Accurate diagnosis of non-Hodgkin lymphoma on whole-slide images using deep learning

Abstract

Digital Pathology is the technique of digitizing histology slides to create high-resolution images. One of the important applications for digital pathology is tissue level classification, such as the identification of the three most common kinds of non-Hodgkin lymphomas; Mantle Cell Lymphoma, Follicular Lymphoma and Chronic Lymphocytic Leukemia, which pose a significant problem for pathologists due to their inherent complexity. In this research, deep learning ideas are combined with an improvement of the CNN algorithm to propose a Non-Hodgkin Lymphoma model which is able, effectively, to categorize these subtypes of non-Hodgkin lymphomas. The proposed model was trained using NIA-curated dataset images and achieved a classification accuracy of 98.7%, about 0.6% better than current classification strategies for non-Hodgkin’s lymphomas based on deep learning.