Improved Transfer Learning Based Deep Learning Model For Breast Cancer Histopathological Image Classification

In recent years, the demand for prompt detection and classification of breast cancer is rising sharply as breast cancer has become leading cancer type among women throughout the world. Convolutional Neural Networks(CNNs) are widely being used for performing mentioned tasks.However, they need a large number of labeled images which may appear to be infeasible for some kinds of medical images data such as mammographic tumor images. To address this difficulty, Transfer Learning becomes convenient. In this paper, we proposed a deep learning model for classifying Benign and Malignant types of breast tumor that trains an InceptionV3 model which pulls out features from the histopathological images of various magnification. These features are then used for classification. Introduced system is validated on BreakHis dataset and gains average validation set accuracy of 99.50%, 98.90%, 98.96% and 98.51% for magnification factor 40X, 100X, 200X and 400X respectively which outperforms all studied baseline models. Different performance metrices such as precision, recall, F
1
score, Specificity have additionally been used for performance estimation purposes.