The amygdala subnuclei: a comparison of 2D-EPI and 3D-EPI sequences at 7 Tesla
Jeanne Leerssen
University of Amsterdam
Supervision by: M.I.C de Haan MSc & Dr. H. Cremers
Abstract
Little is known about the amygdala subnuclei in humans. One reason for this is that commonly used 3 Tesla MRI scanners cannot discern the amygdala subnuclei due to its lack in resolution. High resolution imaging at 7 Tesla makes it possible to research the amygdala subnuclei, however optimal scan sequences for this purpose are not yet established. Therefore, the current study compares two dimensional echo planar imaging (2D-EPI) and segmented three-dimensional echo planar imaging (3D-EPI) sequences at 7 Tesla and aims to find the optimal sequences for the amygdala. Participants performed an emotional face matching task while fMRI data was
collected for both 2D-EPI and 3D-EPI sequences. The results show no differences in Blood Oxygen Level Dependent (BOLD) response between the two scan sequences. The temporal signal to noise ratio (tSNR) was higher in 2D-EPI than in 3D-EPI in the amygdala and its subnuclei, which indicates better signal quality in 2D-EPI. This study is an important first step in future 7 Tesla research involving the amygdala, which could lead to interesting new findings about emotional processing and fear conditioning.
Fear Conditioning and the Amygdala Subnuclei
Research on fear conditioning is a major area of interest within the field of psychology and neuroscience. In fear conditioning a neutral stimulus (CS+) is repeatedly coupled with an aversive unconditioned stimulus (US) such as a shock, while another neutral stimulus (CS-) is presented without the aversive stimulus. After multiple trials the neutral stimulus (CS+) will elicit a fear response, whereas the CS- will not elicit a fear response (MacNamara et al., 2015). On the other hand there is the process of fear extinction, which involves the reduction of a fear response over time. Extinction occurs when the CS+ is repeatedly presented without the US, which gradually reduces the fear response.
There is a substantial amount of support for the theory that fear conditioning and extinction form the basis of anxiety disorders (Lau et al., 2008; Lissek et al., 2005). For example social phobia can develop when a person repeatedly has negative experiences in social situations. After some time, the otherwise neutral social situation itself will elicit a fear response. This example illustrates how fear-learning processes can malfunction and eventually lead to anxiety disorders and disturbances in behavior (Duvarci & Pare, 2014; Graham & Milad, 2011; Indovina, Robbins, Núñez-Elizalde, Dunn, & Bishop, 2011). Extinction on the other hand, can be exploited in treatment of anxiety disorders, such as exposure therapy, to diminish fear responses. It is important to investigate the brain mechanisms involved in fear learning and extinction in order to better understand anxiety disorders and to provide targets for treatment of these disorders.
An important brain area involved in fear conditioning is the amygdala (LeDoux, 2007). The amygdala is extensively explored in research with rodents, which reveals that the amygdala plays a crucial role in fear learning and extinction
(LaBar & Cabeza, 2006; LeDoux, 2012; J. E. LeDoux, Cicchetti, Xagoraris, & Romanski, 1990). Moreover, the different amygdalae subnuclei, such as the lateral nucleus (LA), basolateral nuclei (BA) and the central nucleus (CeA) (Janak & Tye, 2015), have their unique roles during fear learning (LeDoux, 2012; Phelps, Delgado, Nearing, & LeDoux, 2004). As can be seen in Figure 1, the LA is involved in the input of sensory information (Janak & Tye, 2015), while the CeA is involved in the output fear responses or fear expression. The BA is involved in transferring
information from LA to CeA (Duvarci & Pare, 2014).
Figure 1. Simplified representation of the amygdala subnuclei and their role in fear conditioning. CeA: central nucleus, BA: basolateral nucleus, LA: Lateral nucleus.
In human studies the amygdala is also identified as a key structure involved in fear learning and the extinction of fear, however the results in these studies remain inconsistent. Some functional Magnetic Resonance Imaging (fMRI) studies show the effect of fear conditioning on amygdala activation (Phelps et al., 2004), whereas other studies fail to show this effect (Lindner et al., 2015; MacNamara et al., 2015). This
inconsistency is also revealed in a systematic review from Sehlmeyer et al. (2009). This review showed that there was an effect of fear conditioning on amygdala activation in 25 of the 44 studies, and an effect of extinction on amygdala activation in three of the seven studies. On the other hand, two recent meta-analyses indicated that there was no effect of amygdala activation during fear learning (Fullana et al., 2016; Mechias, Etkin, & Kalisch, 2010).
