Deep Learning-Based ECG-Free Cardiac Navigation for Multi-Dimensional and Motion-Resolved Continuous Magnetic...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Clinical assessment of cardiac vitality relies on time-continuous tomographic magnetic resonance imaging (MRI).
• Multi-contrast imaging protocols enhance diagnostic capabilities for pathological tissues.
• Current motion-resolved reconstruction often uses electrocardiograms (ECG) for navigation, adding workflow complexity.

Purpose of the Study:

• To develop and evaluate a sensor-free, deep learning-based navigation method for continuous cardiac MRI.
• To eliminate the need for external sensors like ECG and prior knowledge of heart rates.
• To enable efficient, simultaneous anatomic and functional imaging with multiple contrasts.

Main Methods:

• A deep learning classifier was trained to estimate R-wave timepoints directly from cardiac MRI data.
• The sensor-free navigation approach was evaluated on 3-D, in-vivo, free-breathing continuous cardiac MRI protocols.
• The method was tested with single and multiple imaging contrasts.

Main Results:

• The deep learning navigation achieved an accuracy of over 98% on unseen subjects.
• Image quality was comparable to state-of-the-art ECG-based reconstruction methods.
• The approach successfully enabled ECG-free, continuous cardiac scans.

Conclusions:

• A novel, sensor-free deep learning navigation method significantly improves continuous cardiac MRI workflows.
• This approach reduces reliance on external sensors and manual feature engineering.
• The method offers potential for integration into various continuous imaging sequences, utilizing imaging data for navigation.