Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Contents lists available at ScienceDirect
Psychiatry Research: Neuroimaging journal homepage: www.elsevier.com/locate/psychresns
Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans$ Brian D. Power a,b,n, Fiona A. Wilkes c, Mitchell Hunter-Dickson c, Danielle van Westen d,e, Alexander F. Santillo f, Mark Walterfang g, Christer Nilsson f, Dennis Velakoulis g, Jeffrey C.L. Looi c a
School of Medicine, The University of Notre Dame Australia, Fremantle, Australia Clinical Research Centre, North Metropolitan Health Service - Mental Health, Perth, Australia c Research Centre for the Neurosciences of Ageing, Academic Unit of Psychiatry and Addiction Medicine, Australian National University Medical School, Canberra Hospital, Canberra, Australia d Center for Medical Imaging and Physiology, Skåne University Hospital, Lund, Sweden e Diagnostic Radiology, Department of Clinical Sciences, Lund University, Lund, Sweden f Clinical Memory Research Unit, Department of Clinical Sciences, Lund University, Lund, Sweden g Neuropsychiatry Unit, Royal Melbourne Hospital, Melbourne Neuropsychiatry Centre, Melbourne, Australia b
art ic l e i nf o
a b s t r a c t
Article history: Received 11 July 2014 Received in revised form 14 October 2014 Accepted 4 February 2015
We present a validated protocol for manual segmentation of the thalamus on T1-weighted magnetic resonance imaging (MRI) scans using brain image analysis software. The MRI scans of five normal control subjects were randomly selected from a larger cohort recruited from Lund University Hospital and Landskrona Hospital, Sweden. MRIs were performed using a 3.0T Philips MR scanner, with an eightchannel head coil, and high resolution images were acquired using a T1-weighted turbo field echo (T1 TFE) pulse sequence, with resulting voxel size 1 1 1 mm3. Manual segmentation of the left and right thalami and volume measurement was performed on 28–30 contiguous coronal slices, using ANALYZE 11.0 software. Reliability of image analysis was performed by measuring intra-class correlations between initial segmentation and random repeated segmentation of the left and right thalami (in total 10 thalami for segmentation); inter-rater reliability was measured using volumes obtained by two other experienced tracers. Intra-class correlations for two independent raters were 0.95 and 0.98; inter-class correlations between the expert rater and two independent raters were 0.92 and 0.98. We anticipate that mapping thalamic morphology in various neuropsychiatric disorders may yield clinically useful diseasespecific biomarkers. Crown Copyright & 2015 Published by Elsevier Ireland Ltd. All rights reserved.
Keywords: Biomarker Dementia Neuropsychiatry Neuroimaging Neurodegenerative disorder Thalamus
1. Introduction “Essentially, the cortex must view the world through the thalamus; that is the only view the cortex has.” (Sherman and Guillery, 2006)
☆ WHERE THE WORK WAS CARRIED OUT: Recruitment of patients and imaging was performed at Skåne University Hospital, Lund, Sweden. Image analysis was performed at the Research Centre for the Neurosciences of Ageing, Academic Unit of Psychiatry and Addiction Medicine, Australian National University Medical School, Canberra Hospital, Canberra, ACT, Australia. n Corresponding author at: School of Medicine, The University of Notre Dame Australia, Fremantle, Australia. Tel.: þ 61 8 94330228; fax: þ61 8 94330250. E-mail address: [email protected] (B.D. Power).
1.1. Thalamic morphology in neuropsychiatric disorders using structural magnetic resonance imaging Thalamic morphology has been investigated using structural magnetic resonance imaging (MRI) in a wide range of neuropsychiatric disorders. Meta-analyses have reported thalamic atrophy in schizophrenia (Konick and Friedman, 2001; Adriano et al., 2010) and major depressive disorder (Du et al., 2012), but not in bipolar disorder (Hallahan et al., 2011); increased thalamic volumes have been reported in patients with obsessive–compulsive disorder, with a reduction in volume in subjects treated with antidepressant medication (Gilbert et al., 2000; Atmaca et al., 2007). Thalamic atrophy has been reported both in sporadic Alzheimer's dementia (AD) (de Jong et al., 2008) and in the pre-symptomatic stage of familial AD (Lee et al., 2013; Ryan et al., 2013). Increasing thalamic atrophy has been reported in patients with progressive supranuclear
http://dx.doi.org/10.1016/j.pscychresns.2015.02.001 0925-4927/Crown Copyright & 2015 Published by Elsevier Ireland Ltd. All rights reserved.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
2
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
palsy (PSP) enrolled in longitudinal studies (Whitwell et al., 2011); thalamic atrophy has also been observed in patients with frontotemporal dementia (FTD) (Cardenas et al., 2007; Chow et al., 2008) and Huntington's disease (HD) (Kassubek et al., 2005). Further investigating thalamic morphology may prove fruitful in enhancing disease etiology in the spectrum of neuropsychiatric disorders, and moreover could yield useful biomarkers with clinical utility. 1.2. Anatomical considerations of the thalamus for neuroimaging Arising from the primitive forebrain (and sharing its origins with the telencephalon), the diencephalon can be broadly divided into the thalamus and hypothalamus, although its expanse encompasses a number of other structures (such as the habenular nuclei of the epithalamus). The thalamus can be divided into dorsal and ventral divisions: the dorsal thalamus is composed of nuclei having reciprocal connections with the cerebral cortex and striatum; despite having direct input from the cortex (and in part from the basal ganglia), the ventral thalamus does not generally project to the cortex (Jones, 2007). The ventral thalamus not only has an intimate functional relationship with the dorsal thalamus (see Zikopoulos and Barbas, 2007), but importantly its nuclei (such as the reticular thalamic nucleus and zona incerta) cloak the dorsal thalamus along the intercommissural distance, providing landmarks to assist with the challenge of lateral and inferior boundary definition of the dorsal thalamic mass on neuroimaging. The dorsal thalamus is a paired structure, each roughly the size of a Brazil nut (Bertholletia excelsa) in humans. The thalami have a strategic central position in the brain which is readily identified on neuroimaging, being located at the base of each cerebral hemisphere respectively, between other subcortical structures and the cerebral cortex (Fig. 1). The two distinct dorsal thalami are found on either side of the third ventricle (which defines the thalamic medial boundary). Superiorly the thalami form the floor of the lateral ventricles for the greater part of their rostral-caudal extent. Laterally the dorsal thalamus is cloaked by the reticular thalamic nucleus (creating a fuzzy lateral boundary on MRI, by virtue of its reticulated or net-like appearance) and thence the internal capsule, ultimately separating the thalamus from the basal ganglia. Defining the inferior boundary is problematic throughout the rostro-caudal extent of the thalamus (until the pulvinar is reached), as there are a number of structures associated with the inferior boundary, including the zona incerta (of the ventral thalamus), substantia nigra, the subthalamic nucleus, the red nucleus, and hypothalamus for instance. A discerning feature of the inferior boundary is the presence of numerous white matter tracts surrounding these abovementioned structures; if these white matter tracts can be used to definitively exclude these structures from tracing, then one would have a more accurate description of the inferior dorsal thalamic boundary. Whilst the
caudal pole of the thalamus is clearly defined by the emergence of the pulvinar, defining the rostral pole is perhaps the most challenging and would therefore support a caudal-rostral approach to tracing; the rostral pole requires careful tracing to exclude the hypothalamus, and to be traced in its entirety requires thin slices (e.g., ideally o 1.5 mm). The internal organization of the dorsal thalamus can be conceptualized as having anterior, medial and lateral subdivisions, which are separated by a bowed sheet of myelinated fibers (the internal medullary lamina). Each of these subdivisions are composed of distinct thalamic nuclei, which are defined by characteristic cytoarchitecture and distinct patterns of connectivity, and their proposed involvement in distinct functional modalities (Table 1); of note, at least half of the thalamus is thought to be involved in cortico-cortical communication (the so-called higher order nuclei; see Table 1). 1.3. Development and utility of protocols for manual segmentation of the thalamus A number of different approaches have been used in MRI to examine the thalamus, including image averaging (Andreasen et al., 1994) and edge-finding using anatomical templating (Buchsbaum et al., 1996). By virtue of having automated components, both of these approaches can be useful when dealing with large cohorts, but they can be problematic for precise boundary definition and accommodating individual variability between patients (for review, see Spinks et al., 2002), especially in the context of reported significant inter-individual variability of thalamic nuclei (Uylings et al., 2008). The region of interest approach is therefore the most reliable (albeit labor intensive) approach, involving a trained rater manually segmenting the thalamus in the MRI scan of each participant (Andreasen et al., 1990; Flaum et al., 1995; Gur et al., 1998, Portas et al., 1998; Spinks et al., 2002). There are few widely available validated protocols for thalamic segmentation on MRI providing substantial anatomical detail for replication, and indeed addressing the key challenges of boundary definition. Perhaps the most widely cited is by Portas et al. (1998), who provide a description of the thalamus in 20–21 consecutive coronal 1.5 mm slices in a rostral–caudal direction; whilst the description of surrounding anatomical landmarks is thorough, there is little information regarding precise boundary definition (particularly the challenging lateral and inferior boundaries, and indeed the rostral pole of the thalamic mass). Spinks et al. (2002) describe an approach to manual segmentation of the thalamus and present 12 coronal 1.5 mm slices in their protocol, which they then used to train an artificial network for subsequent automated segmentation (with the goal of making large-scale studies more feasible); they appear to include the nuclei of the ventral thalamus. An advantage of their technique was the utilization of the tricolor image feature in BRAINS2 software, taking advantage of
Fig. 1. Thalamic morphology in 3 slices in caudal-rostral progression on MRI scans (from left to right).
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
3
Table 1 Classification, connections and functions of the dorsal thalamic nuclei. Classification
Nuclei
Principal inputs
Principal outputs
Proposed functional domain
First order (“specific”)
Anterior group Lateral geniculate Medial geniculate Ventral anterior Ventral lateral Ventral medial
Subiculum and presubiculum, mammillary nuclei
Limbic, memory
Retinal ganglion cells
Anterior limbic cortex, cingulate, subiculum, retrosplenial and presubiculum Visual cortex
Vision
Inferior colliculus
Auditory cortex of temporal lobe
Hearing
Substantia nigra
Diffuse prefrontal and cingulate
Motor
Globus pallidus, deep cerebellar nuclei, vestibular nuclei, spinothalamic tract Substantia nigra, taste, vagal, spinal
Motor and premotor cortices
Motor
Diffuse, frontal, cingulate, primary somatosensory and gustatory, medial frontal cortices
Ventral posterior Lateral posterior Lateral dorsal Medial dorsal
Medial and trigeminal lemnisci, spinothalamic tract Primary and secondary somatosensory cortex
Integration of viscero-motor information Somatic sensation of body and face Integration of sensory information Emotional expression Limbic
Higher order (“specific”)
Parietal lobe, superior colliculus and pretectum
Parietal lobe (area 5)
Central lateral
Cingulate gyrus, fornix, pretectum, hypothalamus Amygdaloid nuclear complex, olfactory cortex, hypothalamus, spinal, superior colliculus Superior colliculus and pretectum, temporal, parietal and occipital lobes Spinothalamic, cerebellar nuclei, reticular formation, substantia nigra and superior colliculus
Cingulate gyrus, retrosplenial cortex Prefrontal cortex, including frontal eye field, orbitofrontal and lateral frontal, and basal olfactory structures Temporal, parietal, occipital lobes (area 5, superior temporal gyrus) and prestriate cortex Striatum, motor cortex, somatosensory cortex, parietal cortex, frontal cortex
Centre median Parafascicular Reunions
Cerebellar nuclei, reticular formation, substantia nigra and superior colliculus Periaqueductal gray, reticular formation Fornix
Striatum, parietal and frontal cortex
Integration of sensory information Arousal, eye movement, nociception As for central lateral
Lateral, frontal cortex Hippocampal formation and adjacent regions
As for central lateral Limbic
Paratenial
Brainstem and limbic structures
Medial frontal cortex
Limbic
Pulvinar Intralaminar (“nonspecific”)
Midline (“non specific”)
Adapted from: Jones (2007)(Table 3.1); and Kandel et al. (1991) Principles of Neural Science. 3rd ed., Prentice Hall, Sydney (Table 20–1).
