A deep learning classification task for accurate brain navigation during functional ultrasound imaging

Functional ultrasound imaging is a breakthrough technology for imaging brain activity at high spatiotemporal resolution. As it monitors hemodynamic activity, the resulting images are a non-standard representation of local anatomy. This leads to difficulties when determining the exact anatomical location of the recorded image, which is necessary for correctly interpreting the data. Here we propose a convolutional neural network-based framework for accurately navigating the brain during functional ultrasound imaging solely based on vascular landmarks. Our approach uses an image classification task to identify a suitable set of reference positions, from which the anatomical position of an image can be inferred with a precision of 102 ± 98 μm. Further analysis revealed that the predictions are driven by deep brain areas. The robustness of our approach was validated using an ischemic stroke model. It confirms that functional ultrasound imaging information is sufficient for positioning even when local blood flow is disrupted, as observed in many brain pathologies. ### Competing Interest Statement A.U. is the founder and a shareholder of AUTC company commercializing functional ultrasound imaging solutions for preclinical and clinical research.