Untrained Network for Super-resolution for Non-contrast-enhanced Wholeheart MRI Acquired using Cardiac-triggered...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Three-dimensional (3D) whole-heart magnetic resonance imaging (MRI) is crucial for assessing cardiac anatomy in congenital and acquired heart diseases.
• Current 3D whole-heart MRI protocols are lengthy and prone to banding artifacts, limiting their clinical utility.

Purpose of the Study:

• To validate an unsupervised neural network (SRNN) for super-resolving low-resolution 3D whole-heart MRI images.
• The goal is to reduce MRI acquisition time and enhance image quality.

Main Methods:

• The SRNN was evaluated against bilinear interpolation, AdapSR, and high-resolution ground truth images.
• Qualitative and quantitative assessments, including signal-to-noise ratio (SNR), were performed on data from 30 pediatric patients.
• The SRNN's performance was also compared qualitatively against contrast-enhanced whole-heart images without ground truth.

Main Results:

• The SRNN demonstrated improved image quality, validated by both quantitative and qualitative analyses.
• The network successfully reduced acquisition time by enabling low-resolution scans.
• Results confirmed the SRNN's superiority in enhancing image quality compared to other methods.

Conclusions:

• The SRNN effectively reduces noise and eliminates artifacts in 3D whole-heart MRI.
• It preserves accurate anatomical structures while improving overall image quality.
• The SRNN offers a promising solution for faster and higher-quality cardiac MRI acquisition.