| Summary: | Introduction: Cardiovascular diseases, a leading cause of worldwide morbidity and mortality, are increasingly being managed using state-of-the-art imaging technologies. One such technology, non-contrast computed tomography angiography (CTA), has become a critical tool in this endeavor. However, despite these advancements, challenges such as computational cost and limited spatiotemporal resolution remain, which prevent the full exploitation of the capabilities of these imaging techniques. Recent strides in machine learning, specifically deep learning methodologies, offer promising solutions to these limitations. In this systematic review, we explore the advancements in deep learning (DL) for generating synthetic contrast-enhanced CT images, focusing on techniques that either create or augment contrast without the use of iodinated contrast media. Methods: Utilizing PRISMA guidelines, our methodology involved a thorough search through major databases, selecting studies based on specific inclusion and exclusion criteria related to artificial intelligence’s role in enhancing or mimicking contrast in CT scans. Results: Our findings, derived from 29 articles, highlight significant progress in this field, particularly in non-contrast cardiovascular CT imaging, where DL models like generative adversarial networks (GAN) have shown promise. However, challenges such as image alignment and the need for high-quality paired datasets remain. Conclusion: The review underscores the potential of DL in medical imaging, suggesting a promising future for reducing contrast doses in clinical settings, with further research needed to optimize this innovative approach. © 2024 Universiti Putra Malaysia Press. All rights reserved.
|
|---|