Next-Acceleration-Scale Prediction for Autoregressive MRI Reconstruction
For medical imaging researchers, this work addresses the critical issue of high-acceleration MRI reconstruction, offering a method that preserves high-frequency anatomy and leverages LLM post-training techniques, though it is an incremental improvement over existing autoregressive models.
The paper tackles the problem of MRI reconstruction from highly undersampled measurements, where continuous predictors produce blurry results. By framing reconstruction as autoregressive next-acceleration-scale prediction in a discrete latent space and introducing on-policy privileged information distillation, they achieve sharp reconstructions even under extreme undersampling, with improved performance on the fastMRI benchmark.
MRI reconstruction is an inherently ill-posed inverse problem, since incomplete measurements admit many plausible solutions. This ambiguity becomes more severe under high acceleration, where pixel-domain continuous predictors tend to average over feasible reconstructions and suppress high-frequency anatomy. We address this limitation by moving reconstruction to discrete multi-scale latent space and posing it as autoregressive next-acceleration-scale prediction. Leveraging discrete priors proven effective in visual autoregressive modeling, our method restricts the solution to compact sequences of codebook tokens, enabling sharp reconstructions even from extremely sparse measurements. This discrete autoregressive formulation also aligns naturally with modern large language model post-training techniques. Building on this observation, we introduce on-policy privileged information distillation for visual autoregressive modeling, where a teacher is provided training only privileged context that is unavailable at inference, in our case fully sampled acquisitions, and supervises a student trained on its own rollouts, leading to consistent reconstruction gains. Through extensive experiments on the fastMRI benchmark, we show that our approach delivers improved reconstruction performance across diverse sampling patterns under extreme undersampling. Project website is \hyperlink{https://github.com/yilmazkorkmaz1/discrete-mri-reconstruction-opd}{here}.