Correcting Artifacts in Single Molecule Localization Microscopy Analysis Arising from Pixel Quantum Efficiency Differences in sCMOS Cameras

Abstract

Optimal analysis of single molecule localization microscopy (SMLM) data acquired with a scientific Complementary Metal-Oxide-Semiconductor (sCMOS) camera relies on statistical compensation for its pixel-dependent gain, offset and readout noise. In this work we show that it is also necessary to compensate for differences in the relative quantum efficiency (RQE) of each pixel. We found differences in RQE on the order of 4% in our tested sCMOS sensors. These differences were large enough to have a noticeable effect on analysis algorithm results, as seen both in simulations and biological imaging data. We discuss how the RQE differences manifest themselves in the analysis results and present the modifications to the Poisson maximum likelihood estimation (MLE) sCMOS analysis algorithm that are needed to correct for the RQE differences.

Cega: A Single Particle Segmentation Algorithm to Identify Moving Particles in a Noisy System

“…For algorithms that factor in multiplicative noise, the conversion factors that relate a set of ADUs to effective photons are required quantities. Extensive work on CCD (Young et al, 1998) and sCMOS (Babcock et al, 2019; Huang et al, 2013) sensors has demonstrated that statistical regression with a simplified model can effectively recover Poisson-like statistics for estimators that use multiplicative noise models. Given that a scientific camera produces an output signal S i at pixel i the signal components are defined as: where K i is the effective photon count, G i is the multiplicative gain factor on the Poisson-like signal, O i is the constant mean offset, and R i is an unbiased random electronic signal generated from the camera read noise.…”

Characterizing and correcting camera noises in back-illuminated sCMOS cameras

“…We replaced the camera noise data with our measurement or simulation. The NCS and MLE sCMOS use a gain map to correct FPN, which is considered to be not accurate enough [3, 15]. Therefore, instead of using the gain map, we used the measured gain value of the camera multiplying by the photon response map.…”

“…We use photon response map to replace gain map and/or QE map, so that the calculation can be easier. Note that this treatment is accurate enough for photon signal calculation but would lead to a small bias in electron signal calculation [15]. We further analyze these noise maps by: 1) calculate the SD of relative offset map and photon response map to obtain DSNU and PRNU, respectively; 2) calculate the RMS of read noise map to obtain the amplitude of global read noise of the entire camera; 3) calculate the ratio of the SD to the RMS of the read noise to present RNNU.…”

“…The camera noises in sCMOS cameras (abbreviated as sCMOS noises hereafter) were characterized [5][6][7][8] and their impacts on some applications, such as single molecule localization microscopy (SMLM), were studied [9][10][11][12][13][14]. Noise correction algorithms were integrated into some commercial sCMOS cameras by their manufactories, or developed for several specific imaging scenarios by researchers [11,13,[15][16][17].…”

Cega: a single particle segmentation algorithm to identify moving particles in a noisy system

“…For algorithms that factor in multiplicative noise, the conversion factors that relate a set of ADUs to effective photons are required quantities. Extensive work on CCD (Young et al, 1998) and sCMOS (Babcock et al, 2019;Huang et al, 2013) sensors has demonstrated that statistical regression with a simplified model can effectively recover Poisson-like statistics for estimators that use multiplicative noise models. Given that a scientific camera produces an output signal S i at pixel i the signal components are defined as:…”