three imaging modalities (T1, T2 and PD weighted images) to assist with distinguishing landmarks. Some protocols are unfortunately constrained in their utility given the limited detail provided, and often refer loosely to naturalistic boundaries (Flaum et al., 1995; Gur et al., 1998; Lawrie et al., 1999; Andreasen et al., 1990; Dasari et al., 1999), and others have used an approach according to a pixel-classification method incorporating diencephalic gray matter structures (including thalamus, hypothalamus, and mammillary bodies) (Jernigan et al., 1991). 1.4. Research objective Few high quality thalamic protocols with sophisticated neuroimaging software are widely available for manual segmentation of the human dorsal thalamus on MRI scans. The aim of the current research was to develop and validate an advanced protocol for manual segmentation of the dorsal thalamus on T1-weighted MRI scans using ANALYZE 11.0 software (with 1 mm slice thickness to accommodate detailed slice-by-slice segmentation), to provide detailed descriptions of boundary definition, and to make the protocol widely accessible for further neuroimaging research.
2.2. Image acquisition and volumetric analysis MRIs were performed using a 3.0 Tesla Philips MR scanner, with an eightchannel head coil (Philips Achievas, Philips Medical Systems, Best, The Netherlands). High resolution images were acquired using a T1-weighted turbo field echo (T1 TFE) pulse sequence with parameters set as follows: repetition time (TR) 8 ms; echo time (TE) 4 ms; inversion time (TI) 650 ms; flip angle (FA) 101; number of excitations (NEX) 2; sensitivity/encoding (SENSE) factor 2.5; matrix 240 240; field of view (FOV) 240; resulting voxel size 1 1 1 mm3. There were 175 contiguous coronal slices obtained. The T1-weighted images were anonymised, randomly coded to ensure blinding, and transferred to a MacBook Pro (Apple Inc., Cupertino, CA, USA) computer at the Australian National University Medical School in Canberra, Australia. Of the control scans randomly selected, we excluded images with significant motion artefacts, degraded image quality and radiological pathology in the thalamus (strokes, injuries, lesions). Unilateral manual segmentation of the thalami (left and right) of the five control subjects and calculation of thalamic volumes (left, right and total volume) was initially performed by a single investigator (BDP), with reference to a standard three-dimensional sectional anatomical atlas (Duvernoy, 1999) and stereotactic atlas (Morel, 2007). Segmentation was performed on the right brain, and then on the left brain, yielding two thalamic volumes (left and right) per control. Segmentation was performed using ANALYZE 11.0 (Biomedical Imaging Resource, Mayo Foundation, Rochester, MN, MAYO clinic) software. For subsequent reliability, two other raters (FAW, MH-D) repeated segmentation on the same MRI scans of the same five control subjects, again, yielding two thalamic volumes (left and right) per control.
2. Methods
2.3. Thalamic boundary definition and protocol development
2.1. Participants
The protocol was designed by a neuropsychiatrist with extensive experience in the clinical neuroanatomy of the thalamus (BDP), and with reference to neuroradiological and neurosurgical atlases (Duvernoy, 1999; Morel, 2007). Manual segmentation of the dorsal thalami (left and right) occurred in 28–30 consecutive 1 mm coronal slices (from an average of 175 coronal slices of the human brain); verification occurred in other planes. The protocol proceeds in a caudal to rostral direction (intercommissural plane), starting by tracing the pulvinar as it emerges
The MRI scans of five control subjects were randomly selected from a larger cohort (age-matched and as part of an ongoing study investigating MRI changes in frontotemporal dementia), recruited via Lund University Hospital and Landskrona Hospital, Sweden. All participants gave written consent, and the study was approved by the Regional Ethics Committee for Research.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
4
under the crus fornix, and then proceeding rostrally until the crus cerebri can be seen separating from the body of the pons. The medial border was defined by the third ventricle, and superior cistern in more caudal sections. The lateral border is defined by the reticular thalamic nucleus (of the ventral thalamus). The inferior border was defined at various levels by using either nearby landmarks (e.g., habenular), characteristic white matter tracts proximal to prominent landmarks (i.e., those adjacent to the red nucleus, and subthalamic nucleus), and in more rostral sections the hypothalamic sulcus (hypothalamus, column of fornix). The superior border was defined by the body of the lateral ventricle. The most rostral slices are challenging, characterized by the emergence of the mammillary bodies and hypothalamus. In the most rostral slice the thalamus is visualized at the inferior pole of the head of the caudate, 5–6 mm caudal to the anterior commissure - a prominent landmark. The protocol attempts to use the ventral thalamus (including the reticular thalamic nucleus and zona incerta) for boundary definition, and hence exclude these structures from tracing where possible. 2.4. Intensity settings and inversion using ANALYZE software Using ANALYZE 11.0 (Mayo Clinic, Rochester, MN, USA) the Intensity Loaded Volume - Region of Interest function was used to optimize intensity settings, thereby enhancing contrast between gray and white matter, and assist in boundary definition (minimum value: 50–100; maximum value: 500–600). In addition, the Invert function (making white matter appear dark) was used to facilitate boundary definition on every slice. Each individual slice is presented in the protocol in duplicate; the first image demonstrates thalamic boundary definition for the slice, and the second image is the same slice with intensity settings inverted (see Fig. 2). 2.5. Statistical analysis Reliability of image analysis was performed by measuring intra-class correlations (a two-way mixed effects model for intra-class correlation coefficients using an absolute agreement definition (Shrout and Fleiss, 1979) for each rater between initial segmentation of both left and right thalami of the five subjects, followed by random repeated segmentation of the same scans; reliability was calculated in SPSS 17.0 (IBM Corporation, Somers, NY, USA). Inter-rater reliability of manual segmentation was calculated using a two-way mixed effects model for inter-class correlation coefficients using an absolute agreement definition of an expert rater (BDP) with two other experienced raters (FAW, MH-D), using the same MRI scans of the five subjects.
3. Results A condensed version of the protocol is presented in Table 2 (with reference to slices in Fig. 2), and a complete version is available online as Supplementary material. Fig. 2 shows an example of thalamic segmentation of the right brain on MRI using the protocol. Mean thalamic volumes (right thalamus, left thalamus, total thalamic volume) recorded by the three raters are shown in Table 3. Reliability was conducted by three raters for 10 thalami. Intra-class correlations for two raters were 0.95 (BDP) and 0.98 (FAW) (Table 4). The inter-class correlations were 0.92 (BDP, FAW) and 0.98 (BDP, MH-D) (Table 4).
4. Discussion Few high quality thalamic protocols are widely available for manual segmentation of the human thalamus on MRI scans, and our manual segmentation intra- and inter-class coefficients for reliability above 0.9 compare favorably with previously published manual segmentation methods used in quantitative structural MRI research (Looi et al., 2008; Looi et al., 2009). The protocol presented here represents an advance in the field, as it uses sophisticated neuroimaging software, is undertaken in high resolution images with 1 mm slices, and was developed with reference to both neuroradiological and neurosurgical atlases with an emphasis on boundary definition. A number of other authors have presented protocols; however, these have been constrained by a number of factors including slice thickness and limited information to assist with inferior and lateral boundary definition (Portas
et al., 1998; Spinks et al., 2002). The utilization of the specific function in ANALYZE 11.0 in our protocol to invert intensity settings has the added benefit of enhancing boundary definition, particularly of the lateral and inferior boundaries throughout the rostral-caudal extent of the dorsal thalamus. Few studies have reported a protocol for manual segmentation of the thalamus together with mean thalamic volumes to enable a comparison with volumes reported in our study. van der Werf et al. (2001) reported mean thalamic volumes across the lifespan, but information was limited as segmentation of the thalamus occurred on 8–12 coronal slices. They reported a significant decrease in total thalamic volume with increasing age in healthy subjects (apparent before the onset of volume loss of total brain volume); unadjusted mean values reported were 15.29 cm3 (22– 45 years; n ¼18), 13.41 cm3 (46–60 years; n ¼14), and 12.22 cm3 (61–82 years; n ¼24). Spinks et al. (2002) reported mean values of 7.00 cm3 (S.D. 71.12) for left thalamus and 6.55 cm3 (S.D. 70.54) for right thalamus (8 healthy males controls in their study had a mean age 25.63 years, S.D. 76.55); and Lawrie et al. (1999) reported mean values of 6.60 cm3 (S.D. 70.80) for left thalamus and 6.40 cm3 (S.D. 7 0.60) for right thalamus (in healthy controls, mean age 21.1 years, S.D.7 2.3). Ettinger et al. (2001) (following the protocol by Portas et al. (1998)) reported mean unadjusted thalamic volumes in control subjects (mean age 25.4 years, S. D.75.8) of 15.10 cm3. It may be that the smaller volumes reported in our study reflect the more detailed thalamic boundary definition offered by our protocol, tracing over more slices (by virtue of 1 mm slices), and the exclusion of the zona incerta (together with its surrounding pathways) and sub-thalamic nucleus, for instance. It is also possible that the smaller volumes reported in our study were in part accounted for by the older control subjects in our study (mean age 56, age range 35–72), supported by van der Werf et al. (2001) data suggesting decreased thalamic volume with advancing age. The protocol has sound generalizability as it can be used in a range of image analysis software packages, such as ITK-SNAP (http://www.itksnap.org/pmwiki/pmwiki.php) or MRIcro (http:// www.mccauslandcenter.sc.edu/mricro/mricro/). We envisage that advanced thalamic protocols such as our own will allow for more precise manual segmentation and thalamic boundary definition, and will be necessary to drive technologies to more accurately automatically segment the thalamus, in essence rendering protocols such as ours redundant in the future. To this end we are endeavoring to investigate the correlation between thalamic segmentation using our protocol and existing automatic segmentation techniques of the diencephalon; we anticipate a discrepancy on the basis of the precision, for instance, of the inferior boundary definition in our protocol (in the regions of the subthalamic nucleus and hypothalamus). For instance, a number of studies have used automatic segmentation protocols, and reported larger mean thalamic volumes: left thalamus ¼ 7.98 cm3 and right thalamus¼ 8.07 cm3 in 22 healthy control subjects (Zarei et al., 2010); left thalamus ¼ 7.18 cm3 and right thalamus ¼ 7.79 cm3 in 20 healthy controls (Ryan et al., 2013) and left thalamus ¼7.63 cm3 and right thalamus ¼ 7.68 cm3 in 70 elderly ‘memory complainers' (de Jong et al., 2008). It should be noted that volumetric analysis alone will be limited in its ability to differentiate disease impact on the various thalamic nuclei, and hence structural MRI will need to be combined with other modalities. Shape analysis (e.g., Spherical Harmonics, or SPHARM) is a surface-bases analysis which has the capability to reveal statistically significant regional volume differences (such as surface deflation) in the thalamus, which can provide important information regarding which regions (and by extrapolation, which subdivisions or nuclei) of the thalamus could be affected in disease states (see McKeown et al., 2008; and Janssen et al., 2012).
