Fig. 1. Graphical workflow of our proposed framework that includes a human in the loop.

An image is loaded and the computation backend is prepared. Next, the user selects a ROI that is representative for the complete data set. The noise level is automatically estimated to derive near optimal parameter initialization (see “Parameter estimation” section in Supplementary Figs. 16, 17). Next, the biological expert can optimize the parameter settings at a low latency visualization of the results according to their preferences (typically w.r.t. visualization and/or subsequent segmentation of specific objects). Once the optimal parameters for a specific algorithm are found, the complete data set is ready to be processed. The computationally intensive parts of the workflow are GPU accelerated and indicated with the Quasar logo³⁰.