Advice on postprocessing DySTrack data

Some simple advice on processing data generated with DySTrack, based on our practical experience.

Note

The biggest challenge with these data tends to be their size. For 3D acquisitions of multiple samples over several hours, raw data can be hundreds of GB or even several TB (in particular for AiryScan data).

In consequence, we have generally adopted the convention that deconvolved, 8bit-converted data is the true “raw data” that is actually stored long-term.

One possible downside of this approach is that new methods of deconvolution cannot be applied retroactively.

Common postprocessing steps

1. Deconvolution of mainscan images

   If required/desired, deconvolve your data using the relevant software (e.g. AiryScan Processing for Zeiss, or NIS Deconvolution or Denoising, or any other tool you usually use). Use batch mode to process all time points as one batch.

   Tip

   It can be useful to have a powerful second PC set up with a high-bandwith connection to the microscope PC and an offline-only version of the acquisition software. This means time-consuming processing steps can be run without blocking the microscope.

2. 8bit conversion of each time point

   For most downstream analysis, 256 gray values are sufficient, so to save disk space and improve performance on downstream image analysis tasks, we generally convert images from 16bit to 8bit format using ImageJ macros in batch mode.

   Warning: serious mistakes possible

   Converting images to 8bit must be done with care, as mistakes can render data useless (or even misleading)!

   Crucially, unless all data / time points across an experiment are converted using the same min and max values, intensity will no longer be comparable (if it ever was; absolute intensity measurements in microscopy come with many pitfalls and must be handled with care or avoided)!

   For any data where intensity matters, it is therefore essential to consistently set the same min and max values in the histogram prior to 8bit conversion.

   If these min and max values are chosen poorly, other problems follow. If the range is chosen too small, data may get “cropped” and in consequence effectively look “overexposed”. Conversely, if the range is chosen too large, the histogram of the data may get compressed into far too few gray values, leading to a “filled contour” effect.

   To pick the right min and max values, be sure to check examples at different time points (in case signal increases at later time points), across multiple samples and, where applicable, across all different experimental conditions.

   If this advice is at all unclear, you should probably speak to someone with image analysis experience before processing your data.

3. Maximum z-projection

   We usually use maximum z-projections for visual analysis of time course data (whilst keeping the z-stacks themselves for quantitative analysis). Again, batch processing in ImageJ is recommended.

4. Loading time points as a time course (in ImageJ/Fiji)

   There are many ways of converting individually-saved DySTrack time points into a single “movie”.

   One simple option using standard ImageJ is via File >> Import >> Image Sequence...:

   • Select the folder with your data

   • Use the Filter field to pick out files you want to load. For example, if your maximum z-projection macro adds _zmax.tif as the file ending, type this into the Filter field and you should see that the Count field now displays the correct number of detected files (i.e. the number of time points).

   • Click OK to load the images in a single viewer.

   • If there are multiple channels and/or slices, use Image >> Hyperstacks >> Stack to Hyperstack... and type in the correct values to create A hyperstack.

   In Fiji, using File >> Import >> Bio-Formats and selecting the option Group files with similar names can be used to load data directly as a hyperstack (use e.g. File name | contains | _zmax.tif to get the same file filtering effect).

   Once loaded, you can customize and save the time course as usual.

   Note

   Loading many time points in 3D simultaneously will require a lot of memory and may be slow. Using a virtual stack can help with this, though we tend to just look at projections or at individual time points, depending on what we are looking for.

5. “Deregistration” of DySTracked data

   Sometimes it is of interest to visualize the actual movement of the tracked sample in real-space coordinates (see the bottom panel of the gif on the landing page), for example to measure the velocity or to create a kymograph. We refer to this as “deregistration”.

   The DySTrack repo comes with a Jupyter Notebook that performs deregistration on a dataset, using the coordinates in dystrack_coords.txt and the shape and pixel size of the prescan (must be provided). The notebook also provides some extra options to play around with.

   It is recommended to perform deregistration on z-projected images, as the resulting movies will tend to very large files.

6. High-quality post-registration and tracking

   Depending on the sample, acquisition settings, and image analysis pipeline, samples tracked with DySTrack may not necessarily be perfectly registered across time points. Indeed, DySTrack pipelines should be optimized for speed and robustness, not for high-quality registration.

   Thus, if high-quality registration is required, it is recommended to run it in post, rather than as a DySTrack pipeline during acquisition. This comes with the additional advantage that high-quality main scan data can be used to optimize the registration (as opposed to low-quality prescan data).

   The same applies to get high-quality tracking data (i.e. linked segmentations). It is usually preferable to segment and track objects in the high-quality main scan in post, without relying directly on the DySTrack coordinates. Depending on the smoothness of the raw data and the nature of the tracking algorithm, it may or may not be better to deregister the data (see above) prior to high-quality tracking.