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
5
Fig. 2. Case example of segmentation of the thalamus on consecutive 1-mm slices on MRI, in caudal to rostral progression.
Diffusion tensor imaging will be an important tool given the fact that the various thalamic nuclei can be distinguished by their differential cortical connectivity, allowing the subdivisions and
nuclei of thalamus to be imaged according to preselected cortical areas, for instance (see Jakab et al., 2012; and Zarei et al., 2010). Given that most of the afferent and efferent fibers of the thalamus
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
6
Table 2 Table showing condensed version of the thalamic protocol for manual segmentation of the thalamus on T1-weighted MRI scans (with reference to sample slices in Fig. 2). SLICE
Description of Thalamic Boundary and Related Structures on Coronal Slices
1–3
The oval-shaped pulvinar of the thalamus can be visualized between the crus fornix and the superior cistern (the superior cistern (or quadrigeminal cistern/cistern of the great cerebral vein) is a radiological landmark above the midbrain in the horizontal plane). The pineal gland can be seen centrally, superior to the colliculi. The diagonal crossing of the crus fornix and splenium of the corpus callosum is a characteristic feature (slice 1,2). Duvernoy (1999), Fig 76–77 p154.
4–6
The pulvinar is a distinct mass of gray matter visualized under the corpus callosum, and between the fornix and the superior cistern. The pineal gland can still be seen centrally, superior to the quadrigeminal plate (superior and inferior colliculi), and the caudate can be seen at the lateral edge of the LV. In slice 6 the thalamus is becoming continuous with the midbrain; its inferior and lateral borders are easily defined. A vertical line from the caudate serves as a guide for defining the lateral extremes of the thalamus. Duvernoy (1999), Fig 75–76 p152–154; Morel (2007), Fig 4.1 p56.
7–10
The appearance of the geniculates (MGN, LGN) cause the inferior thalamic border to bulge laterally. A key feature is the posterior commissure (slices 9– 10). The MGN and LGN form its inferior border; the thalamic mass is enclosed by the curved shell of the RTN laterally; the lateral border is the internal capsule. The habenular nucleus emerges, producing a noticeable bulge on the midline. An oblique line from the habenular directed in an inferior-lateral direction at approximately 451 (towards the subiculum) marks the inferior-medial border of the thalamus at this level. Duvernoy (1999), Fig 74–75 p148– 152; Morel (2007), Fig 4.2–4.3 p 57–p58.
11–13
The MD abuts the 3V; the MGN and LGN comprise the inferior thalamic mass; the LGN becomes discontinuous with the thalamic mass in slice 13. The lateral border is defined by the internal capsule, and LV and 3V respectively defining the superior and medial borders. The habenular is less visible in slice 13; to delineate the thalamus from the brainstem in subsequent sections draw a line from the hypothalamic sulcus of 3V to the medial edge of the geniculate complex. Duvernoy (1999), Fig 72–73 p144–146; Morel (2007), Fig 4.3–4.4 p58–59.
14–16
The RN is a characteristic rounded feature at this level; white matter tracts between RN and the inferior border of the thalamic mass provides contrast for boundary definition until the hypothalamus emerges. MGN is no longer evident. LGN is separated from the greater thalamic mass by the internal capsule/ cerebral peduncle (and is no longer visible in slice 16); a line between the LGN and caudate approximates the lateral thalamic boundary. STN emerges inferiorly to the zona incerta. Duvernoy (1999), Fig 69–71 p138–142; Morel (2007), Fig 4.4–4.5 p59–60.
17–20
The LGN has been replaced by the optic tract. The globus pallidus is emerging medial to the putamen. RN and associated white matter tracts remain useful landmarks for definition of the inferior thalamic border; a vertical line from the caudate serves to approximate the lateral thalamic boundary, which is bounded by the internal capsule. The interpeduncular fossa becomes evident. Duvernoy (1999), Fig 67–69 p134–138; Morel (2007), Fig 4.6–4.7 p61–62.
21–23
RN is no longer visible to serve as a landmark; identify the STN superior to the substantia nigra. White matter tracts enveloping the medial aspect of the zona incerta/ STN aide with identification of the inferior thalamic boundary. Duvernoy (1999), Fig 64-66 p128–132; Morel (2007), Fig 4.8–4.9 p63-64.
24–26
The STN is no longer evident at this level (slice 25), and the zona incerta merges with the hypothalamus at its medial border. The thalamic mass is adjacent to the top half of the 3V, whilst the hypothalamus is adjacent to the bottom part (defining the hypothalamic sulcus). The crus cerebri can be seen separating from the body of the pons. The mammillary bodies are distinct (slice 26), and seen medial and adjacent to the optic tract. The caudate enlarges in subsequent rostral sections; as an approximation, the thalamus just overlaps the medial border of the caudate. Duvernoy (1999), Fig 63-64 p126–128; Morel (2007), Fig 4.9–4.11 p64–66.
27–29
The thalamus becomes more discrete in size, and the superior lateral tip of the thalamus abuts the inferior medial tip of the caudate head. Its lateral border is the genu of the internal capsule; the thalamus rests on the third ventricle at the midline. The hypothalamus forms much of the medial border of 3V. The mammillary bodies diminish at this level (slice 28); the thalamus is still present just rostral ( 1–2 mm) to where the mammillary bodies can no longer be visualized. The column of the fornix emerges within the hypothalamus (slice 27), is prominent in slice 29 and is a key landmark in rostral thalamic boundary definition; it has a more medial and superior location in the rostral slices beyond the thalamus, where it can eventually be found superior to the anterior commissure. The most rostral portion of the thalamus is approximately 5–6 mm caudal to the anterior commissure. Duvernoy (1999), Fig 61–62 p122–124; Morel (2007), Fig 4.11–4.12 p66–67.
LV ¼lateral ventricle; 3V ¼ third ventricle; MGN¼ medial geniculate nucleus of the thalamus; LGN¼ lateral geniculate nucleus of the thalamus; RTN ¼ reticular thalamic nucleus; MD ¼medial dorsal nucleus of the thalamus; RN¼ Red nucleus; STN¼ subthalamic nucleus.
Table 3 Table showing mean thalamic volumes recorded by the three raters for the MRI scans of the five participants.
Principal rater (BDP) Secondary rater (FAW) Secondary rater (MH-D)
Right thalamus (N¼ 5)
Left thalamus (N ¼5)
Total thalamus (N ¼ 5)
Mean (mm3)
S.D. (mm3)
Mean (mm3)
S.D. (mm3)
Mean (mm3)
S.D. (mm3)
5670.4
335.7
5592.6
410.9
11263.0
743.6
5803.2
416.0
5510.4
454.0
11313.6
868.2
5663.0
334.2
5656.2
410.8
11319.2
742.1
S.D. ¼standard deviation.
are mapped, utilizing these modalities in exploring the subcortical connectome will clarify the how various neurodegenerative disorders affect the thalamus. This novel approach combining MRI and DTI in schizophrenia has been reported, where images from modalities were superimposed to assist with thalamic boundary definition, with improved reliability (see Kim et al., 2007).
Table 4 Table showing intra- and inter-rater reliability.
Intra-rater Intra-rater Inter-rater Inter-rater
(BDP) (FAW) (BDP, FAW) (BDP, MH-D)
Correlation coefficient
95%CI
0.95 0.98 0.92 0.98
0.79–0.99 0.92–0.99 0.80–0.97 0.92–0.99
Correlation coefficients two-way, mixed effects model intra/interclass correlation, for ten measurements of the thalamus (five each side of the bisected thalamus), based on methods of Shrout and Fleisse (1979); Intra(within)-rater: intraclass correlation, Inter(between)-raters: interclass correlation; 95%CI ¼ 95% of the confidence interval.
There are a number of limitations of this protocol. In a number of slices the lateral geniculate nucleus (LGN) of the dorsal thalamus is discontinuous with the greater dorsal thalamic mass (see slices 13– 15, Fig. 2), and this will pose problems for shape analysis (e.g., SPHARM); in these slices the discontinuous LGN would need to be excluded from tracing for such analysis to proceed. Whilst the protocol attempts to exclude the ventral thalamus and does so reliably with the zona incerta, given the fuzzy lateral border made by the slender reticular thalamic nucleus, it is difficult to be
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
confident that a clear delineation between the dorsal thalamus and reticular nucleus has been achieved in all slices. We argue that the gold standard for segmentation of the thalamus in MRI is manual segmentation by an expert rater. The majority of semi- and fully automated computational segmentation is calibrated by reference to expert manual segmentations. For example, using an automated Bayesian subcortical segmentation method that includes the thalamus, called FIRST, Patenaude et al. (2011) cross-validated their analysis with manually labeled images from the training set. Similarly, other methods such as the use of ADABOOST trains automated analysis software using manually segmented template images for subcortical structures (Looi et al., 2012). The utility of the protocol is to facilitate further neuroimaging studies in the spectrum of neuropsychiatric disorders, with the aim of investigating how the morphology (shape and volume) of the thalamus is involved in the neural networks that underpin normal human cognition, emotion, movement and behavior. Determining how genetic factors are associated with thalamic morphology may yet provide further putative endophenotypes in neuropsychiatric disorders. From a clinical perspective, correlation of thalamic morphology with disease-specific clinical features (i.e., cognitive, emotional or behavioral disturbance, movement disorder) will advance our understanding of the neurobiology of disease, with the potential to identify disease-specific biomarkers (i.e., a measurable characteristic of disease). These biomarkers may prove to have prognostic significance or utility for monitoring treatment response when disease-modifying agents are available.
Contributors BDP conceived the validation study, developed the protocol, and undertook protocol validation with FAW and MH-D. BDP conceived and wrote the first draft of the manuscript, and is guarantor of the validation study and this paper; JCLL coconceived the validation study as the research network coordinator. All co-authors contributed to the paper and research described therein. BDP self-funded travel and infrastructure costs for his portion of the research; FAW, MH-D, DvW, AS, MW, CN, DV and JCLL funded research and infrastructure costs via their respective centres. Owing to space restrictions, some references have been omitted where scientific concepts/knowledge have been essentially established.
Acknowledgment The study was funded through grants from the Swedish Parkinson Fund and the Swedish Science Council (through the Basal Ganglia Disease Linnæus Consortium).
Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.pscychresns.2015. 02.001. References Adriano, F., Spoletini, I., Caltagirone, C., Spalletta, G., 2010. Updated meta-analyses reveal thalamus volume reduction in patients with first-episode and chronic schizophrenia. Schizophrenia Research 123, 1–14. Andreasen, N.C., Ehrhardt, J.C., Swayze, V.W., Alliger, R.J., Yuh, W.T.C., Cohen, G., Ziebell, S., 1990. Magnetic resonance imaging of the brain in schizophrenia. Archives of General Psychiatry 47, 35–44.
