Semi-automatic approach for brain tissue segmentation using MRI (2008)
AUTHORS:
Gonçalves Nicolau,
Vigário Ricardo
BOOKTITLE:
1st INCF Congress of Neuroinformatics: Databasing and Modeling the Brain (Neuroinformatics 2008)
PAGES:
106
@inproceedings{ Goncalves2008-INCF, author = "Gonçalves, Nicolau and Vig{\'a}rio, Ricardo", publisher = "Springer", doi = "10.3389/conf.neuro.11.2008.01.074", responsibleperson = "Nicolau Gon\c{c}alves", title = "Semi-automatic approach for brain tissue segmentation using {MRI}", timestamp = "2009.05.28", booktitle = "1st INCF Congress of Neuroinformatics: Databasing and Modeling the Brain (Neuroinformatics 2008)", address = "Stockholm, Sweden", abstract = "Magnetic resonance imaging is a widely used non-invasive diagnostic tool, which requires expert evaluation to assess the severity of brain lesions. In this paper, a method is presented for semi-automatic detection of those lesions. The goal is to provide a reliable tool for lesion detection, by improving tissue contrast and visualisation. The final objective is to reduce the intensive need for specialists, and allowing a rather systematic follow-up of the lesion evolution, or its treatment. Our approach is based on a combination of machine learning techniques. First, independent component analysis is used to allow for tissue segmentation. Afterwards, self-organising maps are applied, in order to create labels for the subsequent tissue classification. Then, two classification methods are experimented and compared: support vector machines and discriminative clustering, with special emphasis on the latter. The results indicate that the use of discriminative clustering allows for a good tissue classification, with the ability to even detect small isolated lesions, usually not detected in region-growth techniques.", month = "September", note = "Poster", owner = "ngoncalv", flags = "AIRC", file = "NGoncalves-2008-INCF.pdf:NGoncalves-2008-INCF.pdf:PDF", year = "2008", keywords = "brain MRI, discriminative clustering, tissue segmentation, unsupervised classification, MS lesion", impactfactor = "D4", pages = "106" }