A neural network for long-term super-resolution imaging of live cells with reliable confidence quantification

Super-resolution (SR) neural networks transform low-resolution optical microscopy images into SR images. Application of single-image SR (SISR) methods to long-term imaging has not exploited the temporal dependencies between neighboring frames and has been subject to inference uncertainty that is difficult to quantify. Here, by building a large-scale fluorescence microscopy dataset and evaluating the propagation and alignment components of neural network models, we devise a deformable phase-space alignment (DPA) time-lapse image SR (TISR) neural network. DPA-TISR adaptively enhances the cross-frame alignment in the phase domain and outperforms existing state-of-the-art SISR and TISR models. We also develop Bayesian DPA-TISR and design an expected calibration error minimization framework that reliably infers inference confidence. We demonstrate multicolor live-cell SR imaging for more than 10,000 time points of various biological specimens with high fidelity, temporal consistency and accurate confidence quantification. A neural network model improves time-lapse super-resolution imaging of live cells.