Available via license: CC BY 4.0
Content may be subject to copyright.
NON-INVASIVE CANINE ELECTROENCEPHALOGRAPHY (EEG):
A SYSTEMATIC REVIEW
PREP RI NT
Akash Kulgod1, Dirk van der Linden2, Lucas G S França2, Melody Jackson3, and Anna Zamansky4
1Dognosis Technologies
2Northumbria University, Newcastle upon Tyne, UK
3Georgia Institute of Technology, Atlanta, USA
4University of Haifa, Haifa, Israel
August 11, 2023
ABS TRAC T
The emerging field of canine cognitive neuroscience uses neuroimaging tools such as electroen-
cephalography (EEG) and functional magnetic resonance imaging (fMRI) to map the cognitive pro-
cesses of dogs to neural substrates in their brain. Within the past decade, the non-invasive use of
EEG has provided real-time, accessible, and portable neuroimaging insight into canine cognitive
processes. To promote systematization and create an overview of framings, methods and findings
for future work, we provide a systematic review of non-invasive canine EEG studies (N=22), dis-
secting their study makeup, technical setup, and analysis frameworks and highlighting emerging
trends. We further propose new directions of development, such as the standardization of data struc-
tures and integrating predictive modeling with descriptive statistical approaches. Our review ends by
underscoring the advances and advantages of EEG-based canine cognitive neuroscience and the po-
tential for accessible canine neuroimaging to inform both fundamental sciences as well as practical
applications for cognitive neuroscience, working dogs, and human-canine interactions.
Keywords canine science ·cognitive neuroscience ·EEG
Introduction
The multidisciplinary field of cognitive neuroscience is a synthesis of cognitive psychology and neuroscience, aiming
at mapping "elementary cognitive functions onto specific neuronal systems" [3, pg. 613]. Through the deployment of
techniques such as the electroencephalogram (EEG) [98] and functional magnetic resonance imaging (fMRI) [29,105],
neuroscientists are able to perform empirical and quantitative analyses of cognitive processes. For a comprehen-
sive review of the history, methods, and current frameworks of cognitive neuroscience, readers are directed towards
Gazzaniga, Ivry and Mungun’s 2019 treatment of the field [30]. Another multidisciplinary field is canine science,
combining disciplines including evolution, genetics, cognition, ethology, physiology, comparative medicine, and ecol-
ogy [14, 15, 69, 74]. Investigating the recent surge of interest in the scientific study of the domestic dog, Aria and
colleagues [7] found a sixfold increase in the number of studies in canine cognition and behaviour between the years
2006 and 2018, as compared to the preceding period of 1993 to 2005. This interest extends beyond purely veterinary,
pharmaceutical, and basic neuroscience paradigms. Recent studies in canine science span such varied topics as genet-
ics [32], evolutionary neuroscience [36,37] and intelligence [6], as well as investigations into models of epilepsy [68],
aging [51,73] and dementia [95]. Recent developments, such as the rise of open-science multi-team initiatives such as
the ManyDogs Project, which aims to investigate behavioral traits across multiple centers and populations of dogs [71],
and the Working Dog Project, which focuses on improving genetic selection strategies in dog breeding [20], herald
the emerging trend of collaborative consortiums to tackle fundamental questions in the field.
Berns et al. pioneered the use of fMRI in awake, non-restrained dogs in 2012 [8], and in the following year, Kujala and
colleagues were the first to successfully deploy non-invasive EEG with non-sedated dogs [61]. Further developments
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
included the investigation of a range of cognitive processes and their neural underpinnings such as executive func-
tioning [24], visual [25, 61], auditory [5, 17], and olfactory [50] processing, social cognition [23], learning [88] and
sleep [19, 56]. While the field of canine fMRI has received increasing scientific attention, non-invasive canine EEG
has eluded similar treatment. This is despite its noticeable strengths, including high temporal resolution, accessibility,
and real-world applications. For these reasons, we offer a consolidation of the state of EEG-based canine cognitive
neuroscience, providing an overview of the key conceptual framings, methodological approaches, and findings.
To that end, this systematic review contributes the following:
An identification and mapping of 22 non-invasive canine EEG studies based on a thorough literature review and
an tailored database query paired with appropriate exclusion/inclusion criteria.
A systematic analysis of these studies dissecting their research question and participant make-up, technical setups
deployed, dataset properties, and analytical frameworks and findings.
A critical discussion on future avenues for non-invasive canine EEG identifying promising questions that can be
pursued, ideal data practices, integration with other research sub-fields and beneficial methodological and
analytical refinements.
