Multi-Contrast MRI Image Synthesis Using Switchable Cycle-Consistent Generative Adversarial Networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Multi-contrast MRI images are crucial for tissue differentiation but may be unavailable due to scan limitations or artifacts.
• Deep learning, particularly CycleGAN, has advanced medical image synthesis for generating unacquired contrasts from available ones.
• Existing CycleGAN models require substantial computational resources due to dual image generators.

Purpose of the Study:

• To investigate the applicability and performance of switchable CycleGAN for cross-contrast synthesis of pediatric brain MRI images.
• To compare the efficacy of switchable CycleGAN against the original CycleGAN for this specific application.
• To develop a novel switchable CycleGAN model tailored for multi-contrast pediatric MRI synthesis.

Main Methods:

• Development of a switchable CycleGAN model for synthesizing T1-weighted and T2-weighted pediatric brain MRI contrasts.
• Utilizing a large dataset of publicly accessible pediatric structural brain MRI images.
• Conducting quantitative and qualitative experiments to compare switchable CycleGAN with the original CycleGAN.

Main Results:

• The switchable CycleGAN model successfully generated cross-contrast pediatric brain MRI images.
• Experimental results demonstrated that switchable CycleGAN outperformed the original CycleGAN in pediatric MRI synthesis.
• The switchable architecture proved effective and efficient for MRI contrast generation.

Conclusions:

• Switchable CycleGAN is a viable and superior method for cross-contrast pediatric brain MRI synthesis compared to the original CycleGAN.
• This approach offers a promising solution for generating essential MRI contrasts when data acquisition is challenging.
• The study validates the potential of switchable CycleGAN in advanced medical image synthesis applications.