Segmentation of MR images by a fuzzy c-mean algorithm
Authors:
- 19 November 2002
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 21-24
Abstract
Since the manual delineation of the left ventricular (LV) contour in MR images is subject to intra- and inter-observer variations, an automated procedure is proposed. The fuzzy c-mean clustering approach was used to segment the images. The segments were labeled as either the left ventricular lumen or the background by using additionally the information provided by the Hough transform which delivered a rough estimation of the center of the ventricle. To find an optimal agreement between the manually and automatically delineated LV contours, 2700 combinations of different parameters were used in a set of 20 images of patients and normal subjects. An excellent correlation coefficient (r=0.95) was found when parameters were optimized for each individual image. However if the parameter set was fixed for all images, the correlation coefficient decreased to r=0.8. This suggests that a cluster validation measure must be defined for a good performance. This is the topic of further research.Keywords
This publication has 6 references indexed in Scilit:
- Detection and separation of ring-shaped clusters using fuzzy clusteringIEEE Transactions on Pattern Analysis and Machine Intelligence, 1994
- The fuzzy c spherical shells algorithm: A new approachIEEE Transactions on Neural Networks, 1992
- Adaptive fuzzy c-shells clustering and detection of ellipsesIEEE Transactions on Neural Networks, 1992
- Comparative study of Hough Transform methods for circle findingImage and Vision Computing, 1990
- A survey of the hough transformComputer Vision, Graphics, and Image Processing, 1988
- Generalizing the Hough transform to detect arbitrary shapesPattern Recognition, 1981