Information

How are fMRI responses to auditory stimuli measured?

How are fMRI responses to auditory stimuli measured?


We are searching data for your request:

Forums and discussions:
Manuals and reference books:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.

How can fMRI experiments be conducted which measure the response to auditory stimuli (e.g. music) when the apparatus makes such loud, nasty, and distracting noises? Are there quieter MRI tomographs in the meanwhile, or must the stimuli (transfered by earphones) be louder than the noise of the machine? How to measure the effect of a whishper then?


I could have found this article earlier, but as things go, I didn't. But possibly it contains most of what can be said to this question:

Jonathan E. Peellej, Methodological challenges and solutions in auditory functional magnetic resonance imaging, Front Neurosci. 2014; 8: 253


Neural responses to natural and model-matched stimuli reveal distinct computations in primary and nonprimary auditory cortex

Affiliations Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America, Zuckerman Institute of Mind, Brain and Behavior, Columbia University, New York, New York, United States of America, Laboratoire des Sytèmes Perceptifs, Département d’Études Cognitives, ENS, PSL University, CNRS, Paris France

Roles Conceptualization, Funding acquisition, Investigation, Methodology, Resources, Software, Supervision, Writing – original draft, Writing – review & editing

Affiliations Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America, Program in Speech and Hearing Biosciences and Technology, Harvard University, Cambridge, Massachusetts, United States of America, McGovern Institute for Brain Research, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America


MATERIALS AND METHODS

Eight healthy subjects (two males and six females, 19–32 years old) participated in this study, and one subject participated twice on different days. Nine experiments were conducted in total. The subjects provided informed consent, which was approved by the Institutional Review Board of the University of Minnesota. All of the fMRI experiments were performed on a 4T/90-cm bore magnet (Siemens, Erlangen, Germany) system with the Varian INOVA console (Varian Inc., Palo Alto, CA, USA). A single-loop RF surface coil (10 cm diameter) was used. Head motion was limited by a foam-padded holder. At the beginning of the experiment, axial, sagittal, and coronal anatomic MR images were acquired by a conventional T1-weighted TurboFLASH method ( 22 ). On the basis of these images, seven coronal images covering most of the calcarine fissure in V1 were selected to acquire fMRI data using GE-EPI with the following parameters: FOV = 20 × 20 cm 2 , matrix = 64 × 64 in-plane, TE = 22 ms, TR = 1.35 s, slice thickness = 5 mm, 50 ms to acquire each image slice, and 350 ms total to acquire each multislice image set.

To study the interference of acoustic noise on visual cortex activity, a paired-stimuli task paradigm was designed. In this task paradigm (see Fig. 1), the gradient sound generated during each acquisition of a multislice fMRI data set was used as the primary stimulus, and a flashing light following the sound was used as the secondary stimulus, with a variable delay of ISI between the two stimuli. A red LED checkerboard (12.8 cm × 5.9 cm) was used to generate a short (17 ms) visual flashing stimulation. The subjects lay supine in the magnet and viewed the checkerboard display via a mirror housed in a custom-designed holder above their eyes during the fMRI study. They were instructed to fixate their eyes on the center of the display throughout the fMRI study. It has been demonstrated that the neural refractory effect in the human visual cortex during a repeated visual stimulation vanishes when a relatively long delay (≥1 s) between repeated visual stimuli is applied ( 18 ). Therefore, the complication of the neural refractory effect in V1 was eliminated in this study by the use of a long TR of 1.35 s (equivalent to imaging TR). In addition, a task design similar to the conventional block task design for improving the time efficiency of images averaging was allowed and applied in this study. The experimental procedure is illustrated in Fig. 2. Six runs (runs 1–6) were conducted in each fMRI experiment. Each run was composed of seven control periods (40 image sets each) interleaved with six task periods (tasks A–F, 10 image sets each). Six different ISI values (50, 100, 200, 300, 500, and 700 ms) were used in the six task periods, respectively, in a pseudo-randomized order in each run.

Task design used in the paired-stimuli paradigm. Scanner acoustic sound was used as the primary stimulus, and a single flashing light was used as the secondary stimulus, with a delay (ISI) between them. Seven EPI slices covering the primary visual cortex were acquired, generating each scanner acoustic sound (hatched bar), and this image acquisition was repeated 10 times in each task period.

Experimental procedures for each fMRI experiment. Each task block represents 10 presentations of the stimulus. For the analysis of BOLD intensity, the common activated pixels for tasks D, E, F, A, B, and C were generated for run 1. Based on the data from these common activated pixels, six time courses were generated for tasks D, E, F, A, B, and C, respectively. A similar procedure was completed for runs 2–6. Consequently, for each experiment, six time courses were obtained in total for each task. The final quantification for each task was based on the average of the six time courses. For the analysis of the number of activated pixels, corresponding image volumes were averaged across six runs (runs 1–6) for task A. An activated map in V1 was generated based on this averaged image volumes, and the number of activated pixels for task A was calculated on the basis of this generated map. A similar procedure was completed for tasks B–F.

Two parameters—the BOLD signal intensity and the number of activated pixels—were analyzed to investigate the effect of acoustic noise on the visual cortex activation. The data analysis was performed with the use of the STIMULATE software package (MRR, University of Minnesota).

