ISSN 1004-4140
CN 11-3017/P
祁冬, 乔晓春, 施彪, 等. 人工智能在头颈CTA图像后处理和诊断头颈动脉狭窄中的应用价值[J]. CT理论与应用研究(中英文), xxxx, x(x): 1-9. DOI: 10.15953/j.ctta.2023.105.
引用本文: 祁冬, 乔晓春, 施彪, 等. 人工智能在头颈CTA图像后处理和诊断头颈动脉狭窄中的应用价值[J]. CT理论与应用研究(中英文), xxxx, x(x): 1-9. DOI: 10.15953/j.ctta.2023.105.
QI D, QIAO X C, SHI B, et al. The Value of Artificial Intelligence in Post-processing Head and Neck CTA Images and Diagnosing Head and Neck Artery Stenosis[J]. CT Theory and Applications, xxxx, x(x): 1-9. DOI: 10.15953/j.ctta.2023.105. (in Chinese).
Citation: QI D, QIAO X C, SHI B, et al. The Value of Artificial Intelligence in Post-processing Head and Neck CTA Images and Diagnosing Head and Neck Artery Stenosis[J]. CT Theory and Applications, xxxx, x(x): 1-9. DOI: 10.15953/j.ctta.2023.105. (in Chinese).

人工智能在头颈CTA图像后处理和诊断头颈动脉狭窄中的应用价值

The Value of Artificial Intelligence in Post-processing Head and Neck CTA Images and Diagnosing Head and Neck Artery Stenosis

  • 摘要: 目的:探讨人工智能(AI)对头颈动脉CTA图像后处理和诊断头颈动脉狭窄的中的价值。方法:回顾性收集我院2022年11月至2023年10月42例均行头颈CTA和头颈动脉数字减影血管造影(DSA)检查患者的影像学资料。图像后处理及诊断分为AI组和人工组,对比两组在图像后处理用时和图像质量主观评分及识别头颈动脉斑块性质(钙化斑块、非钙化斑块和混合性斑块)的差异。以DSA结果为金标准,对比两组在诊断头颈动脉狭窄的敏感性、特异性、阳性预测值、阴性预测值及准确率差异,并将两组诊断结果与DSA结果进行一致性Kappa检验;采用受试者操作特征曲线(ROC)分析AI组和人工组对头颈动脉狭窄的诊断效能,并采用Z检验比较其差异。结果:AI组后处理及诊断用时为(366.48±18.54)s,较人工组(1291.63±52.27)s缩短了约71.65%,差异有统计学意义;两种方法得到图像质量主观评分差异无统计学意义。人工组共识别头颈动脉斑块 145个,AI组共识别头颈动脉斑块145个,其中AI组准确识别不同性质斑块141个,总准确率为97.24%(141/145)。AI组与人工组对识别头颈动脉钙化斑块、非钙化斑块及混合性斑块差异均无统计学意义,且两组识别斑块性质的一致性较好(Kappa=0.845)。AI组诊断头颈动脉狭窄的敏感性、特异性、阳性预测值、阴性预测值及准确率分别为87.09%(27/31)、81.82%(9/11)、93.10%(27/29)、69.23%(9/13)和85.71%(36/42),与DSA诊断头颈动脉狭窄一致性较好(Kappa=0.792);人工组诊断头颈动脉狭窄的敏感性、特异性、阳性预测值、阴性预测值及准确率分别为90.32%(28/31)、81.82%(9/11)、93.33%(28/30)、75.00%(9/12)和88.10%(37/42),与DSA诊断头颈动脉狭窄一致性较好(Kappa=0.801),并且两组诊断效能差异均无统计学意义。ROC曲线分析结果显示,AI组与人工组诊断头颈动脉狭窄的曲线下面积(AUC)分别为0.845和0.861,差异无统计学意义。结论:AI技术在头颈CTA图像后处理用时、评估头颈动脉狭窄及斑块性质识别方面具有较高的临床价值,可作为医师分析诊断头颈CTA的有效辅助工具。

     

    Abstract: Objective: Discuss the value of Artificial Intelligence (AI) in post-processing head and carotid artery CTA images and diagnosing head and carotid artery stenosis. Methods: The imaging data of 42 patients who all underwent head and neck CTA and digital subtraction angiography (DSA) of head and carotid arteries from November 2022 to October 2023 at our hospital were retrospectively collected. Image post-processing and diagnosis were divided into AI and manual groups to compare the differences in image post-processing time and subjective scores of image quality and identification of head and carotid artery plaque properties (calcified plaque, non-calcified plaque and mixed plaque) between the two groups. To compare the differences in sensitivity, specificity, positive predictive value, negative predictive value and accuracy of the two groups in diagnosing head and neck artery stenosis, using DSA results as the gold standard, and to perform the Kappa test for consistency between the two groups' diagnostic results and DSA results. The diagnostic efficacy of the AI and manual groups on the head and carotid artery stenosis was analyzed by subject operator characteristic curve (ROC) and compared by Z-test. Results: The post-processing and diagnostic time in the AI group was (366.48±18.54) s, which was approximately 71.65% shorter than that in the manual group (1291.63±52.27) s. The difference was statistically significant. The difference between the subjective scores of image quality obtained by the two methods was not statistically significant. A total of 145 head and neck artery plaques were identified in the manual group and 145 in the AI group, of which 141 plaques of different nature were accurately identified in the AI group, with an overall accuracy rate of 97.24% (141/145). There was no statistically significant difference between the AI group and the manual group for identifying calcified plaque, non-calcified plaque and mixed plaque in the head and carotid arteries, and the agreement between the two groups for identifying the nature of plaque was good (Kappa=0.845). The sensitivity, specificity, positive predictive value, negative predictive value and accuracy of the AI group in diagnosing head and neck artery stenosis were 87.09% (27/31), 81.82% (9/11), 93.10% (27/29), 69.23% (9/13) and 85.71% (36/42), respectively, which were in good agreement with DSA in diagnosing head and neck artery stenosis (Kappa= 0.792). The sensitivity, specificity, positive predictive value, negative predictive value and accuracy of the manual group in diagnosing head and neck artery stenosis were 90.32% (28/31), 81.82% (9/11), 93.33% (28/30), 75.00% (9/12) and 88.10% (37/42), respectively, which were in good agreement with DSA in diagnosing head and neck artery stenosis (Kappa=0.801), and the difference in diagnostic efficacy was not statistically significant in either group. The results of the ROC curve analysis showed that the area under the curve (AUC) for the diagnosis of head and neck artery stenosis was 0.845 and 0.861 in the AI and manual groups, respectively, with no significant difference. Conclusion: AI technology has high clinical value in time spent on post-processing head and neck CTA images, assessing head and neck artery stenosis and identifying the nature of carotid plaque, and can be used as a useful aid for physicians in analyzing and diagnosing head and neck CTA.

     

/

返回文章
返回