Together these finding suggest that the majority of the studies with humans do not find fear related amygdala activation, which is in contrast with animal literature. There are two possible explanations for these contradictory results. Firstly, the majority of researchers investigate the amygdala as a whole, without looking at the subnuclei. From rodent research it becomes clear that there are opposite activation effect within subareas of the amygdala. These opposite effects could cancel out the activation when looking at the amygdala as one region. For example within the basolateral nucleus one neuron groups will show excitatory activation during extinction while another neuron group will have an inhibitory effect (Herry et al., 2008). Hence, this might be a reason for the fact that some studies fail to detect Blood Oxygen Level Dependent signal (BOLD) differences in the amygdala in emotional processing research. Second, in emotional processing research there are generally small BOLD effects reported in the amygdala (van der Zwaag, Da Costa, Zurcher, Adams, & Hadjikhani, 2012). These small differences in BOLD signal might not be sufficiently detected with 3 Tesla MRI scanners. Instead the development of high resolution function MRI scanners with a static magnetic field strength of 7 Tesla (Feinberg & Yacoub, 2012) is promising for detecting smaller differences in BOLD signal. The BOLD signal increases significantly with the increase of the magnetic field strength (van der Zwaag et al., 2009). Therefore, the sensitivity to detect
functional activity is higher at 7 Tesla than at a magnetic strength of 3 Tesla
(Theysohn et al., 2013). Moreover, the use of 7 Tesla increases the spatial resolution, which makes it possible to investigate even smaller brain areas (e.g. subnuclei). For instance, Faull, Jenkinson, Clare, and Pattinson (2015) successfully investigated the functional roles of subdivisions of the periaqueductal grey with the use of 7 Tesla, which is impossible with 3 Tesla scanners. In sum, the use of 7 Tesla MRI imaging is very promising for investigating the amygdala subnuclei in humans.
Scan Sequences at 7 Tesla
In order to investigate the amygdala at 7 Tesla, it is important to use scan sequences that are best in detecting amygdala activation. However, to this date researchers have not established the best scan sequences for the amygdala region at 7 Tesla. Therefore the goal of the current study is to determine which scan sequences can best be used for the amygdala. We are interested in two scan sequences: two-dimensional echo planar imaging (2D-EPI) and segmented three-two-dimensional echo planar imaging (3D-EPI). 2D-EPI techniques are most commonly used since the discovery of the BOLD effect (Poser, Koopmans, Witzel, Wald, & Barth, 2010), but a disadvantage of this technique are the long volume acquisition times at high field strength (Jorge, Figueiredo, van der Zwaag, & Marques, 2013). The 3D-EPI can be a good alternative, because this method allows for parallel imaging in two spatial dimensions, which shortens the volume acquisition times (Jorge et al., 2013). Research shows that 3D-EPI has higher spatial signal to noise ratio (sSNR) than 2D-EPI (Poser et al., 2010), but is also more sensitive to physiological noise, such as heart and respiratory rate (Jorge et al., 2013). The contribution of physiological noise can reduce the amount of signal relative to the level of background noise, in other
words the temporal signal to noise ratio (tSNR). In sum, both scan sequences have their advantages and disadvantages and it is not yet clear which sequences are best for the amygdala at 7 Tesla.
In order to compare the scan sequences accurately it is necessary to correct for two sources of noise in the data: physiology and inhomogeneity of the magnetic field. Firstly, physiological processes such as respiratory and heart rate, cause a certain amount of noise in the signal. Movement of the thorax while breathing causes head motion and modulation in the magnetic field (Hutton et al., 2011), which will lead to variation in the signal, or in other words noise. The beating of the heart causes pulsation in blood and cerebrospinal fluid, which also results in periodic signal variations (noise) (Hutton et al., 2011). Since, the amygdala is surrounded by blood vessels (Boubela et al., 2015), it is susceptible to physiological noise. Moreover, the relative amount of physiological noise increases with the increase of field strength (Triantafyllou et al., 2005). Thus, in 7 Tesla fMRI studies on the amygdala, it is really important to remove physiological noise from the data. Research has shown that correcting for these physiological components leads to a significant increase in temporal signal to noise ratio (tSNR) and will increase BOLD sensitivity (Hutton et al., 2011). More specifically, it appears that the tSNR of 3D-EPI data will increase more after physiological noise correction than the tSNR of 2D-EPI data (Jorge et al., 2013). Hence, the physiological correction is needed to accurately compare the two scan sequences. A second source of noise is the difference in the magnetic field at specific locations, in other words the inhomogeneity of the magnetic field. Regions where air and tissue meet, such as the ear canals or air filled cavities (sinuses), often suffer from these inhomogeneities (Poldrack, Mumford, & Nichols, 2011). Local inhomogeneity in the magnetic field cause different artifacts such as signal drop-out
(Hong, To, Teh, Soh, & Chuang, 2015) and spatially distorted MRI images (Poldrack et al., 2011). Since the amygdalae are located closely to sinuses (Sladky et al., 2013), this region suffers from inhomogeneity of the magnetic field (Boubela et al., 2015). Thus, in 7 Tesla studies on the amygdala it is important to correct for inhomogeneity of the magnetic field to avoid displacement and distortion in this region.
