SNRAware: Improved Deep Learning MRI Denoising with SNR Unit Training and G-factor Map Augmentation

2025-03-23

AI Summary

Existing MRI denoising methods suffer from poor generalizability and physically implausible noise modeling. Method: We propose SNRAware—a novel training paradigm featuring (i) an SNR unit module and a G-factor map–driven noise augmentation mechanism that explicitly encodes quantitative SNR and parallel imaging–induced noise distributions in reconstruction; and (ii) a Transformer-CNN dual-backbone architecture jointly optimized with SNR-aware loss and a multi-domain transfer evaluation framework. Results: On in-distribution testing, SNRAware achieves significant PSNR/SSIM improvements. Under out-of-distribution scenarios—including real-time cine and perfusion MRI—contrast-to-noise ratio (CNR) improves by 6.5× and 2.9×, respectively. Moreover, the model successfully transfers across neuro/spinal MRI sequences, field strengths (1.5T/3T), and anatomical regions, demonstrating markedly enhanced cross-sequence, cross-field-strength, and cross-anatomy generalizability.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

To develop and evaluate a new deep learning MR denoising method that leverages quantitative noise distribution information from the reconstruction process to improve denoising performance and generalization. This retrospective study trained 14 different transformer and convolutional models with two backbone architectures on a large dataset of 2,885,236 images from 96,605 cardiac retro-gated cine complex series acquired at 3T. The proposed training scheme, termed SNRAware, leverages knowledge of the MRI reconstruction process to improve denoising performance by simulating large, high quality, and diverse synthetic datasets, and providing quantitative information about the noise distribution to the model. In-distribution testing was performed on a hold-out dataset of 3000 samples with performance measured using PSNR and SSIM, with ablation comparison without the noise augmentation. Out-of-distribution tests were conducted on cardiac real-time cine, first-pass cardiac perfusion, and neuro and spine MRI, all acquired at 1.5T, to test model generalization across imaging sequences, dynamically changing contrast, different anatomies, and field strengths. The best model found in the in-distribution test generalized well to out-of-distribution samples, delivering 6.5x and 2.9x CNR improvement for real-time cine and perfusion imaging, respectively. Further, a model trained with 100% cardiac cine data generalized well to a T1 MPRAGE neuro 3D scan and T2 TSE spine MRI.

Problem

• Develops deep learning MRI denoising using SNR unit training
• Improves denoising performance with noise distribution information
• Enhances generalization across sequences and field strengths

Innovation

• Leverages SNR unit training for MRI denoising
• Uses G-factor map augmentation for noise simulation
• Applies deep learning on diverse synthetic datasets