The analysis of the BOLD signal intensity was based on the common activated pixels in V1 from all six tasks for each run. Based on this criterion, for each run (e.g., run 1 in Fig. 2), the common activated region of interest (ROI) in the visual cortex was generated from the pixels that passed a statistical significance for all six tasks (e.g., tasks D, E, F, A, B, and C), using the period cross-correlation method with a cross-correlation coefficient of ≥0.4. The fMRI time course for each task and two adjacent control periods was then generated based on this common activated ROI (e.g., six time courses were generated for tasks D, E, F, A, B, and C, respectively, in run 1). For each experiment, six time courses were generated for each task (e.g., task A) in total from all six runs (runs 1–6), which were then averaged for that task. The BOLD signal intensity was quantified as the integral of the BOLD signal with the baseline of the control period subtracted from the averaged time course for each task. The results were expressed as the normalization to the average BOLD intensity for all six tasks.

For the number of activated pixels, the analysis for each task (e.g., task A) was based on the activated pixels from that single task (i.e., task A). Therefore, all of the corresponding image volumes from the same task were first averaged in image k-space across the six runs, and then an activation map was generated for each task based on the averaged image volumes using the cross-correlation method with the same cross-correlation coefficient of ≥0.4. As a result, six activation maps were generated for six different tasks in each fMRI experiment. The number of activated pixels was then calculated based on the activation map generated for each task. The ROI covering the calcarine fissure in the visual cortex was determined based on the anatomical images acquired before the fMRI experiment. Only the activated pixels located inside the ROI were counted. The number of activated pixels was also normalized to the average number of activated pixels for all the six tasks.


Contents

Functional magnetic resonance imaging (functional MRI or fMRI) is a specific magnetic resonance imaging (MRI) procedure that measures brain activity by detecting associated changes in blood flow. More specifically, brain activity is measured through low frequency BOLD signal in the brain. [11]

The procedure is similar to MRI but uses the change in magnetization between oxygen-rich and oxygen-poor blood as its basic measure. This measure is frequently corrupted by noise from various sources and hence statistical procedures are used to extract the underlying signal. The resulting brain activation can be presented graphically by color-coding the strength of activation across the brain or the specific region studied. The technique can localize activity to within millimeters but, using standard techniques, no better than within a window of a few seconds. [12]

FMRI is used both in research, and to a lesser extent, in clinical settings. It can also be combined and complemented with other measures of brain physiology such as EEG and NIRS. [13] [14] Arterial spin labeling fMRI can be used as a complementary approach for assessing resting brain functions. [15]

The physiological blood-flow response largely decides the temporal sensitivity, how well neurons that are active can be measured in BOLD fMRI. The basic time resolution parameter is the sampling rate, or TR, which dictates how often a particular brain slice is excited and allowed to lose its magnetization. TRs could vary from the very short (500 ms) to the very long (3 seconds). For fMRI specifically, the haemodynamic response is assumed to last over 10 seconds, rising multiplicatively (that is, as a proportion of current value), peaking at 4 to 6 seconds, and then falling multiplicatively. Changes in the blood-flow system, the vascular system, integrate responses to neuronal activity over time. Because this response is a smooth continuous function, sampling with faster TRs helps only to map faster fluctuations like respiratory and heart rate signals. [16]

While fMRI strives to measure the neuronal activity in the brain, the BOLD signal can be influenced by many other physiological factors other than neuronal activity. For example, respiratory fluctuations and cardiovascular cycles affect the BOLD signal being measured in the brain and therefore are usually tried to be removed during processing of the raw fMRI data. Due to these sources of noise, there have been many experts who have approached the idea of resting state fMRI very skeptically during the early uses of fMRI. It has only been very recently that researchers have become confident that the signal being measured is not an artifact caused by other physiological function. [17]

Resting state functional connectivity between spatially distinct brain regions reflects the repeated history of co-activation patterns within these regions, thereby serving as a measure of plasticity. [18]

In 1992, Bharat Biswal started his work as a graduate student at The Medical College of Wisconsin under the direction of his advisor, James S. Hyde, and discovered that the brain, even during rest, contains information about its functional organization. He had used fMRI to study how different regions of the brain communicate while the brain is at rest and not performing any active task. Though at the time, Biswal’s research was mostly disregarded and attributed to another signal source, his resting neuroimaging technique has now been widely replicated and considered a valid method of mapping functional brain networks. Mapping the brain’s activity while it is at rest holds many potentials for brain research and even helps doctors diagnose various diseases of the brain. [3]

Experiments by neurologist Marcus Raichle's lab at Washington University School of Medicine and other groups showed that the brain's energy consumption is increased by less than 5% of its baseline energy consumption while performing a focused mental task. These experiments showed that the brain is constantly active with a high level of activity even when the person is not engaged in focused mental work (the resting state). His lab has been primarily focused on finding the basis of this resting activity and is credited with many groundbreaking discoveries. These include the relative independence of blood flow and oxygen consumption during changes in brain activity, which provided the physiological basis of fMRI, as well the discovery of the well known Default Mode Network. [19]

Functional Edit

Functional connectivity is the connectivity between brain regions that share functional properties. More specifically, it can be defined as the temporal correlation between spatially remote neurophysiological events, expressed as deviation from statistical independence across these events in distributed neuronal groups and areas. [20] This applies to both resting state and task-state studies. While functional connectivity can refer to correlations across subjects, runs, blocks, trials, or individual time points, resting state functional connectivity focuses on connectivity assessed across individual BOLD time points during resting conditions. [21] Functional connectivity has also been evaluated using the perfusion time series sampled with arterial spin labeled perfusion fMRI. [22] Functional connectivity MRI (fcMRI), which can include resting state fMRI and task-based MRI, might someday help provide more definitive diagnoses for mental health disorders such as bipolar disorder and may also aid in understanding the development and progression of post-traumatic stress disorder as well as evaluate the effect of treatment. [23] Functional connectivity has been suggested to be an expression of the network behavior underlying high level cognitive function partially because unlike structural connectivity, functional connectivity often changes on the order of seconds as in the case of dynamic functional connectivity.

