Access to this full-text is provided by MDPI.
Content available from Animals
This content is subject to copyright.
Citation: Deshpande, G.; Zhao, S.;
Waggoner, P.; Beyers, R.; Morrison, E.;
Huynh, N.; Vodyanoy, V.; Denney, T.S.,
Jr.; Katz, J.S. Two Separate Brain
Networks for Predicting Trainability
and Tracking Training-Related
Plasticity in Working Dogs. Animals
2024,14, 1082. https://doi.org/
10.3390/ani14071082
Academic Editors: Heidi Lyn and
Joclyn Villegas
Received: 29 February 2024
Revised: 28 March 2024
Accepted: 29 March 2024
Published: 2 April 2024
Copyright: © 2024 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
animals
Article
Two Separate Brain Networks for Predicting Trainability and
Tracking Training-Related Plasticity in Working Dogs
Gopikrishna Deshpande 1,2,3,4,5,6,* , Sinan Zhao 1, Paul Waggoner 7, Ronald Beyers 1, Edward Morrison 8,
Nguyen Huynh 1, Vitaly Vodyanoy 8, Thomas S. Denney, Jr. 1,2,3,4 and Jeffrey S. Katz 1,2,3,4
1Auburn University Neuroimaging Center, Department of Electrical & Computer Engineering, Auburn
University, Auburn, AL 36849, USA; szz0011@tigermail.auburn.edu (S.Z.); rjb0018@auburn.edu (R.B.);
nph0013@auburn.edu (N.H.); dennets@auburn.edu (T.S.D.J.)
2Department of Psychological Sciences, Auburn University, Auburn, AL 36849, USA
3Alabama Advanced Imaging Consortium, Birmingham, AL 36849, USA
4Center for Neuroscience, Auburn University, Auburn, AL 36849, USA
5Department of Psychiatry, National Institute of Mental Health and Neurosciences, Bangalore 560029, India
6Department of Heritage Science and Technology, Indian Institute of Technology, Hyderabad 502285, India
7Canine Performance Sciences Program, College of Veterinary Medicine, Auburn University, Auburn,
AL 36849, USA; waggolp@auburn.edu
8Department of Anatomy, Physiology & Pharmacology, Auburn University, Auburn, AL 36849, USA;
morriee@auburn.edu (E.M.); vitalyvodyanoy@auburn.edu (V.V.)
*Correspondence: gopi@auburn.edu; Tel.: +1-334-844-7653
Simple Summary: The expense associated with training detection and service dogs is significant. By
employing resting-state functional resonance imaging technique, a non-invasive method capable of
probing brain function, we can identify the critical brain regions linked to selecting dogs inclined
towards successful training. These biomarkers identified before commencement of training predict
successful trainability and hence reduce training costs by obviating the need to invest in dogs that
are unlikely to be successful. Furthermore, our research extends to elucidating the identified brain
regions in dogs that exhibit homologous functions to those found in the human brain, offering
valuable insights into the evolutionary parallels between humans and our closest animal companions.
Abstract: Functional brain connectivity based on resting-state functional magnetic resonance imaging
(fMRI) has been shown to be correlated with human personality and behavior. In this study, we
sought to know whether capabilities and traits in dogs can be predicted from their resting-state
connectivity, as in humans. We trained awake dogs to keep their head still inside a 3T MRI scanner
while resting-state fMRI data was acquired. Canine behavior was characterized by an integrated
behavioral score capturing their hunting, retrieving, and environmental soundness. Functional scans
and behavioral measures were acquired at three different time points across detector dog training.
The first time point (TP1) was prior to the dogs entering formal working detector dog training. The
second time point (TP2) was soon after formal detector dog training. The third time point (TP3) was
three months’ post detector dog training while the dogs were engaged in a program of maintenance
training for detection work. We hypothesized that the correlation between resting-state FC in the
dog brain and behavior measures would significantly change during their detection training process
(from TP1 to TP2) and would maintain for the subsequent several months of detection work (from
TP2 to TP3). To further study the resting-state FC features that can predict the success of training,
dogs at TP1 were divided into a successful group and a non-successful group. We observed a core
brain network which showed relatively stable (with respect to time) patterns of interaction that
were significantly stronger in successful detector dogs compared to failures and whose connectivity
strength at the first time point predicted whether a given dog was eventually successful in becoming a
detector dog. A second ontologically based flexible peripheral network was observed whose changes
in connectivity strength with detection training tracked corresponding changes in behavior over
the training program. Comparing dog and human brains, the functional connectivity between the
brain stem and the frontal cortex in dogs corresponded to that between the locus coeruleus and left
Animals 2024,14, 1082. https://doi.org/10.3390/ani14071082 https://www.mdpi.com/journal/animals
Animals 2024,14, 1082 2 of 23
middle frontal gyrus in humans, suggestive of a shared mechanism for learning and retrieval of odors.
Overall, the findings point toward the influence of phylogeny and ontogeny in dogs producing two
dissociable functional neural networks.
Keywords: canine; dog; resting state; functional MRI; functional connectivity; comparative biology
1. Introduction
For over tens of thousands of years, dogs (Canis familiaris) have played versatile roles
in human societies, from companionship to performing specific tasks such as detection and
therapy work, underscoring their unique trainability and social intelligence [
1
]. For exam-
ple, sniffer dogs have been trained to detect explosives, as hearing dogs for alerting people
who are deaf to important sounds [
2
], as therapy dogs for supporting language-impaired
children [
3
] and those with stress- and anxiety-related disorders [
4
], etc. Consequently,
the ease with which dogs can be trained to perform various tasks, as well as their general
behavioral capabilities such as hunting and retrieving, become critical parameters for se-
lecting dogs for training. Despite the importance of canine capabilities to human society
and the capital costs incurred in training dogs [
5
,
6
], research into the neural basis for the
behavior of dogs and their trainability is sparse. Such a research endeavor is important for
the following reasons. First, it could potentially lead to procedures involving non-invasive
measurement of canine neural function as a criterion for selecting dogs to be trained,
thereby limiting capital expenditure on less trainable dogs or dogs with less favorable
behavioral capabilities. Training a service dog that is effective in its duties can be quite
costly, with many experts estimating that the expenses could reach figures from $20,000
to $50,000 [
7
]. Therefore, by pinpointing suitable candidates for detection work early, the
average cost of training could be significantly reduced. Second, a scientific account of the
neural structures/processes supporting behavior in dogs could be evaluated with respect
to similar literature in humans and other species for understanding the evolutionary role
of brain–behavior relationships. This comparative evaluation is particularly relevant for
dogs since they are a unique species because they have socially co-evolved with humans
for thousands of years [8].
One of the most widely used non-invasive tools for investigating the neural basis of
behavior in humans is functional magnetic resonance imaging (fMRI), which is based on
the principle that changes in the local concentration of the paramagnetic deoxygenated
hemoglobin due to neural activity leads to enhancement in the magnetic resonance signal.
While functional magnetic resonance imaging (fMRI) has revolutionized our understanding
of the neural basis of behavior in humans, its application to studying the canine brain,
particularly through non-invasive resting-state fMRI (
RS-fMRI
), represents a novel frontier
in comparative neurobiology. Of particular interest to brain–behavioral relationships is the
fMRI signal in the absence of a goal-directed task, known as resting-state fMRI (
RS-fMRI
),
which displays spatially correlated structure-forming distributed brain networks [
9
]. In
humans, the functional connectivity (FC) between brain regions in such networks (mea-
sured through temporal correlation between fMRI time series from those regions), has
been shown to co-vary with various behavioral variables (e.g., cognitive abilities, attention,
working memory, cognitive control) [
10
–
14
]. Szabóet al., 2019 [
15
] and Beckmann et al.,
2020 [
16
] employed resting-state fMRI and the independent component analysis (ICA)
method to demonstrate the presence of resting-state networks (RSNs) in dogs, both in
the awake and anesthetized state, like those found in humans and other animals. While
those studies have been conducted, they have not yet shown the functional properties or
interactions between those networks or any correlation with behavioral data. In our work,
we extended this concept from humans to dogs, and surmised that fMRI-based resting-state
functional connectivity in brain networks obtained from the dog brain would correlate
with canine behavior.
Animals 2024,14, 1082 3 of 23
The methodological challenges involved in measuring fMRI-based functional connec-
tivity in the dog brain are quite daunting. First, head movement poses a big problem for
fMRI since the displacement of the head from one acquisition to the next, if not corrected
for, can appear like a change in image intensity which is unrelated to underlying neural
activity. Therefore, fMRI studies employing animals have either immobilized them [
17
,
18
]
or anesthetized them. The former reduces the comparative validity of the experiment
(e.g., humans are not immobilized) and may make the experiment less ethologically valid,
while the latter has been proved to alter neural activity and connectivity [
19
,
20
]. Therefore,
imaging awake dogs is the best workable option. Recent studies have made strides in this
regard. Both our group [
8
,
21
–
23
] and three other groups [
7
,
24
–
35
] have been successful in
training dogs to keep their head still inside the MRI scanner while fMRI data is acquired.
Specifically, we showed the existence of resting-state brain networks in dogs by scanning
them in a fully conscious and unrestrained state. Critically, we employed optical head
motion tracking with an external camera device to record and account for head motion,
which is inevitable even if the dog is trained to keep its head still. Using this paradigm
enabled us to non-invasively measure whole brain functional connectivity in awake dogs.
The objectives of the current study were twofold. First, we wanted to discover the
resting-state brain networks whose change in the strength of connectivity during a canine
training regimen mirrored corresponding changes in their behavior. Second, we were
interested in investigating whether resting-state brain networks estimated from fMRI data
acquired before the commencement of the training regimen were able to predict whether a
given dog would eventually graduate to become a detector dog or not. In order to achieve
these objectives, we designed a longitudinal experimental paradigm where fMRI data
and behavioral assessments were acquired at multiple time points (TPs) across the time
of participation of dogs in this study. The first time point (TP1) was prior to the dogs
entering formal working detector dog training, but after 1–3 months of MRI training [
21
,
22
]
to keep their head still inside the scanner. The second time point (TP2) was soon after
formal detector dog training which lasted about 3 months. The third time point (TP3) was
three months post detector dog training while the dogs were engaged in a program of
maintenance training for detector dog work. We hypothesized that the correlation between
resting-state FC in the dog brain and behavior measures would significantly change during
their detection training process (from TP1 to TP2) and would maintain for the subsequent
several months of detection work (from TP2 to TP3). This was based on the premise that
detection training would lead to strengthening of certain functional connectivities from
TP1 to TP2, primarily in brain regions/networks involved in processing of reward cues,
reinforcement learning and olfactory processing. Behaviorally, we observe that these func-
tions are engaged during detection training and improve from TP1 to TP2. In addition,
whatever gains are made during detection training are often maintained behaviorally from
TP2 to TP3 and this is referred to as the maintenance period. Maintenance of learned
detection-related behaviors is critical for the operational effectiveness of these dogs. Conse-
quently, we hypothesize that strengthened functional connectivity in the above networks
would then be maintained from TP2 to TP3. Further, the strengthening of functional con-
nectivity in identified paths would also mirror corresponding behavioral improvements
from TP1 to TP2 and subsequent maintenance till TP3. The underlying assumption is
that improvements in behavior are supported by strengthening of connectivity in brain
regions/networks involved in essential aspects of detection training, i.e., the processing
of reward cues, the reinforcement learning, and olfactory processing. This assumption is
supported by the human neuroimaging literature showing a significant correlation between
gains in task performance with increased functional connectivity in regions subserving the
task [
36
]. The study aims to explore whether brain connectivity patterns measured using
fMRI can predict a dog’s success in a training program. By examining brain activity before
training, we hope to identify biomarkers that could be used to select dogs more likely to
become effective working dogs.
Animals 2024,14, 1082 4 of 23
2. Materials and Methods
We recruited forty Labrador retriever dogs (24 males/16 females) with ages in the
range of 12 to 36 months from the Auburn University Canine Performance Sciences Program
and iK9 LLC (www.ik9.com accessed on 29 February 2024).
2.1. Dog Training and Preparation
The dogs for this study came from a working dog acquisition process intended to select
dogs that have the potential to be trained successfully for working tasks. A standardized
assessment test was used for judging the workability of candidate dogs, and that assessment
was also used as a behavioral measure for comparison with fMRI imaging metrics.
Once acquired, the dogs began training for being scanned in the MRI while fully awake
and unrestrained. For this purpose, a full-scale MRI simulator was fabricated. Additionally,
a couple of simulated human knee coils, into which the dog must learn to place and hold
the head, were fabricated for use in training the MRI routine (Figure 1). To prevent any
possible hearing damage, the dogs wore suitable ear protection during the procedure.
Animals 2024, 14, x FOR PEER REVIEW 4 of 24
gions subserving the task [36]. The study aims to explore whether brain connectivity pat-
terns measured using fMRI can predict a dog’s success in a training program. By examin-
ing brain activity before training, we hope to identify biomarkers that could be used to
select dogs more likely to become effective working dogs.
2. Materials and Methods
We recruited forty Labrador retriever dogs (24 males/16 females) with ages in the
range of 12 to 36 months from the Auburn University Canine Performance Sciences Pro-
gram and iK9 LLC (www.ik9.com accessed on 29 February 2024).
2.1. Dog Training and Preparation
The dogs for this study came from a working dog acquisition process intended to
select dogs that have the potential to be trained successfully for working tasks. A stand-
ardized assessment test was used for judging the workability of candidate dogs, and that
assessment was also used as a behavioral measure for comparison with fMRI imaging
metrics.