Methods
Data sources and search query construction
No literature reviews, surveys or meta-analysis of EEG in dogs were available as a starting point for this review. To
begin, we conducted an informal search using Google Scholar in February 2023, searching for ‘EEG in dogs’ and
read through a number of obvious candidates ( [10,60, 61, 66,70,99] [17, 54–56]) to identify commonalities. Of these
papers, we noted that publishers included PLOS One, Springer, Elsevier, and the Royal Society, while we would also
expect papers published in IEEE and ACM venues to be relevant. This allowed us to minimize necessary redundancy
in data sources by employing each publisher’s own search engine, as Scopus indexed all early identified papers, and
papers by expected publishers.
After reading the initial set of studies, we made the following assumptions to aid in the construction of our search
query, to ensure as broad a coverage as possible:
• Dogs are referred to interchangeably as ‘dog*’ or ‘canine*’, even if articles do not necessarily include the full
scope of the canis family; this required an additional exclusion criterion if papers include e.g., Canis lupus
rather than just Canis lupus familiaris.
• EEG is referred to both as ‘EEG’ and ‘Electroencephalography’ or more vaguely hinted at with terms such as
‘brain signals’ or ‘neural processes’ in paper titles, meaning we needed to search through abstracts as well.
• There is no consistent (from title) indication of whether the used EEG technique was invasive or non-invasive,
this required liberal inclusion criteria to include any relevant study and stringent exclusion criteria to manually
filter any invasive EEG study.
• The first non-invasive, non-sedated canine EEG study could be clearly identified as taking place in 2013 [61]
so the year 2010 allowed for all relevant studies to be included.
Based on the above assumptions and reading of the initial papers, we constructed this search query optimized for the
SCOPUS database in March 2023:
( TITLE ( ( canine* OR dog* ) ) AND TITLE-ABS-KEY ( ( eeg OR erp OR electroencephalography ) ) ) AND
PUBYEAR > 2010 AND PUBYEAR < 2024 AND ( LIMIT-TO ( LANGUAGE , "English" ) )
The search query resulted in 205 articles in the SCOPUS database.
Study selection
We first liberally applied the inclusion criterion, followed by a set of three exclusion criteria – see Fig. 1 for an
overview of the entire process and interim study numbers. The inclusion criterion was:
inc1 Studies applying electroencephalography to dogs (canis lupus familiaris)
The following exclusion criteria were used to refine the selection:
2
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
ex1 Studies of other canine species (e.g., wolves, jackals, coyotes) – we built in this potentially redundant exclusion
criteria due to the polysemous use of ‘canine’ in some literature, sometimes referring to its actual meaning of
the Caninae subfamily, but at other times used as a synonym for dogs only.
ex2 Studies employing invasive applications of EEG – we define this as the physical ontology of the subject remaining
unviolated, i.e. the epidermis of the subject dog is not pierced or excised. This excluded any study using
intracranial EEG (iEEG) [83] as well as sub-dermal EEG that uses needle electrodes, a technique first used
by Pellegrino and Sica (2004) [84] in a veterinary context as well as cognition studies such as by Howell and
colleagues. [42]
ex3 Studies employing anaesthesia or other form of sedation – we do include studies of naturally sleeping dogs,
which, indeed, form the majority of the studies conducted in this category
One author applied the inclusion and exclusion criteria over the total set of 205 papers leading to a final selection of 22
papers. To ensure consistent application of the criteria, another author independently coding a randomly selected 10%
subset of the papers. Inter-rater reliability analysis indicated substantial agreement between authors on application of
the inclusion criterion (Cohen’s κ=0.69) criteria, as well as the exclusion criteria (resp. κ=0.69, 1.00, 1.00 for ex1,
ex2, and ex3), leading to the same set of selected papers. We found that the validating author labeled more critically,
which led to some discussion as to whether retrospective studies were to be included; but effectively these had no
impact on the actual selected publications as most were ruled out by both authors on grounds of the exclusion criteria.
Construct & trial
search query
Start Run query on
databases
Scopus
Liberally apply
inclusion criterion
Apply exclusion
criterion ex1
Apply exclusion
criterion ex3
Apply exclusion
criterion ex2
End
205
papers
89
papers
86
papers
23
papers
22
papers
Figure 1: Flowchart of the study selection process and included papers at each step of the process.
Data extraction and analysis
We dissect the obtained studies according to the different workflow stages of a scientific study: research question and
participants, data acquisition, and analysis/findings, respectively. This process was done in entirety manually by one
author. After providing an overview of the 22 studies, we break them down along the following three dimensions -
• Setups deployed to collect data: electrode types, montages, and amplifiers.
• Datasets collected in each of the papers: e.g., availability and quantity of data.
• Analysis frameworks and findings: e.g., pre-processing pipelines (where relevant), and findings.
