About

tomostream is Python module for supporting streaming analysis of tomographic data where all pre-processing and reconstruction procedures are performed in real time while images are collected and the rotary stage is moving. tomostream provides this main functionality:

  • Streaming reconstruction of 3 X-Y-Z ortho-slices through the sample
    The streaming reconstruction engine generates 3 selectable X-Y-Z orthogonal planes and makes them available as an EPICS PV viewable in ImageJ using the EPICS_NTNDA_Viewer plug-in. Projection, dark and flat images used for the reconstruction are taken in real time from a set of PV access variables (pvapy) and stored in a synchronized queue. On each reconstruction call new data are taken from the queue, copied to a circular GPU buffer containing projections for a 180 degrees interval, and then reconstructed.

All tomostream functionalies can be controlled from the tomoStream user interface:

_images/tomoStream.png

tomostream relies on tomoscan_stream_2bm (part of tomoScan) for:

  • Tomography instrument control

  • Projection, dark and flat image broadcast as PV access variables

  • On-demand retake of dark-flat field images

  • On-demand data capturing with saving in a standard hdf5 DXfile file

  • Set a number of projectons (“Pre count”) collected before a triggered data capturing event to be also saved in the same hdf5 file

All tomoscan_stream_2bm functionalies supporting tomostream can be controlled from the tomoScanStream user interface marked in yellow:

_images/tomoScanStream.png

The output of tomostream is a live reconstruction diplaying in ImageJ using the EPICS_NTNDA_Viewer plug-in:

_images/tomoStreamRecon.png

While the sample is rotating is possible to optimize instrument (alignment, focus, sample to detector distance etc.) and beamline (energy etc.) conditions and monitor the effect live on the 3 orthogonal slices. It is also possible to automatically trigger data capturing based on events occurring in the sample and its environment as a result of segmentation or machine learning.