Default mode network Edit

Other resting state networks Edit

Processing data Edit

Many programs exist for the processing and analyzing of resting state fMRI data. Some of the most commonly used programs include SPM, AFNI, FSL (esp. Melodic for ICA), CONN, C-PAC, and Connectome Computation System (CCS).

Methods of analysis Edit

There are many methods of both acquiring and processing rsfMRI data. The most popular methods of analysis focus either on independent components or on regions of correlation.

Independent component analysis Edit

Regional analysis Edit

Other methods for characterizing resting-state networks include partial correlation, coherence and partial coherence, phase relationships, dynamic time warping distance, clustering, and graph theory. [35] [36] [37]

Reliability and reproducibility Edit

Resting-state functional magnetic resonance imaging (rfMRI) can image low-frequency fluctuations in the spontaneous brain activities, representing a popular tool for macro-scale functional connectomics to characterize inter-individual differences in normal brain function, mind-brain associations, and the various disorders. This suggests reliability and reproducibility for commonly used rfMRI-derived measures of the human brain functional connectomics. These metrics hold great potentials of accelerating biomarker identification for various brain diseases, which call the need of addressing reliability and reproducibility at first place. [38]

FMRI with EEG Edit

Many imaging experts feel that in order to obtain the best combination of spatial and temporal information from brain activity, both fMRI as well as electroencephalography (EEG) should be used simultaneously. This dual technique combines the EEG's well documented ability to characterize certain brain states with high temporal resolution and to reveal pathological patterns, with fMRI's (more recently discovered and less well understood) ability to image blood dynamics through the entire brain with high spatial resolution. Up to now, EEG-fMRI has been mainly seen as an fMRI technique in which the synchronously acquired EEG is used to characterize brain activity ('brain state') across time allowing to map (through statistical parametric mapping, for example) the associated haemodynamic changes. [39]

The clinical value of these findings is the subject of ongoing investigations, but recent researches suggest an acceptable reliability for EEG-fMRI studies and better sensitivity in higher field scanner. Outside the field of epilepsy, EEG-fMRI has been used to study event-related (triggered by external stimuli) brain responses and provided important new insights into baseline brain activity during resting wakefulness and sleep. [40]

FMRI with TMS Edit

Transcranial magnetic stimulation (TMS) uses small and relatively precise magnetic fields to stimulate regions of the cortex without dangerous invasive procedures. When these magnetic fields stimulate an area of the cortex, focal blood flow increases at the site of stimulation as well as at distant sites anatomically connected to the stimulated location. Positron emission tomography (PET) can then be used to image the brain and changes in blood flow and results show very similar regions of connectivity confirming networks found in fMRI studies and TMS can also be used to support and provide more detailed information on the connected regions. [41]

Potential pitfalls when using rsfMRI to determine functional network integrity are contamination of the BOLD signal by sources of physiological noise such as heart rate, respiration, [42] [43] and head motion. [44] [45] [46] [47] These confounding factors can often bias results in studies where patients are compared to healthy controls in the direction of hypothesized effects, for example a lower coherence might be found in the default network in the patient group, while the patient groups also moved more during the scan. Also, it has been shown that the use of global signal regression can produce artificial correlations between a small number of signals (e.g., two or three). [48] Fortunately, the brain has many signals. [49]

Research using resting state fMRI has the potential to be applied in clinical context, including use in the assessment of many different diseases and mental disorders. [50]

Disease condition and changes in resting state functional connectivity Edit

    : decreased connectivity [51] : abnormal connectivity [52] : altered connectivity [53][54] and effects of antidepressant treatment: abnormal connectivity [55][56][57][58] and effects of mood stabilizers: abnormal connectivity and network properties [59][60][61][62] : disrupted networks [63] (ADHD): altered "small networks" and thalamus changes [64] : disruption of brain systems and motor network [51] : disruption and decrease/increase in connectivity [65] : altered connectivity [66] : increase/decrease in connectivity [67] : altered connectivity [68][69] : connectivity alterations within corticolimbic circuitry and of insular cortex[70]

Other types of current and future clinical applications for resting state fMRI include identifying group differences in brain disease, obtaining diagnostic and prognostic information, longitudinal studies and treatment effects, clustering in heterogeneous disease states, and pre-operative mapping and targeting intervention. [71] As resting state measurements have no cognitive demands (instead of psychological experiments including tasks), cognitively impaired persons can also be measured easily.


Neonatal Brain Response to Deviant Auditory Stimuli and Relation to Maternal Trait Anxiety

Excessive response to unexpected or “deviant” stimuli during infancy and early childhood represents an early risk marker for anxiety disorders. However, research has yet to delineate the specific brain regions underlying the neonatal response to deviant stimuli near birth and the relation to risk for anxiety disorders. The authors used task-based functional MRI (fMRI) to delineate the neonatal response to deviant stimuli and its relationship to maternal trait anxiety.