For each of these steps, we highlight and synthesize common themes and approaches and use this to provide a consol-
idated outlook on the field and suggestions for future work.
3
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Results
Overview
The selected works are presented in Table 1. The dimensions we use and the overall trends of the studies were as
follows :
•Topic investigated. We label each study with one or more of the following categories -
–visual processing : perception, discrimination or interpretation of images (n=3)
–auditory processing : perception, discrimination or interpretation of sounds (n=1)
–language processing : perception and comprehension of speech (n=3)
–learning : associative learning, memory, and problem-solving (n=4)
–emotion : interpretation of emotionally coded stimuli (n=2)
–social cognition : behavior with con-specifics or humans (n=2)
–sleep : stages and occurrence of patterns during sleep (n=14)
–methodology : approaches to data collection and analysis such as automation or application of machine
learning (n=3)
•Specific research question. We briefly state the research question/hypothesis investigated in each study.
•Number of dogs. The number of participants in each study ranged from 2 [54] - 155 [44].
•Type of dogs. Here we refer to home/laboratory dogs, and their breed (specific/multiple).
•Dog Training. In this category, we describe the nature of training provided to the dogs for data collection
in the study, distinguishing between studies that employed special training and those that did not. We use
special training to refer to a dedicated training process consisting of multiple preparatory sessions, such as
habituation to the equipment, conducted prior to actual data collection.
Overall, four different research centers in North America and Europe have conducted non-invasive canine EEG studies,
with the majority of them coming from Eötvös Loránd University in Budapest, Hungary. The majority of studies
(n=14) recorded EEG from sleeping dogs, taking advantage of the ease of recording higher quality data for longer
periods and the presence of well-defined cross-species EEG structures associated with sleep stages and patterns.
The majority of dogs were companion (home) dogs whose guardians were recruited using surveys, and consisted of
a diverse group of pure and mixed breeds across ages in both male and female animals. Four studies [60, 61, 66, 99]
record from purpose-breed laboratory Beagles and only these dogs underwent extensive training for EEG recording
purposes, whereas companion dogs were habituated to equipment on the same day as recorded sessions.
EEG Setups
To facilitate a systematic and rigorous comparison of studies, we compiled Table 2 outlining the six different canine
EEG setups used by reviewed studies. Some important dimensions of the setups are a) electrodes type and attachment,
which includes the type of electrodes, method of attachment, whether fur was shaved, and the impedance of signals
(a measure of conductivity at the electrode-skin interface); b) the number and montage of electrodes, which refers to
the number of recording channels and the specific arrangement of electrodes; and c) the amplifier model and sampling
rate.
Electrodes type and attachment
Electrodes are made of different conductive materials and are generally categorized along two binaries - gel vs dry,
and active vs passive. The benefit of different electrode types are still debated [39,106], however, with regards to
our review, all studies used passive gel electrodes secured with surgical tape or electrode cream. Four studies shaved
the fur of participant dogs [60, 61, 66, 99], with the goal of ensuring higher quality data by decreasing impedance,
however studies that did not shave fur have achieved comparable levels of impedance. Higher impedance values result
in a lower signal-to-noise ratio (SNR), although optimal electrode impedances vary relative to an amplifier’s input
impedance [85] (but see also Kappenman and Luck [52] for a review). All studies reported keeping impedance values
below 20 kΩ.
