FCTrans UNet: A Hybrid CNN and Transformer Model for Medical Image Segmentations
Haoran Cheng, Mengyu Zhu · 2024
Medical image segmentation plays a pivotal role in isolating the region of interest, significantly advancing the field of medicine, particularly in the diagnosis and treatment of diseases. Convolutional neural networks (CNNs), such as U-Net, have attained significant success in medical image segmentation tasks. However, they are limited in establishing long-range dependencies due to the constrained sensory field of convolutional operations. Recently, researchers have proposed TransUnet to address the limitations of convolutional neural networks in establishing long-term dependencies and global contextual connections. This paper introduces a hybrid network model, feature-concatenate TransUNet (FCTransUNet) to present a improvement to the original TransUNet. To enhance the fusion of features in the encoder and decoder components, a feature fusion module (CSFFM) is introduced. Additionally, a feature extraction module (SFE) is incorporated into the decoder part to bolster feature extraction, thereby improving accuracy in multi-organ image segmentation.