Fast low-dose Computed Tomography image Super-Resolution Reconstruction via Sparse coding and Random Forests

Abstract

X-ray radiation is harmful to human health. Therefore, how to obtain a better quality reconstructed image with low dose scan is a major challenge in the field of computed tomography (CT). This paper proposes a fast low-dose CT super-resolution method based on sparse coding and random forests. By using high-resolution training images and low-resolution training images to obtain high-resolution dictionaries and using back-projection to ensure global consistency, and finally using sparse coding to extract and fuse useful information in low-dose CT images, random forests complete classification. The experimental results show that compared with the dictionary learning method and the traditional interpolation method, the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of CT images obtained by this method are the highest, and the reconstructed images are the most robust and have a fast running speed and training speed.