Abstract
Purpose To introduce SelExNet: a self-supervised framework for two-dimensional spatially selective excitation that jointly optimizes radiofrequency (RF) pulses and gradient waveforms, and extends to multi-channel transmission MRI. Methods Building on prior RF-only and joint RF-gradient optimization approaches, SelExNet couples neural RF and gradient generators with a differentiable Bloch simulator to enable self-supervised pulse optimization without requiring pre-designed target pulses. The framework jointly designs RF pulses and parameterized variable-density spiral gradient waveforms for both single- and multi-channel transmission, with patient-specific adaptation using measured, previously unseen B 0 $$ {B}_0 $$ and B 1 + $$ {B}_1^{+} $$ maps. Results Joint optimization of RF and gradients improved excitation fidelity compared to RF-only optimization. In phantom experiments with synthetic field maps, pretrained pulses showed distortions, whereas fine-tuned pulses restored geometry and uniformity. In vivo studies demonstrated anatomically precise excitation, with fine-tuning improving sharpness and reducing off-target signal. Conclusion The proposed framework enables joint RF-gradient design and extends self-supervised pulse optimization to multi-channel transmission MRI. SelExNet achieves high-fidelity, anatomically precise excitation and demonstrates robustness to field inhomogeneities, offering a scalable pathway for ultra-high field imaging.
Funding Information
  • Natural Sciences and Engineering Research Council of Canada (RGPIN‐2023‐03410)
  • Natural Sciences and Engineering Research Council of Canada (RGPIN‐2023‐04408)
  • Canadian Institutes of Health Research (PJT‐186038)

This publication has 48 references indexed in Scilit: