[Paper Review] B-FIRE: Binning-Free Diffusion Implicit Neural Representation for Hyper-Accelerated Motion-Resolved MRI

Di Xu, Hengjie Liu+12 more|arXiv (Cornell University)|Jan 7, 2026

Advanced Neuroimaging Techniques and Applications0 citations

TL;DR

B-FIRE introduces a binning-free diffusion implicit neural representation framework for hyper-accelerated motion-resolved MRI, enabling instantaneous 3D abdominal anatomy recovery from undersampled non-Cartesian k-space data.

ABSTRACT

Accelerated dynamic volumetric magnetic resonance imaging (4DMRI) is essential for applications relying on motion resolution. Existing 4DMRI produces acceptable artifacts of averaged breathing phases, which can blur and misrepresent instantaneous dynamic information. Recovery of such information requires a new paradigm to reconstruct extremely undersampled non-Cartesian k-space data. We propose B-FIRE, a binning-free diffusion implicit neural representation framework for hyper-accelerated MR reconstruction capable of reflecting instantaneous 3D abdominal anatomy. B-FIRE employs a CNN-INR encoder-decoder backbone optimized using diffusion with a comprehensive loss that enforces image-domain fidelity and frequency-aware constraints. Motion binned image pairs were used as training references, while inference was performed on binning-free undersampled data. Experiments were conducted on a T1-weighted StarVIBE liver MRI cohort, with accelerations ranging from 8 spokes per frame (RV8) to RV1. B-FIRE was compared against direct NuFFT, GRASP-CS, and an unrolled CNN method. Reconstruction fidelity, motion trajectory consistency, and inference latency were evaluated.

Motivation & Objective

• Motivated by the need to recover instantaneous dynamic information in 4D MRI without artifacts from phase-averaged reconstructions.
• Develop a binning-free diffusion-implicit neural representation (INR) framework for motion-resolved MRI.
• Enable reconstruction from highly undersampled non-Cartesian k-space data with motion fidelity.

Proposed method

• Use a CNN-implicit neural representation (INR) encoder-decoder backbone.
• Train with diffusion-based loss that combines image-domain fidelity and frequency-aware constraints.
• Leverage motion-bin training references while performing inference on binning-free undersampled data.
• Apply the method to T1-weighted StarVIBE liver MRI data with accelerations RV8 to RV1.
• Compare against direct NuFFT, GRASP-CS, and an unrolled CNN baseline.

Experimental results

Research questions

• RQ1 Can binning-free diffusion INR reconstruct motion-resolved 4DMRI from extremely undersampled non-Cartesian k-space data?
• RQ2 Does B-FIRE preserve instantaneous motion trajectories better than traditional binning-based approaches?
• RQ3 How does reconstruction fidelity and inference latency of B-FIRE compare to standard methods across different acceleration factors?
• RQ4 Is the method effective on liver MRI data such as StarVIBE with varying undersampling patterns?

Key findings

• B-FIRE enables hyper-accelerated motion-resolved MRI reconstruction on a T1-weighted StarVIBE liver cohort.
• The framework uses a diffusion-augmented INR with an encoder-decoder backbone.
• Inference is performed on binning-free undersampled data and trained with motion-bin references.
• Comparisons include direct NuFFT, GRASP-CS, and an unrolled CNN baseline.
• The study evaluates reconstruction fidelity, motion trajectory consistency, and inference latency.