ImageJ

cell tracking question

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

4 messages Options

Vergara, Leoncio A.

cell tracking question

Dear Image J experts

We are planning to do experiments that require to track the position of about 100 cells during two hrs. in response to different treatments. We are planning to image using an inverted microscope in wide field mode with a CCD camera, we will capture phase contrast and fluorescence frames. For fluorescence we will use either a nuclear dye or one of the cell tracking dyes, or both. Probably we will use the fluorescence signals for the tracking. The images will be collected with a relatively low magnification lens to obtain a large field of view. At this time we are planning to capture just a single field of view at a fixed position at time intervals that will be adjusted depending on the speed of movement of the cells (so no fancy coupling of cell tracking with motorized stage movement). Cell density will be adjusted depending on the requirements of the analysis.

I was wondering if there is a good ready to use solution implemented for automatically tracking these cells.

Any help will be greatly appreciated.

Leoncio

Center for Biomedical Engineering
UTMB
Galveston Texas

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Gabriel Lapointe-4

Re: cell tracking question

Hi Leoncio
Fiji comes with TrackMate and it does a better job in my hand than it's
commercial counterpart Imaris for automated 3D tracking. I strongly suggest
you give it a try.

*Gabriel Lapointe, M.Sc.*
Lab Manager / Microscopy Specialist
Concordia University, Biology Department
7141 Sherbrooke St. West SP 534
Montréal QC H4B 1R6 Canada
Lab : (514) 848-2424 x5988
Office : (514) 848-2424 x3008
Fax : (514) 848-2881
Cell : (514) 278-0247
[hidden email]
cmac.concordia.ca
http://gabriellapointe.ca

2013/2/22 Vergara, Leoncio A. <[hidden email]>

> Dear Image J experts
>
> We are planning to do experiments that require to track the position of
> about 100 cells during two hrs. in response to different treatments. We are
> planning to image using an inverted microscope in wide field mode with a
> CCD camera, we will capture phase contrast and fluorescence frames. For
> fluorescence we will use either a nuclear dye or one of the cell tracking
> dyes, or both. Probably we will use the fluorescence signals for the
> tracking. The images will be collected with a relatively low magnification
> lens to obtain a large field of view. At this time we are planning to
> capture just a single field of view at a fixed position at time intervals
> that will be adjusted depending on the speed of movement of the cells (so
> no fancy coupling of cell tracking with motorized stage movement). Cell
> density will be adjusted depending on the requirements of the analysis.
>
> I was wondering if there is a good ready to use solution implemented for
> automatically tracking these cells.
>
> Any help will be greatly appreciated.
>
> Leoncio
>
> Center for Biomedical Engineering
> UTMB
> Galveston Texas
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Wade Schuette

Re: cell tracking question

In reply to this post by Vergara, Leoncio A.

Leoncio, I did something similar, using the old NIH Image, but it was
22 years ago at what was then Parke-Davis R&D in Ann Arbor.

"/*An Automated High Capacity Data Capture and Analysis System for the
in vitro Assessment of Leukocyte Adhesion Under Shear-stress
Conditions*/", by Joseph Low, Debra Killner, and Wade Schuette, *Journal
of Immunological Methods* 194: 59-70 (1996).

We were tracking neutrophil adhesion to capillary walls in the mesentery
of a live rat, capturing images to a wonderful videodisk recorder that
could replay the video forward or backward at almost any speed, and had
frame-addressable single images.

At that time, and it was long ago, there was no published closed-form
solution to tracking multiple points in 2-D space as they moved more or
less independently. ( I'm assuming you have a 2-dimensional world, or
close to it.)

Of course, our problem was a little more complex, since the rat, being
alive, despite being sedated, kept moving slightly, so we had to
register the whole frame to frame zero before we could even begin to
sort out individual point motions.

I seriously doubt that there is a fully-automated solution, but there
may be a partially automated assistant available.

Our largest problem was trying to determine, confidently, for every
point, for every frame, that each point in frame N+1, at location (x2,
y2) was surely the exact SAME cell as that shown in frame N at location
(x1, y1).

In our case, some cells were moving so rapidly they would arrive from
off-screen, be there in only a few frames, and then be gone. Some
cells would arrive mid study and hang around. Some cells would hang
around, and then suddenly decide to depart. Some cells would have
trajectories that actually intersected so that we had to disentangle
which one went which way after they had "bumped".

