From Compressed-Sensing to Deep Learning MR: Comparative Biventricular Cardiac Function Analysis in a Patient Cohort

Area of Science:

• Cardiovascular Imaging
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Conventional cine (Conv-cine) MRI faces challenges in patients with breath-hold difficulties.
• Compressed sensing (CS) cine offers solutions but has long reconstruction times.
• Artificial intelligence (AI) cine shows potential for rapid cardiac imaging.

Purpose of the Study:

• To compare CS-cine and AI-cine against Conv-cine for biventricular function, image quality, and reconstruction time.
• Evaluate quantitative biventricular parameters and subjective image quality.
• Assess the clinical utility of novel cine techniques.

Main Methods:

• Prospective study involving 70 patients (39±15 years, 54.3% male) using 3T MRI.
• Acquisition via balanced steady-state free precession gradient echo sequences.
• Independent assessment of biventricular parameters and image quality by radiologists; scan and reconstruction times recorded.

Main Results:

• CS-cine and AI-cine showed no significant differences in biventricular function compared to Conv-cine (P > 0.05).
• Both CS-cine and AI-cine significantly reduced scan time compared to Conv-cine.
• AI-cine demonstrated similar image quality to Conv-cine, while CS-cine showed lower scores; AI-cine drastically reduced reconstruction time compared to CS-cine.

Conclusions:

• CS-cine and AI-cine enable whole-heart cardiac cine imaging within a single breath-hold.
• Both techniques can supplement conventional cine MRI for biventricular function assessment.
• AI-cine and CS-cine are beneficial for patients with breath-holding limitations.