4
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Table 1: Overview of the 22 reviewed studies
Study Ref Journal Lab Location Topic Research Question No. of dogs Type of dogs Special training
Törnqvist et. al., 2013 [99] Animal Cognition Jyväskylä Exploratory, Visual
Processing Do ERPs distinguish between human
vs dog faces? 8 Lab-breed; Beagles Yes; 1.5yrs, twice
a week
Kujala et. al., 2013 [61] Plos One Jyväskylä Visual Processing Can non-invasive EEG detect changes
in brain oscillations? 8 Lab-breed; Beagles Yes; 1.5yrs, twice
a week
Kis et. al., 2014 [56] Physiology & Behavior Budapest Sleep Can non-invasive EEG detect changes
in sleep? 22 home dogs; multiple breeds No
Kis et. al., 2017 [55] Nature Scientific Reports Budapest Sleep, Learning Are there associations between sleep
quality and learning ability? 15 home dogs; multiple breeds No
Iotchev et. al., 2017 [43] Nature Scientific Reports Budapest Sleep, Learning Are there features of sleep (spindles)
that predict learning rates? 15 home dogs; multiple breeds No
Levitt et. al., 2018 [66] Journal of Neuroscience Methods Rhode Island Methodology Can ML (SVM) models detect EEG ar-
tifacts across humans and dogs? 9 Lab-breed; Beagles Yes, unspecified
Bunford et. al., 2018 [16] Nature Scientific Reports Budapest Sleep What are some variables that affect
sleep macrostructure? 16 home dogs; multiple breeds No
Iotchev et. al., 2019 [44] Nature Scientific Reports Budapest Sleep How does sex and age affect non-REM
EEG activity? 155 home dogs; multiple breeds No
Iotchev et. al., 2020a [45] Nature Scientific Reports Budapest Sleep, Methodology Are some measures of sleep spindles
better predictors of learning rates? 46 home dogs; multiple breeds No
Gergely et. al., 2020 [31] Animals Budapest Sleep, Methodology What factors affect sleep scoring and
can it be automated with ML? 10 home dogs; multiple breeds No
Iotchev et. al., 2020b [46] Nature Scientific Reports Budapest Sleep, Learning What is the relation between spindle
frequency and learning? 58 home dogs; multiple breeds No
Kiss et. al., 2020 [57] Frontiers of Psychology Budapest Sleep, Social Cogni-
tion Does sleep structure correlate with per-
formance in a cooperation task? 27 home dogs; multiple breeds No
Reicher et. al., 2020 [92] Journal of Sleep Research Budapest Sleep Does canine sleep macrostructure dis-
play the first-night effect? 24 home dogs; multiple breeds No
Kujala et. al., 2020 [60] Nature Scientific Reports Jyväskylä Visual Processing Can ERPs predict visual (faces) stim-
uli? 8 Lab-breed; Beagles Yes; 1.5yrs, twice
a week
Magyari et. al., 2020 [70] Royal Society Open Science Budapest Language Processing Do ERPs distinguish between
known/unknown/nonsense words? 17 home dogs; multiple breeds No
Reicher et. al., 2021a [91] Nature Scientific Reports Budapest Sleep Does sleep macrostructure vary based
on age? 91 home dogs; multiple breeds No
Reicher et. al., 2021b [93] Nature Scientific Reports Budapest Sleep Do dogs display hemispheric asymme-
try during NREM sleep? 19 home dogs; multiple breeds No
Boros et. al., 2021 [10] Current Biology Budapest Language Process-
ing, Learning Do dogs use statistical learning to seg-
ment speech and learn words? 19 home dogs; multiple breeds No
Bálint et. al., 2022 [17] Royal Society Open Science Budapest Auditory Processing Do ERPs distinguish between human
and dog vocalizations? 17 home dogs; multiple breeds No
Karpi´
nski et. al., 2022 [54] Journal of Veterinary Behavior Lublin Exploratory Do ERPs distinguish between voice
commands? 2 home dogs; multiple breeds No
Carreiro et. al., 2022 [18] MDPI Animals Budapest Sleep, Social Cogni-
tion Do sleep features correlate with attach-
ment towards human guardians? 42 home dogs; multiple breeds no
Carreiro et. al., 2023 [19] Nature Scientific Reports Budapest Sleep Does sleep efficiency correlate with
guardian-rated hyperactivity? 86 home dogs; multiple breeds No
5
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Electrodes number and position
Canines have smaller brains than humans and subsequent less space available for electrodes. The reviewed studies
used between one to seven electrodes, along with a ground and reference. In comparison, human EEG studies typically
vary between using 4-256 active channels [65]. However, it is worth noting that dogs have far fewer cortical neurons
than humans, about 500 million [48] compared to 16 billion [38] in humans - a ratio of 1:32. Arguably, this means
that a 4 channel recording in dogs is equivalent to a 128 ’high density’ recording in humans.
Electrode montages in EEG research refers to the "logical, orderly arrangements of electroencephalographic deriva-
tions or channels that are created to display activity over the entire head and to provide lateralizing and localizing
information." [1]. Human EEG research generally uses the 10-20 montage system, formalized in 1957-58 by Herbert
Jasper [49], to standardize electrode montages and ensure replicability across studies. Canine EEG has traditionally
borrowed from the human 10-20 system and the reviewed studies use similar derivations, although setups differed from
each other in the montages used and authors differed in the labels used for specific montage positions. For an outlook
of the montages used in the reviewed studies, see Figure 2. A notable challenge with standardizing electrode montages
in dogs is the remarkable variance in head shape and size across canine breeds. One way to solve this challenge is
to record only from dogs of the same breed and avoid it all altogether, which was the case with the studies recording
only from lab-bred Beagles. [60, 61, 66,99]. Another approach is to use the relative distance between breed-invariant
anatomical markers followed by the Budapest setups [56,70].
