Skip to content

Latest commit

 

History

History
415 lines (313 loc) · 13.5 KB

README.md

File metadata and controls

415 lines (313 loc) · 13.5 KB
Raw

spimstitch

Stitch images taken on a diagonal using the SPIM microscope

Usage

The easiest way to use the pipeline is to run the spimstitch-pipeline.sh script. Pip installs this so it is runnable. These are easy directions for use in the Chung lab, e.g. for paths, /path-to/root-dir/Ex_488_Em_1 and /path-to/root-dir/Ex_562_Em_2 and a standard 4x objective with Y voxel size of 1.8 µm and X step size 2.0 µm (don't type the dollar sign):

$ cd /path-to/root-dir
$ source /home/build/anaconda3/bin/activate chunglab-stack
$ spimstitch-pipeline Ex_488_Em_1
$ spimstitch-pipeline Ex_562_Em_1

You can set the Y voxel size and X step size for the script. This is done with environment variables. For instance, for a X step size of 2.56 µm and Y voxel size of 3.62µm:

$ export X_STEP_SIZE=2.56
$ export Y_VOXEL_SIZE=3.62
$ cd /path-to/root-dir
...

Illumination correction and destriping

The pipeline will automatically create an illumination correction image by default. To override this behavior, set the environment variable, "ILLUM_CORR" to the path to the illumination correction file. For instance,

$ export ILLUM_CORR=/path-to/Ex_488_Em_1-illuc.tiff
...

The default filename for the illumination correction function is the channel name + "-illuc.tiff" (as seen above)

There are two methods for destriping in the pipeline, wavelet and lightsheet. "lightsheet" is used by default, but if you want to use wavelet (which preserves the autofluorescence signal), you can define the "USE_WAVELETS" environment variable to choose that method. You should also specify a background value that will be used as the darkfield value in destriping. A typical value is somewher between 40 and 200.

For wavelet:

$ export USE_WAVELETS=1
$ export BACKGROUND=100 # for instance

Alignment

The pipeline recalculates the Y_VOXEL_SIZE by default. This value is applied to all subvolumes when stitching. If you are processing multiple channels for a pipeline, you should use the alignment from the first for subsequent channels. The alignment is stored in a file whose name is the channel name + "-align.json" (for instance "Ex_488_Em_1-align.json").

It's best not to export this value to ensure that it doesn't inadvertently get applied to an unrelated volume. An example series of invocations:

$ spimstitch-pipeline.sh Ex_488_Em_1
$ ALIGN_FILE=Ex_488_Em_1-align.json spimstitch-pipeline.sh Ex_562_Em_2

Individual commands

dcimg2tif

dcimg2tif converts a .dcimg file to a .tiff stack.

Usage:

dcimg2tif \
  --input <dcimg-file> \
  --output-pattern <output-pattern> \
  [--compression <compression>] \
  [--n-workers <n-workers>] \
  [--rotate-90 <rotate-90>] \
  [--flip-ud] \
  [--start <start>] \
  [--stop <stop>]

where

  • dcimg-file is the name of the .dcimg file to be converted

  • output-pattern is the pattern for filenames for the .tiff files. The z-index of each tiff file is substituted in the pattern, for instance, "img_%05d.tiff".

  • compression is the tiff file compression level: 0-9, default=3

  • n-workers is the number of worker processes to use, default is a single process. dcimg2tif is I/O bound, so it's generally inefficient to use all cores in a machine and a number between 6 and 20 is probably a good choice.

  • rotate-90 the number of 90 degree clockwise rotations of each plane. For the oblique spim machines in the Chung Lab, this number should be "3". Default is 0.

  • flip-ud if present, this flips the image in the Y direction after rotating. This switch should be present for the oblique spim machines in the Chung Lab.

  • start The starting frame to be extracted. Defaults to 0

  • stop One past the last frame to be extracted. Defaults to the number of frames in the .dcimg file.

dcimg2jp2

The dcimg2jp2 command converts a dcimg file to a stack of JPEG 2000 files.

