Cover Article |
Cover Article, NST 2023 issue 4
Low-dose CT (LDCT) effectively reduces the radiation dose received by patients during the CT scanning process, thereby reducing the possible radiation damage to patients. However, low-dose CT technology may introduce noise and artifacts while reducing radiation dose, resulting in blurred images, loss of detail, and compromising the diagnostic value of CT. Therefore, how to improve the image quality of low-dose CT is a vital frontier issue in the field of nuclear medicine.
A novel deep-learning model called Hformer was proposed by Zhang, SY., Wang, ZX., Yang, HB. et al. in articles “Hformer: highly efficient vision transformer for low-dose CT denoising” published in NUCL SCI TECH 34, 61 (2023).
Hformer combines the advantages of the local feature extraction capability of the convolutional neural network and the global feature capture of the Transformer model. The computational effort of the model is also decreased in Hformer by lightweight self-attention computation to improve the performance of the low-dose CT noise reduction model. Efficiency of the model is validated in this study by using the AAPM-Mayo Clinic Low-Dose CT Grand Challenge dataset, and the results demonstrate that Hformer achieves state-of-the-art performance with a PSNR of 33.4405, an RMSE of 8.6956, and an SSIM of 0.9163 without requiring many learning parameters. More details of this study are described in https://doi. org/10.1007/s41365-023-01208-0