Deep Learning-Based Generation of DSC MRI Parameter Maps Using Dynamic Contrast-Enhanced MRI Data

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuroimaging

Background:

• Dynamic susceptibility contrast (DSC) MRI and dynamic contrast-enhanced (DCE) MRI provide valuable perfusion parameters for clinical diagnosis and research.
• Current protocols often require two gadolinium contrast doses when using both DSC and DCE MRI in the same session.
• Developing methods to obtain DSC-derived maps from DCE MRI data can streamline the imaging process and reduce contrast agent administration.

Purpose of the Study:

• To develop and validate deep learning-based methods for synthesizing DSC-derived parameter maps from DCE MRI data.
• To enable the acquisition of both DSC and DCE MRI parameter maps using a single contrast agent dose.

Main Methods:

• A conditional generative adversarial network (cGAN) was designed and trained using a dataset of 64 participants (including brain tumor patients).
• Reference DSC MRI parameter maps were acquired following DCE MRI.
• The performance of the cGAN was evaluated by comparing synthetic DSC maps with ground truth DSC maps using linear regression and Bland-Altman analysis.

Main Results:

• The cGAN successfully synthesized realistic DSC parameter maps from DCE MRI data.
• Synthesized parameters showed similar distributions to ground truth values in healthy controls.
• In brain tumor patients, synthesized parameters in the tumor region demonstrated a strong linear correlation with ground truth values.
• DCE-derived DSC maps visualized regions obscured by susceptibility artifacts in conventional DSC MRI.

Conclusions:

• Deep learning enables the synthesis of DSC-derived parameter maps from DCE MRI data, even in artifact-prone areas.
• This approach holds significant potential for obtaining comprehensive perfusion information (both DSC and DCE) with a single contrast agent injection.
• The findings suggest a more efficient and patient-friendly approach to advanced neuroimaging.