Classification of orbital tumors using convolutional neural networks

Abstract

Abstract Orbital tumors are the most common eye tumors that affect people all over the world. Early detection prevents the progression to other regions of the eye and the body. Also, early identification and treatment could reduce mortality. A computer-assisted diagnosis (CAD) system to help physicians diagnose tumors is in great demand in ophthalmology. In recent years, deep learning has demonstrated promising outcomes in computer vision systems. This work proposes a CAD system for detecting various forms of orbital tumors using convolutional neural networks. The system has three stages: preprocessing, data augmentation and classification. The proposed system was evaluated on two datasets of magnetic resonance imaging (MRI) images containing 1404 MRI T1-weighted images and 1560 MRI T2-weighted images. The results have shown that the system is capable of detecting and classifying the tumor in each image type, and the recognition rate for the T1-weighted image is 98% and for the T2-weighted image is 97%.