TransBTSV2: Towards Better and More Efficient Volumetric Segmentation of Medical Images

Abstract

Transformer, benefiting from global (longrange) information modeling using self-attention mechanism, has been successful in natural language processing and computer vision recently. Convolutional Neural Networks, capable of capturing local features, are unable to model explicit long-distance dependencies from global feature space. However, both local and global features are crucial for dense prediction tasks, especially for 3D medical image segmentation. In this paper, we exploit Transformer in 3D CNN for 3D medical image volumetric segmentation and propose a novel network named TransBTSV2 based on the encoder-decoder structure. Different from our original TransBTS [1], the proposed TransBTSV2 is not limited to brain tumor segmentation (BTS) but focuses on general medical image segmentation, providing a strong and efficient 3D baseline for volumetric segmentation of medical images. As a hybrid CNN-Transformer architecture, TransBTSV2 can achieve accurate segmentation of medical images without any pre-training. With the proposed insight to redesign the internal structure of Transformer and the introduced Deformable Bottleneck Module, a highly efficient architecture is achieved with superior performance. Extensive experimental results on four medical image datasets (BraTS 2019, BraTS 2020, LiTS 2017 and KiTS 2019) demonstrate that TransBTSV2 achieves comparable or better results as compared to the state-of-the-art methods for the segmentation of brain tumor, liver tumor as well as kidney tumor. Code is available at https://github.com/ Wenxuan-1119/TransBTS.

Medical image segmentation based on self-supervised hybrid fusion network

“…We compare our model with the state-of-the-art models on the BraTS 2020 validation set, with experimental data from Li et al (32). The experimental data shows that our model obtains the best results on ET and WT (Table 2).…”

Understanding the brain with attention: A survey of transformers in brain sciences

“…Through the feature fusion and analysis from the Transformer and CNN, TransBTS produced high-resolution segmentation. Subsequently, an improved version of TransBTS called TransBTSV2 was developed, [129] which contained a variable bottleneck module with the Transformer for shape and detail awareness, thereby achieving superior performance. Lyu et al [130] also replaced the bottleneck layer with 12 consecutive Transformers to obtain attention enhancement.…”

Brain Tumor Segmentation Algorithm Based on Asymmetric Encoder and Multimodal Cross-Collaboration

“…In this regard, TransBTS [10] network is proposed, which incorporates Transformer into the 3D CNN encoder-decoder architecture for the first time, enhancing global feature extraction. TransBTSV2 [11] further improves TransBTS by redesigning the Transformer module www.ijacsa.thesai.org and introducing deformable bottleneck module to capture shape-sensitive local features. SwinBTS [12] structure employs 3D SwinTransformer as both encoder and decoder of the network to extract global information from feature maps efficiently, using convolution operation for upsampling and downsampling.…”