Cell finding
Identifying objects in an image is an area that offers a range of
choices. The ideal outcome is to find objects of interest, such as the
nuclei of cells, and reject everything else in the image. The more
sophisticated an algorithm used, the better the result will be. The
trade-off is that with increased processing the slower the scan becomes
because of the time spent making calculations. On the other hand the
user may not be too worried about false identification of objects or
missing too many objects of interest, just wanting to process samples
as quickly as possible with decent results. Therefore options exist to
cover these situations.
Why pre-process?
The result of pre-processing is to generate areas containing objects
whose general features suggest they are what we are looking for. From
this point more sophisticated analysis can be carried out (by the CHARM
algorithm) on these areas to accurately calculate the loci of what we
are after. Without pre-processing the CHARM algorithm has the entire
image to work on and this will require time, leading to a slower scan.
On the other hand, why not pre-process all the time? The answer is that
when illumination conditions aren't ideal the thresholding technique
can fail. Good contrast and uniform illumination is needed for basic
thresholding to work, and when this isn't the case then other methods,
like edge detection, perform better. If objects are being missed with
pre-processing then it might be worth allowing more time for the scan
and unchecking this option.
Pre-process images -
If this is selected then after a snapped image is taken, thresholding
is applied. This essentially separates out bright objects from the rest
of the image. If 'simple threshold' is checked then the numerical value
set by the user on the slider is used to determine which pixels to be
treated as bright (and thus form part of an object) and which to be
ignored. Unchecking this box hands control of threshold setting to the
program which looks at the intensity information in each image to
determine the level to set.
Minimum/Maximum cell diameter
- We may know that the objects we are interested in tend to have a
limited range of sizes. So to narrow down the objects found so far we
can apply a size filter, based on minimum and maximum diameters, to
retain only those which fall between the 2 limits specified.
Threshold - To know
what threshold to set move to an area on the sample and take a camera
snap. Now, moving the threshold slider will show what lies above and
below the threshold limit.
Save cell data - It is
usual that the user will want to save details of the objects found. The
most obvious details are the co-ordinates of the stage needed to bring
the object back into the field of view (at the cross-hairs). Other data
is saved too, such as object size, average intensity, etc. However,
this option may not always be needed, for example when simply the total
number of objects in a region is all that is needed.
Show processing - At
the expense of a small amount of time it is possible to show some steps
in the object-finding procedure as the scan progresses. This may be an
aid in judging whether pre-processing is performing adequately or not.
Simple object-finding can be carried out by identifying the centers of
objects found through thresholding. For well defined and well spaced
single objects this approach can be very successful. When, however,
objects are touching or appear wrongly shaped then the loci assigned to
them will be incorrect. Another method of analysis is needed.
Enable CHARM - An
algorithm called CHARM (compact Hough and radial map) can be used. The
essence of the technique is as follows. Edge detection is performed on
the image, i.e. areas are identified where the intensity changes
sharply (such as at the boundary of a cell). The assumption is then
made that the center of an object of interest lies somewhere along a
line perpendicular to this edge (just as the center of a circle lies on
a line, a radius, which is perpendicular to an edge, a tangent, of the
circle). Now, imagining all the edges forming the boundary of an object
and all the lines perpendicular to these edges, it can be seen that
they should all point in the same direction and cross at the object's
center. This is how object centers are found. There are two marked
advantages of this method over conventional thresholding. Firstly,
absolute brightness of the object is not critical - all that matters is
the degree of contrast between the object and the background. So dimly
lit areas of the field are not such a problem. Secondly, the algorithm
is robust against incomplete information. For example when two objects
join, as in a cell doublet, and part of the boundary is missing, the
radii which remain will still point correctly toward the cells' centers
and indicate 2 points where 'radii crossing' is maximum.
So to get the best object-finding available, tick the 'Enable CHARM'
box. If desired the 'Advanced' button brings up the options to set the
CHARM parameters numerically or use the optimizer to visulaize the
action of the CHARM algorithm based on selected settings.