The current study aims to find the best sequences for investigating the amygdala with high-resolution imaging. We investigate this question by conducting an experiment where people have to match emotional faces (adapted from Hariri, Tessitore, Mattay, Fera, & Weinberger, 2002) while measuring brain activity in the amygdala and the amygdala subnuclei at 7 Telsa. Each participant is presented with the task two times, one time while a 2D-EPI scan is made and the other time a 3D-EPI is made. We will correct the data for physiology and local inhomogeneity. We expect there to be significant amygdala activation when a face is presented compared to a neutral stimuli. More importantly, we will investigate whether the two scan sequences will differ in the amount of BOLD contrast between the face condition and the neutral condition in the amygdala and the subnuclei. In order to assess the quality of the fMRI timeseries, we evaluate temporal signal to noise ratio (tSNR) for the two scan sequences and we will test whether the tSNR is higher in one of the two scan
sequences. The current study is fundamental for research on the amygdala subnuclei using 7 Tesla MRI scanners and could potential lead to a lot of interesting findings about fear conditioning and emotional processing.
Methods
Participants
Four students (mean age = 23, SD = 2.45, range: 20-26 years) participated in this study. All participants reported to have normal sight or corrected to normal sight with contact lenses. The participants were naïve to the content of the emotional face matching task. All participants met the criteria for MRI safety and gave their written informed consent prior to the experiment. The ethics committee of the University of Amsterdam (UvA) approved the study. The participants received 20 Euros as a reward for their participation.
Materials
In this study an emotional face matching task was used adapted from Hariri et al. (2002). The reason for using this task is that looking at fearful or angry faces will elicit amygdala activation (Costafreda, Brammer, David, & Fu, 2008). During the emotional face matching task people are presented with pictures of angry or fearful faces, derived from a standard pictures of facial affect (Ekman & Friesen, 1975). Participants are asked to indicate which of the two faces at the bottom of the screen match the emotional expression of the face at the top of the screen (target face), see Figure 2a. In the neutral condition, people have to indicate which of the two shapes presented match the target shape, see Figure 2b. The task consists of four face blocks and four neutral blocks, which are presented alternately. Each block consists of six trials, where the pictures of either faces or shapes are presented sequentially. Each trial lasted for five seconds. The total duration of the task was five minutes.
Figure 2. Stimuli of the emotional face matching task. The face condition is displayed on the left (a) and the neutral condition on the right (b).
Procedure
Participants are informed about the experiment, are asked to fill in the MRI safety screening form and are asked for their written informed consent. Next, the participant is asked to remove all metal objects and is placed inside the scanner. First the anatomical scans are made and thereafter the functional scans are made while the participant is presented with the emotional face matching task. Each participant will perform the task two times with different scan sequences: one time while a 2D-EPI is made and one time while a 3D-EPI is made. The order of the sequences is counter balanced between participants. Two resting state scans for both scan sequences were made for one participant. The experiment lasted an hour.
MRI data
Image acquisition. The structural and functional MRI images were made
using a 7T Philips MRI scanner at the Spinoza Centre for Neuroimaging. Functional data consisted of two different scan sequences: 2D Gradient Echo Planar Imaging (EPI) pulse sequence (TR = 2000 ms, TE = 23ms, flip angle = 70˚, field of view = 210x51x180, voxel size = 1.5 mm3, slices = 34, volumes = 120) and 3D Gradient
Echo Planar Imaging (EPI) pulse sequence (TR = 2000 ms, TE = 23ms, flip angle = 16˚, field of view = 210x51x180, voxel size = 1.5 mm3, slices = 34, volumes = 120). For both scan sequences (2DEPI and 3DEPI) we also collected 5 volumes in the posterior-anterior direction, which will be used for the correction of the
inhomogeneity of the magnetic field. In addition a 2D whole brain EPI (TR = 2000ms, TE = 23ms, flip angle = 70˚, field of view = 210x139.5x180, voxel size = 1.5 mm3, slices = 93) was collected to aid co-registration from functional to structural space. A T-1 weighted anatomical image was obtained with a Magnetization Prepared with 2 Rapid Gradient Echoes (MP2RAGE) sequence (TR = 6.2 ms, TE = 2.3ms, flip angle = 5˚/7˚, field of view = 205x205x163.84, voxel size = .64 mm3, slices = 256).
Preprocessing. For the correction of local inhomogeneity of the magnetic
field a TOPUP correction was used for the 2D and 3D EPI images for each participant (Andersson, Skare, & Ashburner, 2003; Smith et al., 2004). This method depends on the reversal of the phase encoding direction of the EPI images and computes an inhomogeneity field map (Hong et al., 2015). This field map can be used to correct the EPI distortion in both the task and the resting state data. Research shows that this method provides the most accurate improvements for correcting EPI distortion, compared to other methods (Hong et al., 2015). Next, the fMRI-data was
preprocessed in FEAT in FSL (version 5.0.9, http://fsl.fmrib.ox.ac.uk/fsl/) (Jenkinson, Beckmann, Behrens, Woolrich, & Smith, 2012; Smith et al., 2004; Woolrich et al., 2009) using the following preprocessing steps: 1.) 5% brain/background threshold, 2.) Motion correction (MCFLIRT), 3.) Slice time correction for 2D-EPI, but not for the 3D-EPI, 4.) 2mm smoothing, 5.) High-pass filtering, 6.) Independent component analysis (MELODIC), 7.) Registration to MNI space. To aid the co-registration, the
EPI volumes were first registered to a whole brain EPI volume. Subsequently the whole brain EPI was registered to the anatomical scan of the MP2RAGE (Marques et al., 2010), and the MP2RAGE was registered to MNI-space.