To further complicate matters, much to our surprise, some neutrophils
would bounce repeatedly against the capillary wall, becoming "stickier"
with each bounce, until they finally stopped entirely, stuck, and
penetrated the wall. Fine. We expected that. What we did not expect
was that some neutrophils seemed to have some kind of harpoon-and-reel
mechanism whereby they would hit the wall, slow down, stop, and then,
without rotating, seem to "winch" their way back to where they had
struck. I'm not sure anyone has ever published that, but I know we
didn't. ( Parke-Davis was purchased by Warner-Lambert then sold to
Pfizer many years ago, and everyone left.)

If I had to write tracking code today, I'd probably try to use a
rules-based expert-system, or the equivalent, to do the analysis.
Assuming that the cells / dots / points can be identified, so you have a
data set with frame-number, timestamp, and a list of (x,y) coordinates,
I'd start by looking over, say, ten frames for cells that didn't move,
assign them ID's, see if they remained stationary or almost so for the
rest of the study, and, if so, (1) assign them that trajectory and (2)
REMOVE THEM from your data set. Now you have only 89 cells per
frame, more or less.

Then I'd look, starting at different points in time, again across ten
frames or so, for cells that had a mostly linear motion from frame to
frame, extrapolate those out as far as possible in both directions in
time, label the set, and REMOVE those from your data set. (Try to link
the head of one trajectory to the tail of another while at it.)

If you keep doing that, you can remove most of the cells from
consideration and are left with the ones that are doing something very
strange, accelerating in some direction or something, but you will
also have removed the clutter. If you can assign a maximum speed to the
cells, and you now have cells move than (speed * time-between-frames *
safety-factor) apart, you can somewhat safely assign most of THOSE
remaining cells, at some points in time to their neighbors, and remove
THEM from the set.

Finally you get down to some you need to eyeball, but again, watching
the video or a time series has WAY more information than trying to
simply look at frame 1 and frame-2 and figure out which went where,
which is close to impossible. Some software that lets you click,
label, draw a spline through X points, extrapolate, add to the spline
anything within X-microns of it, etc. would be useful.

Sidebar: I had the experience once, at Cornell, of standing in the
bell-tower and watching the "quad", at a time when dogs were allowed
leash-free on campus. There was a pack of roughly 15 dogs hanging
around waiting for a late-comer. They'd wander around somewhat
independently, then one would find something interesting and somehow
indicate that to the others, The pack would head for them, congregate,
sniff or examine whatever it was excitedly, and after a while get bored
and all wander off on their own again until the next dog found
something. The video of the neutrophils, played at high
speed, showed almost exactly the same behavior. It was cool to watch.

One last thing. After going to all the trouble to get your experiment
in place, if you have the ability, and a nice stable programmable x-y
stage motor, you might want to capture 9 frames at higher resolution
and montage them together to get your analysis frames. You can montage
them and reduce them to a smaller single frame using Image J. Then, if
you find something really interesting going on, you can go back and
"zoom in" on it and get more details.

Serendipity happens. Of course, moving the stage risks losing your
registration and may not be worth the trouble.

Regards,

Wade Schuette (retired).

On 2/22/13 11:46 AM, Vergara, Leoncio A. wrote:

> Dear Image J experts
>
> We are planning to do experiments that require to track the position of about 100 cells during two hrs. in response to different treatments. We are planning to image using an inverted microscope in wide field mode with a CCD camera, we will capture phase contrast and fluorescence frames. For fluorescence we will use either a nuclear dye or one of the cell tracking dyes, or both. Probably we will use the fluorescence signals for the tracking. The images will be collected with a relatively low magnification lens to obtain a large field of view. At this time we are planning to capture just a single field of view at a fixed position at time intervals that will be adjusted depending on the speed of movement of the cells (so no fancy coupling of cell tracking with motorized stage movement). Cell density will be adjusted depending on the requirements of the analysis.
>
> I was wondering if there is a good ready to use solution implemented for automatically tracking these cells.
>
> Any help will be greatly appreciated.
>
> Leoncio
>
> Center for Biomedical Engineering
> UTMB
> Galveston Texas
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

Vergara, Leoncio A.

Re: cell tracking question

Thanks for the detailed response, it was fun and interesting to read. I am enthusiatic about this project. I hope I will be able to implement a good solution for it as you did in your work.

Leoncio

________________________________________
From: ImageJ Interest Group [[hidden email]] on behalf of Wade Schuette [[hidden email]]
Sent: Friday, February 22, 2013 6:23 PM
To: [hidden email]
Subject: Re: cell tracking question

Leoncio, I did something similar, using the old NIH Image, but it was
22 years ago at what was then Parke-Davis R&D in Ann Arbor.

