AI辅助头颈CTA在诊断颅内动脉瘤中的应用  

Application of AI-assisted head and neck CT angiography in diagnosis of intracranial aneurysms

作　　者：邓昊[1] 韦鑫 曹闻挺[1] 艾光勇 余菡 方静 吴博涵 周治明[1] 陈金华 陈维娟[1] DENG Hao;WEI Xin;CAO Wenting;AI Guangyong;YU Han;FANG Jing;WU Bohan;ZHOU Zhiming;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：《第三军医大学学报》2021年第12期1179-1184,共6页Journal of Third Military Medical University

摘　　要：目的探究人工智能(artificial intelligence, AI)在头颈CT血管成像(CT angiography, CTA)诊断颅内动脉瘤中的价值。方法回顾性分析本院2020年8月至2021年1月117例临床疑似颅内动脉瘤患者间隔2周内头颈CTA及数字减影血管造影(digital subtraction angiography, DSA)资料。以DSA结果为金标准,计算AI诊断动脉瘤的敏感度和准确率。记录AI自动测量动脉瘤径线的结果,并与DSA人工测量结果进行一致性检验。结果 117例患者经DSA确诊为颅内动脉瘤105例,共计136个动脉瘤。(1)诊断效能方面,以患者为单位,AI对动脉瘤诊断的敏感度和准确率分别为87.6%和79.5%。以病灶为单位,对总体、<3 mm、3~5 mm和>5 mm的动脉瘤,AI诊断敏感度分别为79.4%、55.2%、81.5%和92.9%,准确率分别为72.0%、53.8%、75.3%和86.3%。(2)定量分析方面,对动脉瘤总体瘤体长径、总体瘤体短径、>5 mm动脉瘤的瘤体长径、<3 mm动脉瘤的瘤体长径和3~5 mm动脉瘤的瘤体长径的测量结果,AI自动测量和DSA人工测量的一致性分别为好、好、较好、差和差(ICC=0.778、0.822和0.722,P均<0.001;ICC=0.113和0.142,P均>0.05)。结论 AI在头颈CTA诊断颅内动脉瘤中具有较高的价值,对瘤体长径>5 mm的径线测量结果可作为参考标准,可以作为辅助诊断工具进一步推广和应用。Objective To explore the clinical value of artificial intelligence(AI) in the diagnosis of intracranial aneurysms by head and neck CT angiography(CTA). Methods A total of 117 patients with clinically suspected intracranial aneurysms admitted in our hospital from August 2020 to January 2021 were enrolled, and their head and neck CTA and digital subtraction angiography(DSA) data were collected and retrospectively analyzed. Taking DSA results as the gold standard, the sensitivity and accuracy of AI in the diagnosis of intracranial aneurysms were calculated. In addition, the results of AI automatic measurement were recorded, and the consistency test was conducted with the results of DSA manual measurement. Results Among the 117 suspected patients, there were 105 cases diagnosed with intracranial aneurysms by DSA, with a total of 136 aneurysms.(1)Diagnostic efficiency: The sensitivity and accuracy of AI for aneurysm diagnosis were 87.6% and 79.5%, respectively, taking patients as unit. While taking lesions as unit, the sensitivity of AI diagnosis was 55.2%, 81.5%, 92.9% and 79.4% for aneurysms <3 mm, 3.5 mm, >5 mm, as well as for the total number of aneurysms, respectively;And the accuracy was 53.8%, 75.3%, 86.3% and 72.0%.(2)Quantitative analysis: The results of AI and DSA showed good consistency in the measurement of the total aneurysm length diameter, total short diameter, and >5 mm aneurysm length diameter(ICC=0.778, 0.822 and 0.722, P<0.001);But poor consistency was seen in the measurement for <3 mm and 3~5 mm aneurysm length diameter(ICC=0.113, 0.142, P>0.05). Conclusion AI has a high value in the application of head and neck CTA in the diagnosis of intracranial aneurysms, with its measurement results to be used as a reference standard in tumors with length diameter >5 mm, it can be further promoted and applied as an auxiliary diagnostic tool.

关 键 词：人工智能 体层摄影术 X线计算机 颅内动脉瘤 诊断 

分 类 号：R445.3[医药卫生—影像医学与核医学] R651.1[医药卫生—诊断学]