Once acquired, the dogs began training for being scanned in the MRI while fully
awake and unrestrained. For this purpose, a full-scale MRI simulator was fabricated. Ad-
ditionally, a couple of simulated human knee coils, into which the dog must learn to place
and hold the head, were fabricated for use in training the MRI routine (Figure 1). To pre-
vent any possible hearing damage, the dogs wore suitable ear protection during the pro-
cedure.
Figure 1. Mock MRI scanner and mock head coil for training dogs.
The training process was based on progressive positive principles reinforcement
learning and was separated into two stages. Throughout the first stage of the training pro-
cess, a recording of the MRI operation sound was played, and the volume of the sound
gradually increased until it was similar to an actual scan. Once a dog put its head within
the knee coil (Figure 2) and remained relatively motionless for approximately 5 min, it
was treated with food rewards. When the dog repeated this performance several times
across the course of an approximately 30 min training session, they were ready for the
next stage.
Figure 1. Mock MRI scanner and mock head coil for training dogs.
The training process was based on progressive positive principles reinforcement
learning and was separated into two stages. Throughout the first stage of the training
process, a recording of the MRI operation sound was played, and the volume of the sound
gradually increased until it was similar to an actual scan. Once a dog put its head within
the knee coil (Figure 2) and remained relatively motionless for approximately 5 min, it was
treated with food rewards. When the dog repeated this performance several times across
the course of an approximately 30 min training session, they were ready for the next stage.
The second stage of training was performed inside the real MRI scanner with the
running of a functional sequence. Transitioning to the actual MRI went smoothly for some
dogs but was more difficult for others. The final target performance for the training was
for a dog to voluntarily enter the MRI scanner, position its head into the knee coil and
remain relatively motionless for an approximately 5 min run and repeat the behavior across
multiple runs within an hour-long session of scanning. The time to train the dogs from
initial training to successful scan in the actual MRI ranged from 12 h to 30 h (on average,
about 18 h), which was divided into several one-hour sessions across days. More details
about the training procedure can be obtained from our previous publications [8,21–23,37].
Animals 2024,14, 1082 5 of 23
Animals 2024, 14, x FOR PEER REVIEW 5 of 24
Figure 2. A dog in the MRI simulator being prompted to place his head in the mock coil.
The second stage of training was performed inside the real MRI scanner with the
running of a functional sequence. Transitioning to the actual MRI went smoothly for some
dogs but was more difficult for others. The final target performance for the training was
for a dog to voluntarily enter the MRI scanner, position its head into the knee coil and
remain relatively motionless for an approximately 5 min run and repeat the behavior
across multiple runs within an hour-long session of scanning. The time to train the dogs
from initial training to successful scan in the actual MRI ranged from 12 h to 30 h (on
average, about 18 h), which was divided into several one-hour sessions across days. More
details about the training procedure can be obtained from our previous publications [8,21–
23,37].
2.2. Longitudinal Experimental Design
To track the changes in functional imaging metrics with time, all of the fMRI scans
and behavioral measures were acquired at multiple time points (TPs) across the time of
participation of the dogs in this study (Figure 3). The first time point (TP1) was prior to
the dogs entering formal working detector dog training, but after 1–3 months of MRI train-
ing to keep their head still inside the scanner. The second time point (TP2) was soon after
formal detector dog training which lasted about 3 months. The third time point (TP3) was
three months post detector dog training while the dogs were engaged in a program of
maintenance training for detector dog work.
Figure 3. A schematic of the longitudinal experimental design.
2.3. Working Dog Assessments
All of the dogs in this project were assessed for their potential to be successfully
trained and employed for working detector dog tasks. The assessment we employed is a
Figure 2. A dog in the MRI simulator being prompted to place his head in the mock coil.
2.2. Longitudinal Experimental Design
To track the changes in functional imaging metrics with time, all of the fMRI scans
and behavioral measures were acquired at multiple time points (TPs) across the time of
participation of the dogs in this study (Figure 3). The first time point (TP1) was prior to the
dogs entering formal working detector dog training, but after 1–3 months of MRI training
to keep their head still inside the scanner. The second time point (TP2) was soon after
formal detector dog training which lasted about 3 months. The third time point (TP3) was
three months post detector dog training while the dogs were engaged in a program of
maintenance training for detector dog work.
Animals 2024, 14, x FOR PEER REVIEW 5 of 24
Figure 2. A dog in the MRI simulator being prompted to place his head in the mock coil.
The second stage of training was performed inside the real MRI scanner with the
running of a functional sequence. Transitioning to the actual MRI went smoothly for some
dogs but was more difficult for others. The final target performance for the training was
for a dog to voluntarily enter the MRI scanner, position its head into the knee coil and
remain relatively motionless for an approximately 5 min run and repeat the behavior
across multiple runs within an hour-long session of scanning. The time to train the dogs
from initial training to successful scan in the actual MRI ranged from 12 h to 30 h (on
average, about 18 h), which was divided into several one-hour sessions across days. More
details about the training procedure can be obtained from our previous publications [8,21–
23,37].
2.2. Longitudinal Experimental Design
To track the changes in functional imaging metrics with time, all of the fMRI scans
and behavioral measures were acquired at multiple time points (TPs) across the time of
participation of the dogs in this study (Figure 3). The first time point (TP1) was prior to
the dogs entering formal working detector dog training, but after 1–3 months of MRI train-
ing to keep their head still inside the scanner. The second time point (TP2) was soon after
formal detector dog training which lasted about 3 months. The third time point (TP3) was
three months post detector dog training while the dogs were engaged in a program of
maintenance training for detector dog work.
Figure 3. A schematic of the longitudinal experimental design.
2.3. Working Dog Assessments
All of the dogs in this project were assessed for their potential to be successfully
trained and employed for working detector dog tasks. The assessment we employed is a
Figure 3. A schematic of the longitudinal experimental design.
2.3. Working Dog Assessments
All of the dogs in this project were assessed for their potential to be successfully trained
and employed for working detector dog tasks. The assessment we employed is a variant of
those widely used across many operational agencies (e.g., U.S. Military, Homeland Security
agencies, law enforcement agencies) for assessing candidate working dogs. The assessment
has two components, performance, and environmental soundness [
38
,
39
]. The performance
element assessed the propensity of the dog to retrieve a thrown object; interest, focus, and
desire to possess a toy reward; the propensity of the dog to use its nose in hunting for
a desired object; the amount of effort and degree of distractibility during retrieving and
hunting games. Utilizing thrown objects in training exercises enables dogs to hone their
olfactory abilities by searching and recognizing a specific scent. Throwing objects can
also serve as a reward-based training method since after successfully retrieving the target
objects, a dog could be rewarded with a treat or praise. This type of positive reinforcement
can help to motivate the dog to continue to work hard and improve its detection skills.
Similarly, engaging in hunting activities allows dogs to exercise their cognitive abilities by
using a range of problem-solving, critical thinking and decision-making skills. Additionally,
Animals 2024,14, 1082 6 of 23
hunting also helps dogs to enhance their physical abilities such as agility and endurance,
which also can be useful for detection work. The environmental soundness element assessed
the extent of startle and ability to recover from sudden loud noises; comfortableness with
and ability to overcome initial difficulty with novel surfaces (such as slick floors), obstacles
(such as open stairs), and surroundings; and reaction to strange/new persons, places, and
busy urban settings. All these activities can help to sharpen dogs’ detection skills, including
their sense of smell, decision-making, focus, and commitment. A composite score from 1
(low proficiency) to 5 (high proficiency) was assigned for each of the measures: retrieve,
hunt, and environmental soundness. The scores for all three measures were summed
to provide one integrated working dog assessment score named “Integrated Behavioral
Score”. This measure was then used for correlation with imaging metrics.
2.4. Data Acquisition
Functional MRI data were acquired using a T2*-weighted single-shot echo planar
imaging (EPI) sequence on a Siemens 3 Tesla Verio scanner (Erlangen, Germany) with
16 axial slices, slice thickness = 3 mm, repetition time (TR) = 1000 ms, echo time (TE) = 29 ms,
field of view (FOV) = 150
×
150 mm
2
, flip angle (FA) = 90 degree, in-plane resolution
2.3
×
2.3 mm, in-plane matrix = 64
×
64, and 200 temporal volumes in each run. Two
resting-state runs were acquired for each dog at each time point. A 15-channel human knee
coil was used as a head coil for the dog brain, and all dogs were trained to keep their heads
in the coil as still as possible (with eyes open) during the scanning. Anatomical images
were acquired using T1-weighted, 3-dimension magnetization-prepared rapid-gradient
echo (
3D-MPRAGE
) sequence for overlay and localization (TR = 1990 ms, TE = 2.85 ms,
FA = 9 deg, FOV = 152
×
152
×
104 mm
3
, in-plane matrix = 192
×
192, number of
partitions = 104, for voxel size = 0.8
×
0.8
×
1.0 mm
3
). All MRI scans included generalized
autocalibrating partially parallel acquisitions (GRAPPA) with an acceleration factor = 2.
As described in our previous publication [
22
], we were able to acquire good-quality
data from the frontal regions of the brain (which are susceptible to distortions) as well.
As described in our previous publications, we employed a single-camera-based exter-
nal optical head motion tracker [
8
,
21
,
23
] during data acquisition. This device provides
motion estimates with a high spatiotemporal resolution and was used in conjunction
with realignment parameters to assess the quality of the data. When we detected motion
amounting to more than one voxel from the external camera, the data were completely
discarded with the assumption that even motion censoring will not be able to retrieve
usable data. When we detected motion amounting to less than one voxel from the external
camera, framewise displacement was estimated using the acquired data post hoc and
motion censoring and interpolation of discarded volumes was employed when framewise
displacement exceeded 0.2 mm. As part of preprocessing, the realignment of scans was
used to estimate 6 motion parameters for each subject (3 translation parameters and 3 ro-
tation parameters). Afterwards, the framewise displacement (FD), an overall measure of
motion, was computed by using these parameters [
40
]. The threshold was selected based
on several literatures [15,21,22].
The longitudinal training and assessment process (Figure 3) was performed in 40 dogs.
However, due to the relatively long period of time that the dogs were expected to be in the
longitudinal training and assessment process, some dogs had to be released from imaging
at different time points. Taken together with data discarded due to motion, we had to reject
data from 10 dogs in TP1, 14 dogs in TP2, and 16 dogs in TP3. Thus, usable data from all
the time points included a total of 154 scans/runs from 30 dogs, with 60 scans from 30 dogs
(17 males/13 females) at TP1, 49 scans from 26 dogs (14 males/12 females) at TP2, and
45 scans from 24 dogs (14 males/10 females) at TP3. Only dogs which had usable data
at all three time points (i.e., 24 dogs) were analyzed in the current study. The data were
incorporated into the “Connectivity–Behavior Correlations across Timepoints” section in
the study. For the predictivity analysis, 30 dogs from TP1 were selected with 13 assessed as
suitable for detection work (7–9 months assessment).
Animals 2024,14, 1082 7 of 23
2.5. Image Preprocessing
The preprocessing of raw RS-fMRI data was performed using SPM12 and the DPARSF
toolbox [
41
,
42
]. The preprocessing steps included slice-timing correction, realignment to
the first functional image (i.e., image-based correction for head motion by aligning 3D
volumes acquired at different time-points), spatial normalization, spatial smoothing with
a Gaussian kernel of 4
×
4
×
4 mm
3
full width at half maximum (FWHM), detrending
and temporal band-pass filtering (in the range 0.01–0.1 Hz) for removing low- and high-
frequency sources of noise. Further, variance due to nuisance factors such as the six head
motion parameters (3 translations and 3 rotations), motion parameters obtained from the
external camera, as well as white matter and cerebrospinal fluid signals were regressed
out from each voxel time series inside the dog brain. Unlike human experiments, spatial
normalization in dogs is not straightforward due to the lack of a general template such as
the MNI template in humans. Existing templates for dogs are derived from less than ten
dogs and thus may not capture the head size variability across different breeds. Therefore,
we used a relatively more accurate two-step spatial normalization method, which was
employed in our previous dog fMRI studies [
8
,
21
–
23
,
37
]. In short, firstly, a good-quality
template was chosen among a pool of dogs (dogs in an anesthetized state are ideal to
minimize moments) in our previous studies [
21
,
22
]. Then, anatomical images of all the
subjects were co-registered in that template and the functional images of all subjects were
co-registered to their respective anatomical images. We could then transition from the
functional images to the selected ones using these steps.
2.6. Characterization of Resting-State Brain Networks
Resting-state networks are defined as the collection of brain regions that are temporally
correlated with each other. In the literature, two predominant approaches have been em-
ployed for characterizing resting-state networks: seed-based connectivity and connectomic
approaches. In the seed-based method, networks are defined based on the strength of
correlation between a seed region and every other region of the brain. Consequently, many
networks can be defined based on different seed regions and these networks have shown
to be correlated with human personality and behavior [
43
–
45
]. For example, the default
mode network (DMN) has been shown to correlate with traits and capabilities [
46
,
47
], and
the reward network (RN) has been shown to be correlated with reward and reinforcement
learning in humans [
43
]. However, seed-based connectivity obtained from predefined ROIs
is limited by the fact that they do not capture interactions between all brain regions. In
order to alleviate this limitation, connectomic approaches have been proposed as a more
holistic measure which captures functional associations between all possible pairs of brain
regions simultaneously [
48
]. Therefore, region-wise resting-state FC was obtained using
Pearson’s correlation between regions taken pair-wise across the entire brain. The regions
themselves were identified by those authors with expertise in canine brain anatomy using
a previously published dog atlas as a guide (http://vanat.cvm.umn.edu/mriBrainAtlas/
accessed on 29 February 2024). For example, if there were N regions inside the brain of a
given dog, then this would result in an N
×
Nregion-wise connectivity matrix for that dog.
