Available online 20 August 2022, 104173

A novel method (3D MTDC-Net) is proposed to automatic brain tumor segmentation.

A MTDC strategy with PMF algorithm is proposed to extract the structure feature from multi-sequence data.

A novel SPC operation is used to extract different dimensions of context semantic information to accelerate convergence of the MTDC-Net.

A MTAU strategy is proposed to increase the weight of semantic information to improve multi-scale tumor segmentation result.

3D MTDC-Net has less parameters about 1.8 million, but it also achieves good segmentation performance.

Glioma is one of the most threatening tumors and the survival rate of the infected patient is low. The automatic segmentation of the tumors by reliable algorithms can reduce diagnosis time. In this paper, a novel 3D multi-threading dilated convolutional network (MTDC-Net) is proposed for the automatic brain tumor segmentation. First of all, a multi-threading dilated convolution (MTDC) strategy is introduced in the encoder part, so that the low dimensional structural features can be extracted and integrated better. At the same time, the pyramid matrix fusion (PMF) algorithm is used to integrate the characteristic structural information better. Secondly, in order to make the better use of context semantical information, this paper proposed a spatial pyramid convolution (SPC) operation. By using convolution with different kernel sizes, the model can aggregate more semantic information. Finally, the multi-threading adaptive pooling up-sampling (MTAU) strategy is used to increase the weight of semantic information, and improve the recognization ability of the model. And a pixel-based post-processing method is used to prevent the effects of error prediction. On the brain tumors segmentation challenge 2018 (BraTS2018) public validation dataset, the dice scores of MTDC-Net are 0.842, 0.892 and 0.819 for core, whole and enhanced of the tumor, respectively. On the BraTS2020 public validation dataset, the dice scores of MTDC-Net are 0.843, 0.896 and 0.813 for the core tumor, whole tumor and enhancing tumor, respectively. Mass numerical experiments show that MTDC-Net is a state-of-the-art network for automatic brain tumor segmentation.

A novel method MTDC-NET for 3D multi-scale brain tumors automatic segmentation. In the encoding stage, MTDC strategy is used to replace common pooling to down-sample the feature graph. In the decoding stage, the SPC operation is used to extract context semantic information from different scales, and the MTAU strategy is used to increase the weight of semantic information. Meanwhile, the dilated-connection is used to replace the linear skip-connection before up-sample operation to recover the detail information better. Finally, a pixel-based post-processing method is used to prevent the effects of error prediction. Experimental results show that the MTDC-Net achieves satisfactory multi-scale brain tumors segmentation performance compared with start-of-the-art methods.

Download : Download high-res image (139KB)Download : Download full-size image

Dilated connect

Multi-threading dilated convolution

Spatial pyramid convolution

Multi-threading adaptive pooling strategy

Brain tumor segmentation

© 2022 Elsevier Inc. All rights reserved.