基于多任务模型的乳腺癌病理图像分类

Breast cancer pathological image classification based on multi-task model

基于卷积神经网络提出了一种多任务模型将乳腺癌组织学图像分为良性与恶性及其子类.该模型是多任务模型,任务一将病理图像分为良性与恶性,任务二将图像分为良性与恶性的子类.模型总的损失函数是两个分类任务损失函数的加权和.该模型采用卷积层和全局平均池化层替代末端全连接层作为分类层,应用数据增强方法提升模型的性能.模型使用乳腺癌病理图像数据集BreaKHis进行消融实验并与VGG16模型进行对比.实验结果显示:提出的模型能够取得更好的性能,在二分类上达到了98.55%~99.52%的分类准确率,在多分类上达到了92.26%~94.85%的分类准确率.

Based on the convolutional neural network,a multi-task model was proposed to divide breast cancer histological images into benign and malignant,and into benign and malignant sub-classes.This model is a multi-task model.In task one,histological images were divided into benign and malignant,and in task two,histological images were divided into benign and malignant subclasses. The total loss function of the model is the weighted sum of the loss functions of the two classification tasks. The loss function of the model is the sum of the loss functions of the two classification tasks. A convolutional layer and a global average pooling layer were used to instead of the end fully connected layer as the classification layer,and data augmentation methods were applied to improve the performance of the model. The breast cancer pathology image dataset BreaKHis was used for ablation experiments,which was compared with the VGG16 model.The experimental results show that the proposed model achieves better performance, with a classification accuracy of 98.55%~99.52% in two classifications, and a classification accuracy of 92.26%~94.85% in multi-classification.