Lose track every 2 frames with LAP

Hello,

I am using the LAP tracking method in the TrackObjects module to track my cells. I have six cells in my movie, and I notice that they are labeled 1-6 at timepoint 1 and then receive the same labels for timepoint 2, just as they should.

However, at timepoint 3, the tracks are lost and they become numbered 7-12. They then have the same numbers for timepoint 4, but at timepoint 5 the tracks are lost and they become 13-18. This pattern continues.

This every-second-image pattern is not reflected in the actual images, in which the cells’ movement is fairly constant.

Any idea what is going on?

Thanks, Daniel

My guess is that it may have to do with your gap settings, but without a pipeline and images it’s impossible to test this. Can you provide those things? Thanks.

Thanks for being to take a look at it! Here is a folder with the raw image, the pipeline, and the tracked output.

https://drive.google.com/drive/folders/0B1lheVBEchkWNDZnX2k5ekYxQ3c?usp=sharing

Thanks again,

Daniel

Getting a permissions error, sorry.

Sorry about that, should be fixed now!

The first thing I’d say is that since you’re running the second phase of the LAP tracking, don’t worry too much about the numbers on that output movie- that graph reflects the initial “phase” of the tracking output, the second phase of LAP will try to stitch those segments together so the numbers on that output image are not necessarily reflective of the final tracking results (other tracking algorithms only have one “phase”, hence the creation on the image in the first place- sorry for the confusion! It’s one of the things that makes LAP the hardest methods to test settings for).

What you should look at is the TrackObjects_Label column in your spreadsheet data- that’ll reflect the true tracking “identity” CP was able to assign after the first phase. TrackObjects_Lifetime is also nice, because it shows you how long a given track has persisted- it looks like 2 of your cells are being tracked for 25 frames, with the others being broken into smaller chunks (you keep all 6 for the first 6 frames, then lose one at frame 7, one at frame 9, the other two at frame 10, though at least 2 of those “lost” cells get tracked for a further 10 consecutive frames under new labels)

Knowing this, you can probably further test new combinations of settings (I think LAP given its ability to suppress merges and splits will probably end up being best for you but you can think about even new tracking modes, Overlap and Measurement might not work but Distance might). This is a really tricky movie though, it’s not that densely time-sampled relative to the movement, the objects are similar looking, and the 6 objects are clustered very close together relative to the distance they travel, so I’m not super hopeful about the ability to get perfect tracking in an automated way. You’re likely going to be forced to balance the throughput of your analysis with how much you can try to minimize those problems in your movies; if you can get better spacing and/or denser time sampling in the acquisition, it’ll make it much more likely you can succeed at an automated tracking strategy, if not it may have to be manual or only semi-automated.

Thanks, that is really good to know! Two follow-up questions:

-I see the TrackObjects_Lifetime column gives me the information you described, but the TrackObjects_Label column is just a series of "1"s. This doesn’t seem right.

-Is there any way to visualize the incomplete tracks so I can manually stitch them together?

-I see the TrackObjects_Lifetime column gives me the information you described, but the TrackObjects_Label column is just a series of "1"s. This doesn’t seem right.

Nope, it’s definitely not right- it means that given how tightly your objects have clumped together, TrackObjects has “merged” them all into one giant mega-object. You’ll need to play around with something in your LAP parameters to see if you can avoid this.

Is there any way to visualize the incomplete tracks so I can manually stitch them together?

You can visualize the segments with CellProfiler’s Tracer program, but there’s no way to “edit” or “correct” the tracks in CellProfiler, sorry. I know editable tracking programs exist in ImageJ/FIJI, and I believe in Icy as well, but it’s not something CellProfiler does.