Traditional electrode montages on humans when translated to dogs might suffer from a higher rate of muscular artifacts
due to the presence of a muscular scalp in dogs, which motivated the authors of the Budapest setups to place electrodes
on the anteroposterior midline, or sagittal crest, of the canine skull (Fz, Cz, and Pz respectively), as it is a bony
ridge that minimizes muscular artifacts. These setups also used either 1 or 2 electrodes close to the eyes (Fp1 and
Fp2) to measure electrooculography (EOG) signals. EOG signals are useful for understanding eye activity, such as
blinks, which can have an effect on recordings from other electrodes. Jyväskylä setups differed in their approach by
positioning electrodes laterally - on the frontal cortex F3 and F4, on the posterior cortex P3 and P4, and the temporal
cortex T3 and T4 (the latter only in the 2020 study [60]).
Finally, the position of the reference electrode plays a key role in recordings as EEG data is derivational and signals
from any electrode are meaningful only in respect to a reference electrode (or electrodes). In the Budapest setups,
either the center of the sagittal crest (Cz) or the bony ridge of the occipital protuberance, or occiput, (Pz) was used as
a reference, given its relative distance from both neural and muscular activity. The Rhode Island setup also used a Cz
reference. Meanwhile, Jyväskylä setups used an electrode placed on the ear as the reference, as did the sole Lublin
study. Figure 2 gives an overview of different potential electrode montages used across the selected studies.
Figure 2: Different electrode montages used by the six technical setups. Electrode labels follow from the human 10-20
system, where letters indicate the lobe — Fp=pre-frontal, F=frontal, P=parietal, T=temporal, and C=center. The left
and right zygomatic arch are also depicted as LZA and RZA. Odd numbers refer to electrode placement on the left
side and even numbers indicate placement on the right side of the brain. Reference electrode positions are highlighted
in orange. It is worth pointing out that large variance amongst canine breeds means the figure is not representative of
all canine individuals.
6
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Amplifier model and sampling rate
Various factors determine amplifier performance including sampling rate, input range, amplifier impedance, band-
width, and the common mode removal ratio (CMRR). A detailed discussion of these factors is outside the scope of
this review (readers are directed towards the 2023 review of EEG systems by Niso and colleagues [79] for a discus-
sion). One feature worth highlighting is the sampling rate which refers to the number of samples that are digitally
acquired per second (measured in Hz). Higher sampling rates are useful primarily to measure higher frequency brain
activity, given by the Nyquist-Shannon theorem [103] that states that for any periodic signal of a given frequency, a
sampling rate higher than two times the frequency of the signal is needed to accurately detect its presence. Sampling
rates of amplifiers used ranged from 250-1024 Hz. As all studies focused analysis on bandwidths between 0-50Hz, all
amplifiers had a sufficient sampling rate.
Table 2: Overview of the six EEG setups used by the reviewed studies.
Setup tag
Papers
Electrode type
Attachment
Fur shaved
Impedance (kω)
Active EL pos
Active EL pos
Ref EL
Ground EL
Amplifier
Sampling rate (Hz)
Setup1 Jyväskylä [60,61, 99] Unilect 40555
Neonatal EEG
electrodes
Bio-adhesive
solid gel Yes 9 ± 4 7 Fp1, Fp2, F3,
F4 P3, P4, Cz
(2013); F3, F4,
T3, T4, P3, P4,
Cz (2020)
Right ear Lower back Embla Titanium-recorder 512
Setup2 Rhode Island [66] Custom built Easy-
Cap EEG with Hy-
drospot EEG Elec-
trodes
Electrode gel Yes N/A 1 Fz Midline (Cz) Left musculus
temporalis Wireless microEEG, Biosignal
group 250
Setup3 Lublin [54] N/A Aqueous KCI Yes N/A 2 F3,F4 Ear N/A 16-channel MindSet-1000 Nolan
Computer Systems 256
Setup4 Budapest 1 [16, 43,44, 55,56, 91] Gold-coated
Ag|AgCl EC2 Grass Elec-
trode Cream No 15 2 Fz, Fp1 Cz Left musculus
temporalis 30-channel Flat Style SLEEP La
Mont Headbox 249
Setup5 Budapest 2 [18, 19,44–46, 57,91, 92] Gold-coated
Ag|AgCl Electrode cream No 20 4 Fz, Cz, Fp1, Fp2 Pz Left musculus
temporalis 25-channel SAM 25R MicroMed
Headbox 1024
Setup6 Budapest 3 [10, 17–19,70, 91] Gold-coated
Ag|AgCl EC2 Grass Elec-
trode Cream No 15 4 Fz, Cz, Fp1, Fp2 Pz LMT 40-channels NuAmps amplifier,
Compumedics Neuroscan 1000
Datasets
The six different technical setups described in the prior section were used to acquire 18 datasets, summarized in Table
3. The discrepancy between number of papers and datasets is because some papers developed novel methodological
approaches [43] or refinements [31, 45] using older datasets, or deploying a different analytical framework on the
same dataset [61]. Many studies use a combination/amalgamation of different data sources so the datasets presented
here should not be assumed to be well-demarcated and unique. Moreover, the overall number of dogs is difficult to
establish, as we observed a considerable overlap of participants between studies and datasets, with the same dogs
being recorded multiple times and the same datasets being reused in different papers. Datasets were categorized into
either epoch or continuous paradigms, based on which, the number of events/minutes of data per dog and per dataset
was calculated. All but one [66] of the awake dogs datasets were epoched and all sleep datasets were continuous.