These connectivity matrices were further used in analyses described below.
2.7. Connectivity–Behavior Correlations across Timepoints
We hypothesized that the correlation between resting-state FC in the dog brain and
behavior measures would significantly change during their detection training process (from
TP1 to TP2), and would maintain for the subsequent several months of detection work
(from TP2 to TP3). This was achieved by two steps (Figure 4): (1) the difference in the FC for
each subject and the path between TP1 and TP2 were correlated with the difference in the
corresponding behavioral measures between TP1 and TP2, respectively. This yielded paths
whose resting-state connectivity differences FC
TP2-TP1
significantly correlated (p< 0.05,
uncorrected) with corresponding differences in the integrated behavioral score IBS
TP2-TP1
.
(2) Among paths satisfying this condition, we retained those paths whose FC significantly
Animals 2024,14, 1082 8 of 23
increased (p< 0.05, uncorrected) from TP1 to TP2, but did not change significantly from
TP2 to TP3. This was based on the premise that detection training would lead to the
strengthening of certain functional connectivities from TP1 to TP2, which would then
maintain those FC levels from TP2 to TP3 during maintenance training. Further, the
identified paths would also mirror corresponding behavioral improvements from TP1
to TP2 and subsequent maintenance to TP3. Given our expectation that the significant
paths identified between TP1 and TP2 will remain consistent between TP2 and TP3, it is
worth examining if any differences in FC paths between TP1 and TP3 (FC
TP3-TP1
) are also
significantly correlated with behavioral changes between TP1 and TP3 (IBSTP3-TP1), hence
supporting our hypothesis. The resting-state brain network resulting from this analysis
represents the flexible networks which change with the detection training process.
Animals 2024, 14, x FOR PEER REVIEW 8 of 24
was obtained using Pearson’s correlation between regions taken pair-wise across the en-
tire brain. The regions themselves were identified by those authors with expertise in ca-
nine brain anatomy using a previously published dog atlas as a guide
(hp://vanat.cvm.umn.edu/mriBrainAtlas/ accessed on 29 February 2024). For example, if
there were N regions inside the brain of a given dog, then this would result in an N × N
region-wise connectivity matrix for that dog. These connectivity matrices were further
used in analyses described below.
2.7. Connectivity–Behavior Correlations across Timepoints
We hypothesized that the correlation between resting-state FC in the dog brain and
behavior measures would significantly change during their detection training process
(from TP1 to TP2), and would maintain for the subsequent several months of detection
work (from TP2 to TP3). This was achieved by two steps (Figure 4): (1) the difference in
the FC for each subject and the path between TP1 and TP2 were correlated with the dif-
ference in the corresponding behavioral measures between TP1 and TP2, respectively.
This yielded paths whose resting-state connectivity differences FC
TP2-TP1
significantly cor-
related (p < 0.05, uncorrected) with corresponding differences in the integrated behavioral
score IBS
TP2-TP1
. (2) Among paths satisfying this condition, we retained those paths whose
FC significantly increased (p < 0.05, uncorrected) from TP1 to TP2, but did not change
significantly from TP2 to TP3. This was based on the premise that detection training would
lead to the strengthening of certain functional connectivities from TP1 to TP2, which
would then maintain those FC levels from TP2 to TP3 during maintenance training. Fur-
ther, the identified paths would also mirror corresponding behavioral improvements
from TP1 to TP2 and subsequent maintenance to TP3. Given our expectation that the sig-
nificant paths identified between TP1 and TP2 will remain consistent between TP2 and
TP3, it is worth examining if any differences in FC paths between TP1 and TP3 (FC
TP3-TP1
)
are also significantly correlated with behavioral changes between TP1 and TP3 (IBS
TP3-TP1
),
hence supporting our hypothesis. The resting-state brain network resulting from this anal-
ysis represents the flexible networks which change with the detection training process.
Figure 4. A schematic of the two steps connectivity–behavior correlation analysis.
2.8. Brain Networks Predictive of Dogs’ Suitability for Detection Work
Networks that statistically correlate with behavioral changes across time do not nec-
essarily guarantee their ability to predict which dogs are suitable for detection work using
pre-training data at TP1. To further study the resting-state FC features that can predict the
Figure 4. A schematic of the two steps connectivity–behavior correlation analysis.
2.8. Brain Networks Predictive of Dogs’ Suitability for Detection Work
Networks that statistically correlate with behavioral changes across time do not neces-
sarily guarantee their ability to predict which dogs are suitable for detection work using
pre-training data at TP1. To further study the resting-state FC features that can predict the
success of training, dogs at TP1 were divided it two groups: the successful group consisting
of 13 dogs which were eventually deemed suitable for detection work at the end of the
longitudinal training and assessment process (7–9 months post-recruitment, Figure 3) and
a non-successful group consisting of 17 dogs which were eventually deemed unsuitable
for detection work. FC paths that were significantly stronger (p< 0.01) in the successful
group compared to the non-successful group at TP1 were determined and used as input
features to the classifier. This could enhance the quality of classification and ensure that
non-discriminatory features are not fed into the classifier.
In order to determine classification accuracy, i.e., the ability of FC features from TP1
identified above to predict whether a given dog would eventually fail or succeed, logistic
regression was used as the training kernel since it performed consistently well. Since this
is a popularly used classifier, we have skipped including a description of it and we refer
to previous work for a detailed description of its underlying principles [
49
]. Also, using
this classifier allowed us to compare our results with a previous study which is similar to
ours [
26
]. The receiver operating characteristic (ROC), which plots the true-positive rate
(TPR) against false-positive rate (FPR) and thus does not depend on a specific threshold,
was generated and the area under the curve (AUC) for the ROC was used as a metric for
performance evaluation. Four-fold cross-validation was employed so that the model could
be built using training data and then be tested using validation data. This minimizes the
chances of model overfitting.
Animals 2024,14, 1082 9 of 23
In order to better understand the functional roles of paths (and corresponding re-
gions) that correlated with behavioral changes due to detection training, we identified the
homologous regions (strictly functionally analogous regions) between dogs and humans
by comparing the similarity of connectivity fingerprints of these regions. We manually
selected 154 human subjects to match the number of scans, gender, and age in dog year
equivalents based on Lebeau’s model [
50
] with our dog group. For more details about data
acquisition and preprocessing of the human data, please refer to previous publications [
51
].
To establish connectivity fingerprints for each subject, we predefined 19 “targets” which
were ROIs covering most of the cortical regions as well as several subcortical regions (Figure
S1, Table S2) that are known to play a crucial role in guiding canine behavior as borne out
by previous literature [52]. Detailed analysis is available in the Supplementary Materials.
All identified connectivity paths were mapped onto the brain surface using Brain-
Net Viewer software (v.1.41) [53].
3. Results
The IBS behavioral score increased significantly from TP1 to TP2 (p< 0.05) and did not
change significantly from TP2 to TP3 (p> 0.05). Also, the difference in scores between TP1
and TP2 (i.e., TP2-TP1) was significantly (p< 0.05) larger than the corresponding difference
in scores between TP2 and TP3 (i.e., TP2-TP3). We identified ten paths in accordance with
our hypotheses (Figure 5and Table 1). These paths satisfied two different criteria. First,
resting-state connectivity difference FC
TP2-TP1
significantly correlated (p< 0.05, uncorrected)
with corresponding differences in the integrated behavioral score IBS
TP2-TP1
(Figure 6).
Second, the strength of these paths significantly increased (p< 0.05, uncorrected) from TP1
to TP2 and then maintained from TP2 to TP3 (p> 0.05) (Figure 7and Table 2). We did
not find any paths that weakened their connectivity strength from TP1 to TP2 and TP3.
Further, the difference in connectivity between TP1 and TP2 (i.e., TP2-TP1) was significantly
(p< 0.05) larger than the corresponding difference in connectivity between TP2 and TP3
(i.e., TP2-TP3). This demonstrates that detection training would lead to the strengthening
of certain functional connectivities from TP1 to TP2, which would then maintain those
FC levels from TP2 to TP3 during maintenance training. The “strengthening” of these
paths involved activity in corresponding ROIs coming in phase (as seen from the positive
correlations) in TP2 and TP3 as compared to being out of phase in TP1 (as evidenced
by negative correlations in TP1). Further, the identified paths mirrored corresponding
behavioral improvements from TP1 to TP2 and subsequent maintenance at TP3.
Animals 2024, 14, x FOR PEER REVIEW 10 of 24
in TP1 (as evidenced by negative correlations in TP1). Further, the identified paths mir-
rored corresponding behavioral improvements from TP1 to TP2 and subsequent mainte-
nance at TP3.
Tab l e 1 . Functional connectivity paths in the dog brain whose FC values satisfied our hypotheses
(i.e., increased significantly from TP1 to TP2 and did not change significantly from TP2 to TP3). Pyri
= pyriform, IPL = inferior parietal region, Hippo = hippocampus, Amy = amygdala, Hypo = hypo-
thalamus, MFG = middle frontal region, Caud = caudate, OB = olfactory bulb, DLPFC = dorsolateral
prefrontal cortex, IFG = inferior frontal region. R and L correspond to right and brain hemispheres,
respectively.
Path No. Path
1 R Pyri ↔ R IPL
2 L Pyri ↔ L IPL
3 L Claustrum/Insula ↔ R IPL
4 L Hippo ↔ L Amy
5 L Hippo ↔ Hypo
6 Brainstem ↔ L MFG
7 R Caud ↔ L Hippo
8 R Claustrum/Insula ↔ OB
9 R DLPFC ↔ L IPL
10 OB ↔ R IFG
Figure 5. Functional connectivity paths in the dog brain whose FC values satisfied our hypotheses.
Pyri = pyriform, IPL = inferior parietal region, Hippo = hippocampus, Amy = amygdala, Hypo =
hypothalamus, MFG = middle frontal region, Caud = caudate, OB = olfactory bulb, DLPFC = dorso-
lateral prefrontal cortex, IFG = inferior frontal region. R and L correspond to right and left brain
hemispheres, respectively.
Figure 5. Functional connectivity paths in the dog brain whose FC values satisfied our hypothe-
ses. Pyri = pyriform, IPL = inferior parietal region, Hippo = hippocampus, Amy = amygdala,
Animals 2024,14, 1082 10 of 23
Hypo = hypothalamus, MFG = middle frontal region, Caud = caudate, OB = olfactory bulb,
DLPFC = dorsolateral prefrontal cortex, IFG = inferior frontal region. R and L correspond to right
and left brain hemispheres, respectively.
Table 1. Functional connectivity paths in the dog brain whose FC values satisfied our hypotheses (i.e.,
increased significantly from TP1 to TP2 and did not change significantly from TP2 to TP3). Pyri = pyri-
form, IPL = inferior parietal region, Hippo = hippocampus, Amy = amygdala, Hypo = hypothalamus,
MFG = middle frontal region, Caud = caudate, OB = olfactory bulb, DLPFC = dorsolateral prefrontal
cortex, IFG = inferior frontal region. R and L correspond to right and brain hemispheres, respectively.
Path No. Path
1 R Pyri ↔R IPL
2 L Pyri ↔L IPL
3 L Claustrum/Insula ↔R IPL
4 L Hippo ↔L Amy
5 L Hippo ↔Hypo
6 Brainstem ↔L MFG
7 R Caud ↔L Hippo
8 R Claustrum/Insula ↔OB
9 R DLPFC ↔L IPL
10 OB ↔R IFG
Animals 2024, 14, x FOR PEER REVIEW 11 of 24
Figure 6. Resting-state connectivity differences between TP2 and TP1 (FCTP2-TP1) in the dog brain
significantly correlated (p < 0.05, uncorrected) with corresponding differences in the integrated be-
havioral score IBSTP2-TP1.
Figure 6. Cont.
Animals 2024,14, 1082 11 of 23
Animals 2024, 14, x FOR PEER REVIEW 11 of 24
Figure 6. Resting-state connectivity differences between TP2 and TP1 (FCTP2-TP1) in the dog brain
significantly correlated (p < 0.05, uncorrected) with corresponding differences in the integrated be-
havioral score IBSTP2-TP1.
Figure 6. Resting-state connectivity differences between TP2 and TP1 (FCTP2-TP1) in the dog brain
significantly correlated (p< 0.05, uncorrected) with corresponding differences in the integrated
behavioral score IBSTP2-TP1.
Animals 2024, 14, x FOR PEER REVIEW 12 of 24
Figure 7. FC paths in the dog brain at each time point whose strength significantly increased (p <
0.05, uncorrected) due to detection training (from TP1 to TP2) and then maintained from TP2 to TP3
(p > 0.05).
Table 2. Differences in p-values for the strength of FC across time points for paths shown in Figure 7.
