SelExNet: A Self-Supervised Physics-Informed Framework for Multi-Channel Joint RF and Gradient Waveform Optimization...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Pulse Sequence Design
• Medical Physics

Background:

• Current MRI techniques often require pre-designed radiofrequency (RF) pulses and gradient waveforms.
• Optimizing RF pulses and gradient waveforms independently can limit excitation fidelity and robustness to field inhomogeneities.
• Self-supervised learning offers a promising avenue for optimizing MRI pulse sequences without predefined targets.

Purpose of the Study:

• To introduce SelExNet, a novel self-supervised framework for 2D spatially selective excitation.
• To enable joint optimization of RF pulses and gradient waveforms for enhanced MRI.
• To extend the framework to multi-channel transmission MRI.

Main Methods:

• SelExNet couples neural RF and gradient generators with a differentiable Bloch simulator.
• It enables self-supervised pulse optimization without requiring pre-designed target pulses.
• The framework jointly designs RF pulses and parameterized variable-density spiral gradient waveforms, adaptable to patient-specific B0 and B1+ maps.

Main Results:

• Joint RF-gradient optimization significantly improved excitation fidelity over RF-only optimization.
• Fine-tuning of pulses restored geometry and uniformity in phantom experiments with synthetic field maps.
• In vivo studies demonstrated anatomically precise excitation, with improved sharpness and reduced off-target signal.

Conclusions:

• SelExNet facilitates joint RF-gradient design and extends self-supervised optimization to multi-channel transmission MRI.
• The framework achieves high-fidelity, anatomically precise excitation robust to field inhomogeneities.
• SelExNet provides a scalable approach for ultra-high field MRI.