Registration Accuracy

For the Medical Device Development Tool submission, we will need to characterize the registration accuracy of eeDAP.

SPIE abstract, poster, and proceedings paper

Over the summer we have expanded our registration accuracy efforts. In August 2017 we submitted an abstract based on this work to present at the SPIE Medical Imaging conference Feb 10-15, 2018. https://spie.org/conferences-and-exhibitions/medical-imaging?SSO=1

T-con 12 June 2017

Qi presented a shorter version of the Pathology Informatics presentation for those that could not make the conference, here is the presentation eeDAP_registration_final.pdf (4 MB, uploaded by Brandon D. Gallas 2 years 7 months ago). The last slide outlines our plans to complete the study. We wanted feedback on the final plans. The group was satisfied with the plans and offered the following feedback.

* Record how long the process takes. * Record any problems encountered during data collection. * Get some feedback from the person collecting the data.

The Philips group asked why we were not performing this study with a phantom slides. Our main response is that the contrast, resolution, and content of a phantom slide is very different from that found in the slides for a study. This appears to limit eeDAP to similar tissue types. We would recommend that the registration accuracy be performed once before an actual data collection study is begun to mitigate any registration issues. This pre-study check allows for the study PI to evaluate the registration accuracy for the slides and FOVs of the study to be performed.

The Philips group is developing a registration accuracy method using phantom slides. Phantom slides might be extremely useful for evaluating the accuracy of the stage motion (no re-registration) and could possibly be automated. This kind of information would be very useful.

Presentation at Pathology Informatics 21 May 2017

Qi presented the registration accuracy method and initial results. The presentation can be found Here.

Initial Description 11 April 2017

Marios and Brandon have two ideas for characterizing registration precision.

  1. 1-4 slides mounted on the stage. A bulls eye reticle. Run eeDAP for some number of cell nuclei or other small feature. A human determines where the small feature lands in the bulls eye.
    • We have started to outline a study along the lines of this idea. Here is a summary.
  2. A micro-bead slide: a mixture of at least two beads, one big (size of a cell, 10 um?) and the other as small as can be made. The small spheres would be rare and would be the anchors/fiducial marks for determining registration precision. Image processing could be used to measure the registration precision. This could be done offline and after-the-fact because eeDAP takes a snap shot of the camera view such that the camera view can be saved.

Created on , Last modified on