FIX in FSL (Salimi-Khorshidi et al., 2014) was used for the physiological correction of the data. This method used Independent Component Analysis (ICA) to decompose the fMRI signal into statistically independent components representing ‘signal’ and ‘noise’. The ‘noise’ components include the effects of motion,
physiology such as heart rate and respiratory rate, and scanner artifacts. FIX identifies these noisy components and regresses the noise out of the data (Salimi-Khorshidi et al., 2014). We used the standard value of 20 for the threshold of removing good versus bad components (recommended by FIX manual). An advantage of this method is that it allows for physiological correction of the data without measuring heart and respiratory rate directly.
Region of interest (ROI) selection. For each participant four ROIs were
created, one for the bilateral amygdala, and three for the subnuclei of the amygdala: basolateral nucleus (BLA), centromedial nucleus (CM) and superficial nucleus (SF). The ROIs of the whole amygdala were created using of the Harvard-Oxford
subcortical structural atlas as provided by FSL. The ROIs of the subnuclei were determined using the Jülich probabilistic anatomical atlas for the CM nucleus, the BLA nucleus and the SF nucleus of the amygdala (Amunts et al., 2005). Only voxels were included in the ROI that had a probability of 50% or higher of belonging to the specific structure. In Figure 3 the different ROIs are depicted in MNI space.
Figure 3. Depiction of the Regions of Interest. BLA: basolateral nucleus, CM: centromedial nucleus, SF: superficial nucleus.
Data analyses. One-tailed one sample t-test (group level) is used to test
whether there is significant BOLD activation in the amygdala for the contrast between the face and the neutral condition. In addition average parameter estimates for this contrast are estimated and converted to percentage BOLD signal change for each ROI. To test whether there is a difference in the amount of signal change between scan sequences, a two-tailed paired samples t-test was used.
tSNR was computed by dividing the mean of the preprocessed images by its standard deviation. The average tSNR values were calculated for each ROI.
Subsequently the difference in tSNR for each ROI is tested by means of a two-tailed paired samples t-test.
Results
Physiology Correction
None of the participants were excluded due to excessive motion (>1.5mm movement). In one participant the 3D-EPI the homogeneity correction failed, therefore the uncorrected images for this scan were used in further analyses. The mean percentages of component that were identified as noise by the FIX command were 17.8% (SD = 22.9) for the 2D-EPI and 17.5% (SD = 12.3) for the 3D-EPI (range: 2.8-53.5%). These percentages did not significantly differ from each other (t(4) = 0.024, p = 0.983). It was evaluated whether removing noisy components, indeed had an effect on the tSNR in the amygdala by conducting a 2 by 2 within subjects Analysis of Variance (ANOVA). The following effects were of interest: the main effect of FIX and the interaction effect between FIX and the type of sequence. As expected, the mean tSNR values in the amygdala were higher for the corrected images (2D: M = 64.89, SD = 27.64; 3D: M = 31.36, SD = 7.09) than for the
the main effect of the FIX correction remained non-significant (Pillai = .597, F(1,3) = 4.424, p =.126), probably due to the small sample size (N=4). The effect of the FIX correction was stronger for the 2D-EPI than for the 3D-EPI, but again the interaction effect did not reach significance (Pillai = .443, F(1,3) = 2,389, p = .220). As
expected, there was a positive correlation between the percentage components
removed and the difference score of the tSNR (corrected minus uncorrected) for 2D (r = .989, p = .005, one-tailed) and for 3D (r = .987, p = .007, one-tailed). The higher the percentage of components that are removed from the data, the bigger the increase of tSNR after the FIX correction. In sum, it seems that the physiological noise correction effectively increases the tSNR.
Emotional face matching task
To check whether amygdala activation indeed followed from the face versus neutral contrast a one-tailed one-sample t-test was computed. As expected, the results show significant BOLD response in the amygdala ROI for the face versus neutral contrast (z = 1.965, p =.027), see Figure 4.
Figure 4. Results from one sample t-test for the contrast between the face and neutral condition for the amygdala ROI.
For each ROI a two-tailed paired-samples t-test was computed to test whether there was a significant difference in percentage BOLD signal change for the
face/neutral contrast between the scan sequences. The assumption of normality for the difference scores was met for all ROIs: whole amygdala (S-W = .816, df = 4, p = .135), BLA nucleus (S-W = .993, df = 4, p = .975), CM nucleus (S-W = .867, df = 4, p = .285) and the SF nucleus (S-W = .990, df = 4, p = .958). Note that with small sample sizes it is difficult to reject the null hypothesis of meeting the normality assumption. The result reveal no significant differences in signal change for the 2D EPI compared to the 3D-EPI for the amygdala (t(3) = -.110, p = .920), the BLA (t(3) = .236, p = .829), the CM(t(3) = -1.893, p = .163) and the SF (t(3) = 1.343, p = .272), see Table 1 and Figure 5. The effect sizes were small for the whole amygdala (d = .055) and the BLA nucleus (d = .118). There were large effect sizes for the CM nucleus (d = .919) and the SF nucleus (d = .671), but these two nuclei showed an opposite effect. In the CM the 3D-EPI has higher signal change, whereas in the SF nucleus the 2D-EPI has higher signal change than the 3D-EPI.
