Data-Driven Retrieval of Effective Point Spread Functions for Super-Resolution Optoacoustic Imaging

Abstract Localization-based super-resolution techniques have revolutionized biomedical imaging by surpassing classical diffraction limits. However, their performance is fundamentally constrained by distortions of the point spread function (PSF) induced by the system and the sample, which are particularly prominent in the case of spatially under-sampling. Here we introduce a data-driven effective point spread function retrieval (DEPR) method that directly learns continuous, field-dependent system responses from experimental point source datasets. Through statistical aggregation of thousands of targets and iterative self-supervised refinement, DEPR captures spatially variant imaging characteristics in situ without prior assumptions or external calibrations. When integrated into the localization optoacoustic tomography (LOT) pipeline, DEPR achieves accurate sub-pixel localization despite under-sampled conditions, thus enhancing resolution while reducing computational burden. We demonstrate its efficacy by in vivo imaging of the murine brain microvasculature using microparticle contrast agents in the first (NIR-I) and second (NIR-II) near-infrared windows, achieving significant improvement in data utilization efficiency and substantial reduction of gridded artifacts compared to conventional approaches. The method resolves vascular structures with ∼41 μm separation across a ∼3 mm imaging depth range. This framework addresses fundamental challenges shared across diverse localization-based imaging modalities, offering a robust and generalizable strategy for high-precision imaging in complex biological systems.