Ultrasound Image Segmentation using a Model of Transformer and DFT

Medical image segmentation is a vital technique for physicians to diagnose and treat certain diseases. Recently, deep learning techniques have been successfully applied to a wide range of applications, such as semantic segmentation. This has led to the widespread application of these techniques in the medical image domain. In this work, we aim to improve the accuracy of deep learning techniques in ultrasound cardiac image segmentation. One notable example of recent advancements is a medical image segmentation model called Multiaxis External Weights UNet (MEW-UNet), which employs a 2D Discrete Fourier Trans-form (DFT) along the three axes of the channel dimension. Building on this existing technique, this work proposes several enhancements to the MEW-UNet architecture by introducing a hybrid approach that incorporates DFT with Transformers. The motivation behind this proposed hybrid architecture is to improve the accuracy and robustness of MEW-UNet for ultrasound image segmentation. The performance of the proposed approach was compared with that of the original MEW-UNet and other state-of-the-art neural networks. Our improved network architectures achieved superior accuracy in terms of DSC and HD95.