Brain Tumor Screening using Adaptive Gamma Correction and Deep Learning

Abstract

Generally, the brain tumor is regarded as one of the most dangerous diseases. It is always too late to detect the brain as the tumors at the early stage are always ignored. In fact, the traditional manual diagnosis process is inefficient. The radiologists have to accomplish a great amount of reading work per day, which can result in weariness and thus lead to misdiagnosis. To liberate radiologists from endless work, a brain tumor screening system based on adaptive gamma correction and deep learning is proposed. The brain images are labeled with non-tumor and tumors, and the radiologists just needs to deal with the brain images labeled with tumors, which can significantly reduce the workload of the radiologists. Firstly, sufficient contrast enhanced T1-weighted brain images are collected. Further, background removal based on iterative threshold and a novel adaptive gamma correction (NAGC) are implemented to generate brain images with similar overall intensity. Finally, data augmentation technologies are applied to enlarge the training set, and convolutional neural network (CNN) is adopted to train the classifier. The results indicate that the accuracy of the proposed system can reach 95.13%.