Low-dose CT Image Restoration using generative adversarial networks

Computed tomography (CT) images are the prominent modality of medical imaging used in medical diagnosis and are now widely applied in clinical diagnosis. However, the excessive X-ray radiation dose associated with CT scans is a potential risk to patients. In this paper, we propose an algorithm that improves the detail and edge preservation performance of low-dose CT images by applying a generative adversarial network structure to a deep learning noise reduction method based on a convolutional neural network. This approach is performed by competing the discriminator neural network and the denoising filter neural network when learning the neural network. In order to preserve the textural details and edges of low-dose CT image, an optimized generator was derived by minimizing the weighted sum of L2 and structural similarity index (SSIM) losses instead of simply using Mean Square Error (MSE) when optimizing neural network training. According to experimental results, the proposed method can effectively suppress noise and remove artifacts compared to the state-of-the-art methods. Experimental results on low-dose CT images show that the proposed method preserves the details of image, reduces noise and has better visual effects.