GEDFormer: Gradient Edge Detection in LDCT Image Denoising Transformer Model

Convolutional neural networks (CNNs) have excelled in deep learning applications, including medical image processing, due to their ability to extract hierarchical feature representations. While CNNs excel at capturing local features, their fixed receptive fields can make it challenging to effectively model global contextual information, essential for certain tasks, such as medical image denoising. To address this, Vision Transformers (ViTs) have emerged as an alternative, leveraging self-attention mechanisms to capture both global and local dependencies within images. This study explores the use of a standalone ViT-based framework for denoising low-dose computed tomography (LDCT) images with a self-guided gradient edge detecting attention module, which aims to preserve critical spatial and frequency details required for accurate diagnostic outcomes. The proposed method is rigorously evaluated by comparing its performance against state-of-the-art traditional CNN models (BM3D, DSC-GAN, RED-CNN) and a hybrid CNN-ViT model (TED-Net). Both numerical data analysis and image inspection are used to demonstrate the efficacy of the ViT-based approach.