EfficientNetv2 Model for Breast Cancer Histopathological Image Classification
Dingming Liu, Weixiao Wang, Xuanrui Wu, Junjie Yang
Abstract
The incidence of breast cancer in women is 24.2% worldwide, ranking first among female cancers. Combination of modern medical image processing technology and deep learning algorithm can be effectively applied to remote diagnosis, instant file access and simplified procedure consultation. In this paper, using Tensorflow Keras backend, and the basic architecture of convolutional neural network based, we compared the accuracy of breast cancer prediction under different pre-trained convolutional neural network models including Efficientnetv2, Resnet_v2, Inception_v3, Mobilenet_v2 and Inception_Resnet_v2 with 9,109 microscopic images of breast tumor tissue from P&D Laboratory in Parana, Brazil. As a whole, Efficientnetv2-b0 to b2 performs as almost well as Mobilenet_ v2 and Inception_Resnet_ v2. Efficientnetv2-b1 did the best in our research with the accuracy score as 0.978. Inception _ v3 was far behind other models with the accuracy score as only 0.739.