TP2 > TP1 TP3 > TP1 TP2 ≠ TP3
R Pyri ↔ R IPL 1.49 × 10
−2
1.76 × 10
−2
9.12 × 10
−1
L Pyri ↔ L IPL 1.83 × 10
−2
5.52 × 10
−3
4.58 × 10
−1
L Claustrum/Insula ↔ R IPL 2.66 × 10
−2
3.06 × 10
−2
9.75 × 10
−1
L Hippo ↔ L Amy 2.23 × 10
−2
3.53 × 10
−2
8.54 × 10
−1
L Hippo ↔ Hypo 7.71 × 10
−3
4.39 × 10
−2
5.18 × 10
−1
Brainstem ↔ L MFG 1.69 × 10
−2
1.42 × 10
−2
9.65 × 10
−1
R Caud ↔ L Hippo 2.29 × 10
−2
4.61 × 10
−2
6.69 × 10
−1
R Claustrum/Insula ↔ OB 3.73 × 10
−2
1.92 × 10
−2
6.79 × 10
−1
R DLPFC ↔ L IPL 2.75 × 10
−6
2.17 × 10
−3
2.62 × 10
−1
OB ↔ R IFG 1.57 × 10
−2
1.51 × 10
−2
8.72 × 10
−1
3.1. Successful vs. Non-Successful Working Dogs
We identified seven paths in the dog brain (Table 3, Figure 8) whose FCs were signif-
icantly stronger in the successful group (n = 13) as compared to the non-successful group
(n = 17) at TP1, TP2, and TP3 (Figure 9), but did not change with training (Table 4). Among
them, six paths were located between Caudate and L MTG.
Table 3. Paths in the dog brain whose FC values were significantly stronger (p < 0.01) in the success-
ful group as compared to the non-successful group at each time point. Caud = Caudate, MTG =
middle temporal region, STG = superior temporal region. R and L correspond to right and left brain
hemispheres, respectively.
p-Value of FC
successful
> FC
non-successful
Path TP1 TP2 TP3
1–6. Caudate ↔ L MTG
3.38 × 10
−4
2.9 × 10
−3
1.9 × 10
−3
7.97 × 10
−4
2.1 × 10
−3
1.6 × 10
−3
8.55 × 10
−5
1.6 × 10
−3
4.1 × 10
−3
9.91 × 10
−5
9.4 × 10
−4
1.7 × 10
−3
Figure 7. FC paths in the dog brain at each time point whose strength significantly increased
(p< 0.05, uncorrected) due to detection training (from TP1 to TP2) and then maintained from TP2 to
TP3 (p> 0.05).
Animals 2024,14, 1082 12 of 23
Table 2. Differences in p-values for the strength of FC across time points for paths shown in Figure 7.
TP2 > TP1 TP3 > TP1 TP2 =TP3
R Pyri ↔R IPL 1.49 ×10−21.76 ×10−29.12 ×10−1
L Pyri ↔L IPL 1.83 ×10−25.52 ×10−34.58 ×10−1
L Claustrum/Insula ↔R IPL 2.66 ×10−23.06 ×10−29.75 ×10−1
L Hippo ↔L Amy 2.23 ×10−23.53 ×10−28.54 ×10−1
L Hippo ↔Hypo 7.71 ×10−34.39 ×10−25.18 ×10−1
Brainstem ↔L MFG 1.69 ×10−21.42 ×10−29.65 ×10−1
R Caud ↔L Hippo 2.29 ×10−24.61 ×10−26.69 ×10−1
R Claustrum/Insula ↔OB 3.73 ×10−21.92 ×10−26.79 ×10−1
R DLPFC ↔L IPL 2.75 ×10−62.17 ×10−32.62 ×10−1
OB ↔R IFG 1.57 ×10−21.51 ×10−28.72 ×10−1
3.1. Successful vs. Non-Successful Working Dogs
We identified seven paths in the dog brain (Table 3, Figure 8) whose FCs were signifi-
cantly stronger in the successful group (n = 13) as compared to the non-successful group
(n = 17) at TP1, TP2, and TP3 (Figure 9), but did not change with training (Table 4). Among
them, six paths were located between Caudate and L MTG.
Table 3. Paths in the dog brain whose FC values were significantly stronger (p< 0.01) in the successful
group as compared to the non-successful group at each time point. Caud = Caudate, MTG = middle
temporal region, STG = superior temporal region. R and L correspond to right and left brain
hemispheres, respectively.
p-Value of FCsuccessful > FCnon-successful
Path TP1 TP2 TP3
1–6. Caudate ↔L MTG
3.38 ×10−42.9 ×10−31.9 ×10−3
7.97 ×10−42.1 ×10−31.6 ×10−3
8.55 ×10−51.6 ×10−34.1 ×10−3
9.91 ×10−59.4 ×10−41.7 ×10−3
2.8 ×10−31.6 ×10−31.9 ×10−3
3.7 ×10−34.2 ×10−31.8 ×10−3
7. L Caud ↔R STG 3.8 ×10−61.9 ×10−34.8 ×10−3
Animals 2024, 14, x FOR PEER REVIEW 13 of 24
2.8 × 10
−3
1.6 × 10
−3
1.9 × 10
−3
3.7 × 10
−3
4.2 × 10
−3
1.8 × 10
−3
7. L Caud ↔ R STG 3.8 × 10
−6
1.9 × 10
−3
4.8 × 10
−3
(a) (b)
Figure 8. FC paths in the dog brain whose strength was significantly stronger in the successful group
(a) as compared to the non-successful group (b) specifically at TP1, TP2, and TP3.
Table 4 . Differences in p-values for the strength of FC across time points for paths shown in Figure
8. None of them were significant.
p-Value of Successful Group p-Value of Non-Successful Group
Path No. TP1 ≠ TP2 TP1 ≠ TP3 TP2 ≠ TP3 TP1 ≠ TP2 TP1 ≠ TP3 TP2 ≠ TP3
1 5.6 × 10
−1
2.3 × 10
−1
9.5 × 10
−2
7.8 × 10
−1
5.2 × 10
−1
4.4 × 10
−1
2 3.8 × 10
−1
3.5 × 10
−1
1.2 × 10
−1
3.4 × 10
−1
8.7 × 10
−1
3.3 × 10
−1
3 6.4 × 10
−1
5.7 × 10
−1
2.9 × 10
−1
5.1 × 10
−1
2.4 × 10
−1
5.1 × 10
−1
4 4.3 × 10
−1
8.7 × 10
−1
4.1 × 10
−1
3.5 × 10
−1
8.9 × 10
−1
5.4 × 10
−1
5 1.7 × 10
−1
6.9 × 10
−1
1.7 × 10
−1
8.8 × 10
−2
5.1 × 10
−1
3.5 × 10
−1
6 6.2 × 10
−2
6.2 × 10
−1
2.6 × 10
−1
6.3 × 10
−2
9.3 × 10
−2
9.6 × 10
−1
7 3.9 × 10
−1
9.1 × 10
−1
4.1 × 10
−1
1.1 × 10
−1
6.6 × 10
−2
5.4 × 10
−1
Figure 8. FC paths in the dog brain whose strength was significantly stronger in the successful group
(a) as compared to the non-successful group (b) specifically at TP1, TP2, and TP3.
Animals 2024,14, 1082 13 of 23
Animals 2024, 14, x FOR PEER REVIEW 13 of 24
2.8 × 10
−3
1.6 × 10
−3
1.9 × 10
−3
3.7 × 10
−3
4.2 × 10
−3
1.8 × 10
−3
7. L Caud ↔ R STG 3.8 × 10
−6
1.9 × 10
−3
4.8 × 10
−3
(a) (b)
Figure 8. FC paths in the dog brain whose strength was significantly stronger in the successful group
(a) as compared to the non-successful group (b) specifically at TP1, TP2, and TP3.
Table 4. Differences in p-values for the strength of FC across time points for paths shown in Figure
8. None of them were significant.
p-Value of Successful Group p-Value of Non-Successful Group
Path No. TP1 ≠ TP2 TP1 ≠ TP3 TP2 ≠ TP3 TP1 ≠ TP2 TP1 ≠ TP3 TP2 ≠ TP3
1 5.6 × 10
−1
2.3 × 10
−1
9.5 × 10
−2
7.8 × 10
−1
5.2 × 10
−1
4.4 × 10
−1
2 3.8 × 10
−1
3.5 × 10
−1
1.2 × 10
−1
3.4 × 10
−1
8.7 × 10
−1
3.3 × 10
−1
3 6.4 × 10
−1
5.7 × 10
−1
2.9 × 10
−1
5.1 × 10
−1
2.4 × 10
−1
5.1 × 10
−1
4 4.3 × 10
−1
8.7 × 10
−1
4.1 × 10
−1
3.5 × 10
−1
8.9 × 10
−1
5.4 × 10
−1
5 1.7 × 10
−1
6.9 × 10
−1
1.7 × 10
−1
8.8 × 10
−2
5.1 × 10
−1
3.5 × 10
−1
6 6.2 × 10
−2
6.2 × 10
−1
2.6 × 10
−1
6.3 × 10
−2
9.3 × 10
−2
9.6 × 10
−1
7 3.9 × 10
−1
9.1 × 10
−1
4.1 × 10
−1
1.1 × 10
−1
6.6 × 10
−2
5.4 × 10
−1
Figure 9. A pictorial spatial representation of the paths in the dog brain shown in Table 3. The thick
line corresponds to multiple paths between L Caud and L MTG while the thin line corresponds to
the single connection between L Caud and R STG. Caud = Caudate, MTG = middle temporal region,
STG = superior temporal region. R and L correspond to right and left brain hemispheres, respectively.
Table 4. Differences in p-values for the strength of FC across time points for paths shown in Figure 8.
None of them were significant.
p-Value of Successful Group p-Value of Non-Successful Group
Path No. TP1 =TP2 TP1 =TP3 TP2 =TP3 TP1 =TP2 TP1 =TP3 TP2 =TP3
15.6 ×10−12.3 ×10−19.5 ×10−27.8 ×10−15.2 ×10−14.4 ×10−1
23.8 ×10−13.5 ×10−11.2 ×10−13.4 ×10−18.7 ×10−13.3 ×10−1
36.4 ×10−15.7 ×10−12.9 ×10−15.1 ×10−12.4 ×10−15.1 ×10−1
44.3 ×10−18.7 ×10−14.1 ×10−13.5 ×10−18.9 ×10−15.4 ×10−1
51.7 ×10−16.9 ×10−11.7 ×10−18.8 ×10−25.1 ×10−13.5 ×10−1
66.2 ×10−26.2 ×10−12.6 ×10−16.3 ×10−29.3 ×10−29.6 ×10−1
73.9 ×10−19.1 ×10−14.1 ×10−11.1 ×10−16.6 ×10−25.4 ×10−1
3.2. Classification Analyses
Considering that the sample size was relatively small, we performed classification
analyses using 1000 iterations of stratified random shuffling with a test size of 25% of
the data (four-fold cross-validation). Classifiers with behavior (integrated behavioral
score) performed above chance with AUC equal to 0.62 (Figure 10, blue). Classifiers using
functional connections in a flexible neural network in the dog brain which changed with
detection training and correlated with corresponding behavioral changes (Table 1) also
performed above chance with AUC = 0.68 (Figure 10, red). However, the best classification
performance was achieved using functional connections within a network of regions in
the dog brain which was significantly stronger in the successful group as compared to the
non-successful group (Table 3) but did not change with training (Table 4). This network
gave an AUC equal to 0.90 (Figure 10, green).
Animals 2024,14, 1082 14 of 23
Animals 2024, 14, x FOR PEER REVIEW 14 of 24
Figure 9. A pictorial spatial representation of the paths in the dog brain shown in Table 3. The thick
line corresponds to multiple paths between L Caud and L MTG while the thin line corresponds to
the single connection between L Caud and R STG..
Caud = Caudate, MTG = middle temporal region,
STG = superior temporal region. R and L correspond to right and left brain hemispheres, respec-
tively.
3.2. Classification Analyses
Considering that the sample size was relatively small, we performed classification
analyses using 1000 iterations of stratified random shuffling with a test size of 25% of the
data (four-fold cross-validation). Classifiers with behavior (integrated behavioral score)
performed above chance with AUC equal to 0.62 (Figure 10, blue). Classifiers using func-
tional connections in a flexible neural network in the dog brain which changed with de-
tection training and correlated with corresponding behavioral changes (Table 1) also per-
formed above chance with AUC = 0.68 (Figure 10, red). However, the best classification
performance was achieved using functional connections within a network of regions in
the dog brain which was significantly stronger in the successful group as compared to the
non-successful group (Table 3) but did not change with training (Table 4). This network
gave an AUC equal to 0.90 (Figure 10, green).
Figure 10. ROC plot for the three classifier models used to predict, using data at TP1, whether a dog
would eventually be suitable for detection work. The straight line in the AUC curve is a reference
line indicating random performance. A classifierʹs ROC curve should lie above this line to be con-
sidered useful for classification tasks.
4. Discussion
Dogs have a unique ability to interact with humans and this ability has led to dogs
working with and assisting humans in various tasks. Therefore, investigations on their
general behavioral capabilities and their related neural bases can inform us about critical
parameters for selecting dogs for training. However, research into the neural basis of the
behavior of dogs, specifically longitudinal investigations, is sparse. This study is the first
to our knowledge to explore the neural processes across different training time points at
in vivo level using resting-state fMRI.
Previous studies have shown the reconfiguration of brain networks during task
learning in humans [54–57]. Such regional network dynamics are consistent with a core–
periphery model wherein certain brain regions show relatively stable (with respect to
Figure 10. ROC plot for the three classifier models used to predict, using data at TP1, whether a dog
would eventually be suitable for detection work. The straight line in the AUC curve is a reference line
indicating random performance. A classifier ’s ROC curve should lie above this line to be considered
useful for classification tasks.