Methods:

The authors used fMRI to measure brain activity evoked by deviant auditory stimuli in 45 sleeping neonates (mean age, 27.8 days 60% female 64% African American). In 41 of the infants, neural response to deviant stimuli was examined in relation to maternal trait anxiety on the State-Trait Anxiety Inventory, a familial risk factor for offspring anxiety.

Results:

Neonates manifested a robust and widespread neural response to deviant stimuli that resembles patterns found previously in adults. Higher maternal trait anxiety was related to higher responses within multiple brain regions, including the left and right anterior insula, the ventrolateral prefrontal cortex, and multiple areas within the anterior cingulate cortex. These areas overlap with brain regions previously linked to anxiety disorders and other psychiatric illnesses in adults.

Conclusions:

The neural architecture sensitive to deviant stimuli robustly functions in newborns. Excessive responsiveness of some circuitry components at birth may signal risk for anxiety and other psychiatric disorders.


METHODS

Generation of Stimuli

The speech sounds were voiceless consonants comprising plosives (/t/, /k/), a fricative (/f/) and an affricate (/t∫/ the phoneme at the start of “cheese”). The plosives (/t/, /k/) are non-continuants that is, are naturally produced with a short post-obstruent unvoiced airflow. The nonspeech mouth sounds comprised four ingressive click sounds: a dental click (/|/), a post-alveolar click (/!/), a lateral click (/∥/) and a bilabial click (/⊙/). Respectively, these are similar to a “tutting” sound (generally written as “tsk-tsk” or “tut-tut”), a “clop,” as in the clip-clop sound made when imitating a trotting horse, a “giddy-up” sound, the click sound made to indicate “get going” or “go faster” (e.g., when on a horse), and a “kissing” sound. These were all produced by a native speaker of British English. Thirty tokens of each sound were used in the experiment, and each token was presented once only (Figure 1).

Speech, ingressive clicks, and SCN sounds share similar amplitude envelopes. Examples of tokens from the speech, ingressive click sounds, and SCN conditions used in the experiment. The top shows the waveform versions of the sounds, whereas the middle shows their spectrotemporal structure in the form of a spectrogram. The bottom shows the amplitude envelope, which describes the mean amplitude of the sounds over time. Note that the three tokens possess a similar amplitude envelope and that the SCN token has a much simpler spectral structure than the speech and click sounds (as shown in the spectrogram).

Speech, ingressive clicks, and SCN sounds share similar amplitude envelopes. Examples of tokens from the speech, ingressive click sounds, and SCN conditions used in the experiment. The top shows the waveform versions of the sounds, whereas the middle shows their spectrotemporal structure in the form of a spectrogram. The bottom shows the amplitude envelope, which describes the mean amplitude of the sounds over time. Note that the three tokens possess a similar amplitude envelope and that the SCN token has a much simpler spectral structure than the speech and click sounds (as shown in the spectrogram).

Sounds were recorded using a solid state recorder (Edirol, R-09HR, Roland, Hosoe-cho, Hamamatsu, Japan) at 24 bits, 96 kHz, and saved as .wav files. The sound files were normalized to the same peak amplitude in Praat (Boersma & Weenink, 2010). Sounds were performed by a native British speaker who produced 30 tokens for each category of speech and ingressive click sound. SCN versions (Schroeder, 1968) were used as the baseline stimuli, and these were generated by multiplying the original waveforms with wide band noise between 50 Hz and 10 kHz.

Behavioral Testing

The stimuli were pretested to ensure that subjects could correctly categorize the sounds as speech or nonspeech. Eight subjects (five men, mean age = 25.7 years) listened to the same trains of sounds used in the fMRI section of this experiment before being asked to decide if the trains of sounds were speech or nonspeech sounds (60 trials in total, 30 speech, and 30 click trials). In a second pretest, the experiment was repeated with individual exemplars of each speech and ingressive sound (80 trials in total, each of the eight sounds was tested 10 times). In both tests, the same token was never presented more than once.

Subjects

Twenty-two healthy right-handed subjects (mean = 26.9 years, 11 men) participated in the present study. All were native English speakers, and we excluded any subjects who had experience with click languages (e.g., those having lived in South Africa). All gave informed consent according to the guidelines approved by the University College London Ethics Committee, who provided local ethics approval for this study.

A 1.5-T Siemens system with a 32-channel head coil was used to acquire 183 T2*-weighted EPI data (3 × 3 × 3 mm 3 , repetition time = 10,000 msec, acquisition time = 3 sec, echo time = 50 msec, flip = 90°) using BOLD contrast. The use of a 32-channel head coil has been shown to significantly enhance signal-to-noise ratio for fMRI in the 1.5-T field (Parikh et al., 2011 Fellner et al., 2009). A sparse scanning protocol was employed to administer the auditory stimuli in the absence of scanner noise. The first two functional volumes were discarded to remove the effect of T1 equilibration. High-resolution T1 anatomical volume images (160 sagittal slices, voxel size = 1 mm 3 ) were also acquired for each subject. During the main experimental run, subjects lay supine in the scanner in the dark and were asked to close their eyes and listen to the sounds played to them. There was no task involved so as to avoid any form of motor priming that a response task, such as a button press, might entail (Figure 2).