7
Andreasen, N.C., Arndt, S., Swayze, V., Cizadlo, T., Flaum, M., O’Leary, D.S., Ehrhardt, J.C., Yuh, W.T.C., 1994. Thalamic abnormalities in schizophrenia visualized through magnetic resonance image averaging. Science 266, 294–298. Atmaca, M., Yildirim, H., Ozdemir, H., Tezcan, E., Poyraz, A.K., 2007. Volumetric MRI study of key brain regions implicated in obsessive–compulsive disorder. Progress in Neuro-psychopharmacology and Biological Psychiatry 31, 46–52. Buchsbaum, M.S., Someya, T., Teng, C.Y., Abel, L., Chin, S., Najafi, A., Haier, R.J., Wu, J., Bunney, W.E., 1996. PET and MRI of the thalamus in never medicated patients with schizophrenia. American Journal of Psychiatry 153, 191–199. Cardenas, V.A., Boxer, A.L., Chao, L.L., 2007. Deformation-based morphometry reveals brain atrophy in frontotemporal dementia. Archives of Neurology 64, 873–877. Chow, T.W., Izenberg, A., Binns, M.A., Freedman, M., Stuss, D.T., Ramirez, J., Black, S.E., 2008. Magnetic resonance imaging in frontotemporal dementia shows subcortical atrophy. Dementia and Geriatric Cognitive Disorders 26, 79–88. Dasari, M., Friedman, L., Jesberger, J., Stuve, T.A., Findling, R.L., Swales, T.P., Schulz, S.C., 1999. A magnetic resonance imaging study of thalamic area in adolescent patients with either schizophrenia or bipolar disorder as compared to healthy controls. Psychiatry Research: Neuroimaging 91, 155–162. de Jong, L.W., van der Hiele, K., Veer, I.M., Houwing, J.J., Westendorp, R.G., Bollen, E.L., de Bruin, P.W., Middlekoop, H.A., van Buchem, M.A., van der Grond, J., 2008. Strongly reduced volumes of putamen and thalamus in Alzheimer's disease: an MRI study. Brain 131, 3277–3285. Du, M.Y., Wu, Q.Z., Yue, Q., Du, M.Y., Wu, Q.Z., Yue, Q., Li, J., Liao, Y., Kuang, W.H., Huang, X.Q., Chan, R.C., Mechelli, A., Gong, Q.Y., 2012. Voxelwise meta-analysis of gray matter reduction in major depressive disorder. Progress in NeuroPsychopharmacology & Biological Psychiatry 36, 11–16. Duvernoy, H.M., 1999. The Human Brian: Surface, Three-Dimensional Sectional Anatomy and MRI, 3rd ed. Springer-Verlag, New York. Ettinger, U., Chitnis, X.A., Kumari, V., Fannon, D.G., Sumich, A.L., O’Ceallaigh, S., Doku, V.C., Sharma, T., 2001. Magnetic resonance imaging of the thalamus in first-episode psychosis. The American Journal of Psychiatry 158, 116–118. Flaum, M., Swayze, V.W., O’Leary, D.S., Yuh, W.T.C., Ehrhardt, J.C., Arndt, S.V., Andreasen, N.C., 1995. Effects of diagnosis, laterality, and gender on brain morphology in schizophrenia. American Journal of Psychiatry 152, 704–714. Gilbert, A.R., Moore, G.J., Keshavan, M.S., Paulson, L.A., Narula, V., MacMaster, F.P., Stewart, C.M., Rosenberg, D.R., 2000. Decrease in thalamic volumes of paediatric patients with obsessive–compulsive disorder who are taking paroxetine. Archives of General Psychiatry 57, 449–456. Gur, R.E., Maany, V., Mozley, P.D., Swanson, C., Bilker, W., Gur, R.C., 1998. Subcortical MRI volumes in neuroleptic-naıve and treated patients with schizophrenia. American Journal of Psychiatry 155, 1711–1717. Hallahan, B., Newell, J., Soares, J.C., Brambilla, P., Strakowski, S.M., Fleck, D.E., Kieseppa, T., Altshuler, L.L., Fornito, A., Malhi, G., McIntosh, A.M., Yurgelun-Todd, D.A., Laba, K.S., Sharma, V., MacQueen, G.M., Murray, R.M., McDonald, C., 2011. Structural magnetic resonance imaging in bipolar disorder: an international collaborative mega-analysis of individual adult patient data. Biological Psychiatry 69, 326–335. Jakab, A., Blanc, R., Berenyi, E.L., Szekely, G., 2012. Generation of individualized thalamus target maps using statistical shape models and thalamocortical tractography. American Journal of Neuroradiology 33, 2110–2116. Janssen, J., Aleman-Gomez, Y., Reig, S., Schnack, H.G., Parellada, M., Graell, M., Moreno, C., Moreno, D., Mateos-Perez, J.M., Udias, J.M., Arango, C., Desco, M., 2012. Regional specificity of thalamic volume deficits in male adolescents with early onset psychosis. The British Journal of Psychiatry 200, 30–36. Jernigan, T.L., Archibald, S.L., Berhow, M.T., Sowell, E.R., Foster, D.S., Hesselink, J.R., 1991. Cerebral structure on MRI, Part I: localization of age-related changes. Biological Psychiatry 29, 55–67. Jones, E.G., 2007. The Thalamus, 2nd ed. Cambridge University Press, Cambridge. Kandel, E.R., Schwartz, J.H., Jessell, T.M., 1991. Principles of Neural Science, 3rd ed Prentice-Hall, Sydney. Kassubek, J., Juengling, F.D., Ecker, D., Landwehrmeyer, G.B., 2005. Thalamic atrophy in Huntington's disease co-varies with cognitive performance: a morphometric MRI analysis. Cerebral Cortex 15, 846–853. Kim, J., Kim, D., Kim, T, Seok, J., Chun, J.W., Oh, M., Park, H., 2007. Volumetric abnormalities in connectivity-based subregions of the thalamus in patients with chronic schizophrenia. Schizophrenia Research 97, 226–235. Konick, L.C., Friedman, L., 2001. Meta-analysis of thalamic size in schizophrenia. Biological Psychiatry 49, 178–186. Lawrie, S.M., Whalley, H., Kestelman, J.N., Abukmeil, S.S., Byrne, M., Hodges, A., Rimmington, J.E., Best, J.J.K., Owens, D.G.C., Johnstone, E.C., 1999. Magnetic resonance imaging of brain in people at high risk of developing schizophrenia. Lancet 353, 30–33. Lee, G.J., Lu, P.H., Medina, L.D., Rodriguez-Agudelo, Y., Melchor, S., Coppola, G., Braskie, M.N., Hua, X., Apostolova, L.G., Leow, A.D., Thompson, P.M., Ringman, J.M., 2013. Regional brain volume differences in symptomatic and presymptomatic carriers of familial Alzheimer's disease mutations. Journal of Neurology, Neurosurgery and Psychiatry 84, 154–162. Looi, J.C.L., Lindberg, O., Liberg, B., Tatham, V., Kumar, R., Maller, J., Millard, E., Sachdev, P., Hogberg, G., Pagani, M., Botes, L., Engman, E., Zhang, Y., Svensson, L., Wahlund, L., 2008. Volumetrics of the caudate nucleus: reliability and validity of a new manual tracing protcol. Psychiaitry Research: Neuroimaging 163, 279–288. Looi, J.C.L., Svensson, L., Lindberg, O., Zandbelt, B.B., Ostberg, P., Orndahl, E., Wahlund, L.-O., 2009. Putaminal volume in frontotemporal lobar degeneration and Alzheimer disease: differential volumesin dementia subtypes and controls. American Journal of Neuroradiology 30, 1552–1560.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
8
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Looi, J.C.L., Rajagopalan, P., Walterfang, M., Madsen, S.K., Thompson, P.M., Macfarlane, M.D., Ching, C., Chua, P., Velakoulis, D., 2012. Differential putaminal morphology in Frontotemporal dementia, Huntington's Disease, Alzheimer's Disease. Australian and New Zealand Journal of Psychiatry 46, 1145–1158. McKeown, M.J., Uthama, A., Abugharbieh, R., Palmer, S., Lewis, M., Huang, X., 2008. Shape (but not volume) changes in the thalami in Parkinson disease. BioMed Central Neurology 8, 8. http://dx.doi.org/10.1186/1471-2377-8-8. Morel, A., 2007. Stereotactic Atlas of the Human Thalamus and Basal Ganglia. Informa Healthcare USA, Inc., New York. Patenaude, B., Smith, S.M., Kennedy, D., Jenkinson, M.A., 2011. Bayesian model of shape and appearance for subcortical brain. NeuroImage 56, 907–922. Portas, C.M., Goldstein, J.M., Shenton, M.E., Hokama, H.H., Wible, C.G., Fischer, I., Kikinis, R., Donnino, R., Jolesz, F.A., MCarley, R.W., 1998. Volumetric evaluation of the thalamus in schizophrenic male patients using magnetic resonance imaging. Biological Psychiatry 43, 649–659. Ryan, N.S., Keihaninejad, S., Shakespeare, T.J., Lehmann, M., Crutch, S.J., Malone, I.B., Thornton, J.S., Mancini, L., Hyare, H., Yousry, T., Ridgway, G.R., Zhang, H., Modat, M., Alexander, D.C., Rossor, M.N., Ourselin, S., Fox, N.C., 2013. Magnetic resonance imaging evidence for presymptomatic change in thalamus and caudate in familial Alzheimer's disease. Brain 136, 1399–1414. Sherman, S.M., Guillery, R.W., 2006. Exploring the Thalamus and Its Role in Cortical Function, 2nd ed MIT Press, Cambridge, MA, USA.
