mustGAN: multi-stream Generative Adversarial Networks for MR Image Synthesis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Multi-contrast MRI protocols offer rich morphological information for diagnosis.
• Scan time and patient motion limit the number and quality of MRI contrasts.
• Image synthesis can generate missing contrasts from existing high-quality MRI data.

Purpose of the Study:

• To develop an advanced multi-stream approach for synthesizing magnetic resonance imaging (MRI) contrasts.
• To improve the quality and quantity of diagnostic information in MRI by overcoming limitations of current protocols.

Main Methods:

• Proposed a novel multi-stream deep learning architecture for MRI synthesis.
• Integrated multiple one-to-one and a many-to-one synthesis streams.
• Employed an adaptive fusion block to combine complementary and shared feature maps for enhanced performance.

Main Results:

• The proposed multi-stream method demonstrated superior performance in quantitative and radiological assessments.
• Outperformed state-of-the-art one-to-one and many-to-one MRI synthesis techniques.
• Successfully synthesized T1-, T2-, PD-weighted, and FLAIR images with high fidelity.

Conclusions:

• The developed multi-stream approach effectively enhances MRI image synthesis.
• This method offers a promising solution for improving diagnostic accuracy in clinical settings.
• Advanced synthesis techniques can overcome practical limitations in multi-contrast MRI acquisition.