Histopathological images classification using aggressive transfer learning

Breast cancer is among the most common and deadly diseases that attack women worldwide. Recent epidemiological statistics show that millions of women are diagnosed with the disease every year, and over half a million women die directly of the disease every year. Proper and timely diagnosis, especially the distinction between benign and malignant breast lesions like calcified residual masses, is critical in increasing the survival rates as well as the therapeutic outcome. The recent advances in artificial intelligence, specifically deep learning, have revolutionized the field of medical imaging and allowed conducting more accurate and valid diagnostic tests. The algorithms help to analyze complex medical data through automated recognition of patterns in histopathological images, helping radiologists and pathologists to make decisions faster and more accurate. Convolutional Neural Networks (CNNs) have shown significant effectiveness in the detection of cancer according to the imagery, due to the ability to learn discriminative attributes on large sets of images and successful classification of tissue samples. In this study, a new aggressive transfer-learning model specifically designed to perform binary classification of breast histopathological images was proposed and assessed on the BreakHis benchmark. The offered solution was more successful in performance indicators in comparison with the current models. In addition, the tested method on the BreaKHis dataset with a clear emphasis on the separation of benign and malignant samples of breast tumors demonstrated competitive results, which justifies the method of aggressive transfer learning as a viable option in the analysis of breast histopathology.