Combined Radiology and Pathology Based Classification of Tumor Types

Abstract

Computer Aided Detection plays a crucial role in the early detection of deadly diseases such as cancer (or) tumor. Pathology and radiology images form the core of tumor diagnosis. Pathology images provide clinical information about the tissues whereas the radiology images can be used for locating the lesions. This work aims at proposing a classification model which categorizes the tumor as oligodendroglioma (benign tumors) (or) astrocytoma (Malignant tumors) using features of both the radiology and pathology images. Dataset from MICCAI Computational Precision Medicine Challenge is used for model building. The proposed model uses dedicated workflows for processing the pathology and radiology images. The feature descriptors of the images are obtained using pre-trained Inception v3 model. The resulting vectors are then used as input to the linear SVM (Support Vector Machine) classification model. The SVM model provided an accuracy of 75% on the blind folded test dataset provided in the competition.