Deep-Learning Based Super-Resolution Functional Ultrasound Imaging of Transient Brain-Wide Neurovascular Activity on a Microscopic Scale

TL;DR

This paper introduces SR-fUS, a deep learning method that enhances functional ultrasound imaging resolution to visualize transient neurovascular activity at a microscopic scale, validated through rat brain experiments.

Contribution

The study presents a novel deep learning-based super-resolution technique for functional ultrasound imaging, surpassing diffraction limits using ultrasound localization microscopy data.

Findings

Achieved 25-μm spatial and 0.2-s temporal resolution in brain imaging.

Successfully visualized transient hemodynamic responses in rat brains.

Validated super-resolution imaging accuracy against two-photon microscopy.

Abstract

Transient brain-wide neuroimaging on a microscopic scale is pivotal for brain research, yet current modalities face challenges in meeting such spatiotemporal requirements. Functional ultrasound (fUS) enables transient neurovascular imaging through red blood cell backscattering, but suffers from diffraction-limited spatial resolution. We hypothesize that deep learning-based super-resolution reconstruction can break through this limitation, introducing super-resolution functional ultrasound (SR-fUS) which leverages ultrasound localization microscopy (ULM) data to achieve super-resolution reconstruction of red blood cell dynamics. By incorporating red blood cell radial fluctuations with uncertainty-driven loss, SR-fUS enables mapping ultrasound Doppler frames to super-resolution blood flow images, achieving 25-{\mu}m spatial and 0.2-s temporal resolution. SR-fUS was applied to image transient hemodynamic responses induced by pain stimulation in rat brains. SR-fUS accuracy in cortical microvasculature during whisker stimulation was further validated by a comparative study with two-photon microscopy.