Generating MIPs and defining ROIs

A top-down maximum-intensity projection (MIP) is often all you need to plan an analysis — to get a quick overview of a very large dataset, or to draw the cropping ROIs that later runs will process. On the Zeiss LLS7 a MIP is written automatically at the end of each acquisition, but you can also generate a deskewed MIP yourself from any supported dataset with lls-pipeline --save-mip.

This walkthrough goes end to end: generate a MIP, draw ROIs on it in Fiji, and feed those ROIs back into a cropping run.

MIP generation is CLI/API only

--save-mip is available on the command line and through the Python API. It is not currently a checkbox in the napari plugin.

1. Generate a deskewed MIP

lls-pipeline --save-mip /path/to/input.tiff --save-dir /path/to/output

The MIP is computed straight from the raw data — each projected pixel is mapped back to its source voxels and the maximum is accumulated along the axial direction. The full deskewed volume is never materialised, so this path is memory-light and fast, which makes it practical on terabyte-scale acquisitions.

What you get and the knobs that matter:

One MIP per timepoint and channel, written using --save-type (tiff, h5 or omezarr).
Smoothness — --mip-interpolation nearest (default; fastest, blocky) or --mip-interpolation linear to blend adjacent scan planes for a smoother MIP.
Geometry aware — works with --skew X/--skew Y and both coverslip-rotation modes, so a MIP from OPM data uses the same --no-coverslip-rotation you deskew with (see the OPM example).
Cropping and deconvolution are ignored on the MIP path — it is a fast whole-frame projection, not a per-ROI product.

2. Draw ROIs on the MIP in Fiji or napari

Defining ROIs in Fiji

Open the MIP in Fiji and draw a rectangle around each region you want to process, adding each to the ROI Manager, then save the ROI Manager as a .zip. If it is only one ROI, this will be saved as a .roi file

ROIs for Zeiss LLS7

For the Zeiss LLS7, the ROIs need to be rotated 90° before use. You can either rotate the MIP image first and then draw, or rotate an existing ROI set with the supplied Fiji macro. The full procedure, with screenshots and the macro, is in Defining ROIs for cropping. This is only for the Zeiss lattice lightsheet 7 as of now (2026).

Defining ROIs in napari

Open the MIP in napari. Create a shapes layer and draw rectangular polygons around each object of interest. Once you are done, you can save the layer as a .csv file which can be accessed by napari-lattice.

Click here for more info on napari shapes layer.

3. Feed the ROIs back into a cropping run

With the rotated ROI file in hand, point a normal (non-MIP) lls-pipeline run at it. This run deskews and crops the full volume — only the ROIs you drew are processed:

# crop_config.yml
input_image: "/path/to/input.tiff"
save_dir: "/path/to/output/cropped"
save_type: "h5"
crop:
  roi_list: "/path/to/rois.zip"

If it is a napari shapes layer saved as a csv

# crop_config.yml
input_image: "/path/to/input.tiff"
save_dir: "/path/to/output/cropped"
save_type: "h5"
crop:
  roi_list: "/path/to/rois.csv"

ROIs and pixel scaling

napari-lattice expects the Fiji ROIs to be in pixel scaling and the napari shapes csv to be in microns. If you need to override this you can use the roi_units argument. You can specify pixel or micron. Refer to ROI section in the CLI page.

lls-pipeline process --yaml-config crop_config.yml

To process only some ROIs, or to run many in parallel, see ROI selection and parallel ROI processing.