NexOP framework optimizes low-field MRI scans, boosting SNR and speed
New deep-learning method cuts scan time while improving image quality on portable 0.3T MRI
Low-field MRI (e.g., 0.3T) offers portable, low-cost imaging but suffers from low signal-to-noise ratio (SNR). The standard fix—repeating acquisitions (NEX)—boosts SNR but doubles or triples scan times. Existing acceleration methods optimize k-space sampling but ignore the NEX dimension, using the same sampling mask for every repetition.
NexOP, developed by researchers from the Technion (Oved and Shimron), jointly optimizes sampling probabilities across both k-space and NEX dimensions under a fixed budget. A deep-learning reconstruction network then fuses multiple low-SNR measurements into a single high-SNR image. Tests on raw 0.3T brain scans show NexOP consistently beats prior methods, producing non-uniform sampling strategies that progressively reduce sampling across repetitions. This efficiently exploits the NEX dimension, enabling faster scans without sacrificing diagnostic quality. Theoretical analysis supports these results, positioning NexOP as a key enabler for affordable, accessible MRI.
- Jointly optimizes sampling density across k-space and NEX repetitions under a fixed sampling budget
- Deep-learning architecture reconstructs a high-SNR image from multiple low-SNR measurements
- Outperforms competing methods on raw 0.3T brain data across multiple acceleration factors and tissue contrasts
Why It Matters
Makes portable low-field MRI faster and higher-quality, advancing affordable healthcare in underserved regions.