Deeptech Learning for Breast Cancer

The only definitive way to diagnose breast cancer is by examining tissue samples collected from biopsy or surgery. The samples are commonly prepared with hematoxylin and eosin (H&E) staining to increase the contrast of structures in the tissue. Traditionally, pathologists examine the tissue on glass slides under a microscope to detect tumor tissue. Diagnosis takes time as pathologists must thoroughly inspect an entire slide at close magnification. Further, pathologists may not notice small tumors. Deep learning methods aim to automate the detection of tumor tissue, saving time and improving the detection rate of small tumors.

Deep learning methods for tumor classification rely on digital pathology, in which whole tissue slides are imaged and digitized. The resulting WSIs have extremely high resolution. WSIs are frequently stored in a multiresolution file to facilitate the display, navigation, and processing of the images.

Our Solution presents classification results as heatmaps that depict the probability that local tissue is tumorous. The localization of tumor regions enables medical pathologists to investigate specific regions and quickly identify tumors of any size in an image.