Rapid reconstruction of highly undersampled, non-Cartesian real-time cine k-space data using a perceptual complex...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Compressed sensing (CS) MRI accelerates imaging but has long reconstruction times.
• Real-time cine MRI is crucial for patients with arrhythmia and dyspnea.
• Faster reconstruction is needed for clinical translation of accelerated MRI.

Purpose of the Study:

• Develop a neural network for faster real-time cine MRI reconstruction.
• Achieve reconstruction times under 1 minute per slice.
• Maintain image quality and accuracy in left ventricular (LV) functional parameters.

Main Methods:

• Introduced a perceptual complex neural network (PCNN) trained on complex-valued MRI data.
• Incorporated a perceptual loss term to minimize incoherent image details.
• Trained and tested PCNN on 40 patients' multi-slice, multi-phase cine MRI data.

Main Results:

• PCNN achieved reconstruction in 25 seconds per slice (80 frames), 166x faster than CS.
• PCNN demonstrated superior data fidelity (SSIM=0.88, NRMSE=0.014) compared to CS.
• LV ejection fraction measurements showed strong correlation (R²=0.92) and good agreement between PCNN and CS.

Conclusions:

• The proposed PCNN enables rapid reconstruction of non-Cartesian real-time cine MRI.
• PCNN offers a clinically viable alternative to GPU-accelerated CS reconstruction.
• This method preserves diagnostic accuracy for LV functional assessment.