Model Adaptation for Image Reconstruction using Generalized Stein's Unbiased Risk Estimator

TL;DR

This paper proposes a Generalized Stein's Unbiased Risk Estimator (GSURE) loss to adapt deep learning image reconstruction models to new acquisition schemes, accounting for measurement noise and reducing overfitting.

Contribution

It introduces a novel GSURE-based loss function for model adaptation in image reconstruction, improving robustness to model mismatch and noise.

Findings

Enhanced reconstruction quality over mean-square error methods

Reduced overfitting in model adaptation

Effective rapid adaptation to new acquisition settings

Abstract

Deep learning image reconstruction algorithms often suffer from model mismatches when the acquisition scheme differs significantly from the forward model used during training. We introduce a Generalized Stein's Unbiased Risk Estimate (GSURE) loss metric to adapt the network to the measured k-space data and minimize model misfit impact. Unlike current methods that rely on the mean square error in kspace, the proposed metric accounts for noise in the measurements. This makes the approach less vulnerable to overfitting, thus offering improved reconstruction quality compared to schemes that rely on mean-square error. This approach may be useful to rapidly adapt pre-trained models to new acquisition settings (e.g., multi-site) and different contrasts than training data