Fig 1.
Comparison of frame averaging and rank filtering on VPD-450 labeled T cells within a lymph node.
(A) Schematic of various filtering methods. (B) 200 μm z-stacks created by (left-right) frame averaging (5-FA), frame averaging followed by spatial rank filtering with 0.7 μm (2 pixel) radius (5-FA-SR), spatiotemporal rank filtering with 0.7 μm (2 pixel) radius (5-FSTR), frame averaging with 6x as many frames (30-FA). Maximum contrast is set to 0.5 x the maximum pixel value in each image to show cells clearly against background. (C) 2D image of a single T cell 100 μm below the lymph node capsule
Fig 2.
Pixel intensity value distributions at high, low, and no signal regions of bead images.
High, low, and no signal regions of fluorescent beads masked to generate empirical distributions of intensity values. Red line indicates exponential fit to distribution tail.
Fig 3.
Performance of rank filters on bead images with different rank parameter values.
(A) Averaged signal to background of low average intensity pixels when rank filtered with various ranks. Error bars indicate standard deviation. (B) Mean values of bead pixels and background pixels used to generate (A). (C) Probability of attenuating high or low signal (false negative) or not attenuating background (false positive) based on binomial distribution cumulative distribution functions (N = 10) and the proportion of background pixels shown in Fig 2. (D) Mapping of intensity values corresponding to different average values compared to rank filter values. (E) Representative images of single sub-resolution beads and pixel linescans across them showing the failure to attenuate background when the rank parameter is set two high and the loss of dim pixels at the edges of the PSF when rank is set too low. Beads are 170 nm in diameter; pixels are 175 nm; theoretical Abbe resolution of the imaging system is 410 nm.
Fig 4.
Characterization of average, temporal rank, and spatiotemporal (3x3 pixel) rank estimators.
(A) Means of sampling distributions generated from dim signal pixels in fluorescent bead images. Error bars indicate standard deviation (B) Normalized mean squared error
Fig 5.
Comparison of spatiotemporal rank filter with frame averaging and frame averaging followed by Gaussian filtering.
Dendritic cell within the lymph node of an XCR1-Venus transgenic mouse. Digitally zoomed region shows that spatiotemporal rank filtering enables high-contrast visualization of a dendritic process without blurring its edges.