Super-resolution and denoising of 4D flow MRI data using Physics-Informed Neural Network

This study introduces an innovative and advanced approach to significantly improve the spatial and temporal resolution of 4D Flow MRI data through the application of a physics-informed neural network (PINN). By synergizing physics-based principles with neural networks, the PINN addresses challenges related to variations and noise in MRI measurements, enhancing the reliability of velocity field predictions. Through rigorous evaluations, the PINN showcases exceptional accuracy in predicting velocity fields for both laminar and turbulent flows within a 2D stenosis model. Furthermore, when applied to Fontan 4D flow MRI data, the PINN effectively mitigates resolution and noise issues, underscoring its potential in enhancing the quality of 4D Flow MRI data. Although promising, this study acknowledges the presence of discrepancies in streamline predictions, particularly in complex patient-specific cases. As such, further refinement and investigation are crucial to optimize the PINN's performance and overcome the remaining limitations. This research represents a significant advancement in 4D Flow MRI, offering the prospect of more reliable and precise predictions in various clinical applications. By continually exploring and developing this novel approach, clinicians and researchers can elevate patient care and understanding of hemodynamic phenomena.