Multi-task Deep Learning for Colon Cancer Grading

Abstract

Automated cancer grading is an important subject of study in digital pathology. In this paper, we introduce a multi-task learning approach to analyze digitized pathology images. The approach performs both classification and regression tasks in combination with a deep convolutional neural network to predict the tumor grade. Employing tissue microarrays (TMAs) and whole slide images (WSI), the proposed method achieved an accuracy of 85.91% in classifying colon tissues into four distinctive pathology classes, including benign and well differentiated, moderately differentiated, and poorly differentiated tumors.