Awaken, Then Scale: Tiny Adaptation for MRI Reconstruction

2026-09-25

AI Summary

This study addresses the inefficiency of few-shot cross-domain adaptation for MRI reconstruction models by proposing a sparse parameter adaptation method based on U-Net and ViT architectures. The core innovations include a sleep-wake mechanism that achieves low-cost model specialization by activating a small fraction of zero-initialized weights, and a dynamic capping strategy that calibrates correction magnitudes and constrains error bounds in new domains without requiring reference images. Experimental results demonstrate that, with only a 1% parameter budget, the proposed approach yields PSNR improvements of 0.43–0.51 dB, significantly outperforming baselines such as LoRA. These findings indicate that the method effectively enhances the cross-domain generalization capability of MRI reconstruction models while maintaining high parameter efficiency.

Abstract

Dormant Awakening (DA) specializes a magnetic resonance imaging (MRI) reconstruction model by fitting a small set of zero weights to one labeled slice. Across fourteen U-shaped convolutional network (U-Net) and vision transformer (ViT) sources, fitting 1,004 or 10,560 weights gives mean peak signal-to-noise ratio (PSNR) gains of .430 and .509 dB. We analyze how scaling the fitted output correction changes evaluation PSNR, then use the calibration correction size to cap changes on new images. The cap bounds changes in root mean squared error (RMSE) relative to the source without an evaluation reference. At 10\% adaptation budget, we compare fixed halving and dynamic capping for DA, unrestricted sparse adaptation and low-rank adaptation (LoRA). The controls increase both measured calibration-evaluation correlations across all six tested architecture-adapter settings. Experiments concern one constructed fastMRI-to-M4Raw shift with fixed source checkpoints and participant panels.

Problem

• MRI reconstruction
• model adaptation
• dormant awakening
• parameter-efficient fine-tuning
• domain shift

Innovation

• Dormant Awakening
• MRI Reconstruction
• Tiny Adaptation
• Low-Rank Adaptation
• Dynamic Capping