Signal intensity informed multi-coil encoding operator for physics-guided deep learning reconstruction of highly...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• First-pass perfusion cardiac MRI (CMR) requires high acceleration, leading to signal intensity and SNR variations.
• Conventional reconstruction methods struggle with these dynamic changes, impacting image quality and generalization.
• Physics-guided deep learning (PG-DL) offers potential but needs adaptation for perfusion imaging challenges.

Purpose of the Study:

• To develop a novel PG-DL reconstruction strategy for accelerated simultaneous multislice (SMS) myocardial perfusion CMR.
• To introduce a signal intensity informed multi-coil (SIIM) encoding operator to address signal variations.
• To improve the generalizability of PG-DL reconstruction across different time-frames and SNR levels.

Main Methods:

• A SIIM encoding operator was developed to capture signal intensity/SNR variations over time.
• This SIIM operator was integrated into a PG-DL reconstruction framework.
• The proposed method was compared against conventional PG-DL, split slice-GRAPPA, LLR, L+S, and ROCK-SPIRiT reconstructions.

Main Results:

• The SIIM-based PG-DL method demonstrated superior performance in highly accelerated (3x SMS, 4x in-plane) free-breathing perfusion CMR.
• Significant noise reduction was observed compared to split slice-GRAPPA.
• Aliasing artifact reduction was superior to LLR, L+S, and conventional PG-DL methods.
• A reader study confirmed the proposed method's superior performance over all comparisons.

Conclusions:

• PG-DL reconstruction with the SIIM encoding operator enhances generalization across varying time-frames and SNRs in accelerated perfusion CMR.
• The SIIM operator effectively handles dynamic signal intensity changes inherent in first-pass perfusion imaging.
• This approach represents a significant advancement in reconstructing high-quality accelerated cardiac MRI.