Real-time 4D MRI reconstruction using DVR-NeMF, a framework for dynamic volumetric reconstruction

In this study, four dynamic magnetic resonance imaging (MRI) sequences were first developed to collect data. On this basis, we presented DVR-NeMF, a neural magnetic field framework that enables real-time, high-dimensional (4D) MRI reconstruction from synchronized dynamic 2D image slices and physiological signals. By embedding spatiotemporal priors into an implicit representation of the imaging space, DVR-NeMF reconstructs temporally consistent 3D volumes over time with high anatomical fidelity. Comprehensive evaluations across a cardiovascular phantom, cardiac dynamic bio-simulators, and living human hearts, including external out-of-distribution validation on the Automated Cardiac Diagnosis Challenge (ACDC) dataset, demonstrated that DVR-NeMF outperforms both autoencoder- and generative adversarial network (GAN)-based baselines in terms of reconstruction accuracy and computational efficiency. Comparative analysis with paired cardiac ultrasound data in terms of key left ventricular function parameters further confirmed its reliability. This work offers a promising paradigm for extending MRI to dynamic, high-dimensional imaging, with potential for real-time functional assessment in clinical settings.