DRT-Unet: A Segmentation Network for Aiding Brain Tumor Diagnosis

Using image segmentation techniques to assist physicians in brain tumor diagnosis is a hot issue computer technology research. Although most networks date have been based on U-Net, the prediction results are depending which not well generalized and need be further improved. As depth of network increases, gradients vanish together with decrease accuracy; meanwhile, large number parameters will cause data redundancy. Moreover, single modality MRI images cannot adequately segment details. Therefore, an improved U-Net model proposed this paper, combines Dilated Convolution-Dense Block-Transformation Convolution-Unet (hereafter referred as DRT-Unet). The adopts combination dilated convolution, dense residual block, transposed convolution. In coding process, convolution block local feature for fusing adopted replace 3 × layers each layer transition used down-sampling. decoding blocks; deconvolution structure up-pooling cascade used. By connecting decoded output features encoded low-level visual features, information loss obtained. experiments paper carried out BraTs2018 BraTs2019 datasets; result, DRT-Unet can effectively lesion regions.