High Resolution TOF-MRA Using Compressed Sensing-based Deep Learning Image Reconstruction for the Visualization of...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Time-of-flight (TOF) MR angiography (MRA) is crucial for visualizing intracranial vasculature.
• Compressed sensing (CS) techniques accelerate MRA acquisition but can impact image quality.
• Deep learning (DL) offers potential for improved image reconstruction in MRA.

Purpose of the Study:

• To evaluate the efficacy of CS-based DL image reconstruction for lenticulostriate artery (LSA) visualization in TOF-MRA.
• To compare the image quality of CS-DL reconstructed MRA with conventional CS-SENSE reconstruction.

Main Methods:

• Five healthy volunteers underwent high-resolution TOF-MRA with full sampling (R-factor=1).
• Images were reconstructed using CS-DL and conventional CS-SENSE at R-factors of 2, 4, and 6.
• Quantitative assessments included visible LSA count, LSA length, and normalized mean squared error (NMSE).
• Qualitative assessments involved evaluating overall image quality and peripheral LSA visibility by two radiologists.

Main Results:

• CS-DL reconstruction showed a significantly higher number of visible LSAs at R-factors 4 and 6 compared to CS-SENSE.
• The length of depicted LSAs was significantly longer with CS-DL at R-factor 6.
• CS-DL yielded significantly lower NMSE values at R-factors 4 and 6.
• Qualitative analysis revealed significantly better overall image quality and peripheral LSA visibility with CS-DL across various R-factors.

Conclusions:

• CS-DL reconstruction effectively preserves and enhances image quality for LSA depiction in TOF-MRA, even at elevated R-factors.
• This method offers a promising alternative to conventional CS-SENSE for accelerated MRA acquisition.
• DL-based reconstruction holds potential for improving diagnostic accuracy in neurovascular imaging.