4. Discussion
Dogs have a unique ability to interact with humans and this ability has led to dogs
working with and assisting humans in various tasks. Therefore, investigations on their
general behavioral capabilities and their related neural bases can inform us about critical
parameters for selecting dogs for training. However, research into the neural basis of the
behavior of dogs, specifically longitudinal investigations, is sparse. This study is the first
to our knowledge to explore the neural processes across different training time points at
in vivo level using resting-state fMRI.
Previous studies have shown the reconfiguration of brain networks during task
learning in humans [
54
–
57
]. Such regional network dynamics are consistent with a core–
periphery model wherein certain brain regions show relatively stable (with respect to time)
patterns of interaction that are necessary for the task performance, while others display
relatively flexible patterns that support learning and changes in task performance [
55
]. In
our work, we extended this concept to the dog training process and hypothesized that
such a core–periphery network may not only mirror changes in behavior with training, but
also predict the success of dog training and could potentially be used for selecting dogs
to be trained.
We identified two systems, one system (the periphery system) consisting of a brain
network which strengthened its connectivity with improvements in canine behavior scores
after detection training. We believe that this is a flexible system which supports learning
related plasticity during the training process. However, the periphery system is not predic-
tive, and its connectivity at the first baseline time point prior to training could not predict
whether a given dog could eventually be trained to become a good detector dog. The other
system we identified did not show significant changes in its connectivity strength during
the training process. However, the strength of connectivity of this core network predicted,
with 90% accuracy, whether a given dog would eventually graduate as a detector dog from
the training regimen. We speculate that the core stable network may be an endophenotype
that is inherited and mainly controlled by genes while the flexible periphery network may
be amended by environmental influences [
58
,
59
]. Below, we discuss this core–periphery
model in greater detail.
Animals 2024,14, 1082 15 of 23
4.1. Flexible Periphery Network Underlying Detection Training
We found significant correlations between behavioral changes and connectivity changes
between the following ROIs related to olfactory processing: L Amy and L Hippo, L Hippo
and Hypo, and L Hippo and R Caud (Figure 11). Previous human studies [
60
–
62
] as well as
dog studies [
21
,
22
] have implicated this set of regions during olfactory processing. Further,
previous studies have shown that functional connectivity of olfaction-related networks may
be reinforced by training, and training-induced behavioral improvement in olfactory per-
formance has been observed in healthy humans [
63
]. Therefore, this sub-network may be
related to an improvement in the olfactory processing capabilities of dogs during detection
training. On the other hand, the insula and inferior frontal gyrus were commonly activated
by visual and odor food cue stimulation in humans in a meta-analysis study on food cue
neuroimaging [
64
]. Considering that during the training process, dogs were reinforced for
successful performance with treats, the increase in FC between the OB and R insula, as well
as between the OB and IFG (Figure 12) might reflect positive reinforcement and the neural
plasticity of conditioning for food-related stimuli.
Animals 2024, 14, x FOR PEER REVIEW 15 of 24
time) paerns of interaction that are necessary for the task performance, while others dis-
play relatively flexible paerns that support learning and changes in task performance
[55]. In our work, we extended this concept to the dog training process and hypothesized
that such a core–periphery network may not only mirror changes in behavior with train-
ing, but also predict the success of dog training and could potentially be used for selecting
dogs to be trained.
We identified two systems, one system (the periphery system) consisting of a brain
network which strengthened its connectivity with improvements in canine behavior
scores after detection training. We believe that this is a flexible system which supports
learning related plasticity during the training process. However, the periphery system is
not predictive, and its connectivity at the first baseline time point prior to training could
not predict whether a given dog could eventually be trained to become a good detector
dog. The other system we identified did not show significant changes in its connectivity
strength during the training process. However, the strength of connectivity of this core
network predicted, with 90% accuracy, whether a given dog would eventually graduate
as a detector dog from the training regimen. We speculate that the core stable network
may be an endophenotype that is inherited and mainly controlled by genes while the flex-
ible periphery network may be amended by environmental influences [58,59]. Below, we
discuss this core–periphery model in greater detail.
4.1. Flexible Periphery Network Underlying Detection Training
We found significant correlations between behavioral changes and connectivity
changes between the following ROIs related to olfactory processing: L Amy and L Hippo,
L Hippo and Hypo, and L Hippo and R Caud (Figure 11). Previous human studies [60–
62] as well as dog studies [21,22] have implicated this set of regions during olfactory pro-
cessing. Further, previous studies have shown that functional connectivity of olfaction-
related networks may be reinforced by training, and training-induced behavioral im-
provement in olfactory performance has been observed in healthy humans [63]. Therefore,
this sub-network may be related to an improvement in the olfactory processing capabili-
ties of dogs during detection training. On the other hand, the insula and inferior frontal
gyrus were commonly activated by visual and odor food cue stimulation in humans in a
meta-analysis study on food cue neuroimaging [64]. Considering that during the training
process, dogs were reinforced for successful performance with treats, the increase in FC
between the OB and R insula, as well as between the OB and IFG (Figure 12) might reflect
positive reinforcement and the neural plasticity of conditioning for food-related stimuli.
(a) (b)
Figure 11. Olfaction-related network in the dog brain (a) and homologous regions in the human
brain (b) which showed significant correlations between behavioral changes and connectivity
changes. Hippo = hippocampus, Amy = amygdala, Hypo = hypothalamus, Caud = caudate. R and L
correspond to right and left brain hemispheres, respectively.
Figure 11. Olfaction-related network in the dog brain (a) and homologous regions in the human brain
(b) which showed significant correlations between behavioral changes and connectivity changes.
Hippo = hippocampus, Amy = amygdala, Hypo = hypothalamus, Caud = caudate. R and L corre-
spond to right and left brain hemispheres, respectively.
Animals 2024, 14, x FOR PEER REVIEW 16 of 24
(a) (b)
Figure 12. Positive reinforcement network in the dog brain (a) and homologous regions in the hu-
man brain (b) which showed significant correlations between behavioral changes and connectivity
changes. OB = olfactory bulb, IFG = inferior frontal region. R and L correspond to right and left brain
hemispheres, respectively.
Paths from the R SFG to L IPL in the periphery network may be part of the fronto-
parietal network (Figure 13). This is consistent with human studies that have shown that
behavioral variables co-vary with connectivity in frontal-parietal networks (FPN) [12,14].
Previous human studies have suggested that brain regions in the frontal and parietal cor-
tices play an important role in cognitive control processes and connectivity within the
FPN directly relates to aention [65–67]. An ICA study has shown a greatly overlapped
frontal-parietal network in macaque and human brains, suggesting an evolutionary pre-
served frontal-parietal system [68]. Moreover, studies related to individual human intel-
ligence found that greater connectivity, especially during task performance, within the
frontal-parietal network was associated with higher intelligence scores [69,70].
(a) (b)
Figure 13. The fronto-parietal network of the dog brain (a) and homologous regions in the human
brain (b) which showed significant correlations between behavioral changes and connectivity
changes. DLPFC = dorsolateral prefrontal cortex, IPL = inferior parietalregion, SFG = superior frontal
region. R and L correspond to right and left brain hemispheres, respectively.
PHG and insula are also known to be involved in familiarity-related judgments [71–
73]. Also, the IPL is known to be associated with familiarity and recollection-related judg-
ments [74]. The increase in the FC within the FPN, between the L insula and IPL, as well
as between the PHG and IPL (Figure 14) with corresponding improvements in behavior
might suggest improved understanding and reaction towards the trainer’s gestures and
commands (via both familiarity and recollection) through learning.
Figure 12. Positive reinforcement network in the dog brain (a) and homologous regions in the
human brain (b) which showed significant correlations between behavioral changes and connectivity
changes. OB = olfactory bulb, IFG = inferior frontal region. R and L correspond to right and left brain
hemispheres, respectively.
Paths from the R SFG to L IPL in the periphery network may be part of the fronto-
parietal network (Figure 13). This is consistent with human studies that have shown that
behavioral variables co-vary with connectivity in frontal-parietal networks (FPN) [
12
,
14
].
Animals 2024,14, 1082 16 of 23
Previous human studies have suggested that brain regions in the frontal and parietal
cortices play an important role in cognitive control processes and connectivity within
the FPN directly relates to attention [
65
–
67
]. An ICA study has shown a greatly over-
lapped frontal-parietal network in macaque and human brains, suggesting an evolutionary
preserved frontal-parietal system [
68
]. Moreover, studies related to individual human
intelligence found that greater connectivity, especially during task performance, within the
frontal-parietal network was associated with higher intelligence scores [69,70].
Animals 2024, 14, x FOR PEER REVIEW 16 of 24
(a) (b)
Figure 12. Positive reinforcement network in the dog brain (a) and homologous regions in the hu-
man brain (b) which showed significant correlations between behavioral changes and connectivity
changes. OB = olfactory bulb, IFG = inferior frontal region. R and L correspond to right and left brain
hemispheres, respectively.
Paths from the R SFG to L IPL in the periphery network may be part of the fronto-
parietal network (Figure 13). This is consistent with human studies that have shown that
behavioral variables co-vary with connectivity in frontal-parietal networks (FPN) [12,14].
Previous human studies have suggested that brain regions in the frontal and parietal cor-
tices play an important role in cognitive control processes and connectivity within the
FPN directly relates to aention [65–67]. An ICA study has shown a greatly overlapped
frontal-parietal network in macaque and human brains, suggesting an evolutionary pre-
served frontal-parietal system [68]. Moreover, studies related to individual human intel-
ligence found that greater connectivity, especially during task performance, within the
frontal-parietal network was associated with higher intelligence scores [69,70].
(a) (b)
Figure 13. The fronto-parietal network of the dog brain (a) and homologous regions in the human
brain (b) which showed significant correlations between behavioral changes and connectivity
changes. DLPFC = dorsolateral prefrontal cortex, IPL = inferior parietalregion, SFG = superior frontal
region. R and L correspond to right and left brain hemispheres, respectively.
PHG and insula are also known to be involved in familiarity-related judgments [71–
73]. Also, the IPL is known to be associated with familiarity and recollection-related judg-
ments [74]. The increase in the FC within the FPN, between the L insula and IPL, as well
as between the PHG and IPL (Figure 14) with corresponding improvements in behavior
might suggest improved understanding and reaction towards the trainer’s gestures and
commands (via both familiarity and recollection) through learning.
Figure 13. The fronto-parietal network of the dog brain (a) and homologous regions in the human
brain (b) which showed significant correlations between behavioral changes and connectivity changes.
DLPFC = dorsolateral prefrontal cortex, IPL = inferior parietalregion, SFG = superior frontal region.
R and L correspond to right and left brain hemispheres, respectively.
PHG and insula are also known to be involved in familiarity-related judgments
[71–73]
.
Also, the IPL is known to be associated with familiarity and recollection-related judg-
ments [
74
]. The increase in the FC within the FPN, between the L insula and IPL, as well
as between the PHG and IPL (Figure 14) with corresponding improvements in behavior
might suggest improved understanding and reaction towards the trainer’s gestures and
commands (via both familiarity and recollection) through learning.
Animals 2024, 14, x FOR PEER REVIEW 17 of 24
(a) (b)
Figure 14. The network related to familiarity and recollection in the dog brain (a) and homologous
regions in the human brain (b) which showed significant correlations between behavioral changes
and connectivity changes. PHG = parahippocampal gyrus, IPL = inferior parietal region, Pyri = pyr-
iform. R and L correspond to right and left brain hemispheres, respectively.
The Locus coeruleus (LC) in the brain stem is the largest repository of Norepineph-
rine (NE) in the human brain [75]. Noradrenergic neurons within LC are widely distrib-
uted and are one of the main ascending pathways from the LC projects to the prefrontal
cortex [76,77]. It has been shown that NE projections from the LC to the cortex support
learning and memory retrieval [78,79]. An animal study has further found that boosting
NE transmission can lead to increased functional connectivity [80]. Thus, the significantly
increased FC (from baseline time point to other TPs post detection training) between the
LC in the brainstem and L MFG (Figure 15) might correspond to the mechanisms of learn-
ing of odors and the retrieval of such memory.
(a) (b)
Figure 15. Learning-related network in the dog brain (a) and homologous regions in the human
brain (b) which showed significant correlations between behavioral changes and connectivity
changes. MFG = middle frontal region. R and L correspond to right and left brain hemispheres,
respectively.
4.2. Stable Core Network for Predicting Training Outcomes
It can be noted that a majority of the paths in the stable core network involve the
caudate. Previous studies have considered the caudate as a part of the reward system and
may reinforce learning in humans as well as dogs [22,30,43]. Studies in humans and mon-
keys have indicated that regions in the temporal cortex respond preferentially to face
recognition [81]. Further, a recent canine study has shown that using fMRI activations
from the caudate, amygdala, and a specialized region in the temporal cortex for face pro-
cessing (known as the dog face area or DFA), the authors were able to predict (through
cross-validation) a given dog’s suitability for assistance work with an accuracy of 80%
Figure 14. The network related to familiarity and recollection in the dog brain (a) and homologous
regions in the human brain (b) which showed significant correlations between behavioral changes and
connectivity changes. PHG = parahippocampal gyrus, IPL = inferior parietal region, Pyri = pyriform.
R and L correspond to right and left brain hemispheres, respectively.
