Transformer With Double Enhancement for Low-Dose CT Denoising

CT image denoising methods for image quality improvement and radiation dose reduction

“…Additionally, two studies adopt Transformer-based approaches. 150,151 with the remaining approaches including filter-based methods as well as hybrid methods. [152][153][154][155][156][157] To be noted that some models are originally developed, 61,62,65,66,68,70,72,73,77,122,153 while some are developed by modifying original models through modifying loss functions, or layers, or extending original models to different domains.…”

“…The model's built-in self -attention mechanism enables it to non-linearly record relationships between segments of the input sequence, making it a good choice for processing high-dimensional CT images. 150 The Transformer model has also shown promise in other fields, such as image processing, and has reached state-of -the-art performance in a variety of natural language processing applications. Additionally, the Transformer model can process each part of the image separately, making it more efficient for handling large images compared to typical convolutional neural networks.…”

“…144 Self -attention is employed in Transformer-based models. 150 Self -attention learning involves calculating attention scores between each pair of elements in the sequence and using these scores to compute weighted sums,which are then used to update the representations of the elements. The key idea is to allow each element to attend to other elements, assigning different degrees of importance to them based on their contextual relevance.…”

“…As shown in Figure 2a, out of 99 publications examined, 61 studies use the models based on CNN, 59–119 while 30 studies are based on GAN 120–149 . Additionally, two studies adopt Transformer‐based approaches 150,151 . with the remaining approaches including filter‐based methods as well as hybrid methods 152–157 .…”

“…with the remaining approaches including filter‐based methods as well as hybrid methods 152–157 . To be noted that some models are originally developed, 61,62,65,66,68,70,72,73,77,122,153 while some are developed by modifying original models through modifying loss functions, or layers, or extending original models to different domains 60,63,64,69,71,72,74–76,78,82,84,102,121,123–125,147 , 150–152,155,158 …”