Deep learning-based noise reduction algorithm using patch group technique in cadmium zinc telluride fusion imaging system: A Monte Carlo simulation study

Abstract

Anatomical and functional fusion imaging systems using cadmium zinc telluride (CZT) are widely used in the field of medical imaging and nuclear-medicine. However, noise is inevitable in the CZT images, and reducing noise is thus crucial for accurate diagnosis of diseases. Among various available techniques for noise reduction in images, deep learning-based noise reduction algorithm using patch group is considered the most efficient method. Therefore, this study is focused on designing deep learning-based noise reduction algorithm using patch group and evaluating it using simulated CZT fusion images with X-ray and gamma ray. We used the Geant4 Application for Tomographic Emission (version 6), which is a Monte Carlo simulation tool, and the normalized noise power spectrum, coefficient of variation (COV), and contrast to noise ratio (CNR) were evaluated. Furthermore, the proposed deep learning-based noise reduction algorithm exhibited better values compared with the conventional noise reduction algorithms. In particular, the COV and CNR values of our algorithm were approximately 8.46 and 1.85 times better than that of the original CZT image. Thus, we successfully demonstrated the feasibility of the proposed deep learning-based noise reduction algorithm using the patch group technique in a CZT fusion imaging system. © 2020 Elsevier GmbH