Statistical techniques for digital pre-processing of computed tomography medical images: A current review

Abstract

Digital pre-processing is a vital stage in the processing of the information contained in multilayer computed tomography images. The purpose of digital pre-processing is the minimization of the effect of image imperfections, which are associated with the noise and artifacts that affect the quality of the images during acquisition, storage, and/or transmission processes. Likewise, there is a wide variety of techniques in specialized literature that address the problem of imperfections, noise, and artifacts present in images. In this study, a comprehensive review of specialized literature on statistical techniques used in the pre-processing of digital images was conducted. The review summarizes updated information from 56 studies conducted over the last 5 years (2018–2022) on the main statistical techniques used for the digital processing of medical images obtained under different modalities, with a special focus on computed tomography. Additionally, the most often used statistical metrics for measuring the performance of pre-processing techniques in the field of medical imaging are described. The most often used pre-processing techniques in the field of medical imaging were found to be statistical filters based on median, neural networks, Gaussian filters based on deep learning, mean, and machine learning applied to multilayer computed tomography images and magnetic resonance images of the brain, abdomen, lungs, and heart, among other organs of the body.