Shrout, P.E., Fleiss, J.L., 1979. Intraclass correlations: uses in assessing rater reliability. Psychological Bulletin 86, 420–428. Spinks, R., Magnotta, V.A., Andreasen, N.C., Albright, K.C., Ziebell, S., Nopoulos, P., Cassell, M., 2002. Manual and automated measurement of the whole thalamus and mediodorsal nucleus using magnetic resonance imaging. Neuroimage 17, 631–642. Uylings, H.B., Amunts, K., Mohlberg, H., Pool, C.W., Zilles, K., Smeets, W. 2008. 3D Probabilistic map of cytoarchitectonically defined human thalamic nuclei in standard MRI. Front. Neuroinform. 1, 3D. van der Werf, Y.D., Tisserand, D.J., Visser, P.J., Hofman, P.A.M., Vuurman, E., Uylings, H.B.M., Jolles, J., 2001. Thalamic volume predicts performance on tests of cognitive speed and decreases in healthy aging. A magnetic resonance imaging-based volumetric analysis. Cognitive Brain Research 11, 377–385. Whitwell, J.L., Avula, R., Master, A., Vemuri, P., Senjem, M.L., Jones, D.T., Jack, C.R., Josephs, K.A., 2011. Disrupted thalamocortical connectivity in PSP: a resting state fMRI, DTI, VBM study. Parkinsonism and Related Disorders 17, 599–605. Zarei, M., Patenaude, B., Damoiseaux, J., Morgese, C., Smith, S., Matthews, P.M., Barkhof, F., Rombouts, S., Sanz-Arigita, E., Jenkinson, M., 2010. Combining shape and connectivity analysis: an MRI study of thalamic degeneration in Alzheimer's disease. NeuroImage 49, 1–8. Zikopoulos, B., Barbas, H., 2007. Circuits for multisensory integration and attentional modulation through the prefrontal cortex and thalamic reticular nucleus in primates. Reviews in the Neurosciences 18, 417–438.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
Contents lists available at ScienceDirect
Psychiatry Research: Neuroimaging journal homepage: www.elsevier.com/locate/psychresns
Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans$ Brian D. Power a,b,n, Fiona A. Wilkes c, Mitchell Hunter-Dickson c, Danielle van Westen d,e, Alexander F. Santillo f, Mark Walterfang g, Christer Nilsson f, Dennis Velakoulis g, Jeffrey C.L. Looi c a
School of Medicine, The University of Notre Dame Australia, Fremantle, Australia Clinical Research Centre, North Metropolitan Health Service - Mental Health, Perth, Australia c Research Centre for the Neurosciences of Ageing, Academic Unit of Psychiatry and Addiction Medicine, Australian National University Medical School, Canberra Hospital, Canberra, Australia d Center for Medical Imaging and Physiology, Skåne University Hospital, Lund, Sweden e Diagnostic Radiology, Department of Clinical Sciences, Lund University, Lund, Sweden f Clinical Memory Research Unit, Department of Clinical Sciences, Lund University, Lund, Sweden g Neuropsychiatry Unit, Royal Melbourne Hospital, Melbourne Neuropsychiatry Centre, Melbourne, Australia b
art ic l e i nf o
a b s t r a c t
Article history: Received 11 July 2014 Received in revised form 14 October 2014 Accepted 4 February 2015
We present a validated protocol for manual segmentation of the thalamus on T1-weighted magnetic resonance imaging (MRI) scans using brain image analysis software. The MRI scans of five normal control subjects were randomly selected from a larger cohort recruited from Lund University Hospital and Landskrona Hospital, Sweden. MRIs were performed using a 3.0T Philips MR scanner, with an eightchannel head coil, and high resolution images were acquired using a T1-weighted turbo field echo (T1 TFE) pulse sequence, with resulting voxel size 1 1 1 mm3. Manual segmentation of the left and right thalami and volume measurement was performed on 28–30 contiguous coronal slices, using ANALYZE 11.0 software. Reliability of image analysis was performed by measuring intra-class correlations between initial segmentation and random repeated segmentation of the left and right thalami (in total 10 thalami for segmentation); inter-rater reliability was measured using volumes obtained by two other experienced tracers. Intra-class correlations for two independent raters were 0.95 and 0.98; inter-class correlations between the expert rater and two independent raters were 0.92 and 0.98. We anticipate that mapping thalamic morphology in various neuropsychiatric disorders may yield clinically useful diseasespecific biomarkers. Crown Copyright & 2015 Published by Elsevier Ireland Ltd. All rights reserved.
Keywords: Biomarker Dementia Neuropsychiatry Neuroimaging Neurodegenerative disorder Thalamus
1. Introduction “Essentially, the cortex must view the world through the thalamus; that is the only view the cortex has.” (Sherman and Guillery, 2006)
☆ WHERE THE WORK WAS CARRIED OUT: Recruitment of patients and imaging was performed at Skåne University Hospital, Lund, Sweden. Image analysis was performed at the Research Centre for the Neurosciences of Ageing, Academic Unit of Psychiatry and Addiction Medicine, Australian National University Medical School, Canberra Hospital, Canberra, ACT, Australia. n Corresponding author at: School of Medicine, The University of Notre Dame Australia, Fremantle, Australia. Tel.: þ 61 8 94330228; fax: þ61 8 94330250. E-mail address: [email protected] (B.D. Power).
1.1. Thalamic morphology in neuropsychiatric disorders using structural magnetic resonance imaging Thalamic morphology has been investigated using structural magnetic resonance imaging (MRI) in a wide range of neuropsychiatric disorders. Meta-analyses have reported thalamic atrophy in schizophrenia (Konick and Friedman, 2001; Adriano et al., 2010) and major depressive disorder (Du et al., 2012), but not in bipolar disorder (Hallahan et al., 2011); increased thalamic volumes have been reported in patients with obsessive–compulsive disorder, with a reduction in volume in subjects treated with antidepressant medication (Gilbert et al., 2000; Atmaca et al., 2007). Thalamic atrophy has been reported both in sporadic Alzheimer's dementia (AD) (de Jong et al., 2008) and in the pre-symptomatic stage of familial AD (Lee et al., 2013; Ryan et al., 2013). Increasing thalamic atrophy has been reported in patients with progressive supranuclear
http://dx.doi.org/10.1016/j.pscychresns.2015.02.001 0925-4927/Crown Copyright & 2015 Published by Elsevier Ireland Ltd. All rights reserved.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
2
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
palsy (PSP) enrolled in longitudinal studies (Whitwell et al., 2011); thalamic atrophy has also been observed in patients with frontotemporal dementia (FTD) (Cardenas et al., 2007; Chow et al., 2008) and Huntington's disease (HD) (Kassubek et al., 2005). Further investigating thalamic morphology may prove fruitful in enhancing disease etiology in the spectrum of neuropsychiatric disorders, and moreover could yield useful biomarkers with clinical utility. 1.2. Anatomical considerations of the thalamus for neuroimaging Arising from the primitive forebrain (and sharing its origins with the telencephalon), the diencephalon can be broadly divided into the thalamus and hypothalamus, although its expanse encompasses a number of other structures (such as the habenular nuclei of the epithalamus). The thalamus can be divided into dorsal and ventral divisions: the dorsal thalamus is composed of nuclei having reciprocal connections with the cerebral cortex and striatum; despite having direct input from the cortex (and in part from the basal ganglia), the ventral thalamus does not generally project to the cortex (Jones, 2007). The ventral thalamus not only has an intimate functional relationship with the dorsal thalamus (see Zikopoulos and Barbas, 2007), but importantly its nuclei (such as the reticular thalamic nucleus and zona incerta) cloak the dorsal thalamus along the intercommissural distance, providing landmarks to assist with the challenge of lateral and inferior boundary definition of the dorsal thalamic mass on neuroimaging. The dorsal thalamus is a paired structure, each roughly the size of a Brazil nut (Bertholletia excelsa) in humans. The thalami have a strategic central position in the brain which is readily identified on neuroimaging, being located at the base of each cerebral hemisphere respectively, between other subcortical structures and the cerebral cortex (Fig. 1). The two distinct dorsal thalami are found on either side of the third ventricle (which defines the thalamic medial boundary). Superiorly the thalami form the floor of the lateral ventricles for the greater part of their rostral-caudal extent. Laterally the dorsal thalamus is cloaked by the reticular thalamic nucleus (creating a fuzzy lateral boundary on MRI, by virtue of its reticulated or net-like appearance) and thence the internal capsule, ultimately separating the thalamus from the basal ganglia. Defining the inferior boundary is problematic throughout the rostro-caudal extent of the thalamus (until the pulvinar is reached), as there are a number of structures associated with the inferior boundary, including the zona incerta (of the ventral thalamus), substantia nigra, the subthalamic nucleus, the red nucleus, and hypothalamus for instance. A discerning feature of the inferior boundary is the presence of numerous white matter tracts surrounding these abovementioned structures; if these white matter tracts can be used to definitively exclude these structures from tracing, then one would have a more accurate description of the inferior dorsal thalamic boundary. Whilst the
caudal pole of the thalamus is clearly defined by the emergence of the pulvinar, defining the rostral pole is perhaps the most challenging and would therefore support a caudal-rostral approach to tracing; the rostral pole requires careful tracing to exclude the hypothalamus, and to be traced in its entirety requires thin slices (e.g., ideally o 1.5 mm). The internal organization of the dorsal thalamus can be conceptualized as having anterior, medial and lateral subdivisions, which are separated by a bowed sheet of myelinated fibers (the internal medullary lamina). Each of these subdivisions are composed of distinct thalamic nuclei, which are defined by characteristic cytoarchitecture and distinct patterns of connectivity, and their proposed involvement in distinct functional modalities (Table 1); of note, at least half of the thalamus is thought to be involved in cortico-cortical communication (the so-called higher order nuclei; see Table 1). 1.3. Development and utility of protocols for manual segmentation of the thalamus A number of different approaches have been used in MRI to examine the thalamus, including image averaging (Andreasen et al., 1994) and edge-finding using anatomical templating (Buchsbaum et al., 1996). By virtue of having automated components, both of these approaches can be useful when dealing with large cohorts, but they can be problematic for precise boundary definition and accommodating individual variability between patients (for review, see Spinks et al., 2002), especially in the context of reported significant inter-individual variability of thalamic nuclei (Uylings et al., 2008). The region of interest approach is therefore the most reliable (albeit labor intensive) approach, involving a trained rater manually segmenting the thalamus in the MRI scan of each participant (Andreasen et al., 1990; Flaum et al., 1995; Gur et al., 1998, Portas et al., 1998; Spinks et al., 2002). There are few widely available validated protocols for thalamic segmentation on MRI providing substantial anatomical detail for replication, and indeed addressing the key challenges of boundary definition. Perhaps the most widely cited is by Portas et al. (1998), who provide a description of the thalamus in 20–21 consecutive coronal 1.5 mm slices in a rostral–caudal direction; whilst the description of surrounding anatomical landmarks is thorough, there is little information regarding precise boundary definition (particularly the challenging lateral and inferior boundaries, and indeed the rostral pole of the thalamic mass). Spinks et al. (2002) describe an approach to manual segmentation of the thalamus and present 12 coronal 1.5 mm slices in their protocol, which they then used to train an artificial network for subsequent automated segmentation (with the goal of making large-scale studies more feasible); they appear to include the nuclei of the ventral thalamus. An advantage of their technique was the utilization of the tricolor image feature in BRAINS2 software, taking advantage of
Fig. 1. Thalamic morphology in 3 slices in caudal-rostral progression on MRI scans (from left to right).
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
3
Table 1 Classification, connections and functions of the dorsal thalamic nuclei. Classification
Nuclei
Principal inputs
Principal outputs
Proposed functional domain
First order (“specific”)
Anterior group Lateral geniculate Medial geniculate Ventral anterior Ventral lateral Ventral medial
Subiculum and presubiculum, mammillary nuclei
Limbic, memory
Retinal ganglion cells
Anterior limbic cortex, cingulate, subiculum, retrosplenial and presubiculum Visual cortex
Vision
Inferior colliculus
Auditory cortex of temporal lobe
Hearing
Substantia nigra
Diffuse prefrontal and cingulate
Motor
Globus pallidus, deep cerebellar nuclei, vestibular nuclei, spinothalamic tract Substantia nigra, taste, vagal, spinal
Motor and premotor cortices
Motor
Diffuse, frontal, cingulate, primary somatosensory and gustatory, medial frontal cortices
Ventral posterior Lateral posterior Lateral dorsal Medial dorsal
Medial and trigeminal lemnisci, spinothalamic tract Primary and secondary somatosensory cortex
Integration of viscero-motor information Somatic sensation of body and face Integration of sensory information Emotional expression Limbic
Higher order (“specific”)
Parietal lobe, superior colliculus and pretectum
Parietal lobe (area 5)
Central lateral
Cingulate gyrus, fornix, pretectum, hypothalamus Amygdaloid nuclear complex, olfactory cortex, hypothalamus, spinal, superior colliculus Superior colliculus and pretectum, temporal, parietal and occipital lobes Spinothalamic, cerebellar nuclei, reticular formation, substantia nigra and superior colliculus
Cingulate gyrus, retrosplenial cortex Prefrontal cortex, including frontal eye field, orbitofrontal and lateral frontal, and basal olfactory structures Temporal, parietal, occipital lobes (area 5, superior temporal gyrus) and prestriate cortex Striatum, motor cortex, somatosensory cortex, parietal cortex, frontal cortex
Centre median Parafascicular Reunions
Cerebellar nuclei, reticular formation, substantia nigra and superior colliculus Periaqueductal gray, reticular formation Fornix
Striatum, parietal and frontal cortex
Integration of sensory information Arousal, eye movement, nociception As for central lateral
Lateral, frontal cortex Hippocampal formation and adjacent regions
As for central lateral Limbic
Paratenial
Brainstem and limbic structures
Medial frontal cortex
Limbic
Pulvinar Intralaminar (“nonspecific”)
Midline (“non specific”)
Adapted from: Jones (2007)(Table 3.1); and Kandel et al. (1991) Principles of Neural Science. 3rd ed., Prentice Hall, Sydney (Table 20–1).
three imaging modalities (T1, T2 and PD weighted images) to assist with distinguishing landmarks. Some protocols are unfortunately constrained in their utility given the limited detail provided, and often refer loosely to naturalistic boundaries (Flaum et al., 1995; Gur et al., 1998; Lawrie et al., 1999; Andreasen et al., 1990; Dasari et al., 1999), and others have used an approach according to a pixel-classification method incorporating diencephalic gray matter structures (including thalamus, hypothalamus, and mammillary bodies) (Jernigan et al., 1991). 1.4. Research objective Few high quality thalamic protocols with sophisticated neuroimaging software are widely available for manual segmentation of the human dorsal thalamus on MRI scans. The aim of the current research was to develop and validate an advanced protocol for manual segmentation of the dorsal thalamus on T1-weighted MRI scans using ANALYZE 11.0 software (with 1 mm slice thickness to accommodate detailed slice-by-slice segmentation), to provide detailed descriptions of boundary definition, and to make the protocol widely accessible for further neuroimaging research.