Table 1. Mean (M) and Standard deviation (SD) of the signal change in % for the face vs. neutral contrast for the amygdala and different subnuclei (bilateral).
Percentage signal change
ROI 2D: M (SD) 3D: M(SD) 95% CI P ES (cohen’s d)
Amygdala .371 (.257) .380 (.163) -.283 - .264 .920 .055 BLA nucleus .259 (.194) .242 (.082) -.209 - .242 .829 .118 CM nucleus -.214 (.381) .104 (.121) -.869 - .233 .163 .919 SF nucleus .398 (.320) .269 (.181) -1.77 - .435 .272 .671 N=4, ROI: region of interest BLA: basolateral, CM: centromedial, SF: superficial. CI: 95% confidence interval of the mean difference, ES: Effect size, cohen’s d.
Table 2. Mean (M) and Standard deviation (SD) of the tSNR for the amygdala and different subnuclei (bilateral)
Temporal signal to noise ratio (tSNR) ROI 2D: M (SD) 3D: M (SD) 95% CI P ES (cohen’s d) Amygdala 64.89 (27.74) 31.36 (7.09) -1.297 – 68.347 .055 1.532 BLA 75.98 (31.50) 35.01 (10.95) 4.057 – 77.882 .039* 1.766 CM 44.92 (16.73) 23.52 (7.13) -2.167 – 77.882 .063 1.445 SF 52.64 (23.57) 27.00 (4.16) -7.471 – 58.747 .091 1.232 N=4, ROI: region of interest BLA: basolateral, CM: centromedial, SF: superficial. * p < .05. CI: 95% confidence interval of the mean difference, ES: Effect size, cohen’s d
Figure 5. Mean percentage signal change and standard error for 2D and 3D EPI. N.s.: not significant. BLA: basolateral nucleus, CM: centromedial nucleus, SF: Superficial nucleus.
For each ROI the tSNR was computed for the task data. A two tailed paired samples t-test was used to test for differences in tSNR for the two scan sequences. The assumption of normality for the difference scores was met for all ROIs: the whole amygdala (S-W = .889, df = 4, p = .378), the BLA nucleus (S-W = .815, df = 4, p = .133), CM nucleus (S-W = .961, df = 4, p = .787) and the SF nucleus (S-W = .925, df = 4, p = .564). The results showed a significant difference in tSNR for the BLA nucleus (t(4) = 3.532, p = .039). The mean tSNR was higher for the 2D EPI than the 3D EPI, see Table 2 and Figure 6. The results from the other ROIs also showed higher mean tSNR for 2D EPI than for the 3D EPI, see Table 2 and Figure 6. The effects were trend significant (p < .1) for the whole amygdala (t(4) = 3.064, p = .055), the CM nucleus (t(4) = 2.890, p = .063) and the SF nucleus (t(4) = 2.464, p = .091). The effect sizes for all the ROIs were large (range Cohen’s d: 1.232 – 1.766).
Figure 6. Mean temporal signal to noise ratio (tSNR) for 2D and 3D-EPI. * p < .05 significance. BLA: basolateral nucleus, CM: centromedial nucleus, SF: Superficial nucleus.
Resting state
Table 2 provides the results for the tSNR of the restingstate scans for one participant. For the 2D-EPI and 3D-EPI the mean and standard deviations of the tSNR is showed for every ROI. For the whole amygdala, and all the amygdala subnuclei the 3D-EPI has a slightly higher mean tSNR than the 2D-EPI. However, statistical
Table 2. Mean and Standard deviation (SD) of temporal signal to noise (tSNR) in resting state scans for the amygdala and different subnuclei (bilateral).
ROI 2D: Mean (SD) 3D: Mean (SD)
Amygdala 29.65 (18.23) 36.49 (14.78)
BLA nucleus 36.11 (21.81) 38.72 (12.84)
CM nucleus 19.07 (5.73) 23.87 (5.62)
SF nucleus 22.63 (11.20) 31.37 (13.13)
N=1, BLA: basolateral, CM: centromedial, SF: superficial.
Conclusion and Discussion
The main goal of the current study was to determine good scan sequences for the amygdala and the amygdala subnuclei for fMRI studies using high-resolution magnetic field of 7 Tesla. This study showed no difference in BOLD signal in the amygdala between the scan sequences for the emotional condition compared to the neutral condition in an emotional face matching task. The quality assessment of the fMRI timeseries, indicated higher tSNR for the 2D-EPI than for the 3D-EPI.This effect was significant for the BLA and trend significant for the whole amygdala, the CM and the SF nuclei, with large effect sizes for all ROIs. Contradictory to the tSNR of the task data, the resting data suggest higher tSNR for the 3D EPI than for the 2D EPI. The mentioned results should be interpreted with caution, because of the small sample size of the current study.
