Extended Data Fig. 3: Characterizing properties of the pre-RL deconvolution.

Simulation validation of SACD and SOFI reconstructions under different imaging conditions. (a-d) From left to right: Wide-field image, 20-frame SACD reconstruction, 20-frame SOFI reconstruction followed by RL deconvolution, and 1000-frame SOFI reconstruction followed by RL deconvolution. Two molecules 160 nm apart were created at a 10 nm size pixel grid. We generated temporal blinking sequences of the fluorescence emitters (Methods), and convolved them with a 220 nm PSF subsequently. The resulting image stack was downsampled 6 times (pixel size 60 nm). The ‘Non-biased’ condition (b) represents that only Poisson noise was considered in the imaging process. The high (b), medium (c), and low (d) conditions denote the corresponding high, medium, and low SNRs and effective on/off contrast ratios (Methods). These conditions involved not only Poisson noise and Gaussian readout noise, but also the smooth fluctuating cytosol and out-of-focus background. In summary, with Poisson noise only, all SOFI (20-frame and 1000-frame) and SACD (20-frame) results achieved comparable performances in high quality. Upon decreases in SNR and contrast, qualities of 20-frame SOFI reconstructions continued to deteriorate while the 1000-frame one performed relatively stable. In comparison, SACD is insensitive to the SNR and contrast conditions and can effectively reconstruct high-fidelity SR images with minimum input. Experiments were repeated ten times independently with similar results. Scale bar: 100 nm.