MS UNet: Multi-scale 3D UNet for Brain Tumor Segmentation

Abstract

AbstractA deep convolutional neural network (CNN) achieves remarkable performance for medical image analysis. UNet is the primary source in the performance of 3D CNN architectures for medical imaging tasks, including brain tumor segmentation. The skip connection in the UNet architecture concatenates multi-scale features from image data. The multi-scaled features play an essential role in brain tumor segmentation. Researchers presented numerous multi-scale strategies that have been excellent for the segmentation task. This paper proposes a multi-scale strategy that can further improve the final segmentation accuracy. We propose three multi-scale strategies in MS UNet. Firstly, we utilize densely connected blocks in the encoder and decoder for multi-scale features. Next, the proposed residual-inception blocks extract local and global information by merging features of different kernel sizes. Lastly, we utilize the idea of deep supervision for multiple depths at the decoder. We validate the MS UNet on the BraTS 2021 validation dataset. The dice (DSC) scores of the whole tumor (WT), tumor core (TC), and enhancing tumor (ET) are \(91.938\%\), \(86.268\%\), and \(82.409\%\), respectively.KeywordsCNNUNetContextual informationDense connectionsResidual inception blocksBrain tumor segmentation