Perception of speech and ingressive click sounds is associated with increased activity in auditory regions. Perception of speech sounds compared with ingressive click sounds (A, white) was associated with increased BOLD activity in left middle and posterior STG (p < .005, cluster threshold = 30). Perception of speech sounds compared with SCN was associated with significant activity in the same regions but extending anteriorly in the left hemisphere (A, black) [Speech vs. SCN: −58 −48 19, −44 −6 −11, 62 −14 −4, 60 −34 6 Speech vs. Ingressive clicks: −66 16 0, 60 −20 −2, −68 −36 8, −22 −32 32]. These activations both lay within cortex identified as speech sensitive by an independent speech localizer run (A, white line). Listening to ingressive click sounds compared with speech sounds was associated with significant activity in prefrontal regions and right occipitoparietal cortex (B, black). [Ingressive clicks vs. SCN: 50 −60 28, −32 −34 8, −32 −20 −10, 42 26 50, 28 8 40, 64 −36 8 Ingressive clicks vs. Speech: 22 32 42, −30 58 0, 44 28 24, 40 10 46, 26 64 14, 44 −64 38]. Neither the comparison of click sounds to speech sounds or to SCN revealed significant activity in mouth motor regions identified by an independent motor localizer run (B, white line). (C) The common activity during the perception of both types of sounds compared with SCN in right STG (p < .005). These data indicate partially separate networks for processing of speech and ingressive click sounds whereby speech sounds are preferentially processed in left middle STG and ingressive click sounds are associated with increased activity in left posterior medial auditory areas known to comprise part of the dorsal “how” pathway. In contrast there is overlapping activity in right superior temporal cortex to both classes of sound. (D) Regions where there is a preferential response to speech in bilateral dorsolateral temporal lobes, with more extensive activation on the left. These activations were identified by the contrast [1 −0.01 −0.99, for Speech > Clicks > SCN, shown in white]. The same contrast for clicks [Clicks > Speech > SCN] did not reveal any effect in speech sensitive auditory areas in left temporal cortex (black).

Perception of speech and ingressive click sounds is associated with increased activity in auditory regions. Perception of speech sounds compared with ingressive click sounds (A, white) was associated with increased BOLD activity in left middle and posterior STG (p < .005, cluster threshold = 30). Perception of speech sounds compared with SCN was associated with significant activity in the same regions but extending anteriorly in the left hemisphere (A, black) [Speech vs. SCN: −58 −48 19, −44 −6 −11, 62 −14 −4, 60 −34 6 Speech vs. Ingressive clicks: −66 16 0, 60 −20 −2, −68 −36 8, −22 −32 32]. These activations both lay within cortex identified as speech sensitive by an independent speech localizer run (A, white line). Listening to ingressive click sounds compared with speech sounds was associated with significant activity in prefrontal regions and right occipitoparietal cortex (B, black). [Ingressive clicks vs. SCN: 50 −60 28, −32 −34 8, −32 −20 −10, 42 26 50, 28 8 40, 64 −36 8 Ingressive clicks vs. Speech: 22 32 42, −30 58 0, 44 28 24, 40 10 46, 26 64 14, 44 −64 38]. Neither the comparison of click sounds to speech sounds or to SCN revealed significant activity in mouth motor regions identified by an independent motor localizer run (B, white line). (C) The common activity during the perception of both types of sounds compared with SCN in right STG (p < .005). These data indicate partially separate networks for processing of speech and ingressive click sounds whereby speech sounds are preferentially processed in left middle STG and ingressive click sounds are associated with increased activity in left posterior medial auditory areas known to comprise part of the dorsal “how” pathway. In contrast there is overlapping activity in right superior temporal cortex to both classes of sound. (D) Regions where there is a preferential response to speech in bilateral dorsolateral temporal lobes, with more extensive activation on the left. These activations were identified by the contrast [1 −0.01 −0.99, for Speech > Clicks > SCN, shown in white]. The same contrast for clicks [Clicks > Speech > SCN] did not reveal any effect in speech sensitive auditory areas in left temporal cortex (black).

Sounds for the main run and instructions for the localizer run were presented using MATLAB with the Psychophysics Toolbox extension (Brainard, 1997), via a Denon amplifier (Denon UK, Belfast, UK) and electrodynamic headphones (MR Confon GmbH, Magdeburg, Germany) worn by the participant. Instructions were projected from a specially configured video projector (Eiki International, Inc., Margarita, CA) onto a custom-built front screen, which the participant viewed via a mirror placed on the head coil.

Each trial was a train of four different speech or click sounds, lasting 3 sec (e.g., /t/–/k/–/t∫/–/f/). The order of sounds was randomized within trial and the ordering of sound category (speech, nonspeech, SCN) was randomized across trials. Across the whole experiment, none of the 30 recorded tokens of each speech/mouth sound were repeated. A ±500 msec onset jitter was used. This main run lasted approximately 30 min.

We carried out a separate localizer run to identify in each subject the cortical regions responsible for executing mouth movements and for speech perception. This employed a block design using a continuous acquisition protocol (repetition time = 3 sec). Subjects were cued via instructions on a screen to execute mouth movements (alternating lip and tongue movements) or to listen to sentences taken from the BKB list (Bench, Kowal, & Bamford, 1979). The baseline condition was silent rest. Each block lasted 21 sec and was repeated four times. This localizer scan lasted approximately 11 min.

