Automatic Brain Tumor Segmentation Using Convolutional Neural Networks with Test-Time Augmentation

Abstract

Automatic brain tumor segmentation plays an important role for diagnosis,\nsurgical planning and treatment assessment of brain tumors. Deep convolutional\nneural networks (CNNs) have been widely used for this task. Due to the\nrelatively small data set for training, data augmentation at training time has\nbeen commonly used for better performance of CNNs. Recent works also\ndemonstrated the usefulness of using augmentation at test time, in addition to\ntraining time, for achieving more robust predictions. We investigate how\ntest-time augmentation can improve CNNs' performance for brain tumor\nsegmentation. We used different underpinning network structures and augmented\nthe image by 3D rotation, flipping, scaling and adding random noise at both\ntraining and test time. Experiments with BraTS 2018 training and validation set\nshow that test-time augmentation helps to improve the brain tumor segmentation\naccuracy and obtain uncertainty estimation of the segmentation results.\n