Finally, we noted the availability of the datasets. No datasets were freely available in their entirety and six datasets
were partially available, either requiring a) additional author authorization on a data repository; b) an email request;
c) missing important metadata; d) missing raw data, i.e. containing only processed data.
Analysis and Findings
We review and consolidate awake and asleep dog studies separately, given their different frameworks, workflows, and
results.
Preprocessing
Preprocessing pipelines in EEG data analysis involve several steps to clean and prepare the data. One of these steps
is artifact removal that aims to identify and remove segments of the signal containing unwanted noise or corruption.
These artifacts can originate from various non-neural sources, such as eye blinks, muscle activity, or external inter-
ference [76]. Artifact removal can be performed using automated methods that apply predefined thresholds, such as
amplitude criteria, to detect and exclude segments with excessive noise. Alternatively, manual inspection techniques
like independent component analysis (ICA) or video monitoring can be employed. The choice between manual and
automated detection methods is a topic of ongoing debate in the field, as each approach has its strengths and limita-
tions. [47]
7
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Reviewed studies differed in their application of pre-processing techniques, which have potential important ramifica-
tions for the analysis and validity of their respective findings. For example, Magyari et al. (2020) [70] tested two
different artifact removal procedures - a multi-level method combining quantitative and qualitative steps, as well as
a single-step approach using filtering and amplitude-based artifact removal. The multi-level approach consisted of
automated amplitude-based rejection, manual video coding of movement and manual inspection of EEG data. The
single-step consisted only of automated artifact removal and filtering. The results showed similar condition differ-
ences between the two cleaning procedures, with a varying percentage of rejected trials. Approximately 75% of the
trials were rejected in the multi-level data cleaning, while 53% were rejected in the amplitude-based procedure. How-
ever, the analysis findings did not differ between preprocessing pipelines, and the authors conclude that both manual
multi-step and automated single-step pipelines were equivalent. Subsequent studies from the Budapest group only use
the automated amplitude-based rejection step.
Kujala et al. [60] differed in their approach by applying a manual inspection of independent component analysis
(ICA) components to mitigate muscular and other artifacts. Artifact-related components were visually identified and
excluded, while a general linear model was used to remove potential electric leakage from the stimulus trigger signal
to EEG channels.
In contrast, Levitt and colleagues [66] trained a support vector machine (SVM) classifier for the automated detection
of EEG artifacts in human, canine, and rodent subjects. The models showed relatively high accuracy across species
in identifying artifacts caused by skeletal and ocular muscles, with an accuracy of 80.57% and an AUC-ROC value of
0.87 for canines specifically.
Wakefulness EEG in Dogs
Awake dog EEG activity accounts for eight of the 22 examined studies in this revision. One of the studies, Levitt et
al., focused on training a SVM model to detect artifacts for preprocessing. Two of the remaining seven studies were
exploratory, with Kujala and colleagues [61] showing for the first time the ability to perform non-invasive EEG in
eight awake dogs in 2013. They observed changes in the power spectrum over the P3/P4 (parieto-occipital) electrodes
during the presentation of a visual stimulus vs rest. Karpinski et. al. [54] recorded pilot data from two companion
dogs during rest and after two different commands, also observing qualitative changes in the power spectrum over the
F3/F4 (frontal) electrodes.
The remaining five studies [10,17,60,70,99] deployed event-related potentials (ERP) frameworks to understand visual,
auditory and language processing. Törnqvist et al. [99] investigated the ERPs of dogs in response to human and dog
faces. The study found that ERPs corresponding to early visual processing were detectable at 75–100 ms from stimulus
onset, and significant differences for dog and human faces could be identified at around 75 ms at posterior sensors.