The results of the tSNR of the task data show higher tSNR for 2D-EPI compared to 3D-EPI. This is contradictory to previous research that concluded that after physiological correction 3D-EPI showed higher tSNR than 2D-EPI (Jorge et al., 2013), whereas before the correction the tSNR of 2D-EPI was higher than 3D-EPI. In our study we used a different method for the correction of physiology than in the study of Jorge et al. (2013). In the study of Jorge et al. (2013) they used either direct
measurements of heart and respiratory rate or principle component analysis from additional resting state data. Whereas in our study we applied principle component analysis based on a standard set of classifiers. Thus, our results could indicate that even after the physiological noise correction, there was still a significant amount of physiological noise in the data. Since 3D-EPI is more sensitive to physiological noise, this will lead to lower tSNR in 3D-EPI. Another possibility is that 2D-EPI indeed has better signal quality in the amygdala. Lastly, in one participant the homogeneity correction was successfully computed in the 2D EPI, but failed in the 3D EPI. The absence of the homogeneity correction results in more noise in the 3D-EPI and thus could be disadvantageous for the tSNR of the 3D EPI. Further research is required to resolve the matter of signal quality differences in 2D and 3D-EPI, where the effects of different physiological correction methods can be compared. The key strengths of this study were the correction for homogeneity of the magnetic field and the correction for physiology. These issues are important since the amygdala is susceptible to
inhomogeneity and physiological induced noise and without these corrections it was impossible to accurately compare the two scan sequences. However, the current study also has some limitations. The main limitation of this study is the small sample size. A sample size of four participants will result in big confidence interval, and thus less precise and reliable mean estimates. Small sample sizes will also reduce the chance of detecting a true effect, in other words it lacks the power to accept the alternative hypothesis. As a result, it could be the case that no statistically significant effect was found when in reality there was a true effect (false negative result). On the other hand the probability of finding a statistically significant effect, when in reality this effect is absent also increases (false positive results) (Button et al., 2013). Therefore it is more difficult to infer conclusions from the results. Another important limitation is the use
of atlas based ROIs. Considering the small size of the amygdala nuclei, atlas based ROIs might not be as accurate in defining the amygdala nuclei. It might be possible that the ROI did not contain all voxels in a specific amygdala nuclei or that the ROI contained voxels that did not belong to the nuclei. Also, the ROIs for the subnuclei slightly overlapped; hence some voxels belonged to two subnuclei rather than one. This is not optimal, especially when the goal is to assess difference in scan sequence for each subnuclei. Therefore this study would benefit from manually creating the ROIs for each subnuclei on the high resolution anatomical scan (MP2RAGE). Since the voxel size is smaller in the high-resolution anatomical scan (0.64mm3), than in the MNI standard brain (1mm3), the high-resolution anatomical scan allows for more accuracy in determining the amygdala and its subnuclei ROI. More specifically, manual selection of the ROI allows for the differentiation the BA and LA subnuclei, which is not yet implemented in the structural atlases that are currently available. This is especially important for research on fear conditioning, because animal studies imply that the LA and the BA have different functional roles. Thus, it makes more sense to differentiate between these nuclei than to select the whole basolateral
complex. Lastly, for this study we were not able to measure heart rate and respiratory rate directly. Although a physiological correction by means of ICA was possible, it might be even better to measure physiology directly and regress the noise out of the data. After removal of physiological noise using direct measurements, it is possible to additionally remove scanner artifacts, movement component and unclassified noise by means of ICA. This method could lead to the optimal quality of the fMRI timeseries.
More research is required to determine the best scan sequences for the
amygdala at a magnetic field strength of 7 Tesla. Future research would benefit from larger sample size and manually selected ROIs. It would also be interesting to assess
whether methods of physiological correction differ in the amount of noise that is reduced. Methods that could be explored are data driven ICA cleaning with standard classifiers or trained classifiers (FIX), direct measurements of physiology
(Retrospective Correction of Physiological Motion Effects, i.e. RETROICOR) or a combination of these methods. Lastly it would be interesting to investigate whether other scan sequences are also good candidates. For example a new imaging technique, multi-band multi-echo imaging (MBME), acquires multiple 2D or 3D slices
simultaneously. This technique has the advantage of high sensitivity, spatial specificity and is less sensitive to physiological noise (Boyacioğlu, Schulz, Koopmans, Barth, & Norris, 2015). Thus MBME has several properties that are beneficial for studies on the amygdala at high-field strengths. Future research could compare MBME with 2D-EPI and 3D-EPI and establish which scan sequence has the best results.
This study is the first to investigate scan sequences for the subnuclei of the amygdala in humans using 7 Tesla. Although the study was unable to show
differences in BOLD response between scan sequences, it is an important first step in future 7 Tesla research involving the amygdala. When optimal scan sequences are assessed for the amygdala subnuclei, this could lead to identifying the different functional roles of the subnuclei in fear conditioning and lead to many other interesting new findings in the field of emotional processing.
References
Amunts, K., Kedo, O., Kindler, M., Pieperhoff, P., Mohlberg, H., Shah, N., . . . Zilles, K. (2005). Cytoarchitectonic mapping of the human amygdala, hippocampal region and entorhinal cortex: intersubject variability and probability maps. Anatomy and embryology, 210(5-6), 343-352.
