Deep Learning for Detection and Classification of Nuclear Protein in Breast Cancer Tissue

Breast cancer is a medical condition in which the cells of the mammary gland grow uncontrollably, creating malignant tumors, capable of dividing, and invading surrounding areas. Currently, the incidence of breast cancer is increasing and getting younger. Medical methods such as Ultrasound, Computed Tomography (CT), X-rays, and Magnetic Resonance Imaging (MRI) are the first options to diagnose breast cancer as well as other diseases in the mammary gland. However, these imaging methods are not enough to accurately determine the status of a breast tumor. To be able to accurately identify breast cancer and assess the extent of the disease, immunohistochemistry will be used to characterize the cells in the tissue sample, including breast cancer cells. In this work, we propose a method to detect and classify nuclear proteins in breast cancer cells on histological images using deep learning techniques. We take advantage of deep learning networks and the transfer learning approach to train network models on the SHIDCB-Ki-67 dataset of Shiraz University of Medical Sciences in Shiraz, Iran. DeepLabv3-MobileNet-V2, DeepLabv3-Xception, and DeepLabv3-DenseNet-121 are used to detect and classify nuclear proteins of breast cancer cells on histological images. Experimental results show that the proposed method with DeepLabv3-DenseNet-121 achieves higher accuracy (98.6%) than DeepLabv3-Xception and DeepLabv3-MobileNet-V2.