Usage:

dcimg2jp2 \
  --input <dcimg-file> \
  --output-pattern <output-pattern> \
  [--n-workers <n-workers>] \
  [--psnr <psnr>]

where:

  • input is the path to the DCIMG file to be converted

  • output-pattern is the pattern for the JPEG 2000 files to be created. dcimg2jp2 will create one level of directories if needed. Example: /path/to/img_%05d.jp2

  • n-workers is the number of worker processes that will be used. Default is to use all cores

  • psnr is a comma-separatedlist of signal-to-noise ratios for the levels in DB. The default is "30,40,50,60,70,80". End the list with "0" if you want lossless compression. A higher number will result in less compression, but higher fidelity.

stack2oblique

The stack2oblique command converts a stack of .tiff files to an de-obliqued precomputed volume, accounting for the 45° tilt of the camera relative to the stage motion.

Usage:

stack2oblique \
  --input <input-pattern> \
  --output <output-path> \
  [--levels <levels>] \
  [--log-level <log-level>] \
  [--n-writers <n-writers>] \
  [--n-workers <n-workers>] \

where

  • input-pattern is a glob expression to collect all of the stack frames, for instance, "Ex_488_Em_1_destriped/img*.tiff" (be sure to put the expression in quotes on the command-line)

  • output-path is the path to the precomputed directory for the output volume.

  • levels is the number of Neuroglancer pyramid levels to be created. A single level (1) is sufficient for use in oblique2stitched, but the default (5) may be better if you want to view the intermediate volume in Neuroglancer.

  • log-level is the level for logging output. Possible values are "DEBUG", "INFO", "WARNING" and "ERROR". The default is "WARNING".

  • n-writers is the number of writer processes to use. The default is 12, unless your computer has fewer CPUs.

  • n-workers is the number of worker processes to use for reading files. The default is 12, unless your computer has fewer CPUs.

oblique2stitched

The oblique2stitched command converts a group of oblique volumes into a single stitched volume. The oblique volumes must be organized in a hierarchy:

/<x>
   /<x>_<y>
     /<z>

where , and are the x, y and z coordinates of the start of the oblique volumes in 10ths of a micron.

Usage:

oblique2stitched \
  --input <input-path> \
  --output <output-path> \
  [--levels <levels>] \
  [--chunk-size <chunk-size>] \
  [--log-level <log-level>] \
  [--n-writers <n-writers>] \
  [--n-workers <n-workers>] \
  [--silent] \
  [--x-step-size <x-step-size>] \
  [--y-voxel-size <y-voxel-size>] \
  [--z-offset <z-offset>] \
  [--output-size <output-size>] \
  [--output-offset <output-offset>]

where

  • input-path is the path to the directory hierarchy of Neuroglancer subvolumes to be stitched

  • output-path is the path to the Neuroglancer volume to be created.

  • levels is the number of Neuroglancer pyramid levels to be created. The default is 5.

  • chunk-size is the size of a precomputed chunk in x,y,z format. The default is 64,64,64 for blockfs, 128,128,128 for ngff.

  • log-level is the level for logging, one of "DEBUG", "INFO", "WARNING" or "ERROR". The default is "WARNING".

  • n-writers is the number of writer processes to use. The default is the lesser of 12 or the number of CPUs on the computer.

  • n-workers is the number of worker processes to use. The default is the lesser of 12 or the number of CPUs on the computer.

  • silent if present will suppress printing of the progress bar

  • x-step-size is the X step size for the stage in microns

  • y-voxel-size is the size of a voxel for the CCD camera in microns

  • z-offset is the offset between subvolumes in the Z direction in pixels.

  • output-size is the size of the volume to create as x,y,z in pixels. The default is the entire volume.

  • output-offset is the offset of the written volume with respect to the subvolumes. The format is x,y,z (in pixels). The default is 0,0,0

oblique-illum-corr

Compute an illumination correction image from one or more .dcimg files. The resulting image can be used as the "--flat" input to pystripe.

The algorithm is as follows:

  • Create a histogram at every pixel of the values from a subset of the frames in all of the .dcimg files that are above the background value

  • At every pixel, take the value at a percentile of the histogram. This percentile should be fairly high (95% to 99.9%) to collect true foreground pixels (e.g. pixels within cells). This is the intermediate image.

  • Fit the intermediate image to the function, A x²+ B x + C y using a RANSAC estimator.

