Fig. 3.

Fast4DReg is relatively resistant to noise. Twelve synthetic 3D video datasets with varying amounts of noise were created and corrected using Fast4DReg, either using maximum- or average-intensity projections. The drift tables were then applied to the original data to assess drift-correction accuracy. (A,B) Schematic illustrating the pipeline used to assess Fast4DReg sensitivity to noise. (C) Example of three noisy datasets used to assess Fast4DReg sensitivity to noise. (D,E) Fast4DReg drift-correction performance for three noisy datasets (C) was assessed using temporal color projections of a selected z-slice (middle of the cell) and kymographs (along the green dashed lines; dimensions, 25 μm × 25 frames). Note that Fast4DReg fails to register the images with an SNR of 1.2 when using maximum-intensity projections (E). (F) Fast4DReg drift-correction performance for the twelve noisy datasets was assessed using image-similarity metrics. The PSNR and PCC between the first and subsequent frames were calculated for each noise amount. For all panels, scale bars: 10 μm.