2.2. Image acquisition and volumetric analysis MRIs were performed using a 3.0 Tesla Philips MR scanner, with an eightchannel head coil (Philips Achievas, Philips Medical Systems, Best, The Netherlands). High resolution images were acquired using a T1-weighted turbo field echo (T1 TFE) pulse sequence with parameters set as follows: repetition time (TR) 8 ms; echo time (TE) 4 ms; inversion time (TI) 650 ms; flip angle (FA) 101; number of excitations (NEX) 2; sensitivity/encoding (SENSE) factor 2.5; matrix 240 240; field of view (FOV) 240; resulting voxel size 1 1 1 mm3. There were 175 contiguous coronal slices obtained. The T1-weighted images were anonymised, randomly coded to ensure blinding, and transferred to a MacBook Pro (Apple Inc., Cupertino, CA, USA) computer at the Australian National University Medical School in Canberra, Australia. Of the control scans randomly selected, we excluded images with significant motion artefacts, degraded image quality and radiological pathology in the thalamus (strokes, injuries, lesions). Unilateral manual segmentation of the thalami (left and right) of the five control subjects and calculation of thalamic volumes (left, right and total volume) was initially performed by a single investigator (BDP), with reference to a standard three-dimensional sectional anatomical atlas (Duvernoy, 1999) and stereotactic atlas (Morel, 2007). Segmentation was performed on the right brain, and then on the left brain, yielding two thalamic volumes (left and right) per control. Segmentation was performed using ANALYZE 11.0 (Biomedical Imaging Resource, Mayo Foundation, Rochester, MN, MAYO clinic) software. For subsequent reliability, two other raters (FAW, MH-D) repeated segmentation on the same MRI scans of the same five control subjects, again, yielding two thalamic volumes (left and right) per control.
2. Methods
2.3. Thalamic boundary definition and protocol development
2.1. Participants
The protocol was designed by a neuropsychiatrist with extensive experience in the clinical neuroanatomy of the thalamus (BDP), and with reference to neuroradiological and neurosurgical atlases (Duvernoy, 1999; Morel, 2007). Manual segmentation of the dorsal thalami (left and right) occurred in 28–30 consecutive 1 mm coronal slices (from an average of 175 coronal slices of the human brain); verification occurred in other planes. The protocol proceeds in a caudal to rostral direction (intercommissural plane), starting by tracing the pulvinar as it emerges
The MRI scans of five control subjects were randomly selected from a larger cohort (age-matched and as part of an ongoing study investigating MRI changes in frontotemporal dementia), recruited via Lund University Hospital and Landskrona Hospital, Sweden. All participants gave written consent, and the study was approved by the Regional Ethics Committee for Research.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
4
under the crus fornix, and then proceeding rostrally until the crus cerebri can be seen separating from the body of the pons. The medial border was defined by the third ventricle, and superior cistern in more caudal sections. The lateral border is defined by the reticular thalamic nucleus (of the ventral thalamus). The inferior border was defined at various levels by using either nearby landmarks (e.g., habenular), characteristic white matter tracts proximal to prominent landmarks (i.e., those adjacent to the red nucleus, and subthalamic nucleus), and in more rostral sections the hypothalamic sulcus (hypothalamus, column of fornix). The superior border was defined by the body of the lateral ventricle. The most rostral slices are challenging, characterized by the emergence of the mammillary bodies and hypothalamus. In the most rostral slice the thalamus is visualized at the inferior pole of the head of the caudate, 5–6 mm caudal to the anterior commissure - a prominent landmark. The protocol attempts to use the ventral thalamus (including the reticular thalamic nucleus and zona incerta) for boundary definition, and hence exclude these structures from tracing where possible. 2.4. Intensity settings and inversion using ANALYZE software Using ANALYZE 11.0 (Mayo Clinic, Rochester, MN, USA) the Intensity Loaded Volume - Region of Interest function was used to optimize intensity settings, thereby enhancing contrast between gray and white matter, and assist in boundary definition (minimum value: 50–100; maximum value: 500–600). In addition, the Invert function (making white matter appear dark) was used to facilitate boundary definition on every slice. Each individual slice is presented in the protocol in duplicate; the first image demonstrates thalamic boundary definition for the slice, and the second image is the same slice with intensity settings inverted (see Fig. 2). 2.5. Statistical analysis Reliability of image analysis was performed by measuring intra-class correlations (a two-way mixed effects model for intra-class correlation coefficients using an absolute agreement definition (Shrout and Fleiss, 1979) for each rater between initial segmentation of both left and right thalami of the five subjects, followed by random repeated segmentation of the same scans; reliability was calculated in SPSS 17.0 (IBM Corporation, Somers, NY, USA). Inter-rater reliability of manual segmentation was calculated using a two-way mixed effects model for inter-class correlation coefficients using an absolute agreement definition of an expert rater (BDP) with two other experienced raters (FAW, MH-D), using the same MRI scans of the five subjects.
3. Results A condensed version of the protocol is presented in Table 2 (with reference to slices in Fig. 2), and a complete version is available online as Supplementary material. Fig. 2 shows an example of thalamic segmentation of the right brain on MRI using the protocol. Mean thalamic volumes (right thalamus, left thalamus, total thalamic volume) recorded by the three raters are shown in Table 3. Reliability was conducted by three raters for 10 thalami. Intra-class correlations for two raters were 0.95 (BDP) and 0.98 (FAW) (Table 4). The inter-class correlations were 0.92 (BDP, FAW) and 0.98 (BDP, MH-D) (Table 4).
4. Discussion Few high quality thalamic protocols are widely available for manual segmentation of the human thalamus on MRI scans, and our manual segmentation intra- and inter-class coefficients for reliability above 0.9 compare favorably with previously published manual segmentation methods used in quantitative structural MRI research (Looi et al., 2008; Looi et al., 2009). The protocol presented here represents an advance in the field, as it uses sophisticated neuroimaging software, is undertaken in high resolution images with 1 mm slices, and was developed with reference to both neuroradiological and neurosurgical atlases with an emphasis on boundary definition. A number of other authors have presented protocols; however, these have been constrained by a number of factors including slice thickness and limited information to assist with inferior and lateral boundary definition (Portas
et al., 1998; Spinks et al., 2002). The utilization of the specific function in ANALYZE 11.0 in our protocol to invert intensity settings has the added benefit of enhancing boundary definition, particularly of the lateral and inferior boundaries throughout the rostral-caudal extent of the dorsal thalamus. Few studies have reported a protocol for manual segmentation of the thalamus together with mean thalamic volumes to enable a comparison with volumes reported in our study. van der Werf et al. (2001) reported mean thalamic volumes across the lifespan, but information was limited as segmentation of the thalamus occurred on 8–12 coronal slices. They reported a significant decrease in total thalamic volume with increasing age in healthy subjects (apparent before the onset of volume loss of total brain volume); unadjusted mean values reported were 15.29 cm3 (22– 45 years; n ¼18), 13.41 cm3 (46–60 years; n ¼14), and 12.22 cm3 (61–82 years; n ¼24). Spinks et al. (2002) reported mean values of 7.00 cm3 (S.D. 71.12) for left thalamus and 6.55 cm3 (S.D. 70.54) for right thalamus (8 healthy males controls in their study had a mean age 25.63 years, S.D. 76.55); and Lawrie et al. (1999) reported mean values of 6.60 cm3 (S.D. 70.80) for left thalamus and 6.40 cm3 (S.D. 7 0.60) for right thalamus (in healthy controls, mean age 21.1 years, S.D.7 2.3). Ettinger et al. (2001) (following the protocol by Portas et al. (1998)) reported mean unadjusted thalamic volumes in control subjects (mean age 25.4 years, S. D.75.8) of 15.10 cm3. It may be that the smaller volumes reported in our study reflect the more detailed thalamic boundary definition offered by our protocol, tracing over more slices (by virtue of 1 mm slices), and the exclusion of the zona incerta (together with its surrounding pathways) and sub-thalamic nucleus, for instance. It is also possible that the smaller volumes reported in our study were in part accounted for by the older control subjects in our study (mean age 56, age range 35–72), supported by van der Werf et al. (2001) data suggesting decreased thalamic volume with advancing age. The protocol has sound generalizability as it can be used in a range of image analysis software packages, such as ITK-SNAP (http://www.itksnap.org/pmwiki/pmwiki.php) or MRIcro (http:// www.mccauslandcenter.sc.edu/mricro/mricro/). We envisage that advanced thalamic protocols such as our own will allow for more precise manual segmentation and thalamic boundary definition, and will be necessary to drive technologies to more accurately automatically segment the thalamus, in essence rendering protocols such as ours redundant in the future. To this end we are endeavoring to investigate the correlation between thalamic segmentation using our protocol and existing automatic segmentation techniques of the diencephalon; we anticipate a discrepancy on the basis of the precision, for instance, of the inferior boundary definition in our protocol (in the regions of the subthalamic nucleus and hypothalamus). For instance, a number of studies have used automatic segmentation protocols, and reported larger mean thalamic volumes: left thalamus ¼ 7.98 cm3 and right thalamus¼ 8.07 cm3 in 22 healthy control subjects (Zarei et al., 2010); left thalamus ¼ 7.18 cm3 and right thalamus ¼ 7.79 cm3 in 20 healthy controls (Ryan et al., 2013) and left thalamus ¼7.63 cm3 and right thalamus ¼ 7.68 cm3 in 70 elderly ‘memory complainers' (de Jong et al., 2008). It should be noted that volumetric analysis alone will be limited in its ability to differentiate disease impact on the various thalamic nuclei, and hence structural MRI will need to be combined with other modalities. Shape analysis (e.g., Spherical Harmonics, or SPHARM) is a surface-bases analysis which has the capability to reveal statistically significant regional volume differences (such as surface deflation) in the thalamus, which can provide important information regarding which regions (and by extrapolation, which subdivisions or nuclei) of the thalamus could be affected in disease states (see McKeown et al., 2008; and Janssen et al., 2012).
