Quantcast
Skip to content

Resolving your Noise Complaints – the Basics of Signal/Noise Ratio in Microscopy

The resolution of any microscope is related to the numerical aperture of the lens and the wavelength of light used to form the image, and can be calculated using Abbe’s law.

This, however, is the ideal situation – the best case scenario. In real life, resolution must be defined in terms of contrast, and there are additional parameters that have an impact on contrast. We have already discussed one of them, i.e. dynamic range. In this article we will look at the related – but not synonymous – concept of signal-to-noise (S/N) ratio.

Why is the S/N ratio important?

In the presence of noise, what we have is essentially a raise of the signal level ‘floor’, below which there is insufficient contrast between specimen features that would enable us to distinguish them as separate. So, the distance required to provide sufficient contrast to resolve two points in an image is increased when noise is taken into account. The S/N ratio also influences imaging capabilities in another way, by reducing the number of gray levels that can be distinguished in an individual image pixel, which in effect establishes a limit on the dynamic range of the image. We often speak about resolution as if noise is zero and the dynamic range is infinite, but in real life noise levels limit the useable dynamic range, contrast and point separation range.

So, what exactly is “noise”?

To begin with, we have to define what we mean by “noise” in the context of this discussion. If we do an immunofluorescence labelling and we do not wash our slides well, we will end up with an increased background. Is this “noise”? In terms of the experiment as a whole, it certainly is – and it will bring about the same results as any other form of noise: reduced dynamic range, reduced resolution etc. But is it “noise” in terms of microscopy? No. It is signal. Whatever exists in our sample is “signal” as far as our microscope is concerned. Maybe rubbish signal, but signal nonetheless, in the optical sense. So, when we discuss about noise in the context of microscopy, we are talking about what happens to the information present in our sample. Whether that information should be present in it in the first place is another story. To give (as usual) an audio analogy: the hiss in an analogue tape recording of a concert is noise; people coughing in the audience is signal.

Where does noise come from?

In microscopy, noise is introduced mainly by the detection electronics.

Detection Electronics

S/N ratio in detection electronics depends on quantum efficiency (incident photon to converted electron), read noise (from the readout electronics), dark noise (from the statistical variation in the number of electrons thermally generated within the detector, independent of the presence of photons, but dependent on device temperature), and shot noise or photon noise (the result of the particle nature of light and the stochastic fluctuation in photon arrival times at the detector, equivalent to the square-root of the signal).

There are many different types of detectors used in different microscopes. Here are three common detectors and their pros and cons when it comes to S/N ratio:

  1. Charged Couple Device (CCD) and Complementary Metal Oxide Semiconductor (CMOS) detectors are both very effective in the visible range. However, they cannot detect single photons, because of the noise introduced by the analogue control electronics, and they cannot be used for single photon timing experiments.
  2. Photo-multiplier tube (PMT) detectors. Single photon sensitivity, large active area, very high gain, low background noise levels (no read noise) and high temporal resolution down to 20 ps. However, the probability of detecting single photons is below 30%.
  3. Avalanche Photo Diode (APD) and Single Photon Avalanche Diode (SPAD) detectors. Higher dark noise than PMTs, counterbalanced by the higher quantum efficiency. Very fast temporal resolution.

There is no “best” type of detector. The choice of detector depends on many parameters – like which microscope it used on (widefield, scanning confocal, spinning disk confocal) and also the applications it is intended for. For example, PMTs and APDs behave differently across the spectrum. There are also hybrid detectors available, which combine the best features from two categories.

A Confocal Consideration: Pinhole Diameter

In confocal systems, pinhole settings also play an indirect role in the final noise levels and can be adjusted, within limits, when noise is an issue. Strictly speaking, to obtain a “true” confocal image, the pinhole should be infinitesimally small. Nothing “infinitesimally small” exists, of course, but trying to use a (realistically) very small pinhole would still be problematic, because such a pinhole would not only reject the out of focus light (see our webinar on confocal microscopy principles for that), but it would also result in a very weak image signal, leading to a low S/N ratio. On the other hand, a very large pinhole allows many photons through, but defies the point of using a confocal. Open it all the way, and you end up with a fancy (and unnecessarily expensive) widefield microscope!!!

So, in practice, it is necessary to adopt an optimum diameter for the pinhole, which depends on the design of the microscope, how it is operated, and the type of specimen. We usually work at “airy 1”, and increase or decrease, within reasonable limits and always taking into account what we gain and what we lose in either direction.

What other measures can you take to improve S/N ratio?

  • The obvious one is to get as bright a signal as possible during staining – but that falls under “staining optimization”.
  • Increasing the excitation intensity is another obvious solution – with obvious limitations. Increasing excitation too much leads to increased bleaching – and phototoxicity, in live imaging- thus compromising the quality of the signal, and the whole experiment. Another side effect is fluorophore saturation, during which fluorescence emission intensity starts behaving in a non-linear fashion after a certain point. This stems from the fact that a significant fraction of the fluorophore molecules has been elevated to the excited state, thus depopulating the ground state, effectively lowering the useable fluorophore concentration.
  • Utilizing image averaging is very helpful. Stochastic noise – by definition – occurs at random pixel coordinates, while signal is always present. During averaging, the system discards pixel values that do not appear consistently over the averaging time, thus significantly lowering the noise level.
  • Accumulation is another useful function: it effectively adds the values over 2 or more scans. Signal is thus increased – but so is noise. One has to carefully combine accumulation with averaging to get a really useful increase in S/N ratio. Reducing the excitation light intensity and repeatedly registering the same field has the added benefit of avoiding phototoxicity and fluorophore saturation. This is because the same amount of radiation causes much less harm when given in many smaller doses than in one large instalment.
  • There are also a number of post-acquisition measures one could take to optimize the image, such as image deconvolution, using the advanced image processing software that is available today.

Having all these parameters in mind is a good starting point towards optimizing S/N ratio when recording microscope images. After some time, choosing what to adjust and how much to adjust it becomes very intuitive for the experienced user. Have a go, and you will be surprised with the huge difference you can have in your images by getting all these parameters just right!

Image Credit: flattop341

1 Comment

  1. Shiraz on July 20, 2016 at 2:07 am

    A pretty good/useful article describing the effects of S/N ratio on image quality throughput. Personally, I’d prefer the accumulation strategy for effective S/N resolution (unless this is not probable, for temporally-sensitive applications, e.g. FLIM, etc) coupled with noise-thresholding; noise (being random) would effectively negate each instance of noise occurrence at a particular pixel over a specified temporal window, while signal would accumulate for the said pixel. Establishing a threshold then allows for the effective nullification of noise resulting in this strategy (though not temporally-sensitive) achieving good S/N ratios.

Leave a Comment





This site uses Akismet to reduce spam. Learn how your comment data is processed.

Scroll To Top
Share via
Copy link