The Advantages of Employing Transfer Learning in the Classification of Breast Cancer Histopathological Images

Digital pathology represents a significant advancement in contemporary medicine, offering enhanced diagnostic capabilities and improved patient outcomes. Pathological examinations, which need particular steps in the diagnostic process, are standard in medical protocols and the law. Today, a new challenge is to use cutting-edge algorithms, like Convolutional Neural Networks (CNN), to classify histological images into different groups. So, the Invasive Ductal Carcinoma (IDC) dataset was used to use some well-known CNNs, such as VGG16, DenseNet169, and EfficientNetV2B3 pre-trained networks, as well as two new custombuilt CNNs with four (CNN1) and five (CNN2) layers. The results show that for a 70% training to 30% testing ratio, CNN1 (0.895), CNN2 (0.882), VGG16 (0.983), DenseNet169 (0.971), and EfficientNetV2B3 (0.979) all got the best results on the test set. The results obtained with pre-trained CNNs are superior to proposed custombuilt CNNs. This outcome denotes the main advantage of leveraging pre-trained CNNs in classifying breast cancer histopathological images.