"/*An Automated High Capacity Data Capture and Analysis System for the
in vitro Assessment of Leukocyte Adhesion Under Shear-stress
Conditions*/", by Joseph Low, Debra Killner, and Wade Schuette, *Journal
of Immunological Methods* 194: 59-70 (1996).

We were tracking neutrophil adhesion to capillary walls in the mesentery
of a live rat, capturing images to a wonderful videodisk recorder that
could replay the video forward or backward at almost any speed, and had
frame-addressable single images.

At that time, and it was long ago, there was no published closed-form
solution to tracking multiple points in 2-D space as they moved more or
less independently. ( I'm assuming you have a 2-dimensional world, or
close to it.)

Of course, our problem was a little more complex, since the rat, being
alive, despite being sedated, kept moving slightly, so we had to
register the whole frame to frame zero before we could even begin to
sort out individual point motions.

I seriously doubt that there is a fully-automated solution, but there
may be a partially automated assistant available.

Our largest problem was trying to determine, confidently, for every
point, for every frame, that each point in frame N+1, at location (x2,
y2) was surely the exact SAME cell as that shown in frame N at location
(x1, y1).

In our case, some cells were moving so rapidly they would arrive from
off-screen, be there in only a few frames, and then be gone. Some
cells would arrive mid study and hang around. Some cells would hang
around, and then suddenly decide to depart. Some cells would have
trajectories that actually intersected so that we had to disentangle
which one went which way after they had "bumped".

To further complicate matters, much to our surprise, some neutrophils
would bounce repeatedly against the capillary wall, becoming "stickier"
with each bounce, until they finally stopped entirely, stuck, and
penetrated the wall. Fine. We expected that. What we did not expect
was that some neutrophils seemed to have some kind of harpoon-and-reel
mechanism whereby they would hit the wall, slow down, stop, and then,
without rotating, seem to "winch" their way back to where they had
struck. I'm not sure anyone has ever published that, but I know we
didn't. ( Parke-Davis was purchased by Warner-Lambert then sold to
Pfizer many years ago, and everyone left.)

If I had to write tracking code today, I'd probably try to use a
rules-based expert-system, or the equivalent, to do the analysis.
Assuming that the cells / dots / points can be identified, so you have a
data set with frame-number, timestamp, and a list of (x,y) coordinates,
I'd start by looking over, say, ten frames for cells that didn't move,
assign them ID's, see if they remained stationary or almost so for the
rest of the study, and, if so, (1) assign them that trajectory and (2)
REMOVE THEM from your data set. Now you have only 89 cells per
frame, more or less.

Then I'd look, starting at different points in time, again across ten
frames or so, for cells that had a mostly linear motion from frame to
frame, extrapolate those out as far as possible in both directions in
time, label the set, and REMOVE those from your data set. (Try to link
the head of one trajectory to the tail of another while at it.)

If you keep doing that, you can remove most of the cells from
consideration and are left with the ones that are doing something very
strange, accelerating in some direction or something, but you will
also have removed the clutter. If you can assign a maximum speed to the
cells, and you now have cells move than (speed * time-between-frames *
safety-factor) apart, you can somewhat safely assign most of THOSE
remaining cells, at some points in time to their neighbors, and remove
THEM from the set.

Finally you get down to some you need to eyeball, but again, watching
the video or a time series has WAY more information than trying to
simply look at frame 1 and frame-2 and figure out which went where,
which is close to impossible. Some software that lets you click,
label, draw a spline through X points, extrapolate, add to the spline
anything within X-microns of it, etc. would be useful.

Sidebar: I had the experience once, at Cornell, of standing in the
bell-tower and watching the "quad", at a time when dogs were allowed
leash-free on campus. There was a pack of roughly 15 dogs hanging
around waiting for a late-comer. They'd wander around somewhat
independently, then one would find something interesting and somehow
indicate that to the others, The pack would head for them, congregate,
sniff or examine whatever it was excitedly, and after a while get bored
and all wander off on their own again until the next dog found
something. The video of the neutrophils, played at high
speed, showed almost exactly the same behavior. It was cool to watch.

One last thing. After going to all the trouble to get your experiment
in place, if you have the ability, and a nice stable programmable x-y
stage motor, you might want to capture 9 frames at higher resolution
and montage them together to get your analysis frames. You can montage
them and reduce them to a smaller single frame using Image J. Then, if
you find something really interesting going on, you can go back and
"zoom in" on it and get more details.

Serendipity happens. Of course, moving the stage risks losing your
registration and may not be worth the trouble.

Regards,

Wade Schuette (retired).

On 2/22/13 11:46 AM, Vergara, Leoncio A. wrote:

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html