- Cutting through the noise
Technique aids calibration and characterization of digital microscopy systems
The emergence of digital technology has enabled researchers to employ microscopy as a quantitative tool because it provides measurements of biological parameters expressed in arbitrary units of signal intensity. To take full advantage of the technique, microscopists must determine the precision of the intensity measurement. Available methods for calibration and standardization of digital microscopes are limited, however, inasmuch as they provide an estimation of a microscope’s precision at only a single level of intensity. Furthermore, they measure only the total noise level — not the characteristics of the noise.
Investigators have reported a technique developed for calibration and characterization of digital microscopes. Shown here are significant intensity levels in the images obtained with two CCD cameras (A,C and B,D, respectively). The investigators segmented the raw images (A,B) to represent the intensity levels with 95 percent confidence (C,D). Images reprinted with permission of the Journal of Microscopy.
In the May issue of Journal of Microscopy, investigators from the University of Silesia in Katowice, Poland, from Purdue University in West Lafayette, Ind., and from Quantitative Imaging Corp. in Burnaby, British Columbia, Canada, described a technique that contributes to improved calibration of digital microscopes by estimating the temporal variability of signal and noise.
The technique involves registration of a time series of images of a stationary biological specimen, enabling separation of monotonic, periodic and random components of pixel intensity changes in time.
“We adapt data-based mechanistic modeling of time series (used, for instance, in stock market analysis) and a photon-transfer technique to perform simultaneous determination of dark, photonic and multiplicative components of noise under conditions of microscope imaging which closely mimic a typical biological imaging experiment,” said Tytus Bernas, who represents both universities. Thus, the researchers can obtain a respective confidence interval, or noise level, for each level of signal.
The method yields results that are less detailed than those provided by available specialized test benches. However, because it employs time series of images acquired from a real biological specimen, researchers can use it to determine the precision of measurements provided by any fluorescence microscope. They also can estimate the precision of a digital microscope using a standard that is composed of uniformly fluorescent polystyrene beads, or even a piece of fluorescent plastic. However, this approach provides only the total measurement noise that corresponds with a single level of fluorescence intensity.
Bernas and his colleagues first validated the calibration technique using test sets of images with known signal and noise characteristics, showing that it provides accurate estimates of multiplicative noise, Poisson noise, additive noise and dark current independently of one another.
They then used it to characterize the performance of two monochromatic CCD cameras made by Quantitative Imaging Corp.: the Rollera XR and the Retiga 4000R. They registered images of endothelial cells obtained with a Nikon wide-field fluorescence microscope that was outfitted with a 40×, 1.3-NA oil-immersion objective and a 100-W Hg arc lamp. They collected time series of 128 images of the cells at 5-s intervals, registering the series for each of the cameras operating at three gain settings and for 0.25- or 0.75-s acquisition times.
Registration of a time series was important for the characterization of the CCDs. Acquiring images at 5-s intervals allowed them to capture periodic components of the signal with periods of 10 s or higher. At the same time, the series of images let them see trends such as those corresponding to photobleaching, which would become apparent after 10 min.
Using the technique, the researchers determined the number of registered photons per pixel for the two cameras (A,C and B,D, respectively), helping to evaluate the practical intensity resolution of the cameras.
The investigators used the measurements to estimate the practical intensity resolution — that is, the dynamic range — of the two cameras. They found that, under similar imaging conditions, the Rollera XR showed a higher total noise level than the Retiga 4000R but that the number of registered photons per pixel also was higher. Therefore, although one may expect higher dynamic range in the images that were registered with the former device, the practical intensity resolution is likely to be better with the latter device.
Bernas emphasized that the calibration and characterization method can be used with any camera and with all digital microscopy modalities, including multidimensional, multispectral, fluorescence lifetime, Raman scattering, single-molecule, and high-content, high-throughput imaging. He added that, besides calibration, “image compression and denoising methods may be developed for particular imaging systems on the basis of our algorithm.”
They are seeking funds to develop the calibration technique further, not only for quantitative digital microscopy but for in vivo imaging of tumors, biomarker tracking or related diagnostics.
MORE FROM PHOTONICS MEDIA