HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Download to citation manager Google Scholar Articles by Hatt, M. Articles by Visvikis, D. PubMed Articles by Hatt, M. Articles by Visvikis, D.

Automatic volume delineation of heterogeneous tumours in PET: Comparison of various methodologies

¹ LaTIM U650, INSERM, Brest, Finistère, France; ² ENST Bretagne, GET-ENST, Brest, Finistère, France

1609

Objectives: Automatic volume delineation methodologies previously suggested for PET are essentially threshold-based. We have developed an automatic segmentation algorithm based on a combination of statistical and fuzzy modelling that was previously shown to perform accurately, independent of image noise and contrast characteristics, on a phantom with spherical lesions and uniform activity distributions. In the present study we investigated the ability of this algorithm, in comparison to others, to delineate inhomogeneous and non-spherical tumours.

Methods: We compared four methods: thresholding (fixed and adaptive), Fuzzy C-Means (FCM), and our FLAB (Fuzzy Locally Adaptive Bayesian) method. In the case of FCM and FLAB, two implementations, using 2 or 3 classes were considered (for homogeneous or heterogeneous objects respectively). 15 simulated tumours generated in 3D using PET images of patient lesions were used in the evaluation. Voxels misclassifications and volume errors were computed.

Results: Threshold-based approaches lead to large errors and variability in the results with large dependence upon noise, contrast or lesion size. Although FCM performed better it was unable to correctly segment small objects and shown reduced accuracy in the delineation of complex shapes. On the other hand FLAB lead to the best results with low errors (<10% volume errors) and good accuracy in delineation even when dealing with heterogeneous activity distribution and in the presence of necrotic lesions.

Conclusions: FLAB outperformed other previously proposed methodologies for the complex task of delineating realistic, non spherical tumours. Furthermore its 3-class version dealt accurately with non-uniform activity distributions, providing a solution for future applications such as "dose painting" in radiotherapy treatment planning.

This Article Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Download to citation manager Google Scholar Articles by Hatt, M. Articles by Visvikis, D. PubMed Articles by Hatt, M. Articles by Visvikis, D.