Keywords: Real-time (RT) steady-state free-precession (SSFP), data synthesizing, cardiac phase regression.
TL;DR: cardiac phase regression from real time 3D+t images
Abstract: Cine steady-state free-precession (SSFP) is the backbone of cardiac MRI, providing visualization of cardiac structure and function over the cardiac cycle, but requires concurrent ECG-gating to combine k-space data over multiple heart beats. However, cine SSFP is limited by a number of factors including arrhythmia, where beat-to-beat variability causes image artifacts. Real-time (RT) SSFP and recent innovations in image reconstruction provides a new potential alternative, capable of acquiring images without averaging over multiple heart beats. However, analysis of cardiac function from this image data can be complex, requiring retrospective analysis of function over multiple cardiac cycles and slices. We propose a deep learning regression method to facilitate cardiac phase detection, leveraging synthetic training approach from historical cine SSFP image data, and evaluate the effectiveness of this approach for detecting cardiac phase on RT SSFP images, manually labeled by expert readers. This combined approach using RT SSFP may have multiple potential advantages over traditional cine SSFP for evaluating cardiac function in patients with arrhythmia or difficulty tolerating long breath holds.