wavelets application for particle detection

[detlefhof]

Hallo, is it possible to use the wavelets technique for particle detection in cell images?

[BruceAmmons]

I suspect it is. It would really depend on what type of particles you are trying to detect and what size ranges they are. Any particular reason you want to use wavelets? There are several methods of detecting particles, and wavelets are probably a little more complicated than necessary.

Do you have a sample image with particles that you want to detect?

Bruce

[detlefhof]

Hi Bruce,
I have just finished a first sketch of a particle tracking program that is able to find the trajectories of several particles in a living cell. ( The particles are vesicles labelled with a fluorescent dye moving near the cell membrane where exocytosis takes place). With a TIRF microscope camera image stacks were recorded (usually stored as .PIC file). The particles are represented by only a few (4 � 16) pixels and the contrast is rather poor. I have started using thresholds (after low-pass filtering and background subtraction) with reasonable results. The disadvantage is that some trajectories are cut in two (or more) pieces because the particle intensity goes down for one ore two images. I also played around with autocorrelation which gives
nice movies showing 3D-looking balls moving around, but the tracking results are not really improved. Especially the separation of partially overlapping particles does not work in both cases.
I heard from other groups working in this field who introduced the wavelet technique in their tracking program with good success. I also found an encouraging paper (http://www.cs.bris.ac.uk/Publications/Papers/1000275.pdf) where an application rather similar to ours is described. They use wavelet filtering for the detection and then a Kalman tracking algorithm to find the trajectories of the particles. I�m not sure whether I�m able to redesign their approach with LABVIW, but at least I want to know more about these techniques. I have attached a folder of TIF images representing a stack, because I�m not sure whether you are able to read a PIC file.

[o_sidla]

Hi Detlef,

the problem with the objects in the image sequence that you provided is certainly that the blobs merge and separate quite often. I think a Kalman Filter could help you estimate trajectories in order to deal wtih such situations. We have implemented a Kalman Framework in LabVIEW and its actually relatively easy to code. Nethertheless my experience suggests that dealing with occulsion of objects, merging and separation in tracking applicationsis a hard problem. You could make use of the LabVIEW Advanced Signal Processing Library which has Wavelets implmented, but I doubt that it can make a solution that much easier. I think that a trajectory estimation with a Kalmann Filter would help you to figure out how objects contintue their motion a
fter a merge/separation. One needs to log where which objects goes, try to find out when objects merge and try to estimate when the separate again.
Good Luck
Oliver Sidla