Graph-based modeling of optical system enables adaptive optics on dynamic samples with self-calibration

Fluorescence microscopy is fundamentally limited by aberrations that degrade resolution and image quality. While adaptive optics can compensate for these distortions, existing approaches face significant limitations: sensor-based methods require complex hardware that increases system complexity and cost, while computational methods rely on idealized models that lead to suboptimal correction when applied to real systems with inevitable imperfections. Moreover, existing computational methods assume static samples, limiting their applicability to live imaging where specimen motion and biological activity occur during acquisition. Here we introduce GRAPHYCS, a computational framework that bridges the gap between models and physical systems through differentiable graph-based modeling with automatic self-calibration and dynamic sample compatibility. In simulations, GRAPHYCS achieves a wavefront RMS error of 0.0252 μm compared to 0.2367 μm for the analytic phase-diversity wavefront sensing method under system non-idealities, and delivers superior aberration-corrected image quality (PSNR: 27.67 dB vs. 23.68 dB). In real-world microscopy experiments, GRAPHYCS demonstrates enhanced sample-induced aberration correction with PSNR of 21.35 dB compared to 19.68 dB for analytic phase-diversity, SSIM of 0.7713 vs. 0.5966, and PCC of 0.9493 vs. 0.8948, while effectively handling spatially varying aberrations across fields of view exceeding 1 mm². Furthermore, GRAPHYCS enables aberration correction in dynamic biological samples, successfully performing simultaneous wavefront sensing and neuronal activity detection in live larval zebrafish brain imaging where conventional phase-diversity methods fail. Overall, GRAPHYCS enables high-resolution imaging across extended regions and in dynamic biological samples without additional hardware complexity, providing a practical solution for aberration correction. Datasets for paper titled " Graph-based modeling of optical system enables adaptive optics on dynamic samples with self-calibration" Synthetic wide-field microscopy data File: Figure2_Simulation.zip Diversity_Images_Ideal.tif Diversity_Images_NonIdeal.tif GT_Object_Image.tif appliedCoeff.txt: applied Zernike coefficients used to generate a set of diversity images Experimental data: System and sample-induced aberrations (wide-field imaging) File: Figure3_Experimental_widefield.zip Diversity_Images_SystemAberration_Lymph.tif Diversity_Images_SampleAberration_Pancreas.tif appliedCoeff.txt: applied Zernike coefficients used to generate a set of diversity images Experimental data: Spatially varying aberration over large field of view (wide-field imaging) File: Figure4_SpatiallyVarying_widefield.zip Diversity_Images_SampleAberration_Pancreas_LargeFoV.tif appliedCoeff.txt: applied Zernike coefficients used to generate a set of diversity images Experimental data: Non-uniform illumination profile for wide-field imaging data File: IlluminationProfile.zip Illumination_Profile_centeralFoV.tif Illumination_Profile_LargeFoV.tif Experimental data: Sample-induced aberration (light-sheet imaging) File: Figure5_Experimental_lightsheet.zip Diversity_Images_SampleAberration_Zebrafish.tif appliedCoeff.txt: applied Zernike coefficients used to generate a set of diversity images Experimental data: Spatially varying aberration over large field of view (light-sheet imaging) File: Figure5_SpatiallyVarying_lightsheet.zip Diversity_Images_SampleAberration_Zebrafish_LargeFoV.tif appliedCoeff.txt: applied Zernike coefficients used to generate a set of diversity images