Conference Paper

A two-step framework for the registration of HE stained and FTIR images

  • Francisco Peñaranda /
  • Valery Naranjo /
  • Rafaél Verdú /
  • Gavin R. Lloyd /
  • Jayakrupakar Nallala /
  • Nick Stone
Conference Proceeding cp
Progress in Biomedical Optics and Imaging - Proceedings of SPIE
  • Volumen: 9703
  • Fecha: 01 January 2016
  • ISSN: 16057422
  • ISBN: 9781628419375
  • Source Type: Conference Proceeding
  • DOI: 10.1117/12.2208869
  • Document Type: Conference Paper
  • Publisher: SPIE
© 2016 SPIE. FTIR spectroscopy is an emerging technology with high potential for cancer diagnosis but with particular physical phenomena that require special processing. Little work has been done in the field with the aim of registering hyperspectral Fourier-Transform Infrared (FTIR) spectroscopic images and Hematoxilin and Eosin (HE) stained histological images of contiguous slices of tissue. This registration is necessary to transfer the location of relevant structures that the pathologist may identify in the gold standard HE images. A two-step registration framework is presented where a representative gray image extracted from the FTIR hypercube is used as an input. This representative image, which must have a spatial contrast as similar as possible to a gray image obtained from the HE image, is calculated through the spectrum variation in the fingerprint region. In the first step of the registration algorithm a similarity transformation is estimated from interest points, which are automatically detected by the popular SURF algorithm. In the second stage, a variational registration framework defined in the frequency domain compensates for local anatomical variations between both images. After a proper tuning of some parameters the proposed registration framework works in an automated way. The method was tested on 7 samples of colon tissue in different stages of cancer. Very promising qualitative and quantitative results were obtained (a mean correlation ratio of 92.16% with a standard deviation of 3.10%).

Author keywords

    Indexed keywords

      Funding details