Brain Tumor Segmentation and Survival Prediction

The paper demonstrates the use of the fully convolutional neural network for glioma segmentation on the BraTS 2019 dataset. Three-layers deep encoder-decoder architecture is used along with dense connection at the encoder part to propagate the information from the coarse layers to deep layers. This architecture is used to train three tumor sub-components separately. Sub-component training weights are initialized with whole tumor weights to get the localization of the tumor within the brain. In the end, three segmentation results were merged to get the entire tumor segmentation. Dice Similarity of training dataset with focal loss implementation for whole tumor, tumor core, and enhancing tumor is 0.92, 0.90, and 0.79, respectively. Radiomic features from the segmentation results predict survival. Along with these features, age and statistical features are used to predict the overall survival of patients using random forest regressors. The overall survival prediction method outperformed the other methods for the validation dataset on the leaderboard with 58.6% accuracy. This finding is consistent with the performance on the test set of BraTS 2019 with 57.9% accuracy.