Thrombus segmentation by texture dynamics from microscopic image sequences

Nicolas Brieu, Jovana Serbanovic-Canic, Ana Cvejic, Derek Stemple, Willem Ouwehand, Nassir Navab, Martin Groher

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

4 Scopus citations

Abstract

The genetic factors of thrombosis are commonly explored by microscopically imaging the coagulation of blood cells induced by injuring a vessel of mice or of zebrafish mutants. The latter species is particularly interesting since skin transparency permits to non-invasively acquire microscopic images of the scene with a CCD camera and to estimate the parameters characterizing the thrombus development. These parameters are currently determined by manual outlining, which is both error prone and extremely time consuming. Even though a technique for automatic thrombus extraction would be highly valuable for gene analysts, little work can be found, which is mainly due to very low image contrast and spurious structures. In this work, we propose to semi-automatically segment the thrombus over time from microscopic image sequences of wild-type zebrafish larvae. To compensate the lack of valuable spatial information, our main idea consists of exploiting the temporal information by modeling the variations of the pixel intensities over successive temporal windows with a linear Markov-based dynamic texture formalization. We then derive an image from the estimated model parameters, which represents the probability of a pixel to belong to the thrombus. We employ this probability image to accurately estimate the thrombus position via an active contour segmentation incorporating also prior and spatial information of the underlying intensity images. The performance of our approach is tested on three microscopic image sequences. We show that the thrombus is accurately tracked over time in each sequence if the respective parameters controlling prior influence and contour stiffness are correctly chosen.

Original languageEnglish
Title of host publicationMedical Imaging 2010
Subtitle of host publicationImage Processing
EditionPART 1
DOIs
StatePublished - 2010
EventMedical Imaging 2010: Image Processing - San Diego, CA, United States
Duration: 14 Feb 201016 Feb 2010

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
NumberPART 1
Volume7623
ISSN (Print)1605-7422

Conference

ConferenceMedical Imaging 2010: Image Processing
Country/TerritoryUnited States
CitySan Diego, CA
Period14/02/1016/02/10

Keywords

  • Dynamic texture
  • Microscopic imaging
  • Segmentation

Fingerprint

Dive into the research topics of 'Thrombus segmentation by texture dynamics from microscopic image sequences'. Together they form a unique fingerprint.

Cite this