基于EMobileNet的肝硬化图像分类网络研究

Study on cirrhosis image classification network based on EMobileNet

为了解决肝硬化病理图像分类准确率低的问题,基于复合卷积神经网络提出了一种更准确的对肝硬化病理图像分类的深度学习模型EMobileNet.EMobileNet是在EfficientNetV2网络的基础上,插入可分离自注意力模块MobileViTv2,并引入四阶龙格-库塔残差块调整MBViTconv模块间的连接方式.肝硬化病理图像数据集由华中科技大学同济医学院附属同济医院提供,实验采用十折交叉验证法对线阵和凸阵图像分别进行训练并分类.对所提出模型进行消融实验并与多种经典网络效果进行对比,结果显示EMobileNet模型在三分类任务上达到了97.89%的总体准确率,显著提高了检测性能,可以取得较高精度的肝硬化病理图像分类效果.

To solve the problem of low classification accuracy of cirrhotic pathological images,this paper proposes a more accurate deep learning model EMobileNet for the classification of cirrhotic pathological images based on a composite convolutional neural network.EMobileNet refers to EfficientNetV2 network with the insertion of the separable self-attention module MobileViTv2,and the introduction of fourth-order Runge-Kutta residual block to adjust the connectivity between MBViTConv modules.The liver cirrhosis pathology image dataset was provided by Tongji Medical College,Huazhong University of Science and Technology,and the experiments were conducted using the ten-fold cross-validation method to train and classify the convex and line array images respectively.The proposed model was subjected to ablation experiments and compared with the effects of various classical networks,and the results showed that EMobileNet achieved an overall accuracy of 97.89%on the triple classification task,significantly improving the detection performance and enabling a higher accuracy in the classification of cirrhotic pathology images.