Performance analysis of breast cancer histopathology image classification using transfer learning models

Abstrak

Convolutional neural networks (CNN) which are deep learning-based methods are being currently successfully deployed and have gained much popularity in medical image analysis. CNN can handle enormous amounts of medical data which makes it possible for accurate detection and classification of breast cancer from histopathological images. In the proposed method, we have implemented transfer learning-based classification of breast cancer histopathological images using DenseNet121, DenseNet201, VGG16, VGG19, InceptionV3, and MobileNetV2 and made a performance analysis of the different models on the publicly available dataset of BreakHis. These networks were pre-trained on the ImageNet database and initialized with weights which are fine-tuned by training with input histopathological images. These models are trained with images of the BreakHis dataset with multiple image magnifications. From the comparative study of these pre-trained models on histopathology images, it is inferred that DenseNet121 achieves the highest breast cancer classification accuracy of 0.965 compared to other models and contemporary methods.