Paper: | SU-AM-PS4.8 |
Session: | Image Segmentation, Retrieval and Analysis |
Time: | Sunday, April 9, 10:50 - 12:10 |
Presentation: |
Poster
|
Title: |
Fuzzy Versus Hard Hidden Markov Chains Segmentation for Volume Determination and Quantitation in Noisy PET Images |
Authors: |
Mathieu Hatt; LaTIM INSERM U560 | | |
| Nicolas Boussion; LaTIM INSERM U560 | | |
| Frederic Lamare; LaTIM INSERM U560 | | |
| Christophe Collet; ENSPS LSIIT UMR CNRS 7005 | | |
| Fabien Salzenstein; InESS CNRS/ULP - UMR 7163 | | |
| Christian Roux; LaTIM INSERM U650 | | |
| Yves Bizais; LaTIM INSERM U560 | | |
| Catherine Cheze-Le Rest; LaTIM INSERM U560 | | |
| Dimitris Visvikis; LaTIM INSERM U560 | | |
Abstract: |
Accurate volume contouring in PET is crucial for quantitation in oncology applications. The objective of our study was to compare the performance of two algorithms for automatic lesion volume delineation that permit noise modelling and have not previously been applied to PET data; namely the Hidden Markov Chains (HMC) model and a novel version: Fuzzy HMC. Both models take into account noise, voxel’s intensity and spatial correlation, in order to classify a voxel as “background” or “functional volume”. In comparison to HMC which only models uncertainty, the novelty of the fuzzy model consists of the inclusion of an estimation of imprecision. The performance of both models was compared using realistic simulated images. Results demonstrate that FHMC performs better than HMC in both quantitation as well as functional volume determination under different imaging conditions. |