Preprocessing and Analyses

Functional data were analyzed using SPM8 (Wellcome Department of Imaging Neuroscience, London, UK) running on Matlab 7.4 (Mathworks, Inc., Sherborn, MA). All functional images were realigned to the first volume by six-parameter rigid body spatial transformation. Functional and structural (T1-weighted) images were then normalized into standard space using the Montreal Neurological Institute (MNI) template. Functional images were then coregistered to the T1 structural image and smoothed using a Gaussian kernel of FWHM at 8 mm. The data were high-pass filtered at 128 Hz. First-level analysis was carried out using motion parameters as regressors of no interest at the single-subject level. A random-effects model was employed in which the data were thresholded at p < .005. Voxelwise thresholding was carried out at 30 voxels to limit potential Type I errors.

Individual contrasts were carried out to investigate the BOLD response to each condition minus the silent rest or SCN, Speech versus Clicks and Clicks versus Speech. These t contrasts were taken up to a second level model. A null conjunction was used to identify significantly active voxels common to more than one condition by importing contrasts at the group level (e.g., Speech > SCN and Clicks > SCN at a threshold of p < .005, cluster threshold of 10). Significant BOLD effects were rendered on a normalized template.

A set of four 10-mm spherical ROIs were created from peak coordinates identified from separate motor and auditory localizer runs. These ROIs lay within left and right superior temporal gyri (STG) and within left and right mouth primary motor cortex (−60 −24 6, 72 −28 10, −53 −12 34, 64 0 28). Mean parameter estimates were extracted for speech and clicks compared with SCN. These are seen in Figure 3.

An additional set of 8-mm spherical ROIs were created from coordinates reported in two previous studies (Pulvermuller et al., 2006 Wilson & Iacoboni, 2006). These studies both reported significant activity in premotor regions during the perception of speech sounds (−62 −4 38, 56 −4 38, −54 −3 46, −60 2 25 Figure 4B). A diameter of 8 mm was chosen here to replicate the analyses done in these previous experiments. In these regions, mean parameter estimates were extracted for speech and clicks compared with SCN.

Finally, two cluster ROIs in ventral sensorimotor cortices were generated by the contrast of all sounds (speech, nonspeech, and SCN) over silent rest. This contrast identified a peak in ventral primary sensorimotor cortex in both hemispheres (Figure 4A). To allow statistical analyses of these data (Kriegeskorte, Simmons, Bellgowan, & Baker, 2009 Vul, Harris, Winkleman, & Pashler, 2008), ROIs were created in an iterative “hold-one-out” fashion (McGettigan et al., 2011), in which the cluster ROIs for each individual participant were created from a group contrast of [All Sounds vs. Rest inclusively masked by the motor localizer] (masking threshold p < .001, cluster threshold = 30) from the other 21 participants. Mean parameter estimates were extracted for speech, clicks, and SCN compared with silent rest.

Left auditory areas preferentially encode speech sounds, but there is no speech specific activity in primary motor cortices. Parameter estimates for speech and ingressive click sounds compared with SCN were calculated within four ROIs generated from peak coordinates from an independent localizer. A and B display the left and right speech ROIs generated from the comparison of listening to sentences against a silent rest condition (FWE = 0.05, cluster threshold = 30) with the parameter estimates displayed below. C and D show the left and right mouth motor ROIs generated from alternating lip and tongue movements compared with silent rest (FWE = 0.05, cluster threshold = 30). Speech sounds were associated with significantly increased activity in left auditory cortex compared with ingressive click sounds. There was nonsignificant difference in levels of activity in right auditory cortex or in the mouth motor regions. In all three of these regions, there was a nonsignificant increase in activity for ingressive click sounds over SCN compared with speech sounds over SCN. Error bars indicate SEM.

Left auditory areas preferentially encode speech sounds, but there is no speech specific activity in primary motor cortices. Parameter estimates for speech and ingressive click sounds compared with SCN were calculated within four ROIs generated from peak coordinates from an independent localizer. A and B display the left and right speech ROIs generated from the comparison of listening to sentences against a silent rest condition (FWE = 0.05, cluster threshold = 30) with the parameter estimates displayed below. C and D show the left and right mouth motor ROIs generated from alternating lip and tongue movements compared with silent rest (FWE = 0.05, cluster threshold = 30). Speech sounds were associated with significantly increased activity in left auditory cortex compared with ingressive click sounds. There was nonsignificant difference in levels of activity in right auditory cortex or in the mouth motor regions. In all three of these regions, there was a nonsignificant increase in activity for ingressive click sounds over SCN compared with speech sounds over SCN. Error bars indicate SEM.

