3 Deep learning approaches in metastatic breast cancer detection
Nazan Kemaloğlu, Turgay Aydoğan, Ecir Uğur Küçüksille · 2021
Breast cancer is one of the most common dangerous and fatal diseases in the world. Breast cancer, which is usually observed in women, is also observed in men, although rarely. Nowadays, thanks to developing technology, the early diagnosis and treatment of breast cancer have resulted in a decrease in mortality due to this disease. Although cancer occurs in a particular organ, it can then spread from that organ to other organs. This condition, called metastasis, is defined as the spread of cancerous cells to other areas directly or through blood-lymph vessels outside the tissue in which they are located. Cancerous cells can move to other parts of the body, settle there and quickly spread to new areas. As they spread to new parts of the body, abnormal cells of the same type develop and continue to have the same name as the first tumor. For example, if breast cancer has spread to the lung, the cells herein are essentially breast cancer cells, and the tumor that is formed in the lung is called “metastatic breast cancer” rather than “lung cancer.” The microscopic examination of adjacent lymph nodes is performed for metastatic breast cancer detection. In this procedure performed by pathologists, the detection of lymph nodes especially with very small tumors is quite difficult and time-consuming. Therefore, computerassisted metastasis detection increases the sensitivity and speed of the procedure. The data obtained from medical images can be processed in models using various artificial intelligence algorithms. The results obtained at the end of this procedure provide guidance to specialists in the diagnosis and treatment of the disease. Therefore, the use of computer vision systems in the diagnosis and treatment of the disease is becoming widespread. Furthermore, the deep learning approach, which has become popular recently, has accelerated this field. Convolutional neural network (CNN), which is a deep learning approach especially used to classify images, is a highly successful model. CNN models are insufficient to determine the properties of objects in the image, such as location, orientation, position and angular value. For this reason, capsule neural networks have been designed in order to obtain the image properties. In this section, breast metastases to axillary lymph nodes, a publicly available breast cancer dataset, were classified with CNN and capsule neural networks. Using the dataset consisting of 130 pathological images of 78 patients available, a study was performed to detect metastases on images.