一种基于多普勒超声与神经网络的脑缺血风险评估方法
A Cerebral Ischemia Risk Assessment Method Based on Doppler Ultrasound and Neural Network
曾泽延 1余龙 2秦旺 2汪昕 1丁晶 1王盛章3
作者信息
- 1. 复旦大学附属中山医院 神经内科,上海 200032
- 2. 复旦大学 航空航天系,生物力学研究所,上海 200433
- 3. 复旦大学 航空航天系,生物力学研究所,上海 200433;复旦大学 工程与应用技术研究院;生物医学工程与技术研究所,上海 200433
- 折叠
摘要
目的 借助多普勒超声测量颈动脉和椎动脉的流速波形,构建神经网络模型,对脑缺血状态进行无创评估.方法 采集已经完善CT灌注(computed tomography perfusion,CTP)及多普勒超声患者的临床数据,通过超声图像采集血流动力学参数,将获得的数据通过神经网络进行训练,最后将结果 与CTP进行验证.结果 共收集 62 例符合纳排标准的患者数据,将其随机分为训练集44例和测试集18例.在训练集中,受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under curve,AUC)、准确率、敏感度、特异度分别为 0.95、0.833、0.923、0.886.在测试集中,ROC的AUC、准确率、敏感度、特异度分别为 0.860、0.889、0.714、1.000.结论 基于多普勒超声及神经网络的模型经过临床验证,对于脑缺血评估的准确性较好,具有早期筛查脑缺血的潜在临床价值.
Abstract
Objective To noninvasively assess the cerebral ischemic status using the velocity profile of the carotid and vertebral arteries measured by Doppler ultrasound and a neural network model.Methods Imaging data were collected from patients who underwent computed tomography perfusion(CTP)and Doppler ultrasound.Hemodynamic parameters were extracted from the ultrasound images.These parameters were used to train a fully connected neural network model.The model was validated using the CTP results.Results Sixty-two eligible patients were included;44 were randomly selected as the training dataset and 18 were designated for validation.In the training set,the area under the curve(AUC)of the receiver operating characteristic,sensitivity,specificity,and accuracy were 0.95,0.833,0.923,and 0.886,respectively.In the test set,the AUC,sensitivity,specificity,and accuracy were 0.860,0.714,1.000,and 0.889,respectively.Conclusions The model based on Doppler ultrasound and neural network was clinically verified and had good accuracy for assessing cerebral ischemia,showing its clinical potential for the early screening of cerebral ischemia.
关键词
血流动力学/脑缺血/CT灌注成像/神经网络/多普勒超声Key words
hemodynamics/cerebral ischemia/computed tomography perfusion/neural network/Doppler ultrasound引用本文复制引用
出版年
2024
NETL
NSTL
万方数据