Effect of MR head coil geometry on deep-learning-based MR image reconstruction

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Traditional MRI reconstruction methods face limitations with geometric coil constraints.
• Deep learning (DL) offers a novel approach to image reconstruction by learning from data.
• Coil overlap in parallel imaging can degrade image quality and reconstruction performance.

Purpose of the Study:

• To evaluate if deep learning (DL) based MRI reconstruction can relax geometric coil constraints compared to traditional methods.
• To assess the impact of increased coil overlap on DL and non-DL reconstruction techniques.
• To compare the performance of DL reconstruction against conjugate gradient SENSE (CG-SENSE) and L1-wavelet compressed sensing (CS).

Main Methods:

• Two sets of head coil geometries (8-channel and 32-channel) were used.
• A DL reconstruction model was developed and compared with CG-SENSE and CS.
• Quantitative metrics and visual assessment were employed to evaluate performance under varying coil overlap.

Main Results:

• DL models significantly outperformed CG-SENSE and CS across all tested scenarios (p < 0.001).
• Increased coil overlap led to performance degradation for all methods, most notably CG-SENSE.
• DL reconstructions demonstrated superior robustness to coil overlap and signal-to-noise ratio (SNR) variations, with minimal performance changes.

Conclusions:

• Deep learning (DL) based MRI reconstruction yields higher quality and more robust images than traditional methods.
• The findings suggest that geometric coil design constraints can be relaxed when employing DL reconstruction techniques.
• DL enables more flexible and potentially simplified coil array designs in future MRI systems.