Deep learning approach for breast cancer staging

Abstract

Whole slide image (WSI) scanning is widely used with the advent of Computer Aided Diagnostics, which introduces the possibility of application of Deep Learning approaches to histopathology images providing exploration in the field of precision medicine. Its ap- plication has been limited due to the high resolution of WSI, which are gigapixel in size, and due to the ‘black-box’ nature of neural nets. We tried to overcome this by introducing a a patch- based approach for breast cancer stage identification based on cancer metastasis to lymph nodes, with an aim to produce interpretable and accurate results that can be verified and studied by pathologists and doctors to help assign treatment to patient. Segmentation is also performed on breast tissue sample to identify the morphological features that can help in cancer grade assignment