REED-NET: Residual Enhanced Encoder-Decoder Network for Low-Dose CT Reconstruction

Increased radiation exposure to the patient is one of the most significant disadvantages of high-dose CT. The ionizing radiation used in CT scans entails the potential risk of damaging DNA and increasing the risk of cancer, especially in the patients those who undergo multiple CT scans. Deep learning (DL) is a useful method for reducing dosage in medical imaging. This is accomplished by optimizing image quality while employing reduced radiation doses. Deep learning models can successfully denoise images, remove artifacts, and improve spatial resolution, allowing images to be acquired at lower dosages without sacrificing diagnostic quality. In the proposed method encoders, decoder networks are combined together and shortcut connections are used to construct the Residual Enhanced Encoder-Decoder CNN (REED-NET) for low-dose CT imaging. We have used the Gaussian Error Linear Unit (GeLU) as the activation function. Training has been performed by using the patch-based approach, our proposed REED-NET has outperformed when compared with the traditional techniques.