Synthesizing MRIs From CT Scans Using Deep Learning Techniques: A Comprehensive Review

Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) are useful imaging techniques used in clinical practice. While CT scans offer advantages in imaging speed, cost, and patient comfort compared to MRI, MRI provides superior image detail in soft tissue structures. This review investigates the application of deep learning algorithms to address this gap by converting CT scans into MRI representations. Research papers focusing on this conversion process were analyzed. The analysis reveals that supervised learning outperforms unsupervised methods in generating realistic images. However, supervised learning struggles with a lack of paired CT-MRI datasets, hindering its effectiveness. Conversely, unsupervised learning offers a solution by generating synthetic paired datasets that can be used for supervised learning to produce high-quality synthetic MRIs. Finally, this review highlights the need for increased availability of paired datasets to further improve MRI synthesis outcomes.