Andersson, J. L., Skare, S., & Ashburner, J. (2003). How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. Neuroimage, 20(2), 870-888.
Boubela, R. N., Kalcher, K., Huf, W., Seidel, E.-M., Derntl, B., Pezawas, L., . . . Moser, E. (2015). fMRI measurements of amygdala activation are confounded by stimulus correlated signal fluctuation in nearby veins draining distant brain regions. Scientific reports, 5, 10499.
Boyacioğlu, R., Schulz, J., Koopmans, P. J., Barth, M., & Norris, D. G. (2015). Improved sensitivity and specificity for resting state and task fMRI with multiband multi-echo EPI compared to multi-echo EPI at 7T. Neuroimage, 119, 352-361.
Button, K. S., Ioannidis, J. P., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S., & Munafò, M. R. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365-376. Costafreda, S. G., Brammer, M. J., David, A. S., & Fu, C. H. (2008). Predictors of
amygdala activation during the processing of emotional stimuli: a meta-analysis of 385 PET and fMRI studies. Brain research reviews, 58(1), 57-70. Duvarci, S., & Pare, D. (2014). Amygdala microcircuits controlling learned fear.
Ekman, P., & Friesen, W. V. (1975). Pictures of facial affect: consulting psychologists press.
Faull, O. K., Jenkinson, M., Clare, S., & Pattinson, K. T. (2015). Functional subdivision of the human periaqueductal grey in respiratory control using 7 tesla fMRI. Neuroimage, 113, 356-364.
doi:10.1016/j.neuroimage.2015.02.026
Feinberg, D. A., & Yacoub, E. (2012). The rapid development of high speed, resolution and precision in fMRI. Neuroimage, 62(2), 720-725. doi:10.1016/j.neuroimage.2012.01.049
Fullana, M. A., Harrison, B. J., Soriano-Mas, C., Vervliet, B., Cardoner, N., Avila-Parcet, A., & Radua, J. (2016). Neural signatures of human fear conditioning: an updated and extended meta-analysis of fMRI studies. Mol Psychiatry, 21(4), 500-508. doi:10.1038/mp.2015.88
Graham, B. M., & Milad, M. R. (2011). The study of fear extinction: implications for anxiety disorders. American Journal of Psychiatry, 168(12), 1255-1265. Hariri, A. R., Tessitore, A., Mattay, V. S., Fera, F., & Weinberger, D. R. (2002). The
amygdala response to emotional stimuli: a comparison of faces and scenes. Neuroimage, 17(1), 317-323.
Herry, C., Ciocchi, S., Senn, V., Demmou, L., Muller, C., & Luthi, A. (2008). Switching on and off fear by distinct neuronal circuits. Nature, 454(7204), 600-606. doi:10.1038/nature07166
Hong, X., To, X. V., Teh, I., Soh, J. R., & Chuang, K.-H. (2015). Evaluation of EPI distortion correction methods for quantitative MRI of the brain at high magnetic field. Magnetic resonance imaging, 33(9), 1098-1105.
Hutton, C., Josephs, O., Stadler, J., Featherstone, E., Reid, A., Speck, O., . . .
Weiskopf, N. (2011). The impact of physiological noise correction on fMRI at 7 T. Neuroimage, 57(1), 101-112. doi:10.1016/j.neuroimage.2011.04.018 Indovina, I., Robbins, T. W., Núñez-Elizalde, A. O., Dunn, B. D., & Bishop, S. J.
(2011). Fear-conditioning mechanisms associated with trait vulnerability to anxiety in humans. Neuron, 69(3), 563-571.
Janak, P. H., & Tye, K. M. (2015). From circuits to behaviour in the amygdala. Nature, 517(7534), 284-292. doi:10.1038/nature14188
Jenkinson, M., Beckmann, C. F., Behrens, T. E., Woolrich, M. W., & Smith, S. M. (2012). Fsl. Neuroimage, 62(2), 782-790.
Jorge, J., Figueiredo, P., van der Zwaag, W., & Marques, J. P. (2013). Signal
fluctuations in fMRI data acquired with 2D-EPI and 3D-EPI at 7 Tesla. Magn Reson Imaging, 31(2), 212-220. doi:10.1016/j.mri.2012.07.001
LaBar, K. S., & Cabeza, R. (2006). Cognitive neuroscience of emotional memory. Nat Rev Neurosci, 7(1), 54-64. doi:10.1038/nrn1825
Lau, J. Y., Lissek, S., Nelson, E. E., Lee, Y., Roberson-Nay, R., Poeth, K., . . . Pine, D. S. (2008). Fear conditioning in adolescents with anxiety disorders: results from a novel experimental paradigm. J Am Acad Child Adolesc Psychiatry, 47(1), 94-102. doi:10.1097/chi.0b01e31815a5f01
LeDoux, J. (2007). The amygdala. Curr Biol, 17(20), R868-874. doi:10.1016/j.cub.2007.08.005
LeDoux, J. (2012). Rethinking the emotional brain. Neuron, 73(4), 653-676. doi:10.1016/j.neuron.2012.02.004
LeDoux, J. E., Cicchetti, P., Xagoraris, A., & Romanski, L. M. (1990). The lateral amygdaloid nucleus: sensory interface of the amygdala in fear conditioning. The Journal of neuroscience, 10(4), 1062-1069.
