Histological Subtype Classification of Gliomas in Digital Pathology Images Based on Deep Learning Approach

摘要

Among primary malignant brain tumors in adults, glioma is the most common one. Diffuse glioma can be classified into seven common pathological subtypes according to histological phenotypes defined in 2016 WHO classification guidance. Compared with grading, subtype classification concernsmore about histological appearance as well as survival and treatment options. A deep learning based framework is proposed to classify seven glioma pathological subtypes. The framework consists of two parts: multiple-tissue detection and subtype classification. The proposed method combinesthe local cellular level information with the global structure information. The framework produced 95.99% classification accuracy on eight types of cellular tissues and 72.66% classification accuracy on seven pathological subtypes.