DE URGENCIAS SUICIDAS.
5.2.2.6. Sin clasificación en coyunturas desencadenantes en estudio.
2.7.1 Brain volumetry
Brain volumetry typically involves segmentation and quantification of various tissue types of the
brain such as GM, WM and CSF. Other Global metrics include total intracranial volume (TIV)
and total brain volume (TBV) which are volume estimates of the intracranial space and the brain
parenchyma respectively. These metrics are called global metrics as they are computed for the
brain as a whole which includes the brain cortex, subcortical regions, cerebellum, brainstem and
for TIV the fluid surrounding these structures. Regional brain volume metrics comprise of volumes
of anatomical brain structures like hippocampus, amygdala, thalamus, nucleus accumbens, etc
which constitute the subcortical structures. Other types of regional brain volumes measures include
lobes and regions that represent various functions like speech, vision, memory etc.
The volume of a brain tissue is measured by integrating the volumes of the voxels belonging
to the region of interest in a structural image. These regions of interest are segmented using manual
or automated methods. Manual brain segmentation is performed by individuals with training in
brain anatomy with the aid of MRI visualization software tools. This is a laborious procedure
which results in inter and intra-operator variability of the estimated volume. Nordenskjöld et al.,
(2013) reported that manual segmentation of TIV of a brain image with 50 image slices can take
up to 25 minutes for an experienced operator. Therefore, automated tools are increasingly used for
brain volumetry. A wide variety of automated methods are available for the segmentation of brain
images. A study by Helms, (2016) provides an extensive review of the available state-of-the-art
21
The whole brain volume metrics are directly computed from the brain images by identifying
the voxels within the intensity range of the tissues. Whereas the regional brain volumes are
computed from the voxels delineated by identifying complex region boundaries that make these
regional structures. In practice, the voxels belonging to these complex brain structures are
identified by carefully registering standard brain atlases onto the brain image. Most of the work in
this thesis is focused on accurately estimating global brain volumes.
2.7.2 Automated Brain Image Segmentation
Brain segmentation involves accurately delineating the brain images into different tissues: GM,
WM, and CSF (Figure 2.9). A variety of automated segmentation methods currently exist of which
Statistical Parametric Mapping (SPM) (Ashburner and Friston, 2005), FreeSurfer
(surfer.nmr.mgh.harvard.edu/) (Dale et al., 1999; Fischl et al., 2002a), and FMRIB Software
Library (FSL) (https://fsl.fmrib.ox.ac.uk/fsl) (Smith et al., 2004) are the most widely used. This
22
Figure 2.9 Brain tissues in a single section of a brain image from (A.) an MRI image, (B.)an average MRI brain tissue probabilistic map.
STATISTICAL PARAMETRIC MAPPING (SPM)
SPM uses an atlas-based Bayesian method to perform brain segmentation. SPM starts by
registering the subject’s brain image onto a tissue probability map (TPM). TPM contains prior probability maps of different tissue classes of the voxels in the standard space template. For linear
and non-linear registration SPM uses a default International Consortium for Brain Mapping space
template (ICBM; Rex et al., 2003) whose coordinates are defined in the Montreal Neurological
Institute (MNI) space. The unified segmentation algorithm also accounts for any biases present in
the image intensity. The segmentation algorithm performs tissue classification, bias correction and
23
The above segmentation yields three types of images: (1) probabilistic tissue maps in the native
space of the subject, (2) probabilistic tissue intensity maps in the template space and (3) modulated
tissue intensity maps in the template space (Figure 2.10). The modulated images preserve the local
volumetric changes in grey and white matter intensity on a voxel-by-voxel basis while spatially
normalizing to the template. The modulated images are used for voxel-based morphometry (VBM)
(Ashburner and Friston, 2000) as they measure subtle changes in the anatomical brain structures
by removing the global affine effects (like scaling, rotation and translation). GM volume (GMV)
and WM volume (WMV) are obtained by integrating the product of the corresponding tissue
probabilities and the voxel volume in the native space of the subject. We will be using VBM to
24
Figure 2.10 SPM brain segmentation pipeline illustrating probabilistic segmentation of GM, WM and CSF and voxel based morphometry (Ashburner and Friston, 2012).
FREESURFER
FreeSurfer utilizes a complex fully automated segmentation method for segmenting the brain into
various cortical and subcortical regions (Fischl et al., 2002a). Segmentation is done by modeling
a segment as an anisotropic Markov random field by using the priors for anatomical regions which
are obtained after registering the image onto a template. FreeSurfer uses recon-all pipeline for
segmentation, which also includes several preprocessing stages such as motion correction, non-
uniform intensity normalization, Talairach transform computation, intensity normalization, skull
25
surface-based analysis of the brain. It creates a vertex map for surfaces which forms the interfaces
(e.g., WM-GM, GM-CSF) between different soft tissues.
FMRIB SOFTWARE LIBRARY (FSL)
FSL uses the SIENAX algorithm to perform brain extraction and tissue segmentation. SIENAX
first creates a brain mask for the T1 image using FSL’s brain extraction tool (BET) (Jenkinson et
al., 2002b). BET performs brain segmentation by growing a tessellated mesh inside the cranium
until the boundary between brain and CSF is reached. After brain extraction, brain tissues are
segmented using FSL’s Automated Segmentation Tool (FAST) (Zhang et al., 2001). FAST uses hidden Markov random field segmentation with expectation maximization algorithm and segments
the brain into GM and WM segmentations.