Deep Transfer Learning for Breast Cancer Classification

According to GLOBOCAN 2020, there were 10.3 million cancer-related deaths and 19.3 million new cases of the disease worldwide. Female breast cancer represented 11.7% of all cases, or 2.26 million cases, globally [10, 31]. In Africa, year 2020, 85 787 women died from breast cancer out of 186598 new cases, 12.5% of world’s death by breast cancer [31]. In the same year, we diagnosed 2262419 new cases of breast cancer in the world [31]. Through the use of computer-aided procedures, the accuracy of diagnosis may be significantly boosted. One of the most efficient methods to detect breast cancer is a pathologist’s examination of histopathological pictures under a microscope [9, 12]. Since the invention of digital image scanners, picture identification has improved [3]. The practice of digital pathology has advanced significantly, in part because whole slide image (WSI) scanners enable quick and affordable diagnosis [8]. Breast cancer of the Invasive Ductal Carcinoma (IDC) kind comes from a dysfunction of milk duct’s cells of the breast and eventually spreads to the breast tissue nearby. It is an example of cancer type where early detection and accurate classification can significantly improve patient outcomes. The results for patients can be considerably improved by IDC early identification and precise categorization. On the basis of histopathological scans, IDCs have been accurately classified using machine learning approaches [1, 26]. Diagnosing pathological imaging is difficult and necessitates manual skills and a microscope. Due to the limitations of human cognition, this time-consuming approach frequently results in mistakes. One might anticipate time savings and decreased mistake rates with an automated system [26]. One of the main challenges in analyzing cancer images using deep learning algorithms is the limited availability of data [23, 29]. Deep learning algorithms require large amounts of data to train effectively, and the lack of data can lead to overfitting or underfitting of the model [23]. Many models have been created to detect malignant cells for quite some time, and deep learning models have been used in that regard. [24]. But, one can ask that: can deep learning algorithms accurately predict the presence of cancer tumor in image data? In this paper, we are going to use Vision Transformers, VGG, and ResNets to classify images of breast cancer.