Advanced noise reduction and feature enhancement in medical CT brain imaging using deep learning

Abstract

Computed Tomography (CT) is a powerful diagnostic tool used in modern medicine, which helps a doctor provide accurate diagnostic reports and plan the individual’s treatment. However, doing multiple CT scans on one individual poses health risks like cellular damage, deoxyribonucleic acid (DNA) mutation, tissue damage and lastly development of cancer over time. As a result, concerns regarding radiation exposure have led to the development of low-dose Computed Tomography (LDCT) techniques. However, LDCT suffers from low image quality with increased noise, causing the diagnostic to be less accurate. In this study, we provide a novel approach to enhance these LDCT images. Our work is leveraged upon deep learning techniques involving Generative Adversarial Networks (GAN). Thus, through rigorous experimentation on a diverse set of clinical CT images, we found the best model for enhancing LDCT images which holds details of anatomical structure and at the same time minimizes noise. This causes more accurate diagnostics, reducing the health hazard of patients. Thus, this approach paves the way for more accurate diagnostics while minimizing radiation exposure.