基于AlexNet的桥小脑角脑膜瘤和听神经瘤MRI图像的识别研究  

MRI Image Recognition of Cerebellopontine Angle Meningioma and Acoustic Neuroma Based on AlexNet

作　　者：江华 于同刚[1] 吴丽琼 丁建 胡小洋 刘颖 JIANG Hua;YU Tong-gang;WU Li-qiong;DING Jian;HU Xiao-yang;LIU Ying(Shanghai Gamma Hospital,Shanghai 200235;University of Shanghai for Science and Technology,Shanghai 200093)

机构地区：[1]上海伽玛医院,上海200235 [2]上海理工大学,上海200093

出　　处：《中国医疗器械信息》2022年第1期44-47,共4页China Medical Device Information

摘　　要：目的:使用卷积神经网络及磁共振图像鉴别脑膜瘤与听神经瘤,研究影响识别的原因并提升准确率。方法:采集388位患者的增强后T1WI影像,将其进行筛选和扩充,分别将原始数据集与扩充后数据集对卷积神经网络AlexNet进行训练并输出预测结果,对预测结果进行分析。结果:扩充数据集模型的训练准确率为0.8026,明显高于原数据集模型的训练准确率的0.7526。结论:数据集越大对识别的准确率有显著的提高,使用磁共振图像训练的卷积神经网络用于肿瘤识别具有实际的临床意义。Objective:To identify meningioma and acoustic neuroma by convolution neural network and magnetic resonance imaging,and to study the factors affecting the recognition and improve the accuracy.Methods:In this study,388 patients with enhanced T1WI images were selected and expanded.The original dataset and the expanded dataset were used to train the convolution neural network AlexNet,and the prediction results were output and analyzed.Results:The training accuracy of the extended dataset model was 0.8026,which was significantly higher than that of the original dataset model(0.7526).Conclusion:The larger the dataset is,the higher the recognition accuracy is.Convolution neural network trained by MRI has practical clinical significance in tumor recognition.

关 键 词：桥小脑角脑膜瘤 听神经瘤 深度学习 卷积神经网络 

分 类 号：R445.2[医药卫生—影像医学与核医学]