Auditory-sensitive sensorimotor regions do not discriminate between speech and ingressive click sounds. The whole-brain contrast of all sounds compared with rest revealed significant activity in bilateral auditory cortices and ventral sensorimotor cortices (A, transparent white). Using this contrast, masked inclusively by the motor localizer (A, black), cluster ROIs were generated in both left and right hemispheres (A, white). Mean parameter estimates were extracted for these two regions using an interactive “leave-one-out” approach (see Methods), and these are displayed in the bottom left. The only significant comparison was that of [Speech > Rest] compared with [SCN > Rest] [Speech > Rest] compared with [Clicks > SCN] was not significantly different. To investigate whether there may be regions in premotor cortex that are specifically activated during the perception of speech compared with other sounds, we then generated 8-mm spherical ROIs on the basis of the coordinates reported in two previous studies Wilson and Iacoboni (2006) represented in B by solid white circles (−62 −4 38 and 56 −4 38), and Pulvermuller et al. (2006) represented by dotted white lines in the left hemisphere involved in movement and perception of lip and tongue movements (−54 −3 46 and −60 2 25, respectively). Mean parameter estimates for these five regions are plotted below for speech sounds compared with SCN and for ingressive clicks compared with SCN. There were no significant differences in any of these regions between the mean response to speech sounds and ingressive clicks demonstrating that activity in these areas is not specific to speech sounds. This was also the case for all subpeaks identified by the motor localizer. Error bars indicate SEM.

Auditory-sensitive sensorimotor regions do not discriminate between speech and ingressive click sounds. The whole-brain contrast of all sounds compared with rest revealed significant activity in bilateral auditory cortices and ventral sensorimotor cortices (A, transparent white). Using this contrast, masked inclusively by the motor localizer (A, black), cluster ROIs were generated in both left and right hemispheres (A, white). Mean parameter estimates were extracted for these two regions using an interactive “leave-one-out” approach (see Methods), and these are displayed in the bottom left. The only significant comparison was that of [Speech > Rest] compared with [SCN > Rest] [Speech > Rest] compared with [Clicks > SCN] was not significantly different. To investigate whether there may be regions in premotor cortex that are specifically activated during the perception of speech compared with other sounds, we then generated 8-mm spherical ROIs on the basis of the coordinates reported in two previous studies Wilson and Iacoboni (2006) represented in B by solid white circles (−62 −4 38 and 56 −4 38), and Pulvermuller et al. (2006) represented by dotted white lines in the left hemisphere involved in movement and perception of lip and tongue movements (−54 −3 46 and −60 2 25, respectively). Mean parameter estimates for these five regions are plotted below for speech sounds compared with SCN and for ingressive clicks compared with SCN. There were no significant differences in any of these regions between the mean response to speech sounds and ingressive clicks demonstrating that activity in these areas is not specific to speech sounds. This was also the case for all subpeaks identified by the motor localizer. Error bars indicate SEM.


Sadato, N. et al. Nature 380, 526–528 (1996).

Weeks, R. et al. J. Neurosci. 20, 2664–2672 (2000).

Kujala, T. et al. Trends Neurosci 23, 115–120 (2000).

Cox, R. W. Computers and Biomedical Research 29, 162–173 (1996).

Talairach, J. & Tournoux, P. Co-Planar Stereotaxic Atlas of the Human Brain (Thieme Medical, New York, 1988).

Christman, S. Cerebral Asymmetries in Sensory and Perceptual Processing (Elsevier Science B.V., 1997).

Westbury, C. F. et al. Cereb. Cortex 9, 392–405 (1999).

Rademacher, J. et al. Neuroimage 13, 669–683 (2001).

Martinez, A. et al. Nat. Neurosci. 2, 364–369 (1999).

Gandhi, S. P. et al. Proc. Natl. Acad. Sci. USA 96, 3314–3319 (1999).

Petitto, L.A. et al. Proc. Natl. Acad. Sci. USA 97, 13961–13966 (2000).

Nishimura, H. et al. Nature 397, 116 (1999).

Calvert, G. A. et al. Science 276, 593–596 (1997).

Neville, H. J. Ann. NY Acad. Sci. 608, 71–91 (1990).

Baumgart, F. et al. Nature 400, 724–726 (1999).


Methods

Subjects

All subjects were privately owned pet dogs (for details see Supplementary Table S1). The sample of subjects used for the behavioural preference test (Experiment 3) consisted of 24 dogs. Twenty of those had been used for the fMRI task (Experiment 1), and 15 for the eye-tracking test (Experiment 2).

Dog–human relationship

To evaluate the intensity and probable quality of the dog–human relationship, we conducted a caregivers’ survey (N = 15 14 females, 1 male) to assess the dogs’ age at the time when they have adopted them and how many hours per day the caregiver and the familiar person (N = 15 6 females, 9 males) on average actively spent with the dog during the week and on the weekends (see Supplementary Table S1a).

Ethical statement

All reported experimental procedures were reviewed and approved by the institutional ethics and animal welfare committee in accordance with the GSP guidelines and national legislation (ETK-21/06/2018, ETK-31/02/2019, ETK-117/07/2019) based on a pilot study at the University of Vienna (ETK-19/03/2016-2, ETK-06/06/2017). The dogs’ human caregivers gave written consent to participate in the studies before the tests were conducted. Additionally, informed consent was given for the publication of identifying images (see Fig. 4, S2 Supplementary Movie S1, S2) in an online open-access publication.

The portrait of a subjects’ caregiver shown with neutral (middle), happy (left) and angry expression (right). The respective video is shown in Supplementary Movie S1, S2.

Stimuli

We created short (3 s) videos showing human faces that are changing emotional facial expressions (see Fig. 4), transforming (morphing) from neutral to either happy or angry expression (see Movie S1, S2). The face pictures were taken from the human caregiver of each dog, a familiar person, and a stranger (for details, see Supplementary Material).