Usage:

oblique-illum-corr \
  --output <output-file> \
  [--intermediate-output <intermediate-output-file>] \
  [--n-frames <n-frames>] \
  [--n-bins <n-bins>] \
  [--values-per-bin <values-per-bin>] \
  [--min-samples <min-samples>] \
  [--percentile <percentile>] \
  [--background <background>] \
  [--rotate-90 <rotate-90>] \
  [--flip-ud] \
  <dcimg-file> [<dcimg-file>...]

where

  • output-file is the name of the .tiff file to be written

  • intermediate-output is the name of a .tiff file that holds the values at the percentile of the per-pixel histogram.

  • n-frames is the number of frames to select at random from among all the frames of all the dcimg files. The default is all of them.

  • n-bins is the number of bins per pixel in the histogram. This number should be high enough to capture the dynamic range, but low enough not to use too much memory. The memory consumed is 4 * image-width * image-height * n-bins. The default is 1024.

  • values-per-bin is the number of intensity values per bin. The dynamic range of the histogram is n-bins * values-per-bin* with values above this being clipped. For instance, if n-bins is 4 and values-per-bin is 1024, the dynamic range is 0 to 4095.

  • min-samples is the minimum number of samples to take in each RANSAC round. The default is 20.

  • percentile is the percentile value to take from the histogram of pixel values. For instance, a percentile of "95" will take the 95%th brightest value at each pixel. The default is 95.

  • background is the background cutoff. Pixels aren't included in the histogram if they fall below this number. The default is 150.

  • rotate-90 the number of 90 degree clockwise rotations of each plane. For the oblique spim machines in the Chung Lab, this number should be "3". Default is 0.

  • flip-ud if present, this flips the image in the Y direction after rotating. This switch should be present for the oblique spim machines in the Chung Lab.

  • dcimg-file is the path to a .dcimg file to include in the calculation. Multiple .dcimg files can be specified on the command line.

oblique-align

oblique-align estimates a y pixel size by aligning subvolumes. The algorithm is to find bright points at random in the overlapping regions between subvolumes and perform a gradient descent of the Pearson correlation coefficient between the two overlapping regions.

The result is the median value among all alignment estimates that have a final Pearson correlation coefficient greater than a cutoff value. The estimates and the result are written to a .json file. The voxel size can be read out of this file using the command:

Y_VOXEL_SIZE=`python -c "import json;print(json.load(open('"<align-file>"'))['voxel_size'])"`

Usage:

oblique-align \
  --input <precomputed-path> \
  --output <align-file> \
  [--voxel-size <y-voxel-size>] \
  [--x-step-size <x-step-size>] \
  [--is-oblique] \
  [--n-cores <n-workers>] \
  [--sigma <sigma>] \
  [--sample-count <sample-count>] \
  [--window-size <window-size>] \
  [--blob-detection-window-size <blob-detection-window-size>] \
  [--border-size <border-size>] \
  [--min-correlation <min-correlation>]

where

  • precomputed-path is the root path to the precomputed volumes. All precomputed subvolumes in all subdirectories under this will be used to compute the estimate. See oblique2stitched for details on directory layout.

  • align-file is the JSON file containing the calculated alignment.

  • y-voxel-size the nominal y voxel size. The alignment will improve upon this value.

  • x-step-size the x step size in microns.

  • is-oblique this flag should be specified for volumes created by stack2oblique.

  • n-cores the number of processes to use when doing the gradient descents.

  • sigma the smoothing sigma for the blob detector and for the image fed into the gradient descent. The value is in microns. The default is 2.5

  • sample-count the number of samples to be taken per overlapped region. The default is 20.

  • window-size the window size for the Pearson correlation calculation as "x,y,z". The default is 21,21,21. All values must be odd.

  • blob-detection-window-size the window size for the blob detector. Random points are chosen in the overlap region and blobs are detected within this window around the points. The format is x,y,z and the default is 64,64,64.

  • border-size is a border to each side of the target for the gradient descent. A smaller number fetches fewer pixels initially, but a larger number fetches more pixels, allowing for a longer travel along the gradient before fetching more pixels. The format is x,y,z and the default is 10,10,10.

  • min-correlation is the minimum allowed Pearson correlation coefficient. All estimates whose correlation coefficient is below this value are discarded. The default is .95.