Semi-supervised Image-to-Image translation for robust image registrationDownload PDF

Apr 09, 2021 (edited Apr 20, 2021)MIDL 2021 Conference Short SubmissionReaders: Everyone
  • Keywords: Semi-supervised image-to-image translation, GAN, Image registration
  • TL;DR: A semi-supervised image-to-image translation made possible an automated, reliable multi-modal image registration of thousands of microscopy marmoset brain images
  • Abstract: The Japan Brain/MINDS Project aims at studying the neural networks controlling higher brain functions in the marmoset. As part of it, we develop an image processing pipeline for marmoset brain imaging data, where various microscopy images of different modalities need to be co-registered. In initial experiments, multi-modal image registration frequently failed due to an erroneous initialization. Our data set includes images of Nissl stained brain sections, backlit images as well as images of neural tracer injections using two-photon microscopy. More than 10000 high-resolution 2D images required co-registration, a large amount that demands a reliable automation process. We implemented a semi-supervised image-to-image translation which allowed a robust image alignment initialization. With such an initial alignment, all images can be successfully registered using a state-of-the-art multi-modal image registration algorithm.
  • Paper Type: validation/application paper
  • Primary Subject Area: Image Registration
  • Secondary Subject Area: Transfer Learning and Domain Adaptation
  • Paper Status: original work, not submitted yet
  • Source Code Url:
  • Data Set Url: Data will be made publicly available in 2021/ 2022
  • Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
  • Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
4 Replies