VTBIS: Vision Transformer for Biomedical Image Segmentation

Abstract

In this paper, we propose a novel network named Vision Transformer for Biomedical Image Segmentation (VTBIS). Our network splits the input feature maps into three parts with 1 × 1, 3 × 3 and 5 × 5 convolutions in both encoder and decoder. Concat operator is used to merge the features before being fed to three consecutive transformer blocks with attention mechanism embedded inside it. Skip con- nections are used to connect encoder and decoder trans- former blocks. Similarly, transformer blocks and multi scale architecture is used in decoder before being linearly pro- jected to produce the output segmentation map. We test the performance of our network using Synapse multi-organ segmentation dataset, Automated cardiac diagnosis chal- lenge dataset, Brain tumour MRI segmentation dataset and Spleen CT segmentation dataset. Without bells and whis- tles, our network outperforms most of the previous state of the art CNN and transformer based models using Dice score and the Hausdorff distance as the evaluation met- rics.

Tasks

Reproductions