Dictionary-based reconstruction of spatio-temporal 3D magnetic field images from a quantum diamond microscope

Understanding how electric currents flow in complex biological and electronic systems requires three-dimensional magnetic imaging with high spatial and temporal resolution. However, reconstructing current sources from measured magnetic fields is challenging in multilayer and dynamically evolving environments, where signal overlap and noise render conventional two-dimensional inversion approaches unreliable. Here, we demonstrate widefield nitrogen-vacancy (NV) center magnetometry for time-resolved three-dimensional magnetic source localization. Using a custom multilayer micro-coil platform that emulates localized, time-varying neuronal-like currents, we acquire magnetic field maps with micrometer-scale spatial and millisecond temporal resolution via per-pixel lock-in detection. Source localization is performed using a sparsity-promoting least absolute shrinkage and selection operator (LASSO) framework that incorporates experimentally measured magnetic field basis maps as structured spatial priors. Applied to a simulated neuronal dictionary of 6250 neurons, the method enables robust identification of sparse and correlated source configurations. These results establish a general framework for dynamic three-dimensional magnetic source localization in complex multilayer systems.