MRI image synthesis for fluid-attenuated inversion recovery and diffusion-weighted images with deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Magnetic Resonance Imaging (MRI) often requires multiple sequences, increasing scan time.
• Synthesizing different MRI contrasts from existing scans can enhance diagnostic information.
• Deep learning offers potential for generating synthetic MRI contrasts.

Purpose of the Study:

• To synthesize fluid-attenuated inversion recovery (FLAIR) and diffusion-weighted images (DWI) using a deep conditional adversarial network.
• To evaluate the performance of synthesizing FLAIR and DWI from T1- and T2-weighted MRI inputs.
• To assess the utility of composite MRI inputs for efficient multi-contrast image synthesis.

Main Methods:

• A deep conditional adversarial network, comprising a generator and discriminator, was employed.
• T1-weighted, T2-weighted, and composite MRI images were used as input data.
• DICOM images were converted to 8-bit RGB images, with specific channels assigned to T1 and T2 weighted data.

Main Results:

• The model using composite MRI input demonstrated superior performance for DWI synthesis (lowest rMAE, highest MI).
• For FLAIR synthesis, T2-weighted inputs yielded more accurate results than T1-weighted inputs.
• The composite input approach proved efficient for multi-contrast MRI synthesis.

Conclusions:

• The proposed framework effectively synthesizes FLAIR and DWI from T1/T2-weighted MRI, enhancing versatility and quality.
• Composite MRI inputs facilitate efficient multi-contrast image generation, potentially reducing the need for extra scans.
• This deep learning approach offers a promising method for improving MRI data acquisition and analysis.