🤖 AI Summary
This work addresses the limitations of traditional MRI reconstruction methods, which discretize both image content and coil sensitivities, resulting in high memory consumption and poor structural awareness—particularly challenging in highly undersampled dynamic 3D cardiac imaging. The authors propose the first model-driven framework based on continuous neural fields, representing magnetization and coil sensitivities through tensor products of univariate neural fields to yield continuously differentiable representations. Integrated with a differentiable MRI physical forward model, this approach overcomes the constraints of discrete formulations. Evaluated under extreme undersampling conditions with acceleration factors up to 16, the method substantially outperforms existing model-driven techniques, effectively preserving fine anatomical details and temporal motion consistency.
📝 Abstract
Conventional MRI reconstruction methods treat images and coil sensitivities as discrete objects, leading to high memory demands and limited structural awareness that hamper effective regularization. These limitations hinder accurate reconstruction in highly undersampled scenarios, such as dynamic 3D cardiac magnetic resonance (CMR). We introduce a discretization-free, memory-efficient, model-based framework for dynamic 2D and 3D MRI reconstruction from highly undersampled data. We represent magnetization and coil sensitivities as continuous objects -- differentiable functions -- using tensor products of univariate neural fields. This tensor product structure enables scalable optimization in high-dimensional spatiotemporal settings. Our method outperforms state-of-the-art model-based reconstructions in dynamic 2D and 3D MR settings, preserving structure and motion even under aggressive undersampling (e.g., acceleration factor 16).