Experiment 1: fMRI task

Before the experiment, dogs had received extensive training by a professional dog trainer to habituate to the scanner environment (sounds, moving bed, ear plugs etc. see 108 ). For data acquisition, awake and unrestrained dogs laid down in prone position on the scanner bed with the head inside the coil, but could leave the scanner at any time using a custom-made ramp. The dog trainer stayed inside the scanner room throughout the entire test trial (run) outside of the dog’s visual field. Data acquisition was aborted if the dog moved extensively, or left the coil. After the scan session, the realignment parameters were inspected. If overall movement exceeded 3 mm, the run was repeated in the next test session. To additionally account for head motion, we calculated the scan-to-scan motion for each dog, referring to the frame wise displacement (FD) between the current scan t and its preceding scan t-1. For each scan exceeding the FD threshold of 0.5 mm, we entered an additional motion regressor to the first-level GLM design matrix 115,116 . On average, 3.3% (run 1) and 9.8% (run 2) scans were removed (run 1:

26/270 scans). If more than 50% of the scans exceeded the threshold, the entire run was excluded from further analyses. This was the case for one run (56%/151 scans). We truncated a run for one dog to 190 scans due to excessive motion because the dog was not available for another scan session.

The task alternated between the morph videos (500 × 500 pixels) and a black fixation cross in the centre of the screen that served as visual baseline (3–7 s jitter, mean = 5 s white background) each run started and ended with 10 s of visual baseline. The presentation order of the morph videos was randomized, but the same human model × emotion combination (i.e., angry stranger) was never directly repeated. The task was split into two runs with a duration of 4.5 min (270 volumes) each, but with a short break in-between if dogs completed both runs within one session. One run contained 60 trials (30 per emotion 20 trials per human model). Scanning was conducted with a 3 T Siemens Skyra MR-system using a 15-channel human knee-coil. Functional volumes were acquired using an echo planar imaging (EPI) sequence (multiband factor: 2) and obtained from 24 axial slices in descending order, covering the whole brain (interleaved acquisition) using an echo planar imaging (EPI) sequence (multiband factor: 2) with a voxel size of 1.5 × 1.5 × 2 mm 3 and a 20% slice gap (TR/TE = 1000 ms/38 ms, field of view = 144 × 144 × 58 mm 3 ). An MR-compatible screen (32 inch) at the end of the scanner bore was used for stimulus presentation. An eye-tracking camera (EyeLink 1000 Plus, SR Research, Ontario, Canada) was used to monitor movements of the dogs during scanning. The structural image was acquired in a prior scan session with a voxel size of 0.7 mm isotropic (TR/TE = 2100/3.13 ms, field of view = 230 × 230 × 165 mm 3 ). Data analysis and statistical tests are described in the Supplementary Material.

Experiment 2: Eye-tracking task

The eye-tracking task consisted of two tests (Experiment 2a and b) of four trials each, with at least seven days between them. In each trial the morph video of the human caregiver was presented together with either a stranger (Experiment 2a) or a familiar person (Experiment 2b). Both videos were shown with the same, either happy (two trials) or angry (two trials), facial expression. The location (left, right) of the caregiver as well as the emotion (happy, angry) was counterbalanced across the four trials of each test. The dogs went through a three-point calibration procedure first and then received two test trials in a row. At the beginning of each trial the dog was required to look at a blinking white trigger point (diameter: 7.5 cm) in the centre of the screen to start the 15-s (5 × 3 s) stimulus presentation. After a 5–10 min break, this sequence was repeated once. The dogs were rewarded with food rewards at the end of each two-trial block. Data analysis and statistical tests are described in the Supplementary Material.

Experiment 3: Behavioural preference task

The behavioural preference tests consisted of the measuring of the dogs’ movement patterns inside a rectangular arena facing two videos that were presented simultaneously on two computer screens. The screens were placed opposite to the arena entrance, at a distance of 165 cm on the floor, 135 cm apart from each other (for more details, see Supplementary Material). The dog entered the arena centrally through a tunnel with a trap door and could then move freely for the whole duration of stimulus presentation (10 × 3 s, continuous loop). Like in Experiment 2, the experiment consisted of two tests (Experiment 3a and b) of four trials each, with 1-min breaks between trials and at least seven days between the two experiments. The morph videos were shown in the exact same order and on the same sides (left, right) as in Experiment 2. After each trial, the experimenter called the dog back and went to the corridor outside the test room until the onset of the next trial. The dog was rewarded with a few pieces of dry food at the end of each experiment.

First, we manually cut out the period of stimuli presentation (30 s test trial) from the experiment recordings and then analysed the obtained videos with K9-Blyzer, a software tool which automatically tracked the dog and detected its body parts to analyse the potential behavioural preferences of the dogs towards the different displayed stimuli. Based on the dogs’ body part tracking data (head location, tail and centre of mass in each frame), the system was configured to produce measurements of specified parameters (areas of interest, dogs’ field of view, dog-screen distance) related to the dogs’ stimuli preference. We specified six parameters related to the right and left side/ screen preference (mapped to caregiver, stranger, familiar person), which are described in Supplementary Table S2. The details of the data analysis and statistical values are also provided in the Supplementary Material section.


Watch the video: fMRI - How it Works and What its Good For (July 2022).


Comments:

  1. Hinto

    Anything similar.



Write a message