Beyond Sparsity: Weight Location and Network Context in Pruned MRI Reconstruction

2026-09-25

AI Summary

This study addresses the limitation of existing MRI reconstruction network pruning methods that focus solely on sparsity while neglecting weight positions and contextual information. We systematically investigate how weight retention strategies in U-Net and Vision Transformer (ViT) architectures affect reconstruction quality. Employing a controlled editing evaluation approach, we move beyond the conventional sparsity-centric perspective to reveal the critical roles of architecture-specific descriptors and surrounding trained weights. Our findings demonstrate that, at equivalent removal ratios, retaining high-resolution computational features yields superior performance. Furthermore, while pre-trained weight energy enhances reconstruction quality, mask rankings can reverse under varying contextual conditions. These results confirm that sparsity alone is an insufficient metric for effective pruning, thereby establishing a novel paradigm for model compression in MRI reconstruction.

Abstract

Pruning reduces the number of weights in magnetic resonance imaging (MRI) reconstruction networks. Equal sparsity, however, can retain weights with different computational roles and different compatibility with the trained network. We study these effects across 120 convolutional U-Net and vision transformer models using controlled edits evaluated before retraining. In U-Net, preserving high-resolution computation improves reconstruction at equal deletion counts across all tested sparsity levels. Equal-operation controls reveal additional sensitivity of the first convolution, which operation count alone cannot explain. In transformers, retained pretrained weight energy correlates with quality within sparsity levels, yet the same masks reverse their reconstruction ranking when the surrounding trained weights change. At 90\% sparsity, matching intermediate output scales largely removes a high-energy-mask penalty while retaining an advantage for the network's original mask. Thus sparsity alone does not characterize reconstruction quality: architecture-specific descriptors are informative, but mask quality can still depend on the surrounding trained network.

Problem

• MRI reconstruction
• network pruning
• sparsity
• weight location
• network context

Innovation

• Network Pruning
• MRI Reconstruction
• Weight Location
• Network Context
• Sparsity