Automatic Diagnosis of Endometrial Cancer With Deep Learning

Diagnosing endometrial cancer early is critical to maximizing the survival rate. Currently, the best diagnostic method is manual inspection under microscope by histologists. This method can be slow, which can negatively affect survival rates. Automating diagnosis has great potential to drastically speed up the diagnostic pipeline, but has its unique challenges. Automating endometrial cancer diagnosis from biopsy slides is challenging due to the sparse availability of images as well as the gigapixel resolution of the pathology images. This thesis presents a method to automatically compress and classify endometrial biopsy slides into benign, neoplastic, and hyperplastic classes.