Transformer-Based Medical Image Fast Segmentation Network

This paper presents a Transformer - based Swift Medical Image Segmentation Network, with the objective of enhancing the precision and velocity of medical image segmentation. The self - attention mechanism is employed to effectively extract image features and strengthen the comprehension of intricate structures. The system is composed of three components: the feature extraction module, the Transformer encoder, and the segmentation decision - making unit. In the feature extraction part, the input image undergoes processing by Convolutional Neural Network (CNN), and a multitude of low - level features are acquired. Subsequently, these features are input into the Transformer encoder, which utilizes the self - attention mechanism to capture long - distance dependencies, thus elevating the segmentation accuracy. Eventually, with the assistance of a fully-connected layer, the segmentation decision-making section generates the final segmentation map. To verify the effectiveness of the proposed method, simulations are performed on numerous public medical image datasets. The findings indicate that the model delivers exceptional performance in the segmentation of CT and MRI images. Based on the Brain Tumor Segmentation Challenge (BRATS) dataset, this model attains a Dice coefficient of 0.89, outperforming traditional segmentation algorithms. Moreover, the reasoning speed of the model can reach up to 30 frames per second, which markedly enhances the processing efficiency.