基于迁移学习和改进EfficientNet-B0的脑肿瘤分类算法  

Brain Tumor Classification Algorithm Based on Transfer Learning and Improved EfficientNet-B0

作　　者：王勇[1] 杨义龙 范晓晖 周雷 孔祥勇[1] WANG Yong;YANG Yilong;FAN Xiaohui;ZHOU Lei;KONG Xiangyong(School of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093

出　　处：《电子科技》2025年第4期46-51,共6页Electronic Science and Technology

基　　金：国家自然科学基金(61906121)。

摘　　要：针对现有脑肿瘤分类模型和方法复杂度高以及识别率低等问题,文中提出一种基于改进EfficientNet-B0的模型用于3种脑肿瘤分类。在数据预处理阶段,使用ROI(Region of Interest)特征裁剪出脑肿瘤图像的关键特征区域,并按肿瘤类型扩增数据集。根据卷积网络设计思想重新设计了EfficientNet中的MBConv(Mobile Inverted Bottleneck Convolution)模块,在首步卷积后引入卷积注意力CBAM(Convolutional Block Attention Module)。为了更完整地进行迁移学习,在不修改原始输出结构的基础上外接3个神经元用于脑肿瘤的三分类。改进网络模型具有更低的复杂度,可更好地适应肿瘤病灶的识别。文中利用迁移学习方法在公开数据集figshare-Brain Tumor Dataset上进行微调。实验结果表明,改进模型在该公共数据集上分类准确率为99.67%,相较于原始EfficientNet-B0网络提升了约3.1百分点。In view of the problems of high complexity and low recognition rate of existing brain tumor classification models and methods,an model based on improved EfficientNet-B0 is proposed for the classification of three brain tumors.In the data preprocessing stage,the ROI(Region of Interest)feature is used to cut out the key feature regions of brain tumor images,and the data set is expanded according to tumor type.The MBConv(Mobile Inverted Bottleneck Convolution)module in EfficientNet is redesigned according to the convolutional network design idea,and the CBAM(Convolutional Block Attention Module)is introduced after the first convolution step.In order to carry out transfer learning more completely,three neurons are attached to the brain tumors without modifying the original output structure.The improved network model has lower complexity and better adapts to the identification of tumor lesions.The transfer learning method is used to fine-tune the public data set figshare-Brain Tumor Dataset.Experimental results show that the improved model achieves a classification accuracy of 99.67%on the public data set,which is about 3.1 percentage points higher than the original EfficientNet-B0 network.

关 键 词：脑肿瘤分类 深度学习 卷积神经网络 阈值化处理 类平衡 EfficientNet ECA注意力机制 CBAM注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]