Super-Resolution of Cardiac MR Cine Imaging using Conditional GANs and Unsupervised Transfer Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• High-resolution cardiac MRI is limited by long acquisition and breath-hold times.
• Standard 2D SSFP sequences yield anisotropic images, hindering analysis.
• Existing methods struggle with clinical applicability and data requirements.

Purpose of the Study:

• To develop a robust super-resolution algorithm for generating isotropic cardiac MR images.
• To improve the accuracy of downstream cardiac image analysis tasks.
• To enable high-quality isotropic imaging without extended acquisition protocols.

Main Methods:

• A novel adversarial learning super-resolution algorithm based on conditional GANs.
• Incorporation of an optical flow component to guide image synthesis.
• End-to-end unsupervised transfer learning for clinical applicability.

Main Results:

• Synthesized 3D isotropic, anatomically plausible cardiac MR images.
• Outperformed state-of-the-art methods qualitatively and quantitatively.
• Improved ventricle segmentation (DSC 0.95/0.81) and non-rigid registration (DSC 0.75 to 0.86).

Conclusions:

• The proposed SR method effectively generates isotropic cardiac MR images.
• Super-resolved images enhance accuracy in cardiac quantification and motion tracking.
• This approach offers a clinically viable solution for improved cardiac MRI analysis.