Classifying breast cancer using transfer learning models based on histopathological images

Deep learning algorithms are designed to learn from the data, where these require large amount of training dataset for accurate prediction. Recent studies have depicted that transfer learning-based DL approaches perform accurately in a variety of applications to create Computer-Aided Design (CAD) systems. These systems are used for the early detection and analysis of diseases such as lung cancer, brain tumor, and breast cancer using various modalities. Instead of developing neural network models from the scratch, pre-trained models are frequently utilized for DL-based tasks in computer vision as they diminish time. The effectiveness of transfer learning models without applying augmentation and preprocessing techniques to automate the classification of tumors is explained in this work. Seven transfer learning models (LENET, VGG16, DarkNet53, DarkNet19, ResNet50, Inception, and Xception) are implemented on BreakHis dataset for the tumor classification, where Xception computed the best accuracy of 83.07%. Further, to attain the accuracy with unbalanced dataset, a new parameter named Balanced Accuracy (BAC) is best computed by DarkNet53 (87.17%). This study will facilitate the researchers and medical practitioners to choose an accurate model for the classification of tumor with unbalanced dataset. It will aid medical professionals to efficiently and precisely classify the disease.