Preparing images to work with NBLAST

[Greg Jefferis]

Both the command line and now the online versions of NBLAST (https://github.com/jefferislab/NBLAST_on-the-fly/pull/65 thanks to josesho) can carry out searches based on image data. However 3 simple preprocessing steps are recommended 

1. Threshold
   
2. Downsample
   
3. Skeletonise
   
4. Save as compressed Nrrd

All these operations can be carried out in ImageJ/Fiji. A fifth option, 

5. Cropping the image

cannot be carried out in Fiji

Details:

1. Threshold image so that foreground pixels are 255 and background 0 (i.e. Dark Background). You can use Image ... Adjust ... Threshold with the Dark Background option ticked

2. Downsample images to somewhere around 1 x 1 x 1 µm to 1.5 x 1.5 x 1.5 µm for fly neurons. You can use Image ... Scale for this purpose. It is critical that whatever scaling method you use adjusts the voxels size of the image appropriately (you can fix it yourself before saving the image with Image ... Properties)

3. The image will be prepared for searching by the R function nat::dotprops. All non-zero voxels will result in a vector being generated in this representation. Skeletonisation is required to turn thicker lines into one voxel wide lines which are optimal for NBLAST, increasing the speed and sensitivity of queries.

You can use Plugins ... Skeleton ... Skeletonize 2D/3D for this purpose

4. You can save the image as a compressed nrrd. It will be much smaller (typically 100-1000x) smaller by holding down the Alt key before you choose the File ... Save as ... Nrrd option. 

5.  Cropping to the smallest region containing non-zero voxels may also be helpful, but Fiji does not handle non-zero origins so the cropped image. For these purposes I have use the unu command line tool, CMTK's convertx tool or a wrapper for convertx in R AutoCropNrrd.

Best wishes,

Greg.