The Locus coeruleus (LC) in the brain stem is the largest repository of Norepinephrine
(NE) in the human brain [
75
]. Noradrenergic neurons within LC are widely distributed and
are one of the main ascending pathways from the LC projects to the prefrontal cortex [
76
,
77
].
It has been shown that NE projections from the LC to the cortex support learning and
memory retrieval [
78
,
79
]. An animal study has further found that boosting NE transmission
Animals 2024,14, 1082 17 of 23
can lead to increased functional connectivity [
80
]. Thus, the significantly increased FC (from
baseline time point to other TPs post detection training) between the LC in the brainstem
and L MFG (Figure 15) might correspond to the mechanisms of learning of odors and the
retrieval of such memory.
Animals 2024, 14, x FOR PEER REVIEW 17 of 24
(a) (b)
Figure 14. The network related to familiarity and recollection in the dog brain (a) and homologous
regions in the human brain (b) which showed significant correlations between behavioral changes
and connectivity changes. PHG = parahippocampal gyrus, IPL = inferior parietal region, Pyri = pyr-
iform. R and L correspond to right and left brain hemispheres, respectively.
The Locus coeruleus (LC) in the brain stem is the largest repository of Norepineph-
rine (NE) in the human brain [75]. Noradrenergic neurons within LC are widely distrib-
uted and are one of the main ascending pathways from the LC projects to the prefrontal
cortex [76,77]. It has been shown that NE projections from the LC to the cortex support
learning and memory retrieval [78,79]. An animal study has further found that boosting
NE transmission can lead to increased functional connectivity [80]. Thus, the significantly
increased FC (from baseline time point to other TPs post detection training) between the
LC in the brainstem and L MFG (Figure 15) might correspond to the mechanisms of learn-
ing of odors and the retrieval of such memory.
(a) (b)
Figure 15. Learning-related network in the dog brain (a) and homologous regions in the human
brain (b) which showed significant correlations between behavioral changes and connectivity
changes. MFG = middle frontal region. R and L correspond to right and left brain hemispheres,
respectively.
4.2. Stable Core Network for Predicting Training Outcomes
It can be noted that a majority of the paths in the stable core network involve the
caudate. Previous studies have considered the caudate as a part of the reward system and
may reinforce learning in humans as well as dogs [22,30,43]. Studies in humans and mon-
keys have indicated that regions in the temporal cortex respond preferentially to face
recognition [81]. Further, a recent canine study has shown that using fMRI activations
from the caudate, amygdala, and a specialized region in the temporal cortex for face pro-
cessing (known as the dog face area or DFA), the authors were able to predict (through
cross-validation) a given dog’s suitability for assistance work with an accuracy of 80%
Figure 15. Learning-related network in the dog brain (a) and homologous regions in the human brain
(b) which showed significant correlations between behavioral changes and connectivity changes.
MFG = middle frontal region. R and L correspond to right and left brain hemispheres, respectively.
4.2. Stable Core Network for Predicting Training Outcomes
It can be noted that a majority of the paths in the stable core network involve the
caudate. Previous studies have considered the caudate as a part of the reward system
and may reinforce learning in humans as well as dogs [
22
,
30
,
43
]. Studies in humans and
monkeys have indicated that regions in the temporal cortex respond preferentially to face
recognition [
81
]. Further, a recent canine study has shown that using fMRI activations from
the caudate, amygdala, and a specialized region in the temporal cortex for face processing
(known as the dog face area or DFA), the authors were able to predict (through cross-
validation) a given dog’s suitability for assistance work with an accuracy of 80% [
28
]. Thus,
it is not surprising that we found multiple paths between the L Caud and L MTG in the
stable core network that were stronger in the successful group at TP1. Further, the strength
of connectivity of these paths at TP1 was able to predict the success of training with an
accuracy of 90% using a classifier with cross-validation comparable to the one used by
Berns et al., 2016 [28].
The main role of the STG is to process sound stimuli [
82
]. During the training process,
dogs were reinforced for successful performance with treats and verbal rewards (the trainer
praised the dog—“Good dog” or “Yes”). Stronger connectivity between the L Caud and R
STG in the successful group might suggest that the dogs that were able to associate human
verbal praises with rewards might have a better chance to be trained as detection dogs.
One study [
83
] shows the significant role of human voices in shaping canine emotional
experiences, revealing parallels in the dogs’ reactions to basic emotional expression in
human vocalizations like those observed in humans. This indicates a profound interplay
between human speech patterns and canine emotional comprehension.
It should be noted that the caudate was found to be a part of both the core and
periphery networks (L Caud in the stable core network and R Caud in flexible periphery
network). The factor that prohibits dogs from successful training is believed to be their
fearfulness/anxiety towards novel/complex environments [
84
]. Previous studies have
suggested that dopamine D4 receptors influence canine fearfulness, anxiety and impulsivity
related traits [
85
,
86
]. Dopamine D4 receptors have been found to be concentrated in the
caudate [
87
]. Since the expression of D4 receptors in the caudate is controlled by specific
genes, they might influence fearfulness-, anxiety-, and impulsivity-related traits in dogs,
thereby influencing baseline neural connectivity between the caudate and other brain
regions. This might be a potential mechanism by which the caudate may be involved in
Animals 2024,14, 1082 18 of 23
the stable core network that predicts whether dogs can be trained to become successful
working dogs. On the other hand, the role played by the caudate in reinforcement-based
learning in humans is known [
88
,
89
]. Specifically in dogs, Berns et al. found that the
caudate was significantly more active when the dogs were exposed to different types of
reward stimuli [
26
–
28
,
30
,
90
]. We found an increase in FC between the R caudate and
L hippo in the periphery network with detection training. Therefore, our findings also
support the role of the caudate in learning, reward, and training-related neural plasticity.
Taken together, we identified a core-periphery organization in the dog brain and
these systems responded differently to the detection training process, which appear to
represent ontogeny and phylogeny. The flexible periphery network is ontological, as it
changed with corresponding behavioral changes due to training/learning that were mainly
located in regions implicated in odor processing. Biologically, network flexibility might be
driven by physiological processes that facilitate the participation of corresponding regions
in multiple functional communities while learning new tasks. On the other hand, the
caudate-based core network is driven more by phylogeny because of its stability across
detection training. Such a stable network may contain information about intrinsic learning
ability for individuals which can successfully predict the outcome of training. Our result
suggests that, upon replication and refinement, fMRI-based resting-state brain connectivity
may assist in choosing dogs that are more easily trainable for performing detection tasks
before they enter the training regimen. They also suggest that good working dogs learn
efficiently, and they can be trained well in what we want them to learn.
4.3. Insights from Human Homology
A basic challenge in animal neuroimaging is to compare and explain brain functions
across species, especially at the voxel level. Interpretation often relies on the assumption
that putative homologous areas are functionally similar [
91
]. However, this assumption is
not always valid [
92
] since putative homology is generally not established on any rigorous
statistical premise. Previous studies have suggested that specific functions in an area of
one species may be shifted to other regions in other species [
93
]. In this study, we sought
to identify the homologous areas by comparing the connectivity fingerprints of regions
between humans and dogs [
94
,
95
]. This was carried out because there is relatively abundant
literature about the functional roles of brain regions in the human brain than in the dog
brain. The approach of matching connectivity fingerprints is a viable technique and has
been used in a number of studies previously [96,97].
We identified several regions that do not functionally correspond to their putative
homologous regions. For example, the R pyriform, L pyriform, and R dorsolateral prefrontal
cortex in dogs shared similar connectivity profiles as the R parahippocampal gyrus, L
parahippocampal gyrus, and R superior frontal gyrus in humans. Moreover, the LC in the
human brainstem was identified to be homologous to a specific region of the dog brainstem.
With the help of previous human literature and based on the homology established using
the connectivity fingerprint matching procedure, we speculated that the connectivity
between the brain stem and the frontal cortex in dogs corresponded to that between the
LC and L MFG in humans and this might underlie mechanisms of learning of odors and
retrieval of such memory. Despite the lack of abundant literature into the neural basis of
behaviors of dogs and their trainability, projecting dog brain regions identified here onto the
human brain and evaluating similar literature on humans may help us better understand
the evolutionary role of brain–behavior relationships. This comparative evaluation is
particularly relevant for dogs since they are a rare species which has socially co-evolved
with humans for thousands of years.
5. Limitations and Future Research
A few limitations of this work are noteworthy. Although the number of dogs is
comparatively large, the sample size is still considered small in an absolute sense which
did not allow us to use stringent p-value thresholds that would reduce the false-positive
Animals 2024,14, 1082 19 of 23
rate. Given that the field of canine imaging is in its infancy, we believe that standards
applicable to human imaging should not be applied to canine imaging at this point in
time. Once procedures used for training dogs to keep their head still inside the scanner
are mastered and a workforce with those skills is developed, it will become feasible to
increase the sample size in the future and that will allow more conservative statistical
thresholds to be used. Until that happens, accepting a higher false-positive rate in canine
imaging results will nurture discovery science in this nascent field. For example, the
identified dog core-periphery model could provide a window into the evolutionary role of
brain–behavior relationships and provides great potential for answering questions about
phylogeny and ontogeny in future studies. It should be noted that no significant gender
differences were found in the identified two systems in our study; however, future studies
should aim to replicate our findings in different breeds of dog. Further, we have compared
our classification results with those from a previous study [
28
]. However, here we studied
detection dogs while they focused on service dogs. Therefore, the comparison of our results
with those reported by Berns et al. is qualitative at best. As the study is exclusively on
labradors, it is important to note that the findings may not be applicable to other breeds of
dogs due to distinct genetic and physiological characteristics that vary between different
breeds. Therefore, we should consider diversity in the dog population and include a
broader range of breeds in the future studies to increase the generalizability of the findings.
This study’s behavioral assessments were conducted over a relatively short period,
while prior research has highlighted the importance of longer intervals for observing
meaningful variations in the scores [
39
]. To evaluate the consistency of behavior over an
extended period, using a longer interval of assessment may be necessary. In addition, in a
study that tracks individuals over an extended period, the lack of control for maturation
can impede the accuracy of determining cause-and-effect relationships between study
variables [
98
]. Therefore, to achieve more reliable results, we should include a control
group that does not participate in the training and take age into account as a potential
confounding variable.
6. Conclusions
In this paper, through the analysis of resting-state fMRI data collected from dogs
undergoing training for detection tasks in different time points, we have identified core
brain regions that may aid in the early prediction of dogs’ success in training. Additionally,
our investigation into shared brain regions and functions between humans and canines
significantly contributes to our understanding of interspecies cognition and behavior. These
findings highlight the promising avenues potential for future research in this area.
Supplementary Materials: The following supporting information can be downloaded at: https://www.
mdpi.com/article/10.3390/ani14071082/s1, Figure S1: A pictorial spatial representation of “target”
regions in human brain and dog brain; Figure S2: A schematic illustration of the connectivity fingerprint
matching approach in our study; Figure S3: A pictorial spatial representation of homologous regions
in the human brain identified in Table. S3; Figure S4: A pictorial spatial representation of homologous
regions in the human brain identified in Table. S4; Table S1: Matched age for dogs and human subjects;
Table S2: Montreal Neurological Institute (MNI) coordinates of “target” regions in humans; Table S3:
Regions of the dog brain connected by paths identified above (in Table 1) and corresponding homologous
regions in the human brain with their MNI coordinates; Table S4: Regions of the dog brain connected
by paths identified above (in Table 3) and corresponding homologous regions in the human brain with
their MNI coordinates.
Author Contributions: Conceptualization: G.D., P.W., E.M., V.V., T.S.D.J. and J.S.K.; methodol-
ogy: G.D., J.S.K. and P.W.; software. G.D., S.Z. and R.B.; validation: G.D.; formal analysis, G.D.
and S.Z.; investigation, G.D., S.Z., P.W. and J.S.K.; resources, P.W.; data curation, G.D., S.Z. and
R.B.;
writing—original
draft preparation, G.D., S.Z., P.W., R.B., E.M., N.H., V.V., T.S.D.J. and J.S.K.:
writing—review
and editing; G.D., S.Z., P.W., R.B., E.M., N.H., V.V., T.S.D.J. and J.S.K.: visualization,
G.D. and S.Z.; supervision, G.D., J.S.K., P.W. and T.S.D.J.; project administration, G.D., J.S.K., P.W.
Animals 2024,14, 1082 20 of 23
and T.S.D.J.; funding acquisition, G.D., P.W., E.M., V.V., T.S.D.J. and J.S.K. All authors have read and
agreed to the published version of the manuscript.
Funding: Funding from Auburn University’s Intramural Grant and from the Defense Advanced
Research Projects Agency (government contract/grant number W911QX-13-C-0123) is gratefully
acknowledged. The views, opinions, and/or findings contained in this article are those of the authors
and should not be interpreted as representing the official views or policies, either expressed or
implied, of the Defense Advanced Research Projects Agency, US Department of Defense, or the
Federal Government of the United States.
Institutional Review Board Statement: Approval was obtained from the Auburn University Institu-
tional Animal Care and Use Committee (protocol number: 2011-1933) for performing this study.
Informed Consent Statement: All the dogs were owned by the Canine Performance Sciences program
at Auburn University. Informed consent was obtained from CPS for using their dogs in the study.
The author PW is affiliated with CPS.
Data Availability Statement: Since this project was funded by the US Department of Defense,
we cannot deposit the data in a public repository due to contractual obligations. However, indi-
viduals interested in accessing the data can request them from the authors and we can approach
DARPA (Defense Advanced Research Projects Agency) to provide permission to share the data on an
individual basis.