Lindner, K., Neubert, J., Pfannmoller, J., Lotze, M., Hamm, A. O., & Wendt, J. (2015). Fear-potentiated startle processing in humans: Parallel fMRI and orbicularis EMG assessment during cue conditioning and extinction. Int J Psychophysiol, 98(3 Pt 2), 535-545. doi:10.1016/j.ijpsycho.2015.02.025 Lissek, S., Powers, A. S., McClure, E. B., Phelps, E. A., Woldehawariat, G., Grillon,
C., & Pine, D. S. (2005). Classical fear conditioning in the anxiety disorders: a meta-analysis. Behav Res Ther, 43(11), 1391-1424.
doi:10.1016/j.brat.2004.10.007
MacNamara, A., Rabinak, C. A., Fitzgerald, D. A., Zhou, X. J., Shankman, S. A., Milad, M. R., & Phan, K. L. (2015). Neural correlates of individual differences in fear learning. Behav Brain Res, 287, 34-41.
doi:10.1016/j.bbr.2015.03.035
Marques, J. P., Kober, T., Krueger, G., van der Zwaag, W., Van de Moortele, P.-F., & Gruetter, R. (2010). MP2RAGE, a self bias-field corrected sequence for improved segmentation and T 1-mapping at high field. Neuroimage, 49(2), 1271-1281.
Mechias, M. L., Etkin, A., & Kalisch, R. (2010). A meta-analysis of instructed fear studies: implications for conscious appraisal of threat. Neuroimage, 49(2), 1760-1768. doi:10.1016/j.neuroimage.2009.09.040
Phelps, E. A., Delgado, M. R., Nearing, K. I., & LeDoux, J. E. (2004). Extinction learning in humans: role of the amygdala and vmPFC. Neuron, 43(6), 897-905. doi:10.1016/j.neuron.2004.08.042
Poldrack, R. A., Mumford, J. A., & Nichols, T. E. (2011). Handbook of functional MRI data analysis: Cambridge University Press.
Poser, B. A., Koopmans, P. J., Witzel, T., Wald, L. L., & Barth, M. (2010). Three dimensional echo-planar imaging at 7 Tesla. Neuroimage, 51(1), 261-266. Salimi-Khorshidi, G., Douaud, G., Beckmann, C. F., Glasser, M. F., Griffanti, L., &
Smith, S. M. (2014). Automatic denoising of functional MRI data: combining independent component analysis and hierarchical fusion of classifiers.
Neuroimage, 90, 449-468. doi:10.1016/j.neuroimage.2013.11.046
Sehlmeyer, C., Schoning, S., Zwitserlood, P., Pfleiderer, B., Kircher, T., Arolt, V., & Konrad, C. (2009). Human fear conditioning and extinction in neuroimaging: a systematic review. PLoS One, 4(6), e5865.
doi:10.1371/journal.pone.0005865
Sladky, R., Baldinger, P., Kranz, G. S., Tröstl, J., Höflich, A., Lanzenberger, R., . . . Windischberger, C. (2013). High-resolution functional MRI of the human amygdala at 7T. European journal of radiology, 82(5), 728-733.
Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E., Johansen-Berg, H., . . . Flitney, D. E. (2004). Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage, 23, S208-S219.
Theysohn, N., Qin, S., Maderwald, S., Poser, B. A., Theysohn, J. M., Ladd, M. E., . . . Tendolkar, I. (2013). Memory-related hippocampal activity can be measured robustly using FMRI at 7 tesla. J Neuroimaging, 23(4), 445-451.
doi:10.1111/jon.12036
Triantafyllou, C., Hoge, R. D., Krueger, G., Wiggins, C. J., Potthast, A., Wiggins, G. C., & Wald, L. L. (2005). Comparison of physiological noise at 1.5 T, 3 T and
7 T and optimization of fMRI acquisition parameters. Neuroimage, 26(1), 243-250. doi:10.1016/j.neuroimage.2005.01.007
van der Zwaag, W., Da Costa, S. E., Zurcher, N. R., Adams, R. B., Jr., & Hadjikhani, N. (2012). A 7 tesla FMRI study of amygdala responses to fearful faces. Brain Topogr, 25(2), 125-128. doi:10.1007/s10548-012-0219-0
van der Zwaag, W., Francis, S., Head, K., Peters, A., Gowland, P., Morris, P., & Bowtell, R. (2009). fMRI at 1.5, 3 and 7 T: characterising BOLD signal changes. Neuroimage, 47(4), 1425-1434.
doi:10.1016/j.neuroimage.2009.05.015
Woolrich, M. W., Jbabdi, S., Patenaude, B., Chappell, M., Makni, S., Behrens, T., . . . Smith, S. M. (2009). Bayesian analysis of neuroimaging data in FSL.
