基于ResUNet与PSPNet的CT血管造影自动分割模型可辅助诊断颈动脉粥样硬化斑块  

The automatic segmentation model of CT angiography based on ResUNet and PSPNet assists in the diagnosis of carotid atherosclerotic plaques with accurate precision

作　　者：郭中萍 顾艳 GUO Zhongping;GU Yan(Lianyungang Clinical College of Nanjing Medical University/Lianyungang First People's Hospital,Lianyungang 222061,China)

机构地区：[1]南京医科大学连云港临床医学院/连云港市第一人民医院,江苏连云港222061

出　　处：《分子影像学杂志》2025年第4期479-483,共5页Journal of Molecular Imaging

基　　金：连云港市卫健委面上基金(202204)。

摘　　要：目的探讨基于ResUNet与PSPNet的CT血管造影(CTA)自动分割模型在帮助评估颈动脉粥样硬化斑块中的价值。方法回顾性纳入647例于2020年10月~2022年10月在连云港市第一人民医院行头颈部CTA检查的颈动脉粥样硬化斑块形成患者,按7∶1.5∶1.5的比例随机分为训练集(n=475)、验证集(n=86)及测试集(n=86)。训练集中放射科医师标记的图像用于开发基于ResUNet与PSPNet的自动分割模型,在验证集、测试集使用精确度、敏感度、召回率等参数评估模型对颈动脉斑块的诊断性能。结果在训练集中,自动分割模型在动脉粥样硬化斑块分割方面显示出良好的性能;验证集和测试集的结果进一步验证了其实用性。对测试集进行不同斑块类型亚组分析,结果显示基于CTA图像的深度学习模型对不同斑块类型显示出良好的斑块诊断准确性。结论基于ResUNet与PSPNet的自动分割模型辅助诊断颈动脉粥样硬化斑块的准确性较高,具有临床可行性。Objective To explore the value of the automatic segmentation model of computed tomography angiography(CTA)based on ResUNet and PSPNet in helping to evaluate carotid atherosclerotic plaques.Methods This study retrospectively included 647 patients with carotid atherosclerotic plaque formation who underwent head and neck CTA examinations.They were randomly divided into training set(n=475),validation set(n=86)and test set(n=86)at a ratio of 7:1.5:1.5.The images marked by radiologists in the training set were used to develop the automatic segmentation model based on ResUNet and PSPNet.Parameters such as precision,sensitivity,and recall were used in the validation set and the test set to evaluate the diagnostic performance of the model for carotid plaques.Results In the training set,the automatic segmentation model had already demonstrated good performance in the segmentation of atherosclerotic plaques.Its practicability was further verified in the validation set and the test set.In addition,a subgroup analysis of different plaque types was conducted on the test set,and the results showed that the deep learning model based on CTA images demonstrated good plaque diagnostic accuracy for different plaque types.Conclusion The automatic segmentation model based on ResUNet and PSPNet has relatively high accuracy in assisting the diagnosis of carotid atherosclerotic plaques and is clinically feasible.

关 键 词：颈动脉斑块 深度学习 卒中 

分 类 号：R743.3[医药卫生—神经病学与精神病学]