ZXY DimensionOrder Data Storage

[CPhillips]

Greetings,

I am working on an Optical Coherence Tomography (OCT) project and we are considering using OME(big)TIFFs to store the data. As the current project is designed, Z indicates depth into the sample and with scan with X as the fast axis and Y as the slow axis. Therefore we end up with ZX images stacked along the Y axis.

Is there a recommended way to store the data in a way that is congruent with our current definitions, or would we be better off redefining the axis definitions in our program such that depth is Y, fast axis is X, and slow axis is Z? I simply want to minimize the mental load for users so they don't have to transpose the dimensions in their head when analyzing the data.

Thanks,
Conner

[sbesson]

Hi Conner

I am working on an Optical Coherence Tomography (OCT) project and we are considering using OME(big)TIFFs to store the data. As the current project is designed, Z indicates depth into the sample and with scan with X as the fast axis and Y as the slow axis. Therefore we end up with ZX images stacked along the Y axis.

Thanks for starting the discussion. Your use case is actually similar to a few other imaging modalities where the data acquisition does not fit the convential XY[other] order e.g. OPT or lightsheet.

Is there a recommended way to store the data in a way that is congruent with our current definitions, or would we be better off redefining the axis definitions in our program such that depth is Y, fast axis is X, and slow axis is Z? I simply want to minimize the mental load for users so they don't have to transpose the dimensions in their head when analyzing the data.

An example of how such things can be represented with the current technology can be illustrated by this IDR dataset. This is a lightsheet 5D volume (XYZ, time and channels) with orthogonal projections alongside XY, XZ, and YZ. Each projection is internally stored using and represented as a separate 4D image and the 4 images in the fileset have implicit spatial relationships between themselves.

It would be interesting to hear more about your pipeline and the users expectation. Would these images be available via viewing tools? What type of downstream analysis are your performing?

Best,
Sebastien

[CPhillips]

Hi Sebastian,

I will take a look at the dataset you provided and reply back once I have had the chance.

The solution for the old system was saving the IFDs in the order they were obtained (ZX along Y axis) and generating an ImageJ ImageDescription tag that allows ImageJ to read the IFDs as a stack, e.g.:

Code: Select all

ImageJ=1.43u
images=1024
slices=1024
unit=um
spacing=10
loop=false

As it stands the tools are mostly used for image filtering and qualitative visual defect detection by an operator. Automatic processing and defect detection, or at least ROI marking will likely be investigated down the road.