The use of the Kalman filter in the automated segmentation of EIT lung images

A. Zifan, P. Liatsis, B. E. Chapman

Research output: Contribution to journalArticlepeer-review

8 Scopus citations

Abstract

In this paper, we present a new pipeline for the fast and accurate segmentation of impedance images of the lungs using electrical impedance tomography (EIT). EIT is an emerging, promising, non-invasive imaging modality that produces real-time, low spatial but high temporal resolution images of impedance inside a body. Recovering impedance itself constitutes a nonlinear ill-posed inverse problem, therefore the problem is usually linearized, which produces impedance-change images, rather than static impedance ones. Such images are highly blurry and fuzzy along object boundaries. We provide a mathematical reasoning behind the high suitability of the Kalman filter when it comes to segmenting and tracking conductivity changes in EIT lung images. Next, we use a two-fold approach to tackle the segmentation problem. First, we construct a global lung shape to restrict the search region of the Kalman filter. Next, we proceed with augmenting the Kalman filter by incorporating an adaptive foreground detection system to provide the boundary contours for the Kalman filter to carry out the tracking of the conductivity changes as the lungs undergo deformation in a respiratory cycle. The proposed method has been validated by using performance statistics such as misclassified area, and false positive rate, and compared to previous approaches. The results show that the proposed automated method can be a fast and reliable segmentation tool for EIT imaging.

Original languageBritish English
Pages (from-to)671-694
Number of pages24
JournalPhysiological Measurement
Volume34
Issue number6
DOIs
StatePublished - Jun 2013

Keywords

  • EIT
  • ill-posed problems
  • impedance
  • Kalman filter
  • regularization
  • segmentation

Fingerprint

Dive into the research topics of 'The use of the Kalman filter in the automated segmentation of EIT lung images'. Together they form a unique fingerprint.

Cite this