Data preprocessing

Cropping and scaling

If the voxel size of your data is very small, there may not be enough memory on your machine to run the registrations. It may also lead to undesirable processing times. If this is the case, you may want to scale your data.

After the data has been reconstructed, there is usually quite a bit of empty space around the embryo, which should be removed in order to reduce volume file size.

Both these tasks can be done if the reconstructions are processed using HARP

When processing our E14.5 embryo data, we scale the voxel size to 14μm and our E18.5 data to 28μm

Volume size

Population average

Before make a population average, all input volumes must of the same dimensions. The following command will pad volumes to the largest extent each dimensions across all the volumes.

To pad all volumes (in place) to the same size use the following. (see Padding) for more options

$ lama_pad_volumes.py -i inputs_dir --clobber

Registration of the baseline and mutants

Before running job_runner or lama_reg, all the baseline and mutant volumes, as well as population average target and associated volumes (eg. mask and labels) must all have the same shape as the population average.

$ lama_pad_volumes.py -i wt_dir mut_dir target_dir --clobber

Bit depth

The images should all be the same bit-depth. The volumes that come off the CT scanner can often be 16bit. To save memory and CPU time, we can cast the volumes to 8bit before running LAMA See casting volumes