Exploring Siamese Networks in Self-Supervised Fast MRI Reconstruction

2025-01-18

AI Summary

This paper addresses fast MRI reconstruction without paired fully-sampled references. We propose SiamRecon, the first self-supervised Siamese framework for this task. Methodologically, we innovatively adapt the Siamese architecture to the k-space domain: physically grounded k-space data augmentation generates matched image pairs, while an alternating optimization scheme—inspired by the EM algorithm—mitigates representation collapse, enabling implicit self-supervision. Crucially, SiamRecon requires no fully-sampled ground-truth labels and learns strong reconstruction priors solely from undersampled k-space measurements. Evaluated on single-coil brain and multi-coil knee datasets, SiamRecon achieves state-of-the-art performance among self-supervised MRI reconstruction methods, significantly outperforming existing unsupervised and self-supervised approaches.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Search and Optimization: Sampling/Simulation-based Search
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Reconstructing MR images using deep neural networks from undersampled k-space data without using fully sampled training references offers significant value in practice, which is a self-supervised regression problem calling for effective prior knowledge and supervision. The Siamese architectures are motivated by the definition"invariance"and shows promising results in unsupervised visual representative learning. Building homologous transformed images and avoiding trivial solutions are two major challenges in Siamese-based self-supervised model. In this work, we explore Siamese architecture for MRI reconstruction in a self-supervised training fashion called SiamRecon. We show the proposed approach mimics an expectation maximization algorithm. The alternative optimization provide effective supervision signal and avoid collapse. The proposed SiamRecon achieves the state-of-the-art reconstruction accuracy in the field of self-supervised learning on both single-coil brain MRI and multi-coil knee MRI.

Problem

• Siamese architecture
• undersampled k-space data
• magnetic resonance imaging reconstruction

Innovation

• SiamRecon
• Siamese architecture
• self-supervised learning