Semi-automatic approach for brain tissue segmentation using MRI
-
1
Helsinki University of Technology, Finland
Introduction: Magnetic resonance imaging (MRI) is a widely used non-invasive imaging and diagnostic tool, which requires expert evaluation to assess the severities of brain lesions. We present a method for semi-automatic identification of brain tissues in MRI. Our approach uses self-organizing maps (SOMs) for voxel labelling, which are used to seed the discriminative clustering (DC) classification algorithm.
This approach greatly improves upon [1].
Data: We used simulated brain MRI from the BrainWeb database [2], with 1 mm3 of spatial resolution. T1, T2 and PD sequence images, with 3% of noise and no intensity nonuniformity were used. The only tissues retained for analysis were white-matter, grey-matter and cerebrospinal fluid.
Methods: SOMs perform a lattice projection that preserves similarity information from the input space, through competitive and Hebbian learning rules [3]. After training multiple randomly initialised SOMs, voxels consistently clustered together were selected as representing stable properties in the image input space. This is followed by a very limited human interaction step, grouping reliable clusters corresponding to identical tissue types.
The final classification of the voxels is done with DC [4]. It's auxiliary supervising data are the labels found with SOM. The goal in DC is to partition the input data space into clusters that are local and homogeneous in terms of their auxiliary data. Locality is enforced by defining the clusters as Voronoi regions in the primary data space, while homogeneity is enforced by searching for different sets of partitions, capable of predicting the auxiliary data with distributional prototypes for each region. The DC classification was done with three prototypes, since this was the number that led to the best visual representation and we have only three tissue types.
DC allows for memberships in different clusters, as well as overlaps between the grey levels of different tissue voxels.
Results and Discussion: The percentage of wrongly classified voxels in the final classification was 3.42%. As can be seen in Fig.1, all the tissues present are well classified, and only the borders contain voxels that are composed of different tissues (note the darker colour therein). These results demonstrate that the methodology proposed here allows for a suitable tissue classification, and opens up good perspectives for classifying other tissues such as brain lesions. Also real images, where borders between tissues are not so evident, should become easier to analyse.
References
1. ] E. Karp and R. Vig?rio, "Unsupervised MRI Tissue Classification by Support Vector Machines", in Proceeding 2nd IASTED Int. Conf. on Biomedical Engineering, Austria, 2004, pp. 88-91
2. http://www.bic.mni.mcgill.ca/brainweb
3. T. Kohonen, "Self-Organizing Maps", 3rd ed. Springer, 2001
4. S. Kaski, J. Sinkkonen, and A. Klami, "Discriminative clustering", Neurocomputing, vol. 69, pp. 18-41, 2005
Conference:
Neuroinformatics 2008, Stockholm, Sweden, 7 Sep - 9 Sep, 2008.
Presentation Type:
Poster Presentation
Topic:
Neuroimaging
Citation:
Gonçalves
N and
Vigário
R
(2008). Semi-automatic approach for brain tissue segmentation using MRI.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2008.
doi: 10.3389/conf.neuro.11.2008.01.074
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
28 Jul 2008;
Published Online:
28 Jul 2008.
*
Correspondence:
Nicolau Gonçalves, Helsinki University of Technology, Espoo, Finland, ngoncalv@cis.hut.fi