtained by AI still require manual correction to reduce missed diagnoses and misjudgement.更多还原