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
5
Fig. 2. Case example of segmentation of the thalamus on consecutive 1-mm slices on MRI, in caudal to rostral progression.
Diffusion tensor imaging will be an important tool given the fact that the various thalamic nuclei can be distinguished by their differential cortical connectivity, allowing the subdivisions and
nuclei of thalamus to be imaged according to preselected cortical areas, for instance (see Jakab et al., 2012; and Zarei et al., 2010). Given that most of the afferent and efferent fibers of the thalamus
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
6
Table 2 Table showing condensed version of the thalamic protocol for manual segmentation of the thalamus on T1-weighted MRI scans (with reference to sample slices in Fig. 2). SLICE
Description of Thalamic Boundary and Related Structures on Coronal Slices
1–3
The oval-shaped pulvinar of the thalamus can be visualized between the crus fornix and the superior cistern (the superior cistern (or quadrigeminal cistern/cistern of the great cerebral vein) is a radiological landmark above the midbrain in the horizontal plane). The pineal gland can be seen centrally, superior to the colliculi. The diagonal crossing of the crus fornix and splenium of the corpus callosum is a characteristic feature (slice 1,2). Duvernoy (1999), Fig 76–77 p154.
4–6
The pulvinar is a distinct mass of gray matter visualized under the corpus callosum, and between the fornix and the superior cistern. The pineal gland can still be seen centrally, superior to the quadrigeminal plate (superior and inferior colliculi), and the caudate can be seen at the lateral edge of the LV. In slice 6 the thalamus is becoming continuous with the midbrain; its inferior and lateral borders are easily defined. A vertical line from the caudate serves as a guide for defining the lateral extremes of the thalamus. Duvernoy (1999), Fig 75–76 p152–154; Morel (2007), Fig 4.1 p56.
7–10
The appearance of the geniculates (MGN, LGN) cause the inferior thalamic border to bulge laterally. A key feature is the posterior commissure (slices 9– 10). The MGN and LGN form its inferior border; the thalamic mass is enclosed by the curved shell of the RTN laterally; the lateral border is the internal capsule. The habenular nucleus emerges, producing a noticeable bulge on the midline. An oblique line from the habenular directed in an inferior-lateral direction at approximately 451 (towards the subiculum) marks the inferior-medial border of the thalamus at this level. Duvernoy (1999), Fig 74–75 p148– 152; Morel (2007), Fig 4.2–4.3 p 57–p58.
11–13
The MD abuts the 3V; the MGN and LGN comprise the inferior thalamic mass; the LGN becomes discontinuous with the thalamic mass in slice 13. The lateral border is defined by the internal capsule, and LV and 3V respectively defining the superior and medial borders. The habenular is less visible in slice 13; to delineate the thalamus from the brainstem in subsequent sections draw a line from the hypothalamic sulcus of 3V to the medial edge of the geniculate complex. Duvernoy (1999), Fig 72–73 p144–146; Morel (2007), Fig 4.3–4.4 p58–59.
14–16
The RN is a characteristic rounded feature at this level; white matter tracts between RN and the inferior border of the thalamic mass provides contrast for boundary definition until the hypothalamus emerges. MGN is no longer evident. LGN is separated from the greater thalamic mass by the internal capsule/ cerebral peduncle (and is no longer visible in slice 16); a line between the LGN and caudate approximates the lateral thalamic boundary. STN emerges inferiorly to the zona incerta. Duvernoy (1999), Fig 69–71 p138–142; Morel (2007), Fig 4.4–4.5 p59–60.
17–20
The LGN has been replaced by the optic tract. The globus pallidus is emerging medial to the putamen. RN and associated white matter tracts remain useful landmarks for definition of the inferior thalamic border; a vertical line from the caudate serves to approximate the lateral thalamic boundary, which is bounded by the internal capsule. The interpeduncular fossa becomes evident. Duvernoy (1999), Fig 67–69 p134–138; Morel (2007), Fig 4.6–4.7 p61–62.
21–23
RN is no longer visible to serve as a landmark; identify the STN superior to the substantia nigra. White matter tracts enveloping the medial aspect of the zona incerta/ STN aide with identification of the inferior thalamic boundary. Duvernoy (1999), Fig 64-66 p128–132; Morel (2007), Fig 4.8–4.9 p63-64.
24–26
The STN is no longer evident at this level (slice 25), and the zona incerta merges with the hypothalamus at its medial border. The thalamic mass is adjacent to the top half of the 3V, whilst the hypothalamus is adjacent to the bottom part (defining the hypothalamic sulcus). The crus cerebri can be seen separating from the body of the pons. The mammillary bodies are distinct (slice 26), and seen medial and adjacent to the optic tract. The caudate enlarges in subsequent rostral sections; as an approximation, the thalamus just overlaps the medial border of the caudate. Duvernoy (1999), Fig 63-64 p126–128; Morel (2007), Fig 4.9–4.11 p64–66.
27–29
The thalamus becomes more discrete in size, and the superior lateral tip of the thalamus abuts the inferior medial tip of the caudate head. Its lateral border is the genu of the internal capsule; the thalamus rests on the third ventricle at the midline. The hypothalamus forms much of the medial border of 3V. The mammillary bodies diminish at this level (slice 28); the thalamus is still present just rostral ( 1–2 mm) to where the mammillary bodies can no longer be visualized. The column of the fornix emerges within the hypothalamus (slice 27), is prominent in slice 29 and is a key landmark in rostral thalamic boundary definition; it has a more medial and superior location in the rostral slices beyond the thalamus, where it can eventually be found superior to the anterior commissure. The most rostral portion of the thalamus is approximately 5–6 mm caudal to the anterior commissure. Duvernoy (1999), Fig 61–62 p122–124; Morel (2007), Fig 4.11–4.12 p66–67.
LV ¼lateral ventricle; 3V ¼ third ventricle; MGN¼ medial geniculate nucleus of the thalamus; LGN¼ lateral geniculate nucleus of the thalamus; RTN ¼ reticular thalamic nucleus; MD ¼medial dorsal nucleus of the thalamus; RN¼ Red nucleus; STN¼ subthalamic nucleus.
Table 3 Table showing mean thalamic volumes recorded by the three raters for the MRI scans of the five participants.
Principal rater (BDP) Secondary rater (FAW) Secondary rater (MH-D)
Right thalamus (N¼ 5)
Left thalamus (N ¼5)
Total thalamus (N ¼ 5)
Mean (mm3)
S.D. (mm3)
Mean (mm3)
S.D. (mm3)
Mean (mm3)
S.D. (mm3)
5670.4
335.7
5592.6
410.9
11263.0
743.6
5803.2
416.0
5510.4
454.0
11313.6
868.2
5663.0
334.2
5656.2
410.8
11319.2
742.1
S.D. ¼standard deviation.
are mapped, utilizing these modalities in exploring the subcortical connectome will clarify the how various neurodegenerative disorders affect the thalamus. This novel approach combining MRI and DTI in schizophrenia has been reported, where images from modalities were superimposed to assist with thalamic boundary definition, with improved reliability (see Kim et al., 2007).
Table 4 Table showing intra- and inter-rater reliability.
Intra-rater Intra-rater Inter-rater Inter-rater
(BDP) (FAW) (BDP, FAW) (BDP, MH-D)
Correlation coefficient
95%CI
0.95 0.98 0.92 0.98
0.79–0.99 0.92–0.99 0.80–0.97 0.92–0.99
Correlation coefficients two-way, mixed effects model intra/interclass correlation, for ten measurements of the thalamus (five each side of the bisected thalamus), based on methods of Shrout and Fleisse (1979); Intra(within)-rater: intraclass correlation, Inter(between)-raters: interclass correlation; 95%CI ¼ 95% of the confidence interval.
There are a number of limitations of this protocol. In a number of slices the lateral geniculate nucleus (LGN) of the dorsal thalamus is discontinuous with the greater dorsal thalamic mass (see slices 13– 15, Fig. 2), and this will pose problems for shape analysis (e.g., SPHARM); in these slices the discontinuous LGN would need to be excluded from tracing for such analysis to proceed. Whilst the protocol attempts to exclude the ventral thalamus and does so reliably with the zona incerta, given the fuzzy lateral border made by the slender reticular thalamic nucleus, it is difficult to be
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
confident that a clear delineation between the dorsal thalamus and reticular nucleus has been achieved in all slices. We argue that the gold standard for segmentation of the thalamus in MRI is manual segmentation by an expert rater. The majority of semi- and fully automated computational segmentation is calibrated by reference to expert manual segmentations. For example, using an automated Bayesian subcortical segmentation method that includes the thalamus, called FIRST, Patenaude et al. (2011) cross-validated their analysis with manually labeled images from the training set. Similarly, other methods such as the use of ADABOOST trains automated analysis software using manually segmented template images for subcortical structures (Looi et al., 2012). The utility of the protocol is to facilitate further neuroimaging studies in the spectrum of neuropsychiatric disorders, with the aim of investigating how the morphology (shape and volume) of the thalamus is involved in the neural networks that underpin normal human cognition, emotion, movement and behavior. Determining how genetic factors are associated with thalamic morphology may yet provide further putative endophenotypes in neuropsychiatric disorders. From a clinical perspective, correlation of thalamic morphology with disease-specific clinical features (i.e., cognitive, emotional or behavioral disturbance, movement disorder) will advance our understanding of the neurobiology of disease, with the potential to identify disease-specific biomarkers (i.e., a measurable characteristic of disease). These biomarkers may prove to have prognostic significance or utility for monitoring treatment response when disease-modifying agents are available.
Contributors BDP conceived the validation study, developed the protocol, and undertook protocol validation with FAW and MH-D. BDP conceived and wrote the first draft of the manuscript, and is guarantor of the validation study and this paper; JCLL coconceived the validation study as the research network coordinator. All co-authors contributed to the paper and research described therein. BDP self-funded travel and infrastructure costs for his portion of the research; FAW, MH-D, DvW, AS, MW, CN, DV and JCLL funded research and infrastructure costs via their respective centres. Owing to space restrictions, some references have been omitted where scientific concepts/knowledge have been essentially established.
Acknowledgment The study was funded through grants from the Swedish Parkinson Fund and the Swedish Science Council (through the Basal Ganglia Disease Linnæus Consortium).
Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.pscychresns.2015. 02.001. References Adriano, F., Spoletini, I., Caltagirone, C., Spalletta, G., 2010. Updated meta-analyses reveal thalamus volume reduction in patients with first-episode and chronic schizophrenia. Schizophrenia Research 123, 1–14. Andreasen, N.C., Ehrhardt, J.C., Swayze, V.W., Alliger, R.J., Yuh, W.T.C., Cohen, G., Ziebell, S., 1990. Magnetic resonance imaging of the brain in schizophrenia. Archives of General Psychiatry 47, 35–44.
