Journal of Neuroscience Methods 182 (2009) 110–122 Contents lists available at ScienceDirect Journal of Neuroscience Methods journal homepage: www.elsevier.com/locate/jneumeth Gaussian mixture model-based segmentation of MR images taken from premature infant brains Harri Merisaari a,b,∗ , Riitta Parkkola b,c , Esa Alhoniemi a , Mika Teräs b , Liisa Lehtonen d , Leena Haataja d , Helena Lapinleimu d , Olli S. Nevalainen a a Department of Information Technology and Turku Centre for Computer Science (TUCS), FI-20014 University of Turku, Finland b Turku PET Centre, Turku University Central Hospital, FI-20521 Turku, Finland c Department of Radiology, University of Turku, FI-20521 Turku, Finland d Department of Pediatrics, University of Turku, FI-20521 Turku, Finland article info Article history: Received 22 December 2008 Received in revised form 25 May 2009 Accepted 27 May 2009 Keywords: Premature infant MRI Cerebrospinal fluid segmentation Watershed method abstract Segmentation of Magnetic Resonance multi-layer images of premature infant brain has additional chal- lenges in comparison to normal adult brain segmentation. Images of premature infants contain lower signal to noise ratio due to shorter scanning times. Further, anatomic structure include still greater vari- ations which can impair the accuracy of standard brain models. A fully automatic brain segmentation method for T1-weighted images is proposed in present paper. The method uses watershed segmenta- tion with Gaussian mixture model clustering for segmenting cerebrospinal fluid from brain matter and other head tissues. The effect of the myelination process is considered by utilizing information from T2- weighted images. The performance of the new method is compared voxel-by-voxel to the corresponding expert segmentation. The proposed method is found to produce more uniform results in comparison to three accustomary segmentation methods originally developed for adults. This is the case in particular when anatomic forms are still under development and differ in their form from those of adults. © 2009 Elsevier B.V. All rights reserved. 1. Introduction Magnetic Resonance Imaging (MRI) is a non-invasive method for analysing human anatomy. The method is based on quantum properties of nuclei. When exposed to a high magnetic field, small magnetic fields of the nuclei in the target object become aligned to the high magnetic field. Then, Radio Frequency (RF) pulses are sent to the object in order to increase the energy state of nuclei. After the pulse is switched off, the energy states decrease which is mea- sured by a scanner. Depending on the parameters of this process such as the frequency and timing of RF pulses, 3D images repre- senting different characteristics of tissues can be acquired, such as so-called T1- and T2-weighted MR image modalities. Hydrogen proton 1 H is a commonly observed nucleus in medical MR imag- ing, since it is associated with water molecules and human body contains plenty of water in its tissues. In brain imaging, the inter- est is at bone of the skull, brain gray matter (GM), brain white matter (WM) and cerebrospinal fluid (CSF). Head tissues have char- acteristic intensity values in MR images depending on, for example, how much water the tissue contains. This information can be used ∗ Corresponding author at: Signalistinkatu 17 c 26, 20360 Turku, Finland. Tel.: +358 40 577 9999. E-mail address: harri.merisaari@utu.fi (H. Merisaari). in separation and localization of the different tissues. For exten- sive information about physical properties of MR scanning, see (Dhawan, 2003a). Image segmentation into the four main regions (GM, WM, CSF, non-brain) is a fundamental step in the analysis of MR brain images. Segmentation of MR images is useful, for example, as a preprocess- ing step for a large variety of image processing applications, such as normalization and co-registration of images for comparative statis- tical analysis. In the present study we are interested in MR images of premature infant brain, as a part of a research project for long range follow up of the development of premature infants. While the main interest here lies in the volume of CSF, 1 we want also to extract the WM and GM. Segmentation of brain MR images of premature infants can be considered to be significantly more demanding than segmentation of normal adult brains. The MR scanning times are shorter than with adults because longer scanning times might not be comfortable for infants unless they are asleep. Because of shorter scanning time the signal to noise ratio is smaller. 2 1 In segmentation software the CSF segment is sometimes defined to contain all the regions not belonging to either GM or WM. 2 Head motion by the patient is an other source for bad images. We suppose in the present work that this kind of material has been previously filtered out. 0165-0270/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jneumeth.2009.05.026