Back to Physics: Operator-Guided Generative Paths for SMS MRI Reconstruction

Simultaneous multi-slice (SMS) imaging with in-plane undersampling enables highly accelerated MRI but yields a strongly coupled inverse problem with deterministic inter-slice interference and missing k-space data. Most diffusion-based reconstructions are formulated around Gaussian-noise corruption and rely on additional consistency steps to incorporate SMS physics, which can be mismatched to the operator-governed degradations in SMS acquisition. We propose an operator-guided framework that models the degradation trajectory using known acquisition operators and inverts this process via deterministic updates. Within this framework, we introduce an operator-conditional dual-stream interaction network (OCDI-Net) that explicitly disentangles target-slice content from inter-slice interference and predicts structured degradations for operator-aligned inversion, and we instantiate reconstruction as a two-stage chained inference procedure that performs SMS slice separation followed by in-plane completion. Experiments on fastMRI brain data and prospectively acquired in vivo diffusion MRI data demonstrate improved fidelity and reduced slice leakage over conventional and learning-based SMS reconstructions.

💡 Research Summary

Simultaneous multi‑slice (SMS) MRI accelerates data acquisition by exciting several slices within a single readout, but the resulting measurements contain deterministic slice‑to‑slice interference (generated by CAIPI phase modulation) and, when combined with in‑plane undersampling, a severe loss of k‑space data. Traditional linear reconstructions (SENSE, GRAPPA) explicitly model these operators but struggle to recover high‑frequency details under high multiband (MB) factors and strong in‑plane acceleration, leading to slice leakage. Recent diffusion‑based methods treat the problem as Gaussian‑noise denoising and enforce the SMS physics only through external data‑consistency projections; this creates a mismatch because the actual degradation is deterministic, not stochastic, and the projection step cannot fully correct structured aliasing.

The authors propose an “operator‑guided” generative framework that replaces the stochastic noise model with a deterministic degradation trajectory defined by the known acquisition operators. For each reconstruction stage they define an operator AΩ (Ω∈{M,U}) – A_M models the CAIPI‑induced superposition for a target slice, while A_U is the Cartesian undersampling mask. The clean k‑space k* and the physically consistent corrupted state yΩ = AΩ(k*) are linked by a monotonic schedule αt∈