Background estimation
The background in your image is information you want to remove in the final result. It is normally composed of many constant signals, plus non-specific stainings all over the sample which are affected by Photon Noise and therefore are not constant at all. (See BackGround for more details).
You may be also interested in considering some low intensity regions in the image as background, in order to enhance high intensity objects.
Background estimation for deconvolution
In the deconvolution wizard in Huygens Essential and Huygens Professional, estimating the BackGround is part of the deconvolution procedure. This mode of background estimation can also be set under deconvolution parameters in the Professional Operation Window or the Batch Processor template. There are multiple Background Modes, available for background estimation, which are described on this page below.
After the estimation you can indicate a percentage of correction to be applied to the automatic value. For example, set this to zero to accept the estimated value as is, or set it to -10 to reduce the estimated value in 10%. If you manually modify the background absolute value, the change will be also reflected in the percentage correction. This relative background value is especially useful for timeseries, which intensity could decay due bleaching.
In Huygens Professional, there are multiple ways you can estimate the BackGround:
Automatic estimation
The mean background can be estimated automatically in Huygens Professional with the Estimate background tool in the Analysis menu of the Operations window. Use a search area of 0.7 micron radius. If your image does not contain large open areas, decrease the radius.
This tool reports the found value to the Task report window along with the coordinates of the area in which the value was found, so you that can confirm this value visually in e.g. the Twin Slicer.
The reported value is a conservative estimate of the average background. For instance, when there is appreciable bleaching there is high probability that the tool will find the last recorded, darkest layer. You can exclude such layers from the search by cropping the image.
The Estimate background tool tries to evade off focus light of the object in conventional images. Still, it is a good idea to inspect the location were the tool reported the lowest average value.
This automatic estimation will be executed also if you run a deconvolution command like cmle or qmle with any Background Mode other than manual, but you can correct the estimated values by a percentage as defined by the Background Per Channel parameters.
Selecting a region of interest
You can have more control if you extract the interesting portion of the image to a new image by using the Intelligent Cropper, and the ask for the statistics of that independent chunk (Edit > Statistics): you will obtain among other things the average value. Still, the cropper only allows you to crop or extract regions of cuboid shape.
But the ROI selection in the Object Analyzer can be used for more complex shapes. When you have a ROI defined anywhere in the image and click the "Analyze all" button, also information about the ROI is printed on the table, among others (by default, but you can change this) the sum intensity and the number of voxels. By dividing these two figures you can easily get an average value for the intensity inside the ROI.
Because defining a ROI in the background where not objects are normally found may by tricky, you may want to activate in the MIP segmentation group the same channel you are analyzing, so that you have a visual reference on where to define the ROI.
See Object Analyzer ROI.
Tuning background estimation
A normal deconvolution procedure usually requires tuning the background value until the results fit your experimental necessities. As you can restart the deconvolution to use different parameters, you can test and compare which is the best background value. When you find it, you can apply it (in absolute terms, or as a percentage of correction to the automatically averaged one) to all your images.
A deconvolution example where the background is tuned until it fulfills the user's experimental requirements is detailed in Tuning Huygens Deconvolution. In this example, Huygens Professional is used instead of Huygens Essential. That is why some Tcl Huygens commands are shown as reference. But don't pay attention to them if you don't want to, what matters is the idea that tuning the background can take you to the results you want.
Background estimation within the Colocalization Analyzer
For the colocalization analyzer there are three methods to estimate the background:
- Gaussian minimum (which is equal to the method explained above)
- Costes method
- Optimized method
The big difference between the gaussian minimum and the costes/optimised methods is that the gaussian minimum only looks at the information of a single channel. Costes and optimised method use the information of two channels to find the optimal background setting.
Costes method
This method is based on the estimation method explained in article Automatic and Quantitative Measurement of Protein-Protein Colocalization in Live Cells by Sylvain V. Costes, Dirk Daelemans, Edward H. Cho, Zachary Dobbin, George Pavlakis and Stephan Lockett, Biophysical Journal, volume 86, June 2004 (3993-4003).
This method calculates the regression line of the 2D-histogram or 2D scatterplot and each point on this line is a combination of backgrounds for the red and green channel. Starting with the highest point on the regression line, the position is decreased over the regression line, until the Pearson coefficient of the background is zero. In other words, until the Pearson coefficient of the voxels that are below the background threshold is zero.
Optimized method
This method continues with the Costes method without the assumption that the ideal background threshold combination is on the regression line. By using an iterative process, the entire histogram is searched to find the point such that the Pearson of the background is zero (or closest to).
The assumption that is made for both the Costes as the Optimised method is that the noise in a constant background is Poission distributed. The Pearson coefficient of two channels with Poission noise is zero.
To see how the values of these methods are used when setting a threshold in the Colocalization Analyzer, please visit the section on Threshold handling on our Colocalization Theory page.