A Two-Stage Atrous Convolution Neural Network for Brain Tumor Segmentation and Survival Prediction

The chapter introduces a sophisticated two-stage neural network model for brain tumor segmentation and survival prediction. The model employs atrous convolutions and asymmetric U-Net architecture to accurately identify tumor core, enhancing tumor structures, and whole tumor regions within multi-parametric MRI scans. The segmentation results are then used to extract radiomic features, which are employed in survival prediction using the Extra Trees algorithm. The model demonstrates promising results in both segmentation and survival prediction tasks, showcasing the potential of advanced neural networks in medical imaging and patient outcome prediction.