The analysis of the cerebral blood flow is an important diagnostic modality for the detection of pathologies such as strokes or aneurysms. Recent works have demonstrated that the cerebral blood flow of a patient can be estimated at the microscopic scale using ultrafast ultrasound localization microscopy (ULM) of intravenously injected microbubbles. This technique relies on the detection and tracking of these microbubbles across time frames, and requires accurate sub-pixel localization of microbubbles centers. Current state-of-the-art methods accomplish this by correlating the image of the microbubbles with a discrete representation of the point spread function (PSF), i.e., the imaging response of a single microbubble, and detecting local maxima via thresholding and local refinements. The goal of this project is to investigate methods for improving this sub-pixel local maxima detection. A first step will be to investigate deconvolution methods for the microbubble images via an inverse problem formulation with sparsity-promoting regularization to get an initial estimate of the microbubble locations. Sub-pixel refinement will then be performed using spline-based continuous representations of the images and the PSF. This continuous modeling will allow us to leverage gradient-based methods to estimate local maxima at a continuous level. A thorough study will be conducted to evaluate each step of the pipeline and to quantify the resolution limit of the adopted method, i.e., the minimum separation between microbubbles that can be resolved.
Reference: [1] Demené, C., Robin, J., Dizeux, A. et al. Transcranial ultrafast ultrasound localization microscopy of brain vasculature in patients. Nat Biomed Eng 5, 219–228 (2021).