Deformation-encoding Deep Learning Transformer for High-Frame-Rate Cardiac Cine MRI.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Processing

Background:

• Cardiac MRI (cMRI) is crucial for heart assessment.
• Increasing frame rate in cardiac cine MRI improves temporal resolution but can extend scan times or reduce spatial resolution.
• Deep learning offers potential solutions for enhancing image acquisition parameters.

Purpose of the Study:

• To develop and validate a transformer-based deep learning model for increasing cardiac cine MRI frame rates.
• To maintain spatial resolution and scan time while improving temporal information.
• To compare the model's performance against traditional interpolation methods.

Main Methods:

• A transformer-based deep learning model was trained on a large retrospective dataset (5840 patients) from multiple centers and vendors.
• The model's interpolation performance was evaluated using root mean square error (RMSE) against linear and bicubic methods.
• A prospective study assessed reader preference between actual and model-interpolated high-frame-rate cines (50 fps).

Main Results:

• The deep learning model generated artifact-free interpolated cardiac cine images.
• Model-based interpolation showed significantly lower RMSE compared to linear and bicubic methods in both internal and external testing.
• Reader studies indicated noninferiority, with a majority of readers showing "no preference" between actual and interpolated 50-fps cines.

Conclusions:

• A transformer-based deep learning model effectively increases cardiac cine frame rates.
• The developed model preserves spatial resolution and scan time, yielding image quality comparable to actual high frame rates.
• This deep learning approach holds promise for advancing cardiac MRI diagnostic capabilities.