Title: A comprehensive review on MRI to CT and MRI to PET image synthesis using deep learning

Authors: M.S. Meharban; M.K. Sabu; T. Santhanakrishnan

Addresses: Department of Computer Application, Cochin University of Science and Technology, Kochi, Kerala, India ' Department of Computer Application, Cochin University of Science and Technology, Kochi, Kerala, India ' Defence Research and Development Organisation (DRDO), Naval Physical and Oceanographic Laboratory (NPOL), Kochi, India

Abstract: Image synthesis is the process of generating a synthetic image with desired qualities. Although CT and PET images are suffering from ionising radiation, MRI images are free from such radiation. Due to this fact, we need a system to generate synthetic CT and PET images from MRI images. The system will be helpful to avoid such ionising radiation from CT and PET and makes a better patient treatment workflow. This work reviewed various deep learning synthetic CT and synthetic PET generation methods. More than 75 papers were selected from PubMed and ScienceDirect databases from 2017 to 2021. Recently, CycleGAN variants have produced better results with no need for paired data. However, an effective evaluation measure was not available to evaluate the efficacy of the proposed works. Additional blind tests involving radiologists are required to evaluate the visual quality of the synthesised image.

Keywords: computed tomographic; CT; positron emission tomography; PET; magnetic resonance imaging; MRI; generative adversarial networks; GAN.

DOI: 10.1504/IJBET.2023.134586

International Journal of Biomedical Engineering and Technology, 2023 Vol.43 No.3, pp.207 - 232

Received: 19 May 2022
Accepted: 13 Nov 2022
Published online: 30 Oct 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article