The“scientific” CMOS (sCMOS) camera architecture fundamentally differs from CCD and EMCCD cameras. In digital CCD and EMCCD cameras, conversion from charge to the digital output is generally through a single electronic chain, and the read noise and the conversion factor from photoelectrons to digital outputs are highly uniform for all pixels, although quantum efficiency may spatially vary. In CMOS cameras, the charge to voltage conversion is separate for each pixel and each column has independent amplifiers and analog-to-digital converters, in addition to possible pixel-to-pixel variation in quantum efficiency. The “raw” output from the CMOS image sensor includes pixel-to-pixel variability in the read noise, electronic gain, offset and dark current. Scientific camera manufacturers digitally compensate the raw signal from the CMOS image sensors to provide usable images. Statistical noise in images, unless properly modeled, can introduce errors in methods such as fluctuation correlation spectroscopy or computational imaging, for example, localization microscopy using maximum likelihood estimation. We measured the distributions and spatial maps of individual pixel offset, dark current, read noise, linearity, photoresponse non-uniformity and variance distributions of individual pixels for standard, off-the-shelf Hamamatsu ORCA-Flash4.0 V3 sCMOS cameras using highly uniform and controlled illumination conditions, from dark conditions to multiple low light levels between ~20 to ~1,000 photons / pixel per frame to higher light conditions. We further show that using pixel variance for flat field correction leads to errors in cameras with good factory calibration.
Quantitative fluorescent imaging requires optimization of the complete optical system, from the sample to the detector.
Such considerations are especially true for precision localization microscopy such as PALM and (d)STORM where the
precision of the result is limited by the noise in both the optical and detection systems. Here, we present a Camera
Simulation Engine (CSE) that allows comparison of imaging results from CCD, CMOS and EM-CCD cameras under
various sample conditions and can accurately validate the quality of precision localization algorithms and camera
performance. To achieve these results, the CSE incorporates the following parameters: 1) Sample conditions including
optical intensity, wavelength, optical signal shot noise, and optical background shot noise; 2) Camera specifications
including QE, pixel size, dark current, read noise, EM-CCD excess noise; 3) Camera operating conditions such as
exposure, binning and gain. A key feature of the CSE is that, from a single image (either real or simulated "ideal") we
generate a stack of statistically realistic images. We have used the CSE to validate experimental data showing that
certain current scientific CMOS technology outperforms EM-CCD in most super-resolution scenarios. Our results
support using the CSE to efficiently and methodically select cameras for quantitative imaging applications. Furthermore,
the CSE can be used to robustly compare and evaluate new algorithms for data analysis and image reconstruction. These
uses of the CSE are particularly relevant to super-resolution precision localization microscopy and provide a faster,
simpler and more cost effective means of system optimization, especially camera selection.
Light microscopy imaging is being transformed by the application of computational methods that permit the detection of
spatial features below the optical diffraction limit. Successful localization microscopy (STORM, dSTORM, PALM,
PhILM, etc.) relies on the precise position detection of fluorescence emitted by single molecules using highly sensitive
cameras with rapid acquisition speeds. Electron multiplying CCD (EM-CCD) cameras are the current standard detector
for these applications. Here, we challenge the notion that EM-CCD cameras are the best choice for precision localization
microscopy and demonstrate, through simulated and experimental data, that certain CMOS detector technology achieves
better localization precision of single molecule fluorophores. It is well-established that localization precision is limited
by system noise. Our findings show that the two overlooked noise sources relevant for precision localization microscopy
are the shot noise of the background light in the sample and the excess noise from electron multiplication in EM-CCD
cameras. At low light conditions (< 200 photons/fluorophore) with no optical background, EM-CCD cameras are the
preferred detector. However, in practical applications, optical background noise is significant, creating conditions where
CMOS performs better than EM-CCD. Furthermore, the excess noise of EM-CCD is equivalent to reducing the
information content of each photon detected which, in localization microscopy, reduces the precision of the localization.
Thus, new CMOS technology with 100fps, <1.3 e- read noise and high QE is the best detector choice for super resolution precision localization microscopy.