Another study [60] deployed a similar experimental paradigm, with the addition of emotionally valenced faces and
objects, and detected a group-level response sensitive to emotional expressions at 130–170 ms, and the differentiation
of faces from objects occurring at 120–130 ms. The authors also trained a support vector machine (SVM) classifier
on the EEG data to discriminate between responses to pairs of images. The classification accuracy was highest for
humans/dogs vs. scrambled images, with the most informative time interval being between 100–140 ms and 240–280
ms after the presentation of stimuli.
Two studies investigated facets of language processing using ERP paradigms [10,70]. Magyari et. al [70] explored the
group-level ERPs of dogs listening to known, unknown, and nonsense words, finding a significant difference in ERP
values between known and nonsense words at 650–800 ms. They also found a positive association between the word
usage frequency by the dog’s guardian and the individual dog’s ERP effects, providing evidence that the association
may arise from familiarity with the words. The second study to deploy an ERP paradigm [10] investigated the neural
processes underlying speech segmentation. The authors examined ERPs from the presentation of artificial words, after
participants had been exposed to a continuous speech stream that differed in the distribution of words . In their study,
two important factors were considered: transitional probability and word frequency. Transitional probability refers
to the likelihood of one sound or word following another in a sequence, reflecting the statistical regularities of the
language. Word frequency, on the other hand, represents how often a word occurs in a given language. The results
of the study showed that the ERPs exhibited an early effect (220-470 ms) related to transitional probability and a late
component (590-790 ms) modulated by both word frequency and transitional probability, indicating the involvement
of multiple cognitive processes in speech segmentation.
Finally, Balint et al. [17] investigated the auditory processing of 17 dogs in response to human and dog vocalizations.
They found that, similar to humans, dogs exhibited differential ERP responses based on the species of the vocalizer.
Specifically, within the time window of 250-650 ms after stimulus onset, ERPs were more positive for human vocaliza-
tions compared to dog vocalizations. Furthermore, a later time window of 800-900 ms demonstrated an ERP response
that also reflected the species of the vocalizer. These results highlight the existence of species-specific processing
8
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
of vocalizations in dogs and provide insights into the neural mechanisms underlying their perception of human and
non-human vocalizations.
Sleep EEG in Dogs
EEG recordings of sleeping dogs were examined in 14 of the 22 studies described in this review. In 2014, Kis
and colleagues pioneered the use of non-invasive EEG to record brain activity in sleeping dogs, providing a tool to
investigate fundamental questions about sleep architecture in those [56]. While most studies relied on human coding
and analysis, automated techniques using algorithms to find specific patterns called spindles [43, 45] were developed
and refined. Furthermore, machine learning models, including logistic regression (LogReg), gradient boosting trees
(GBTs), as well as convolutional neural networks (CNNs), were also deployed and validated to predict sleep stages in
dogs [31].
Another study investigated the impact of pre-sleep activity, timing, and location on sleep macro-structure, such as
the duration of sleep and the transitions between sleep stages [16]. The authors discovered that the intensity of pre-
sleep activity and the location and timing of sleep sessions had interactive effects on sleep macrostructure. Pre-sleep
intensive activity and night-time sleeping were associated with more time spent in both non-rapid eye movement
(NREM) and rapid eye movement (REM) sleep. Furthermore, they found that dogs sleeping in a location outside
their home were less likely to experience REM sleep. A later study by Reicher and colleagues [92] investigated the
well-known first-night adaptation effect seen in humans and found that it also manifests in dogs, albeit with marked
differences. The first-night adaptation effect refers to the recurring observation that the first recorded sleep session in
humans differs from all subsequent recordings, as it is marked by the necessity to adapt to the recording conditions.
In dogs, a significant difference was observed between the first and third recordings, with dogs spending more time
in sleep and having a shorter latency to drowsiness in session 3 than in session 1. Reicher and colleagues [93]
also investigated whether dogs exhibit functional hemispheric asymmetry, a phenomenon in which the right and left
hemispheres of an individual displays differential activity during a cognitive process, frequently observed during sleep
in aquatic mammals [72]. They found a complex asymmetry contingent on the recording session, sleep cycle, and type
of frequency, with some similarities but also many differences between canines and humans.
In addition to exploring fundamental and comparative questions, researchers have also investigated the impact of
biological variables such as age, sex, and weight on sleep activity. The impact of age on sleep macrostructure is
significant, showing correlations with the power of some frequency bands [91]. Specifically, past 8 months of age,
older dogs had higher powers of alpha, beta and gamma frequencies, and lower delta frequencies, compared to younger
dogs. Another approach has involved the measurement of spindles, which are phasic bursts of thalamo-cortical activity
that appear in the cortex as transient oscillations in the sigma range (typically defined in humans as 9-16 Hz) [21].
