Super-Resolution Hyperpolarized ¹³C Imaging of Human Brain Using Patch-Based Algorithm

Spatial resolution of metabolic imaging with hyperpolarized ¹³C-labeled substrates is limited owing to the multidimensional nature of spectroscopic imaging and the transient characteristics of dissolution dynamic nuclear polarization. In this study, a patch-based algorithm (PA) is proposed to enhance spatial resolution of hyperpolarized ¹³C human brain images by exploiting compartmental information from the corresponding high-resolution ¹H images. PA was validated in simulation and phantom studies. Effects of signal-to-noise ratio, upsampling factor, segmentation, and slice thickness on reconstructing ¹³C images were evaluated in simulation. PA was further applied to low-resolution human brain metabolite maps of hyperpolarized [1-¹³C] pyruvate and [1-¹³C] lactate with 3 compartment segmentations (gray matter, white matter, and cerebrospinal fluid). The performance of PA was compared with other conventional interpolation methods (sinc, nearest-neighbor, bilinear, and spline interpolations). The simulation and the phantom tests showed that PA improved spatial resolution by up to 8 times and enhanced the image contrast without compromising quantification accuracy or losing the intracompartment signal inhomogeneity, even in the case of low signal-to-noise ratio or inaccurate segmentation. PA also improved spatial resolution and image contrast of human ¹³C brain images. Dynamic analysis showed consistent performance of the proposed method even with the signal decay along time. In conclusion, PA can enhance low-resolution hyperpolarized ¹³C images in terms of spatial resolution and contrast by using a priori knowledge from high-resolution ¹H magnetic resonance imaging while preserving quantification accuracy and intracompartment signal inhomogeneity.