CoilDrop-MRI matches supervised MRI quality with self-supervised coil-dropout training

CoilDrop-MRI, a self-supervised reconstruction using coil-wise dropout, matches supervised quality without fully sampled training data. It outperforms existing self-supervised methods across multi-site, multi-field-strength (0.3T-3T) and multi-modality MRI.

• CoilDrop-MRI integrates coil dropout in both image-domain (SENSE) and k-space (SPIRiT) unrolled architectures, better leveraging inter-coil correlations.
• The method is extended to multi-shot, phase-corrected diffusion MRI, demonstrating versatility.
• It exhibits strong data efficiency and robust generalization across field strengths and imaging sequences.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.