基于改进ResNeXt的肺癌病理图像分类

Lung cancer pathological image classification based on improved ResNeXt

针对肺癌病理图像的自动分类,提出一种改进的卷积神经网络,引入一种新型卷积结构。以ResNeXt残差网络作为基础网络模型,使用新型卷积involution替代部分传统卷积层,解决传统卷积核的感受野小和通道之间信息比较冗余的问题。实验结果表明,该网络模型在LC25000数据集的肺癌病理图像分类任务中,肺良性图像、肺腺癌图像和肺鳞状细胞癌图像准确率分别达到100.00%、99.47%、99.47%,整个数据集准确率达到99.47%,表明改进的网络模型可以提高对肺癌病理图像分类的准确率。

Aiming at the automatic classification of pathological images of lung cancer,an improved convolutional neural network was introduced and a new convolution structure was proposed.ResNeXtresidual network was used as the basic network model and a new type of convolution involution was used to replace some of the traditional convolutional layers,which solved the problems of small receptive fields of traditional convolution kernels and relatively redundant information between channels.Experimental results show that in the lung cancer pathological image classification task of the LC25000 dataset,the accuracies of the lung benign,lung adenocarcinoma and lung squamous cell carcinoma images of the network model reach 100.00%,99.47%,and 99.47%,respectively.The accuracy of entire dataset reaches 99.47%.Therefore,it shows the improved model can improve the accuracy of lung cancer pathological image classification.