Beyond Diffusion: Consistency Models for One‑Step, High‑Fidelity MRI Reconstruction

Mary-Brenda Akoda, Chen Qin

Keywords: MRI reconstruction, diffusion models, consistency models, inverse problems, cardiac MRI

Abstract: Magnetic resonance imaging (MRI) provides excellent soft-tissue contrast, but suffers from long acquisition times, limiting throughput and increasing patient discomfort. Recent diffusion-based generative models have achieved state-of-the-art reconstruction fidelity for undersampled MRI, but typically require hundreds to thousands of neural function evaluations (NFEs), which severely limits their practicality in time-sensitive clinical settings. We introduce C-MORE (Consistency-Model-based One-step REconstruction for MRI), to our knowledge, the first one-step consistency-model-based framework for accelerated MRI reconstruction. C-MORE investigates an unconditional one-step consistency model prior and then in one NFE, tackles the inverse problem by leveraging measurement-guided encoding and tunable physics-based refinement updates, thus eliminating multistep diffusion sampling while still providing a controllable trade-off between runtime and reconstruction fidelity. On the CMR$\times$Recon dataset comprising multiple cardiac contrasts and both single- and multi-coil acquisitions, C-MORE achieves higher reconstruction quality than recent state-of-the-art diffusion-based samplers, while reconstructing each image in $0.18-0.52$ seconds on average ($\approx22-193\times$ speed-up in just 1 NFE). These results establish C-MORE as a practical blueprint for real-time high-fidelity MRI reconstruction across multiple contrasts and acquisition settings.

Primary Subject Area: Image Acquisition and Reconstruction

Secondary Subject Area: Generative Models

Registration Requirement: Yes

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 162