Acknowledgments: We thank Julie Rodiek and Wayne Duggan for facilitating data acquisition during
the study.
Conflicts of Interest: The authors declare no conflicts of interest.
References
1.
Miyadera, K.; Acland, G.M.; Aguirre, G.D. Genetic and phenotypic variations of inherited retinal diseases in dogs: The power of
within- and across-breed studies. Mamm. Genome 2012,23, 40–61. [CrossRef] [PubMed]
2.
Sachs-Ericsson, N.; Hansen, N.K.; Fitzgerald, S.G. Benefits of assistance dogs: A review. Rehabil. Psychol. 2002,47, 251–277.
[CrossRef]
3.
Gee, N.R.; Harris, S.L.; Johnson, K.L. The role of therapy dogs in speed and accuracy to complete motor skills tasks for preschool
children. Anthrozoos 2007,20, 375–386. [CrossRef]
4.
Yount, R.A.; Olmert, M.D.; Lee, M.R. Service dog training program for treatment of posttraumatic stress in service members. U.S.
Army Med. Dep. J. 2012, 63–69. Available online: http://www.ncbi.nlm.nih.gov/pubmed/22388685 (accessed on 1 April 2012).
5.
Cobb, M.; Branson, N.; McGreevy, P.; Lill, A.; Bennett, P. The advent of canine performance science: Offering a sustainable future
for working dogs. Behav. Process. 2015,110, 96–104. [CrossRef] [PubMed]
6. Slabbert, J.M.; Odendaal, J.S.J. Early prediction of adult police dog efficiency-A longitudinal study. Appl. Anim. Behav. Sci. 1999,
64, 269–288. [CrossRef]
7. Berns, G.S.; Cook, P.F. Why Did the Dog Walk Into the MRI? Curr. Dir. Psychol. Sci. 2016,25, 363–369. [CrossRef]
8.
Thompkins, A.M.; Deshpande, G.; Waggoner, P.; Katz, J.S. Functional Magnetic Resonance Imaging of the Domestic Dog: Research,
Methodology, and Conceptual Issues. Comp. Cogn. Behav. Rev. 2016,11, 63–82. [CrossRef]
9.
Biswal, B.; Yetkin, F.Z.; Haughton, V.M.; Hyde, J.S. Functional connectivity in the motor cortex of resting human brain using
echo-planar. Magn. Reson. Med. 1995,34, 537–541. [CrossRef]
10.
Buckner, R.L.; Andrews-Hanna, J.R.; Schacter, D.L. The brain’s default network: Anatomy, function, and relevance to disease.
Ann. N. Y. Acad. Sci. 2008,1124, 1–38. [CrossRef]
11. Gilbert, D.T.; Wilson, T.D. Prospection: Experiencing the future. Science 2007,317, 1351–1354. [CrossRef] [PubMed]
12.
Hampson, M.; Driesen, N.R.; Skudlarski, P.; Gore, J.C.; Constable, R.T. Brain connectivity related to working memory performance.
J. Neurosci. Off. J. Soc. Neurosci. 2006,26, 13338–13343. [CrossRef] [PubMed]
13.
Leech, R.; Kamourieh, S.; Beckmann, C.F.; Sharp, D.J. Fractionating the default mode network: Distinct contributions of the
ventral and dorsal posterior cingulate cortex to cognitive control. J. Neurosci. Off. J. Soc. Neurosci. 2011,31, 3217–3224. [CrossRef]
[PubMed]
14.
Seeley, W.W.; Menon, V.; Schatzberg, A.F.; Keller, J.; Glover, G.H.; Kenna, H.; Reiss, A.L.; Greicius, M.D. Dissociable intrinsic
connectivity networks for salience processing and executive control. J. Neurosci. Off. J. Soc. Neurosci. 2007,27, 2349–2356.
[CrossRef] [PubMed]
15.
Szabó, D.; Czeibert, K.; Kettinger, Á.; Gácsi, M.; Andics, A.; Miklósi, Á.; Kubinyi, E. Resting-state fMRI data of awake dogs (Canis
familiaris) via group-level independent component analysis reveal multiple, spatially distributed resting-state networks. Sci. Rep.
2019,9, 15270. [CrossRef] [PubMed]
16.
Beckmann, K.M.; Wang-Leandro, A.; Dennler, M.; Carrera, I.; Richter, H.; Bektas, R.N.; Steiner, A.; Haller, S. Resting state networks
of the canine brain under sevoflurane anaesthesia. PLoS ONE 2020,15, e0231955. [CrossRef] [PubMed]
Animals 2024,14, 1082 21 of 23
17.
Chen, G.; Wang, F.; Dillenburger, B.C.; Friedman, R.M.; Chen, L.M.; Gore, J.C.; Avison, M.J.; Roe, A.W. Functional magnetic
resonance imaging of awake monkeys: Some approaches for improving imaging quality. Magn. Reson. Imaging 2012,30, 36–47.
[CrossRef] [PubMed]
18.
Kulkarni, P.; Stolberg, T.; Sullivan, J.M.; Ferris, C.F. Imaging evolutionarily conserved neural networks: Preferential activation of
the olfactory system by food-related odor. Behav. Brain Res. 2012,230, 201–207. [CrossRef] [PubMed]
19.
Lu, H.; Zuo, Y.; Gu, H.; Waltz, J.A.; Zhan, W.; Scholl, C.A.; Rea, W.; Yang, Y.; Stein, E.A. Synchronized delta oscillations correlate
with the resting-state functional MRI signal. Proc. Natl. Acad. Sci. USA 2007,104, 18265–18269. [CrossRef]
20.
Wang, K.; Van Meer, M.P.A.; Van Der Marel, K.; Van Der Toorn, A.; Xu, L.; Liu, Y.; Viergever, M.A.; Jiang, T.; Dijkhuizen, R.M.
Temporal scaling properties and spatial synchronization of spontaneous blood oxygenation level-dependent (BOLD) signal
fluctuations in rat sensorimotor network at different levels of isoflurane anesthesia. NMR Biomed. 2011,24, 61–67. [CrossRef]
21.
Jia, H.; Pustovyy, O.M.; Waggoner, P.; Beyers, R.J.; Schumacher, J.; Wildey, C.; Barrett, J.; Morrison, E.; Salibi, N.; Denney, T.S.; et al.
Functional MRI of the olfactory system in conscious dogs. PLoS ONE 2014,9, e86362. [CrossRef] [PubMed]
22.
Jia, H.; Pustovyy, O.M.; Wang, Y.; Waggoner, P.; Beyers, R.J.; Schumacher, J.; Wildey, C.; Morrison, E.; Salibi, N.; Denney, T.S.; et al.
Enhancement of odor-induced activity in the canine brain by zinc nanoparticles: A functional MRI study in fully unrestrained
conscious dogs. Chem. Sens. 2016,41, 53–67. [CrossRef]
23.
Kyathanahally, S.P.; Jia, H.; Pustovyy, O.M.; Waggoner, P.; Beyers, R.; Schumacher, J.; Barrett, J.; Morrison, E.E.; Salibi, N.;
Denney, T.S.; et al. Anterior-posterior dissociation of the default mode network in dogs. Brain Struct. Funct. 2015,220, 1063–1076.
[CrossRef] [PubMed]
24.
Andics, G.A.; Gácsi, M.; Faragó, T.; Szabó, D.; Miklósi, Á.; Nieto-Castanon, A. Neural mechanisms for lexical processing in dogs.
Science 2016,353, 1030–1032. [CrossRef] [PubMed]
25.
Andics, G.M.; Faragó, T.; Kis, A.; Miklósi, Á. Voice-sensitive regions in the dog and human brain are revealed by comparative
fMRI. Curr. Biol. 2014,24, 574–578. [CrossRef] [PubMed]
26.
Berns, G.S.; Brooks, A.M.; Spivak, M. Functional MRI in awake unrestrained dogs. PLoS ONE 2012,7, e38027. [CrossRef]
[PubMed]
27.
Berns, G.S.; Brooks, A.M.; Spivak, M. Scent of the familiar: An fMRI study of canine brain responses to familiar and unfamiliar
human and dog odors. Behav. Process. 2015,110, 37–46. [CrossRef] [PubMed]
28. Berns, G.S.; Brooks, A.M.; Spivak, M.; Levy, K. Functional Mri in Awake Dogs Predicts Suitability for Assistance Work. Sci. Rep.
2016,90, 43704. [CrossRef] [PubMed]
29.
Bunford, N.; Andics, A.; Kis, A.; Miklósi, Á.; Gácsi, M. Canis familiaris as model for non-invasive comparative neuroscience.
Trends Neurosci. 2017,40, 438–452. [CrossRef]
30.
Cook, P.F.; Prichard, A.; Spivak, M.; Berns, G.S. Awake Canine fMRI Predicts Dogs’ Preference for Praise Versus Food. Soc. Cogn.
Affect. Neurosci. 2016,11, 1853–1862. [CrossRef]
31.
Cook, P.F.; Spivak, M.; Berns, G. Neurobehavioral evidence for individual differences in canine cognitive control: An awake fMRI
study. Anim. Cogn. 2016,19, 867–878. [CrossRef] [PubMed]
32.
Cuaya, L.V.; Hernández-Pérez, R.; Concha, L. Our faces in the dog’s brain: Functional imaging reveals temporal cortex activation
during perception of human faces. PLoS ONE 2016,11, e0149431. [CrossRef] [PubMed]
33.
Dilks, D.D.; Cook, P.; Weiller, S.K.; Berns, H.P.; Spivak, M.; Berns, G.S. Awake fMRI reveals a specialized region in dog temporal
cortex for face processing. PeerJ 2015,3, e1115. [CrossRef] [PubMed]
34.
Huber, L.; Lamm, C. Understanding dog cognition by functional magnetic resonance imaging. Learn. Behav. 2017,45, 101–102.
[CrossRef] [PubMed]
35.
Ramaihgari, B.; Pustovyy, O.M.; Waggoner, P.; Beyers, R.J.; Wildey, C.; Morrison, E.; Salibi, N.; Katz, J.S.; Denney, T.S.;
Vodyanoy, V.J.; et al. Zinc nanoparticles enhance brain connectivity in the canine olfactory network: Evidence from an fMRI study
in unretsrained awake dogs. Front. Vet. Med. 2018,5, 127. [CrossRef] [PubMed]
36.
Thompson, T.W.; Waskom, M.L.; Gabrieli, J.D. Intensive Working Memory Training Produces Functional Changes in Large-scale
Frontoparietal Networks. J. Cogn. Neurosci. 2016,28, 575–588. [CrossRef] [PubMed]
37.
Thompkins, A.M.; Lazarowski, L.; Ramaiahgari, B.; Gotoor, S.S.R.; Waggoner, P.; Denney, T.S.; Deshpande, G.; Katz, J.S. Dog–
human social relationship: Representation of human face familiarity and emotions in the dog brain. Anim. Cogn. 2021,24, 251–266.
[CrossRef] [PubMed]
38.
Rocznik, D.; Sinn, D.L.; Thomas, S.; Gosling, S.D. Criterion analysis and content validity for standardized behavioral tests in a
detector-dog breeding program. J. Forensic Sci. 2015,60, S213–S221. [CrossRef]
39.
Sinn, D.L.; Gosling, S.D.; Hilliard, S. Personality and performance in military working dogs: Reliability and predictive validity of
behavioral tests. Appl. Anim. Behav. Sci. 2010,127, 51–65. [CrossRef]
40.
Power, J.D.; Barnes, K.A.; Snyder, A.Z.; Schlaggar, B.L.; Petersen, S.E. Spurious but systematic correlations in functional
connectivity MRI networks arise from subject motion. Neuroimage 2012,59, 2142–2154. [CrossRef]
41.
Chao-Gan, Y.; Yu-Feng, Z. DPARSF: AMATLABToolbox for “Pipeline” Data Analysis of Resting-State fMRI. Front. Syst. Neurosci.
2010,4, 13. [CrossRef] [PubMed]
42.
Friston, K.J.; Holmes, A.P.; Worsley, K.J.; Poline, J.P.; Frith, C.D.; Frackowiak, R.S. Statistical parametric mapping in functional
imaging: A general linear approach. Hum. Brain Mapp. 1995,2, 189–210. [CrossRef]
Animals 2024,14, 1082 22 of 23
43.
Bluhm, R.; Williamson, P.; Lanius, R.; Théberge, J.; Densmore, M.; Bartha, R.; Osuch, E. Resting state default-mode network
connectivity in early depression using a seed region-of-interest analysis: Decreased connectivity with caudate nucleus. Psychiatry
Clin. Neurosci. 2009,63, 754–761. [CrossRef] [PubMed]
44.
Fair, D.A.; Cohen, A.L.; Dosenbach, N.U.F.; Church, J.A.; Miezin, F.M.; Barch, D.M.; Raichle, M.E.; Petersen, S.E.; Schlaggar, B.L.
The maturing architecture of the brain’s default network. Proc. Natl. Acad. Sci. USA 2008,105, 4028–4032. [CrossRef] [PubMed]
45.
Sheline, Y.I.; Raichle, M.E.; Snyder, A.Z.; Morris, J.C.; Head, D.; Wang, S.; Mintun, M.A. Amyloid Plaques Disrupt Resting State
Default Mode Network Connectivity in Cognitively Normal Elderly. Biol. Psychiatry 2010,67, 584–587. [CrossRef] [PubMed]
46.
