Role of fuzzy pre-classifier for high performance LI/MA segmentation in B-mode longitudinal carotid ultrasound images

Annu Int Conf IEEE Eng Med Biol Soc. 2010:2010:4719-22. doi: 10.1109/IEMBS.2010.5626390.

Abstract

The automated segmentation of the carotid artery wall from ultrasound images is required for an accurate measurement of the artery intima-media thickness. Segmentation accuracy of automated techniques is usually limited by noise and artifacts. In 2005, the authors developed a methodology called CULEX whose performance was noise sensitive. The final stage of CULEX segmentation was fuzzy clustering of the pixels, to detect the lumen-intima (LI) and media-adventitia (MA) carotid wall interfaces. In this paper, we show the effect of a fuzzy Mamdani-type pre-classifier used to improve the segmentation performance. Thanks to the Mamdami fuzzy pre-classifier, we optimized the de-fuzzyfication threshold, increasing the LI and MA performance by 62% and 3.5%, respectively. The obtained segmentation errors (55.6 ± 69.4 microm for LI and 34.4 ± 24.4 microm for MA), validated against human tracings and on a 200-images dataset containing a mixture of healthy and plaque vessels.

Publication types

  • Evaluation Study

MeSH terms

  • Algorithms
  • Carotid Arteries / diagnostic imaging*
  • Carotid Artery Diseases / diagnostic imaging*
  • Fuzzy Logic*
  • Humans
  • Image Enhancement / methods*
  • Image Interpretation, Computer-Assisted / methods*
  • Pattern Recognition, Automated / methods*
  • Reproducibility of Results
  • Sensitivity and Specificity
  • Ultrasonography / methods*