1 Laboratory of Applied Computational Imaging, Centre Énergie Matériaux Télécommunications, Institut National de la Recherche Scientifique, Université du Québec, 1650 Boulevard Lionel-Boulet, Varennes, Québec J3X1P7 Canada.
2 Department of Physics, Concordia University, 7141 Sherbrooke St. W., Montreal, Québec H4B1R6 Canada.
3 Department of Biology, Concordia University, 7141 Sherbrooke St. W., Montreal, Québec H4B1R6 Canada.

Real-time dynamic imaging of microbubbles is crucial for understanding their microscale biophysical interactions and advancing ultrasound therapy. Despite progress in time-resolved optical imaging, existing techniques still face trade-offs between acquisition speed, spatial resolution, affordability, and system complexity. Here, we introduce compressed optical-streaking dark-field ultrahigh-speed microscopy (COSDUM), a compact imaging platform that synergistically combines compressed sensing, streak imaging, dark-field microscopy, and deep learning. COSDUM compressively records megahertz acoustic microbubble dynamics over a wide field of view in a snapshot and reconstructs spatially resolved dynamics using a convolutional neural network-based algorithm. Using COSDUM, we captured stable cavitation, nonlinear oscillations, post-excitation free oscillations, and inertial collapse across microbubbles whose radii range from 0.5 to 2.1 µm. Applying COSDUM to microbubble-cell interaction in whole blood, we observed, for the first time, interplay between vibrating microbubbles and blood cells, including microbubble-driven platelet dynamics and highly asymmetric microbubble deformation and conformation around an adjacent red blood cell.

Supplementary information: The online version contains supplementary material available at 10.1186/s43074-026-00232-8.