Severed Trajectories Using Trackmate

Hi, before I address my problem I want to say that I appreciate your efforts developing and answering the possible solutions to the community’s issues.

To give you a little perspective, we are working with a sessile droplet and we are currently addressing our efforts trying to reconstruct the internal flow of the droplet over time; evaporation, symmetry of the droplet, thermodynamic conditions, among others are parameters that are not easy to stabilize and they have an impact in our observations.

Its an outstanding complicated problem trying to resolve the internal flows that we are addressing with an experimental equipment developed in our lab. Our experiments are made with a droplet that has 1mm of diameter and 1 micron fluorescent particles at a given density. The microscope is capable of recording 3D+t images using a piezoelectric device coupled to the objective that changes the height of the focal plane.

In regards to the trajectories obtained using the trackmate, we stumbled to the problem that the trajectories are severed at least once for an important percentage of the particles. We have tried to preprocess our images but we cannot get rid of the problem, some of the transformations we have tried are the FFT bandpass, deconvolution, contrast enhancement, combination of these routines and the results does not improve.

The parameters selected to run the trackmate were:

-DoG of radius of 2 and 3 pixels and not sub-pixel localization

-The LAP Tracker with a Max Distance of 30 pixels and not allowing Gap Closing, Segmented Merging nor Segment Splitting.

In the images attached to this request it is shown two panels of the same particle. The right panel shows the original image and the left panel shows the resulting severed trajectory, changing from blue to green. The yellow squared is just there to highlight the particle that we are following.

In the original images, it can be seen that the particle, although it changes slightly of height (focal plane), the intensity and the area remain relatively the same, the displacement made by the particle is within the parameters selected, despite all of the above considerations, the trajectory of the particle is finished by the algorithm at t=4 and starts a new trajectory with the same particle at t=5.

We do not know what is the problem, we are trying to modify some internal parameters of the plugin but so far we haven’t been able to minimize the problem.

We are hoping that you can point to us where do you think is the problem so we can focus our efforts.

Here is the link to the images

Could you post the raw images somewhere we could access them? Uploading them on the forum did not work.

Oh thanks, here is the link…

I would need the raw data. On the movie capture you send, the problem is apparent but I cannot find the reason why. Indeed, tracking it gives this:


Dear Jose, dear Gabriel,

Unfortunately I cannot bring better news.

I tried fine-tuning the parameters and I met the death valley of tracking.

I thought your images would be very difficult to cope with. They are 3D objects, with a large Z dimension. The spots we want to track are about 6 pixels in diameter, but this is also the Z-spacing between your z-slices. I though we would loose them and that this large deltaZ would forbid their detection. Plus, the image apparently is not acquired with a confocal (am I correct?) as we see the defocus shadows of each spots in the z planes above and below it.

Yet, it worked rather ok, and we can detect most of the spots.

I tried then with tracking, and of course I could not get 100% perfect results.

I think it is impossible to.

In the tracking methods offered by TrackMate you have to specify a maximal search radius. This radius must be larger than the max displacement of a single spot, otherwise the tracking algorithm will break the track in two for large displacement.

At the same time, if you make this search radius too large, there are good chances that the algorithm will be confused by other spots if they are close to the target spots.

For the spots that are close to the surface of the ball, this is what happens: They move a lot, and they are very dense. So of course the tracking algorithm is confused.

I am doubtful that a tracking algorithm can give you 100% accuracy in this situation. Certainly, TrackMate does not ship the tracking algorithms that can harness your data.

My next steps would be to

  • try other tools such as the SpotTracker of Icy (

  • or depending no whether your scientific question allows for it, semi-automatically follow a few spots

  • or change the imaging conditions (fewer spots? faster imaging rate?)

Good luck.

I will post a copy of my answer to the forum in case people can comment.

Best regards