MRI reconstruction risk predicted from intermediate image steps
MRI Reconstruction Risk Estimation from Inference Dynamics
Computer Vision and Pattern Recognition
Summary
MRI scans are made by building an image step-by-step, but not all scans are equally accurate. The authors study the images produced during this stepwise process to estimate how much error the final image might have. They find that features extracted from these intermediate images can predict the final reconstruction error well, even when tested on new data or different methods. This approach helps understand and anticipate the quality of MRI images before the process finishes.
What this means in practice
- •For medical imaging teams: Estimate MRI image quality early during reconstruction to flag potentially unreliable scans for review or repeated acquisition.
- •For mri system developers: Integrate trajectory-based risk predictors into MRI software for adaptive reconstruction strategies that adjust based on predicted errors.
Authors
Tamir Shor, Alex Bronstein
Abstract
Iterative MRI reconstructors produce intermediate estimates before returning a final image. We study these inference trajectories as a model-internal representation of realized reconstruction risk, measured by log-NMSE. Model-native states are projected to magnitude images, summarized by a fixed 16-dimensional descriptor, and mapped to risk using a source-trained ridge readout. Across three reconstruction backbones, trajectory features are strongly predictive of reconstruction risk on held-out source data. Under zero-shot target shift, their behavior is backbone-dependent, exhibiting both preserved positive association and readout reversal. Conditional analyses show that trajectory information can remain associated with error beyond an initialization-only predictor. A prefix analysis further shows that useful trajectory information can emerge before reconstruction completes. Repeated-mask and low-shot calibration studies indicate that trajectories primarily encode persistent case difficulty, while their orientation and calibration can change under domain shift.