7
Andreasen, N.C., Arndt, S., Swayze, V., Cizadlo, T., Flaum, M., O’Leary, D.S., Ehrhardt, J.C., Yuh, W.T.C., 1994. Thalamic abnormalities in schizophrenia visualized through magnetic resonance image averaging. Science 266, 294–298. Atmaca, M., Yildirim, H., Ozdemir, H., Tezcan, E., Poyraz, A.K., 2007. Volumetric MRI study of key brain regions implicated in obsessive–compulsive disorder. Progress in Neuro-psychopharmacology and Biological Psychiatry 31, 46–52. Buchsbaum, M.S., Someya, T., Teng, C.Y., Abel, L., Chin, S., Najafi, A., Haier, R.J., Wu, J., Bunney, W.E., 1996. PET and MRI of the thalamus in never medicated patients with schizophrenia. American Journal of Psychiatry 153, 191–199. Cardenas, V.A., Boxer, A.L., Chao, L.L., 2007. Deformation-based morphometry reveals brain atrophy in frontotemporal dementia. Archives of Neurology 64, 873–877. Chow, T.W., Izenberg, A., Binns, M.A., Freedman, M., Stuss, D.T., Ramirez, J., Black, S.E., 2008. Magnetic resonance imaging in frontotemporal dementia shows subcortical atrophy. Dementia and Geriatric Cognitive Disorders 26, 79–88. Dasari, M., Friedman, L., Jesberger, J., Stuve, T.A., Findling, R.L., Swales, T.P., Schulz, S.C., 1999. A magnetic resonance imaging study of thalamic area in adolescent patients with either schizophrenia or bipolar disorder as compared to healthy controls. Psychiatry Research: Neuroimaging 91, 155–162. de Jong, L.W., van der Hiele, K., Veer, I.M., Houwing, J.J., Westendorp, R.G., Bollen, E.L., de Bruin, P.W., Middlekoop, H.A., van Buchem, M.A., van der Grond, J., 2008. Strongly reduced volumes of putamen and thalamus in Alzheimer's disease: an MRI study. Brain 131, 3277–3285. Du, M.Y., Wu, Q.Z., Yue, Q., Du, M.Y., Wu, Q.Z., Yue, Q., Li, J., Liao, Y., Kuang, W.H., Huang, X.Q., Chan, R.C., Mechelli, A., Gong, Q.Y., 2012. Voxelwise meta-analysis of gray matter reduction in major depressive disorder. Progress in NeuroPsychopharmacology & Biological Psychiatry 36, 11–16. Duvernoy, H.M., 1999. The Human Brian: Surface, Three-Dimensional Sectional Anatomy and MRI, 3rd ed. Springer-Verlag, New York. Ettinger, U., Chitnis, X.A., Kumari, V., Fannon, D.G., Sumich, A.L., O’Ceallaigh, S., Doku, V.C., Sharma, T., 2001. Magnetic resonance imaging of the thalamus in first-episode psychosis. The American Journal of Psychiatry 158, 116–118. Flaum, M., Swayze, V.W., O’Leary, D.S., Yuh, W.T.C., Ehrhardt, J.C., Arndt, S.V., Andreasen, N.C., 1995. Effects of diagnosis, laterality, and gender on brain morphology in schizophrenia. American Journal of Psychiatry 152, 704–714. Gilbert, A.R., Moore, G.J., Keshavan, M.S., Paulson, L.A., Narula, V., MacMaster, F.P., Stewart, C.M., Rosenberg, D.R., 2000. Decrease in thalamic volumes of paediatric patients with obsessive–compulsive disorder who are taking paroxetine. Archives of General Psychiatry 57, 449–456. Gur, R.E., Maany, V., Mozley, P.D., Swanson, C., Bilker, W., Gur, R.C., 1998. Subcortical MRI volumes in neuroleptic-naıve and treated patients with schizophrenia. American Journal of Psychiatry 155, 1711–1717. Hallahan, B., Newell, J., Soares, J.C., Brambilla, P., Strakowski, S.M., Fleck, D.E., Kieseppa, T., Altshuler, L.L., Fornito, A., Malhi, G., McIntosh, A.M., Yurgelun-Todd, D.A., Laba, K.S., Sharma, V., MacQueen, G.M., Murray, R.M., McDonald, C., 2011. Structural magnetic resonance imaging in bipolar disorder: an international collaborative mega-analysis of individual adult patient data. Biological Psychiatry 69, 326–335. Jakab, A., Blanc, R., Berenyi, E.L., Szekely, G., 2012. Generation of individualized thalamus target maps using statistical shape models and thalamocortical tractography. American Journal of Neuroradiology 33, 2110–2116. Janssen, J., Aleman-Gomez, Y., Reig, S., Schnack, H.G., Parellada, M., Graell, M., Moreno, C., Moreno, D., Mateos-Perez, J.M., Udias, J.M., Arango, C., Desco, M., 2012. Regional specificity of thalamic volume deficits in male adolescents with early onset psychosis. The British Journal of Psychiatry 200, 30–36. Jernigan, T.L., Archibald, S.L., Berhow, M.T., Sowell, E.R., Foster, D.S., Hesselink, J.R., 1991. Cerebral structure on MRI, Part I: localization of age-related changes. Biological Psychiatry 29, 55–67. Jones, E.G., 2007. The Thalamus, 2nd ed. Cambridge University Press, Cambridge. Kandel, E.R., Schwartz, J.H., Jessell, T.M., 1991. Principles of Neural Science, 3rd ed Prentice-Hall, Sydney. Kassubek, J., Juengling, F.D., Ecker, D., Landwehrmeyer, G.B., 2005. Thalamic atrophy in Huntington's disease co-varies with cognitive performance: a morphometric MRI analysis. Cerebral Cortex 15, 846–853. Kim, J., Kim, D., Kim, T, Seok, J., Chun, J.W., Oh, M., Park, H., 2007. Volumetric abnormalities in connectivity-based subregions of the thalamus in patients with chronic schizophrenia. Schizophrenia Research 97, 226–235. Konick, L.C., Friedman, L., 2001. Meta-analysis of thalamic size in schizophrenia. Biological Psychiatry 49, 178–186. Lawrie, S.M., Whalley, H., Kestelman, J.N., Abukmeil, S.S., Byrne, M., Hodges, A., Rimmington, J.E., Best, J.J.K., Owens, D.G.C., Johnstone, E.C., 1999. Magnetic resonance imaging of brain in people at high risk of developing schizophrenia. Lancet 353, 30–33. Lee, G.J., Lu, P.H., Medina, L.D., Rodriguez-Agudelo, Y., Melchor, S., Coppola, G., Braskie, M.N., Hua, X., Apostolova, L.G., Leow, A.D., Thompson, P.M., Ringman, J.M., 2013. Regional brain volume differences in symptomatic and presymptomatic carriers of familial Alzheimer's disease mutations. Journal of Neurology, Neurosurgery and Psychiatry 84, 154–162. Looi, J.C.L., Lindberg, O., Liberg, B., Tatham, V., Kumar, R., Maller, J., Millard, E., Sachdev, P., Hogberg, G., Pagani, M., Botes, L., Engman, E., Zhang, Y., Svensson, L., Wahlund, L., 2008. Volumetrics of the caudate nucleus: reliability and validity of a new manual tracing protcol. Psychiaitry Research: Neuroimaging 163, 279–288. Looi, J.C.L., Svensson, L., Lindberg, O., Zandbelt, B.B., Ostberg, P., Orndahl, E., Wahlund, L.-O., 2009. Putaminal volume in frontotemporal lobar degeneration and Alzheimer disease: differential volumesin dementia subtypes and controls. American Journal of Neuroradiology 30, 1552–1560.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
8
B.D. Power et al. / Psychiatry Research: Neuroimaging ∎ (∎∎∎∎) ∎∎∎–∎∎∎
Looi, J.C.L., Rajagopalan, P., Walterfang, M., Madsen, S.K., Thompson, P.M., Macfarlane, M.D., Ching, C., Chua, P., Velakoulis, D., 2012. Differential putaminal morphology in Frontotemporal dementia, Huntington's Disease, Alzheimer's Disease. Australian and New Zealand Journal of Psychiatry 46, 1145–1158. McKeown, M.J., Uthama, A., Abugharbieh, R., Palmer, S., Lewis, M., Huang, X., 2008. Shape (but not volume) changes in the thalami in Parkinson disease. BioMed Central Neurology 8, 8. http://dx.doi.org/10.1186/1471-2377-8-8. Morel, A., 2007. Stereotactic Atlas of the Human Thalamus and Basal Ganglia. Informa Healthcare USA, Inc., New York. Patenaude, B., Smith, S.M., Kennedy, D., Jenkinson, M.A., 2011. Bayesian model of shape and appearance for subcortical brain. NeuroImage 56, 907–922. Portas, C.M., Goldstein, J.M., Shenton, M.E., Hokama, H.H., Wible, C.G., Fischer, I., Kikinis, R., Donnino, R., Jolesz, F.A., MCarley, R.W., 1998. Volumetric evaluation of the thalamus in schizophrenic male patients using magnetic resonance imaging. Biological Psychiatry 43, 649–659. Ryan, N.S., Keihaninejad, S., Shakespeare, T.J., Lehmann, M., Crutch, S.J., Malone, I.B., Thornton, J.S., Mancini, L., Hyare, H., Yousry, T., Ridgway, G.R., Zhang, H., Modat, M., Alexander, D.C., Rossor, M.N., Ourselin, S., Fox, N.C., 2013. Magnetic resonance imaging evidence for presymptomatic change in thalamus and caudate in familial Alzheimer's disease. Brain 136, 1399–1414. Sherman, S.M., Guillery, R.W., 2006. Exploring the Thalamus and Its Role in Cortical Function, 2nd ed MIT Press, Cambridge, MA, USA.
Shrout, P.E., Fleiss, J.L., 1979. Intraclass correlations: uses in assessing rater reliability. Psychological Bulletin 86, 420–428. Spinks, R., Magnotta, V.A., Andreasen, N.C., Albright, K.C., Ziebell, S., Nopoulos, P., Cassell, M., 2002. Manual and automated measurement of the whole thalamus and mediodorsal nucleus using magnetic resonance imaging. Neuroimage 17, 631–642. Uylings, H.B., Amunts, K., Mohlberg, H., Pool, C.W., Zilles, K., Smeets, W. 2008. 3D Probabilistic map of cytoarchitectonically defined human thalamic nuclei in standard MRI. Front. Neuroinform. 1, 3D. van der Werf, Y.D., Tisserand, D.J., Visser, P.J., Hofman, P.A.M., Vuurman, E., Uylings, H.B.M., Jolles, J., 2001. Thalamic volume predicts performance on tests of cognitive speed and decreases in healthy aging. A magnetic resonance imaging-based volumetric analysis. Cognitive Brain Research 11, 377–385. Whitwell, J.L., Avula, R., Master, A., Vemuri, P., Senjem, M.L., Jones, D.T., Jack, C.R., Josephs, K.A., 2011. Disrupted thalamocortical connectivity in PSP: a resting state fMRI, DTI, VBM study. Parkinsonism and Related Disorders 17, 599–605. Zarei, M., Patenaude, B., Damoiseaux, J., Morgese, C., Smith, S., Matthews, P.M., Barkhof, F., Rombouts, S., Sanz-Arigita, E., Jenkinson, M., 2010. Combining shape and connectivity analysis: an MRI study of thalamic degeneration in Alzheimer's disease. NeuroImage 49, 1–8. Zikopoulos, B., Barbas, H., 2007. Circuits for multisensory integration and attentional modulation through the prefrontal cortex and thalamic reticular nucleus in primates. Reviews in the Neurosciences 18, 417–438.
Please cite this article as: Power, B.D., et al., Validation of a protocol for manual segmentation of the thalamus on magnetic resonance imaging scans. Psychiatry Research: Neuroimaging (2015), http://dx.doi.org/10.1016/j.pscychresns.2015.02.001i
