基于深度学习算法对颅内动脉瘤进行自动检测和分割  

Automatic Detection and Segmentation of Intracranial Aneurysm Based on Deep Learning Algorithm

作　　者：郭东方 郝家伟 吴彦林 鄂亚军[1] 王光磊 李志强[1] GUO Dongfang;HAO Jiawei;WU Yanlin(Department of Interventional Radiology&Vascular Surgery,the Affiliated Hospital of Hebei University,Baoding,Hebei Province 071000,P.R.China)

机构地区：[1]河北大学附属医院介入血管外科,保定071000

出　　处：《临床放射学杂志》2023年第1期14-19,共6页Journal of Clinical Radiology

摘　　要：目的应用深度学习算法自动检测和分割CTA图像中的颅内动脉瘤。方法搜集行头颅CTA检查的颅内动脉瘤患者原始CTA图像。对原始CTA图像进行预处理后,输入到U-net神经网络进行检测和分割。以人工标记为参考标准。结果在测试集中,直径>4 mm的动脉瘤,敏感度为92.0%,Dice中位数为0.82;直径>5 mm的动脉瘤,敏感度为93.1%,Dice中位数为0.84;直径>6 mm的动脉瘤,敏感度为95.5%,Dice中位数为0.87。与人工标记结果具有高度一致性。结论深度学习算法可以自动检测和分割直径>4 mm的动脉瘤,不仅可以节省人力和时间成本,在一定程度上为动脉瘤破裂造成蛛网膜下腔出血的预防及颅内动脉瘤治疗的选择提供参考依据。Objective To use deep learning algorithm to detect and segment intracranial aneurysms in CTA images automatically.Methods Original CTA images of patients with intracranial aneurysms who underwent cranial CTA were collected.The original CTA image is preprocessed and input to U-net neural network for detection and segmentation.Take manual marking as reference standard.Results In the test set,the sensitivity of aneurysms larger than 4mm was 92.0%,with a median Dice of 0.82.For aneurysms larger than 5mm in diameter,the sensitivity was 93.1%,and the median Dice was 0.84.For aneurysms larger than 6mm in diameter,the sensitivity was 95.5%,with a Dice median of 0.87.The results are highly consistent with manual marking.Conclusion The deep learning algorithm proposed in this paper can automatically detect and segment aneurysms larger than 4mm in diameter,which can not only save labor and time costs,but also provide reference for the prevention of subarachnoid hemorrhage caused by ruptured intracranial aneurysm and the selection of intracranial aneurysm treatment.

关 键 词：颅内动脉瘤 蛛网膜下腔出血 CT血管造影 深度学习 

分 类 号：R743.3[医药卫生—神经病学与精神病学]