Fox, M.D.; Raichle, M.E. Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging. Nat.
Rev. Neurosci. 2007,8, 700–711. [CrossRef] [PubMed]
47.
Raichle, M.E.; MacLeod, A.M.; Snyder, A.Z.; Powers, W.J.; Gusnard, D.A.; Shulman, G.L. A default mode of brain function. Proc.
Natl. Acad. Sci. USA 2001,98, 676–682. [CrossRef] [PubMed]
48.
van den Heuvel, M.P.; Stam, C.J.; Boersma, M.; Hulshoff Pol, H.E. Small-world and scale-free organization of voxel-based
resting-state functional connectivity in the human brain. NeuroImage 2008,43, 528–539. [CrossRef] [PubMed]
49. Cox, D.R. The regression analysis of binary sequences (with discussion). J. R. Stat. Soc. Ser. B 1958,20, 215–242.
50.
Patronek, G.J.; Waters, D.J.; Glickman, L.T. Comparative longevity of pet dpgs and humans: Implication for gerontology research.
J. Gerontol. Biol. Sci. 1997,52A, B171–B178. [CrossRef]
51.
Smith, S.M.; Andersson, J.; Auerbach, E.J.; Beckmann, C.F.; Bijsterbosch, J.; Douaud, G.; Duff, E.; Feinberg, D.A.; Griffanti, L.;
Harms, M.P.; et al. Resting-state fMRI in the Human Connectome Project. NeuroImage 2013,80, 144–168. [CrossRef] [PubMed]
52.
Horowitz, A. Domestic Dog Cognition and Behavior: The Scientific Study of Canis familiaris; Springer: Berlin/Heidelberg, Germany, 2014.
[CrossRef]
53.
Xia, M.; Wang, J.; He, Y. BrainNet Viewer: A Network Visualization Tool for Human Brain Connectomics. PLoS ONE 2013,8, e68910.
[CrossRef] [PubMed]
54.
Bassett, W.N.F.; Porter, M.A.; Mucha, P.J.; Carlson, J.M.; Grafton, S.T. Dynamic reconfiguration of human brain networks during
learning. Proc. Natl. Acad. Sci. USA 2011,108, 7641–7646. [CrossRef] [PubMed]
55.
Bassett, D.S.; Wymbs, N.F.; Rombach, M.P.; Porter, M.A.; Mucha, P.J.; Grafton, S.T. Task-Based Core-Periphery Organization of
Human Brain Dynamics. PLoS Comput. Biol. 2013,9, 16. [CrossRef] [PubMed]
56.
Braun, U.; Schäfer, A.; Walter, H.; Erk, S.; Romanczuk-Seiferth, N.; Haddad, L.; Bassett, D.S. Dynamic reconfiguration of frontal
brain networks during executive cognition in humans. Proc. Natl. Acad. Sci. USA 2015,112, 11678–11683. [CrossRef] [PubMed]
57.
Chai, L.R.; Mattar, M.G.; Blank, I.A.; Fedorenko, E.; Bassett, D.S. Functional Network Dynamics of the Language System. Cereb.
Cortex 2016,26, 4148–4159. [CrossRef] [PubMed]
58.
Ekman, M.; Derrfuss, J.; Tittgemeyer, M.; Fiebach, C.J. Predicting errors from reconfiguration patterns in human brain networks.
Proc. Natl. Acad. Sci. USA 2012,109, 16714–16719. [CrossRef] [PubMed]
59.
Gollo, L.L.; Roberts, J.A.; Cocchi, L. Mapping how local perturbations influence systems-level brain dynamics. NeuroImage 2017,
160, 97–112. [CrossRef] [PubMed]
60.
Poellinger, A.; Thomas, R.; Lio, P.; Lee, A.; Makris, N.; Rosen, B.R.; Kwong, K.K. Activation and habituation in olfaction—An
fMRI study. NeuroImage 2001,13, 547–560. [CrossRef]
61.
Savic, I.; Gulyas, B.; Larsson, M.; Roland, P. Olfactory functions are mediated by parallel and hierarchical processing. Neuron 2000,
26, 735–745. [CrossRef]
62.
Vedaei, F.; Oghabian, M.A.; Firouznia, K.; Harirchian, M.H.; Lotfi, Y.; Fakhri, M. The Human Olfactory System: Cortical Brain
Mapping Using fMRI. Iran. J. Radiol. 2016,in press. [CrossRef]
63.
Kollndorfer, K.; Fischmeister, F.P.S.; Kowalczyk, K.; Hoche, E.; Mueller, C.A.; Trattnig, S.; Schöpf, V. Olfactory training induces
changes in regional functional connectivity in patients with long-term smell loss. NeuroImage Clin. 2015,9, 401–410. [CrossRef]
64.
Huerta, C.I.; Sarkar, P.R.; Duong, T.Q.; Laird, A.R.; Fox, P.T. Neural bases of food perception: Coordinate-based meta-analyses of
neuroimaging studies in multiple modalities. Obesity 2014,22, 1439–1446. [CrossRef]
65.
Dosenbach, N.U.F.; Fair, D.A.; Cohen, A.L.; Schlaggar, B.L.; Petersen, S.E. A dual-networks architecture of top-down control.
Trends Cogn. Sci. 2008,12, 99–105. [CrossRef]
66.
Markett, S.; Reuter, M.; Montag, C.; Voigt, G.; Lachmann, B.; Rudorf, S.; Elger, C.E.; Weber, B. Assessing the function of the fronto-
parietal attention network: Insights from resting-state fMRI and the attentional network test. Hum. Brain Mapp. 2014,35, 1700–1709.
[CrossRef]
67. Olsson, A.; Ochsner, K.N. The role of social cognition in emotion. Trends Cogn. Sci. 2008,12, 65–71. [CrossRef]
68.
Ghahremani, M.; Hutchison, R.M.; Menon, R.S.; Everling, S. Frontoparietal Functional Connectivity in the Common Marmoset.
Cereb. Cortex 2016,27, 3890–3905. [CrossRef]
69.
Hearne, L.J.; Mattingley, J.B.; Cocchi, L. Functional brain networks related to individual differences in human intelligence at rest.
Sci. Rep. 2016,6, 32328. [CrossRef]
70.
Song, M.; Zhou, Y.; Li, J.; Liu, Y.; Tian, L.; Yu, C.; Jiang, T. Brain spontaneous functional connectivity and intelligence. NeuroImage
2008,41, 1168–1176. [CrossRef]
71.
Montaldi, D.; Spencer, T.J.; Roberts, N.; Mayes, A.R. The neural system that mediates familiarity memory. Hippocampus 2006,
16, 504–520. [CrossRef]
Animals 2024,14, 1082 23 of 23
72.
Vilberg, K.L.; Rugg, M.D. Dissociation of the neural correlates of recognition memory according to familiarity, recollection, and
amount of recollected information. Neuropsychologia 2007,45, 2216–2225. [CrossRef]
73. Woodruff, C.C.; Johnson, J.D.; Uncapher, M.R.; Rugg, M.D. Content-specificity of the neural correlates of recollection. Neuropsy-
chologia 2005,43, 1022–1032. [CrossRef]
74.
Vilberg, K.L.; Rugg, M.D. Memory retrieval and the parietal cortex: A review of evidence from a dual-process perspective.
Neuropsychologia 2008,46, 1787–1799. [CrossRef]
75.
Herregodts, P.; Ebinger, G.; Michotte, Y. Distribution of monoamines in human brain: Evidence for neurochemical heterogeneity
in subcortical as well as in cortical areas. Brain Res. 1991,542, 300–306. [CrossRef]
76.
Sara, S.J. The locus coeruleus and noradrenergic modulation of cognition. Nature Reviews. Neuroscience 2009,10, 211–223.
[CrossRef]
77.
Sara, S.J.; Bouret, S. Orienting and Reorienting: The Locus Coeruleus Mediates Cognition through Arousal. Neuron 2012,
76, 130–141. [CrossRef]
78. Harley, C.W. Norepinephrine and dopamine as learning signals. Neural Plast. 2004,11, 191–204. [CrossRef]
79.
Murchison, C.F.; Zhang, X.-Y.; Zhang, W.-P.; Ouyang, M.; Lee, A.; Thomas, S.A. A distinct role for norephinephrine in memory
retrieval. Cell 2004,117, 131–142. [CrossRef]
80.
Guedj, C.; Monfardini, E.; Reynaud, A.J.; Farnè, A.; Meunier, M.; Hadj-Bouziane, F. Boosting Norepinephrine Transmission
Triggers Flexible Reconfiguration of Brain Networks at Rest. Cereb. Cortex 2016,27, 4691–4700. [CrossRef]
81.
Pinsk, M.A.; Arcaro, M.; Weiner, K.S.; Kalkus, J.F.; Inati, S.J.; Gross, C.G.; Kastner, S. Neural representations of faces and body
parts in macaque and human cortex: A comparative FMRI study. J. Neurophysiol. 2009,101, 2581–2600. [CrossRef]
82.
Langers, D.R.; van Dijk, P.; Backes, W.H. Lateralization, connectivity and plasticity in the human central auditory system.
Neuroimage 2005,28, 490–499. [CrossRef]
83.
Siniscalchi, M.; d’Ingeo, S.; Fornelli, S.; Quaranta, A. Lateralized behavior and cardiac activity of dogs in response to human
emotional vocalizations. Sci. Rep. 2018,8, 77. [CrossRef]
84.
Overall, K.L. That Dog Is Smarter Than You Know: Advances in Understanding Canine Learning, Memory, and Cognition. Top.
Companion Anim. Med. 2011,26, 2–9. [CrossRef]
85.
ChaeYoung, L.; ChangHoon, K.; SooAn, S.; DaeSung, S.; JooHyun, K.; ChanKyu, P. The dopamine D4 receptor polymorphism
affects the canine fearfulness. Anim. Cells Syst. 2008,12, 77–83. [CrossRef]
86.
Wan, M.; Hejjas, K.; Ronai, Z.; Elek, Z.; Sasvari-Szekely, M.; Champagne, F.A.; Miklósi, Á.; Kubinyi, E. DRD4 and TH gene
polymorphisms are associated with activity, impulsivity and inattention in Siberian Husky dogs. Anim. Genet. 2013,44, 717–727.
[CrossRef]
87.
Primus, R.J.; Thurkauf, A.; Xu, J.; Yevich, E.; Mcinerney, S.; Shaw, K.; Tallman, J.F.; Gallager, D.W., II. Localization and
Characterization of Dopamine D4 Binding Sites in Rat and Human Brain by Use of the Novel, D4 Receptor-Selective Ligand [3H]
NGD 94-1. J. Pharmacol. Exp. Ther. 1997,282, 1020–1027.
88.
Haruno, M. A Neural Correlate of Reward-Based Behavioral Learning in Caudate Nucleus: A Functional Magnetic Resonance
Imaging Study of a Stochastic Decision Task. J. Neurosci. 2004,24, 1660–1665. [CrossRef]
89.
Haruno, M.; Kawato, M. Heterarchical reinforcement-learning model for integration of multiple cortico-striatal loops: fMRI
examination in stimulus-action-reward association learning. Neural Netw. 2006,19, 1242–1254. [CrossRef]
90.
Cook, P.F.; Spivak, M.; Berns, G.S. One pair of hands is not like another: Caudate BOLD response in dogs depends on signal
source and canine temperament. PeerJ 2014,2, e596. [CrossRef]
91.
Van Essen, D.C.; Dierker, D.L. Surface-based and probabilistic atlases of primate cerebral cortex. Neuron 2007,56, 209–225.
[CrossRef]
92.
Orban, G.A.; Claeys, K.; Nelissen, K.; Smans, R.; Sunaert, S.; Todd, J.T.; Wardak, C.; Durand, J.B.; Vanduffel, W. Mapping the
parietal cortex of human and non-human primates. Neuropsychologia 2006,44, 2647–2667. [CrossRef] [PubMed]
93. Striedter, G.F. Brain homology and function: An uneasy alliance. Brain Res. Bull. 2002,57, 239–242. [CrossRef] [PubMed]
94.
Mars, R.B.; Verhagen, L.; Gladwin, T.E.; Neubert, F.X.; Sallet, J.; Rushworth, M.F.S. Comparing brains by matching connectivity
profiles. Neurosci. Biobehav. Rev. 2016,60, 90–97. [CrossRef] [PubMed]
95.
Passingham, R.E.; Stephan, K.E.; Kotter, R. The anatomical basis of functional localization in the cortex. Nat. Rev. Neurosci. 2002,
3, 606–616. [CrossRef] [PubMed]
96.
Hutchison, R.M.; Gallivan, J.P. Functional coupling between frontoparietal and occipitotemporal pathways during action and
perception. Cortex 2016,98, 8–27. [CrossRef] [PubMed]
97.
Sallet, J.; Mars, R.B.; Noonan, M.P.; Neubert, F.-X.; Jbabdi, S.; O’Reilly, J.X.; Filippini, N.; Thomas, A.G.; Rushworth, M.F. The
organization of dorsal frontal cortex in humans and macaques. J. Neurosci. Off. J. Soc. Neurosci. 2013,33, 12255–12274. [CrossRef]
98.
Reicher, V.; Bunford, N.; Kis, A.; Carreiro, C.; Csibra, B.; Kratz, L.; Gácsi, M. Developmental features of sleep electrophysiology in
family dogs. Sci. Rep. 2021,11, 22760. [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.
Available via license: CC BY 4.0
Content may be subject to copyright.
