Self-Supervised Noise Adaptive MRI Denoising via Repetition to Repetition (Rep2Rep) Learning

Purpose: This work proposes a novel self-supervised noise-adaptive image denoising framework, called Repetition to Repetition (Rep2Rep) learning, for low-field (<1T) MRI applications. Methods: Rep2Rep learning extends the Noise2Noise framework by training a neural network on two repeated MRI acquisitions, using one repetition as input and another as target, without requiring ground-truth data. It incorporates noise-adaptive training, enabling denoising generalization across varying noise levels and flexible inference with any number of repetitions. Performance was evaluated on both synthetic noisy brain MRI and 0.55T prostate MRI data, and compared against supervised learning and Monte Carlo Stein's Unbiased Risk Estimator (MC-SURE). Results: Rep2Rep learning outperforms MC-SURE on both synthetic and 0.55T MRI datasets. On synthetic brain data, it achieved denoising quality comparable to supervised learning and surpassed MC-SURE, particularly in preserving structural details and reducing residual noise. On the 0.55T prostate MRI dataset, a reader study showed radiologists preferred Rep2Rep-denoised 2-average images over 8-average noisy images. Rep2Rep demonstrated robustness to noise-level discrepancies between training and inference, supporting its practical implementation. Conclusion: Rep2Rep learning offers an effective self-supervised denoising for low-field MRI by leveraging routinely acquired multi-repetition data. Its noise-adaptivity enables generalization to different SNR regimes without clean reference images. This makes Rep2Rep learning a promising tool for improving image quality and scan efficiency in low-field MRI.

B1+ mapping near metallic implants using turbo spin echo pulse sequences

Purpose: To propose a B1+ mapping technique for imaging of body parts containing metal hardware, based on magnitude images acquired with turbo spin echo (TSE) pulse sequences. Theory and Methods: To encode the underlying B1+, multiple (two to four) TSE image sets with various excitation and refocusing flip angles were acquired. To this end, the acquired signal intensities were matched to a database of simulated signals which was generated by solving the Bloch equations taking into account the exact sequence parameters. The retrieved B1+ values were validated against gradient-recalled and spin echo dual angle methods, as well as a vendor-provided turboFLASH-based mapping sequence, in gel phantoms and human subjects without and with metal implants. Results: In the absence of metal, phantom experiments demonstrated excellent agreement between the proposed technique using three or four flip angle sets and reference dual angle methods. In human subjects without metal implants, the proposed technique with three or four flip angle sets showed excellent correlation with the spin echo dual angle method. In the presence of metal, both phantoms and human subjects revealed a narrow range of B1+ estimation with the reference techniques, whereas the proposed technique successfully resolved B1+ near the metal. In select cases, the technique was implemented in conjunction with multispectral metal artifact reduction sequences and successfully applied for B1+ shimming. Conclusion: The proposed technique enables resolution of B1+ values in regions near metal hardware, overcoming susceptibility-related and narrow-range limitations of standard mapping techniques.

fastMRI: An Open Dataset and Benchmarks for Accelerated MRI

Accelerating Magnetic Resonance Imaging (MRI) by taking fewer measurements has the potential to reduce medical costs, minimize stress to patients and make MRI possible in applications where it is currently prohibitively slow or expensive. We introduce the fastMRI dataset, a large-scale collection of both raw MR measurements and clinical MR images, that can be used for training and evaluation of machine-learning approaches to MR image reconstruction. By introducing standardized evaluation criteria and a freely-accessible dataset, our goal is to help the community make rapid advances in the state of the art for MR image reconstruction. We also provide a self-contained introduction to MRI for machine learning researchers with no medical imaging background.