Multimodal Brain Tumor Segmentation and Survival Prediction Using Hybrid Machine Learning

Abstract

In this paper, we propose a UNet-VAE deep neural network architecture for brain tumor segmentation and survival prediction. UNet-VAE architecture has shown great success in brain tumor segmentation in the multimodal brain tumor segmentation (BraTS) 2018 challenge. In this work, we utilize the UNet-VAE to extract high dimension features, then fuse with hand-crafted texture features to perform survival prediction. We apply the proposed method to the BraTS 2019 validation dataset for both tumor segmentation and survival prediction. The tumor segmentation result shows dice score coefficient (DSC) of 0.759, 0.90, and 0.806 for enhancing tumor (ET), whole tumor (WT), and tumor core (TC), respectively. For the feature fusion-based survival prediction method, we achieve 56.4% classification accuracy with mean square error (MSE) 101577, and 51.7% accuracy with MSE 70590 for training and validation, respectively. In testing phase, the proposed method for tumor segmentation achieves average DSC of 0.81328, 0.88616, and 0.84084 for ET, WT, and TC, respectively. Moreover, the model offers accuracy of 0.439 with MSE of 449009.135 for overall survival prediction in testing phase.