ST-SRPerf: Continuous spatiotemporal representation for perfusion MRI super-resolution through neural ODE and...

Perfusion magnetic resonance imaging (MRI), an advanced imaging technique, captures dynamic blood flow in organs such as the brain, breast, and prostate by acquiring sequential MR images following contrast agent administration. This noninvasive approach provides crucial insights into tissue vascularity and permeability, essential for tumor analysis and understanding neurodegenerative diseases. However, achieving high spatial and temporal resolution in perfusion MRI remains challenging due to inherent trade-offs, often resulting in inaccuracies in perfusion parameter estimation. We introduced SpatioTemporal Super-Resolution framework for Perfusion MRI (ST-SRPerf), a novel spatiotemporal super-resolution framework that enhances both spatial and temporal resolution in perfusion MRI. ST-SRPerf integrates neural ordinary differential equations for continuous temporal trajectory calculation and implicit neural representations for continuous spatial implicit representation. This integration allows the transformation of low-resolution perfusion MRI sequences into high-resolution outputs across arbitrary spatial and temporal dimensions. Comprehensive experiments on brain dynamic contrast-enhanced MRI (DCE-MRI), dynamic susceptibility contrast MRI (DSC-MRI), and breast DCE-MRI datasets demonstrate that ST-SRPerf significantly outperforms existing methods in image quality and perfusion parameter estimation. The enhancements are evident across various experimental scenarios and downsampling factors, showcasing notable improvements in both spatial and temporal dimensions. ST-SRPerf represents a significant advancement in perfusion MRI super-resolution techniques, offering a promising tool for enhancing image resolution and diagnostic accuracy in clinical practice.