Prime Scientific CMOS Camera Processing Tools for Super-Resolution Microscopy

Author: Dan Croucher, Applications Team Manager, Teledyne Photometrics


Prime Scientific CMOS cameras provide the highest levels of sensitivity which make them ideal for low-light imaging applications such as single molecule fluorescence and super-resolution microscopy. The Prime cameras come equipped with in-line processing features which can be enabled to improve point localization performance for localization based super-resolution imaging. These processing features, PrimeEnhance™ and PrimeLocate™, can improve signal-to-noise ratios to increase localization accuracy as well as aiding with falsepositive localized events and data management.

Shot Noise Impact on Localization Accuracy

When single-molecule events are detected, they create a distribution, and in a perfect world, this distribution for a point of light would be identical with each measurement. However, real-life imaging requires dealing with a variety of noise sources which can affect the intensity distribution of a point and therefore have a large impact on the localization accuracy.

The primary noise source, shot noise, adds a variability to the intensity which follow a Poisson distribution. As a mean intensity measurement, the distribution created by the single-molecule will have the appropriate intensity spread across the pixels, but individual measurements will contain some randomness which can have a large impact on localization performance. This can be especially true when the distribution, which typically spans a 3x3 pixel array, is spread across a larger area and further increases the impact of the shot noise on the localization precision.

As discussed in the PrimeEnhance Technical Note, PrimeEnhance reduces the effects of shot noise on images through real-time image-processing, providing an noise reduction factor of 3x to 5x. By reducing the pixel-topixel variation caused by shot noise, there may be advantages to using PrimeEnhance during super-resolution acquisitions to improve localization accuracy.

Evaluation Prime-Enhance Impact on Localization Accuracy

The Shim Group at Korea University is an interdisciplinary lab covering from physical chemistry to biophysics and cell biology. Using the GattaPAINT80 nano-rulers (surface-immobilized DNA origami with three localization points evenly spaced with 80nm gaps), the Shim group ran some super-resolution localization performance comparisons between a 1024x1024 EMCCD camera, Prime 95B BSI CMOS camera, and a Prime95B BSI CMOS camera with PrimeEnhance enabled to see how the different technologies fared against each other.

The following parameters were used to do the comparison:

  • Exposure time: 20ms
  • 30,000 frame acquisition
  • TIRF Illumination
  • 647nm laser, ~2 kW/cm2 intensity

512x512 EMCCD
Figure 1. 1024x1024 EMCCD, 128x128 array, 130µm pixel size, EM Gain 30X (A) Teledyne Photometrics Prime 95B, 150x150 array, 110µm pixel size, PrimeEnhance disabled (B) Teledyne Photometrics Prime 95B, 150x150 array, 110µm pixel size, PrimeEnhance enabled (C).
Figure 2a
Figure 2b
Figure 2c
Figure 2d
Figure 2. 1024x1024 EMCCD localized output (A) 1024x1024 EMCCD achieves a 13.3nm localization accuracy (B) Teledyne Photometrics Prime 95B localized output, PrimeEnhance disabled (C) Teledyne Photometrics Prime 95B, PrimeEnhance disabled achieves a 9.2nm localization accuracy (D) Teledyne Photometrics Prime 95B localized output PrimeEnhance enabled with same threshold applied to other images. The same thresholding level allows some noise and false localizations to appear (E) Teledyne Photometrics Prime 95B, PrimeEnhance enabled with the same threshold has a lower accuracy due to an increased number of false events and noise present in the image (F) Teledyne Photometrics Prime 95B localized output PrimeEnhance enabled with 2X threshold. Higher threshold causes false-positive events to be filtered out (G) Teledyne Photometrics Prime 95B, PrimeEnhance enabled with 2X threshold achieves a 7.5nm localization accuracy (H).

The experimental results for these tests show that the Teledyne Photometrics Prime 95B camera was able to provide 30% better localization (13.3nm vs 9.2nm) accuracy when compared to the EMCCD camera. There are also indications that PrimeEnhance can further increase the localization accuracy by another 15-20% when compared to the accuracy without PrimeEnhance (7.5nm vs 9.2nm) and a potential 40% improvement (7.5nm vs 13.3nm) when compared to EMCCD cameras.

It must be pointed out that a higher filtering threshold is recommended to ensure that any extra noise and other image processing effects are prevented from appearing in the image and affecting the output. Tests indicate that a 2x to 3X higher threshold may be required.

Prime-Locate to Pre-Process Background and Improve Data-Management

Super-resolution microscopy requires sparsity in data to ensure that single fluorophores near each other are able to be temporally separated from each other. This causes that image output to be primarily background populated with sparse single points of illumination. Since much of the background image will need to be filtered out during the localization process, there is an opportunity to reduce the amount of data acquired by preprocessing the background and removing it from images. Additionally, this presents an opportunity to decrease the amount of time required to process a typical dataset consisting of tens of thousands of frames.

PrimeLocate enables this pre-processing by identifying the single-molecules in an image and transferring just these regions to the computer for localization. PrimeLocate does not localize the points, and allows users to select their preferred algorithms to localize the image data.

Figure 3.
Figure 3. PrimeLocate dynamically identifies the single-molecules which are present in the image on a frame by frame basis (A) PrimeLocate removes the background around the regions and only transfers the single-molecules on a frame by frame basis (B).
Figure 4a
Figure 4b

Figure 4. A 1024x1024 acquisition of 16-bit data for 1000 frames returns 2GB of data, which can be compressed down to 1.1GB (A) A 1024x1024 acquisition of 16-bit data for 1000 frames using PrimeLocate returns 2GB of data, which can be compressed down to 0.27GB (B).

When doing a side by side acquisition and comparing the ability to compress the data, a traditional dataset would achieve a ~45% compression ratio. Using PrimeLocate, the compression ratio can be increased to ~85% providing a significant amount of reduction in required storage space.

PrimeLocate is already providing real-world utility with reduction of data-glut caused by scientific CMOS cameras which acquire extremely large amounts of data very quickly.

Prime-Enhance and Prime-Locate for Super-Resolution Microscopy

The Prime 95B scientific CMOS cameras are available with features that can potentially improve localization microscopy performance. Initial tests indicated that a 20% to 40% improvement of performance is possible in localization accuracy, depending on the imaging technology currently being used. Increased thresholding settings will be required to ensure that proper results are generated.

PrimeLocate can provide up to 85% reduction in acquired data-set size, allowed for easier management and access of super-resolution data, which can easily get into the many tens of thousands of images.