基于光学相干断层扫描血管成像技术智能预测先天性心脏病围术期转归的临床研究

A clinical study on intelligent prediction of perioperative outcomes in congenital heart disease based on optical coherence tomography angiography technology

目的探索基于光学相干断层扫描血管成像(OCTA)智能预测先天性心脏病(CHD)围术期转归的可行性,为CHD围术期评估提供无创便捷的辅助工具。方法前瞻性收集2017年5月至2021年5月于我院行CHD矫正术患者的临床资料和OCTA图像。根据是否发生术后过度失血或围术期复合不良结局标注OCTA图像。标注图像经数据增强后,训练深度学习预测模型,使用测试集判断模型的性能。结果202例CHD患者中,发生术后过度失血的患者有45例(22.3%),发生围术期复合不良结局的有58例(28.7%)。在测试集中,术后过度失血预测模型的受试者工作特性曲线下面积(AUC)为0.82,灵敏度和特异度分别为0.90和0.75,准确率为0.78;围术期复合不良结局预测模型的AUC为0.81,灵敏度和特异度分别为0.83和0.80,准确率为0.81。结论术前OCTA图像联合人工智能可高效预测CHD围术期转归。

Objective To explore the feasibility of developing an intelligent algorithm for predicting the perioperative outcomes in patients with congenital heart disease(CHD)based on optical coherence tomography(OCTA),and to provide a non-invasive and convenient tool for perioperative assessment and outcome prediction of CHD.Methods We prospectively collected the clinical data and preoperative OCTA images of patients undergoing congenital cardiac surgery in our hospital during the period of May 2017 to May 2021.OCTA images were labeled according to whether there was excessive postoperative bleeding or adverse perioperative composite outcomes.Data augmentation was used for the OCTA images with prognostic annotations.A deep learning outcome prediction model was trained,and then a test set was used to judge the performance of the model.Results Of 202 CHD patients,45(22.3%)developed excessive postoperative bleeding and 58(28.7%)underwent adverse perioperative composite outcomes.In the test set,the area under the receiver operating characteristic curve(AUC)of the model for predicting postoperative excessive bleeding was 0.82,the sensitivity and specificity were 0.90 and 0.75 respectively,and the accuracy was 0.78;the AUC of the adverse perioperative composite outcome prediction model was 0.81,the sensitivity and specificity were 0.83 and 0.80 respectively,and the accuracy was 0.81.Conclusions Preoperative OCTA images combined with artificial intelligence can accurately predict the perioperative outcomes of CHD.