A Review of GAN-Based Brain Image Synthesis Using MRI and CT for Deep Learning Applications-R1

Abstract

The generation of synthetic medical images, particularly for brain imaging using magnetic resonance imaging (MRI) and computed tomography (CT), has gained significant attention due to its potential to enhance diagnostic accuracy, reduce healthcare costs, and minimize patient exposure to radiation. Generative Artificial Intelligence (AI) models, especially Generative Adversarial Networks (GANs), have demonstrated exceptional promise in addressing key challenges in synthetic image generation. This review examines the role of GANs in producing synthetic brain images from magnetic resonance imaging (MRI) and computed tomography (CT) data, emphasizing their ability to generate realistic and diverse datasets for the training of advanced machine learning algorithms. Particular attention is given to issues such as the necessity for large, well-annotated datasets, the impact of paired versus unpaired data, dataset size, and the effectiveness of various GAN architectures and other deep learning (DL) techniques in brain modality translation. Furthermore, we compare the performance of different models using widely adopted metrics, including mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM), based on literature published between 2017 and 2023.

Article Content

Keywords

Article Info