人工智能在判断急性脑梗死半暗带中的应用

Application of artificial intelligence in judging penumbra in acute cerebral infarction

目的研究人工智能在判断急性脑梗死(ACI)半暗带中的应用价值。方法选择南昌市第一医院2021年1月至12月收治的200例ACI患者作为研究对象,分别于梗死后1、3 d时对所有受试者均开展T2FLAIR、扩散加权成像(DWI)、动脉自旋标记灌注成像(ASL)、弥散峰度成像(DKI)以及多b值DWI检查,分析核心梗死区域与半暗带相关参数[包括脑血容量(CBV)、脑血流量(CBF)、平均通过时间(MTT)以及达峰时间(TTP)]。比较不同成像技术脑梗死后1、3 d时的核心梗死区域以及半暗带符合率,比较不同深度学习模型预测核心梗死区域以及半暗带符合率情况。结果核心梗死区域CBV、CBF以及MTT水平均低于半暗带,而TTP水平高于半暗带,差异有统计学意义(P<0.05);CBV、CBF、MTT以及TTP联合预测半暗带的曲线下面积、灵敏度、特异度以及约登指数均高于上述四项指标单独预测,差异有统计学意义(P<0.005);T2FLAIR、DWI、ASL、DKI以及多b值DWI诊断脑梗死后1、3 d时的核心梗死区域以及半暗带符合率比较,差异无统计学意义(P>0.005);DeepLab v3深度学习模型预测核心梗死区域以及半暗带符合率均高于RefineNet、PSPNet深度学习模型,差异有统计学意义(P<0.017)。结论人工智能在判断ACI半暗带中的应用价值较高,可有效提高ACI半暗带的检出率以及预测准确性。

Objective To study and analyze the application value of artificial intelligence in judging penumbra of acute cerebral infarction(ACI).Methods A total of 200 ACI patients admitted to the the Frist Hospital of Nanchang from January to December 2021 were selected as research objects.T2FLAIR,diffusion-weighted imaging(DWI),arterial spin label perfusion imaging(ASL),diffuse kurtosis imaging(DKI),and multi-b-value DWI were performed on all subjects at 1,3 d after infarction respectively.The parameters related to the cerebral blood volume(CBV),cerebral blood flow(CBF),mean transit time(MTT),and time to peak(TTP)were analyzed.The coincidence rate of core infarct area and penumbra was compared at 1,3 d after cerebral infarction with different imaging techniques.In addition,different deep learning models were established according to relevant parameters to compare the coincidence rates of core infarction region and penumbra predicted by different deep learning models.Results The levels of CBV,CBF and MTT of core infarction region were lower than those in penumbra,while TTP was higher than those in penumbra,the differences were statistically significant(P<0.05).ROC curve analysis showed that the area under curve,sensitivity,specificity and Jorden index of the penumbra predicted by CBV,CBF,MTT and TTP were higher than those predicted by the above four indexes alone,the differences were statistically significant(P<0.005).There were no significant differences in the coincidence rates of core infarct area and penumbra on T2FLAIR,DWI,ASL,DKI and multi-b-value DWI at 1,3 d after cerebral infarction diagnosis(P>0.005).DeepLab v3 deep learning model predicted higher coincidence rate of core infarct area and penumbra area than RefineNet and PSPNet deep learning model,the differences were statistically significant(P<0.017).Conclusion The application value of artificial intelligence in judging ACI penumbra is high,which can effectively improve the detection rate and prediction accuracy of ACI penumbra.