基于卷积神经网络GoogLeNet算法构建颅内动脉瘤诊断模型

Construction of intracranial aneurysm diagnostic model based on GoogLeNet algorithm of convolutional neural network

目的评价基于卷积神经网络的GoogLeNet算法在颅内动脉瘤自动分类诊断中的应用效果。方法本项研究回顾性收集了2020年1月至2023年1月在西南医科大学附属医院进行头部CT扫描的234例颅内动脉瘤患者和正常对照者的计算机断层扫描血管造影图像作为研究对象,采用Pytorch框架构建基于GoogLeNet算法的卷积神经网络模型,并使用He初始化方法和Adam优化器进行模型参数初始化和优化,采用交叉熵作为损失函数,并使用批标准化和dropout技术进行模型训练和防止过拟合。结果基于GoogLeNet算法构建的颅内动脉瘤诊断模型在测试集上获得了较高的准确度和较低的损失函数值,受试者工作特征曲线显示训练集的曲线下面积为0.891,测试集为0.851,证明了该模型在颅内动脉瘤诊断中具有很好的应用前景。结论基于卷积神经网络的GoogLeNet算法可以有效地应用于颅内动脉瘤诊断,并且具有较高的准确度和较低的损失函数值,可以为颅内动脉瘤的早期诊断和治疗提供参考依据。

Objective To explore the application effect of GoogLeNet algorithm based on convolutional neural network(CNN)in automatic classification and diagnosis of intracranial aneurysms.Methods Computed tomography angiography images of 234 patients with intracranial aneurysms and normal controls who underwent head CT scanning at the Affiliated Hospital of Southwest Medical Uni-versity between January 2020 and January 2023 were retrospectively collected for this study.A convolutional neural network model based on the GoogLeNet algorithm was constructed using the PyTorch framework.The model parameters were initialized and optimized using the He initialization method and the Adam optimizer.Cross-entropy was used as the loss function,and batch normalization and dropout techniques were employed for model training and to prevent overfitting.Results The intracranial aneurysm diagnostic model based on the GoogLeNet algorithm in this study achieved high accuracy and low loss function value on the test set.The ROC curve showed that the AUC of the training set was 0.891 and the test set was 0.851,which proved that the model had a good application pros-pect in the diagnosis of intracranial aneurysms.Conclusion The GoogLeNet algorithm based on convolutional neural network could be effectively applied to the diagnosis of intracranial aneurysms with high accuracy and low loss function value,which provided reference for the early diagnosis and treatment of intracranial aneurysms.