In 2017, Iotchev and colleagues [43] developed an algorithm to quantify sleep spindles in dogs and subsequent work
has discovered associations between the frequency, density, and amplitude of spindles with the age and sex of canine
participants. It is worth highlighting the observation that an increase in age was associated with a decrease in the
density and amplitude of slow spindles [44].
Another theme explored by researchers in dog EEG is the relationship between sleep activity and other cognitive
processes. For example, Kis and colleagues [55] observed a connection between sleep activity and learning rates,
specifically that increased beta and decreased delta activity during REM sleep were related to higher performance on a
novel learning task. Which could be related to the processed described on Iotchev’s work, on the potential significance
of spindle activity for learning and memory processes [43,45,46]. These studies suggest that learning gain (increases in
performance on cognitive tasks between sessions inter-spaced by sleep) is correlated with measures of spindle density.
While the authors acknowledged potential confounders with demographic variables, the studies suggest a possible
causal role of spindles in the consolidation of memory.
On a different association with sleep quality, Carreiro et al. [19] investigated the relation between sleep activity and
owner-rated hyperactivity and found that dogs rated as more hyperactive and impulsive demonstrated less total sleep
time, a reduced percentage of REM sleep, and lower spindle density compared to dogs rated as less hyperactive and
impulsive. Moreover, owner-rated hyperactivity and impulsivity were associated with increased wakefulness after
sleep onset and greater sleep fragmentation.
Finally, two studies explored the relation between sleep activity and features of human-canine interaction - specifically
cooperation and attachment. Kiss et al. [57] deployed an experimental paradigm testing the ’audience effect’ between
dogs and their human guardians, which relates to the difference in task performance based on the presence of visual
attention. Spectral sleep analysis revealed associations between REM and non-REM power activity and susceptibility
to the audience effect. In other words, the willingness of participant dogs to follow task instructions, irrespective
of whether their guardian was looking at them, was associated in a trait-like manner with alpha, beta, theta, delta
frequency bands power during sleep. Carreiro et al. [18] used an adapted form of the Strange Situation Task (SST) [2]
9
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
to index attachment levels of canine subjects and investigated whether derived attachment scores correlated with sleep
activity features. They found associations between the level of attachment and the duration of NREM sleep, as the
activity in certain frequency bands.
Discussion
Review of Findings
Temporal nature of dog cognitive processes in wakefulness
Non-invasive EEG uses electrodes placed on the scalp that pick up signals that are the end product of the integration
of postsynaptic potentials of hundreds of thousands of neurons traversing from the brain across tissue, bone, muscle,
skin and hair. This leads to a measure of brain activity with a low spatial resolution but high temporal resolution [22].
Effectively, this means that the inferences from awake canine EEG data are related to fine-grained features of temporal
activity. This is what we saw from the five studies that deployed hypothesis-driven ERP analysis frameworks to
investigating facial, vocalization and speech comprehension, with the significant time-windows displayed in Table 4.
Meaningful inferences can be derived from a wide range of times (between 30-950ms) post onset of a stimulus, as
previously highlighted by a study on the potential of machine learning (ML) models such as Support Vector Machines
(SVMs) in predicting stimulus categories based on activity in such time-windows [60].
Relationship between sleep and physiological traits in dogs
In contrast to wakefulness dog studies, the reviewed sleep studies took a different approach in their analysis framework.
While the high temporal resolution of EEG was occasionally leveraged in the spindle studies [44,45], the main focus
was on investigating associations between more general sleep stages and patterns and psychological traits. Some of
the significant correlations found are highlighted in Table 5.
Comparisons Between Dog and Human EEG
Non-invasive EEG with humans has a rich literature, and reviewed studies often used comparative framings to generate
hypotheses or provide explanatory models. An overview of some of the overlapping components are provided in Table
6. The N1 component, well-studied in human subjects [11], appears to also be present in canines. Kujala (2013)
[61] noted a deflection at 75ms in response to visual stimuli, earlier than typically observed in human studies. This
component, observed primarily in posterior channels (P3/P4), appears to differentiate between human and dog faces.
Boros (2021) [10] observed a N100 effect at electrode Fz as opposed to electrode Cz for word stimuli. A face-sensitive
component, representing a holistic representation of a face, appears at 170ms post stimulus for humans [108]. Kujala
(2020) [60] identified emotional expression-dependent effects between 127–170ms from stimulus onset, suggesting
face processing in dogs may be connected with the processing of the affective content of the stimulus. The word-
familiarity effect, visible in human infants between 200-400ms [75] post word-onset, may also be present in dogs. ERP
differences between WORDS and NONSENSE conditions appeared between 650-800ms post word-onset, towards the
end of the words [70].
The P300 is a component of the event-related potential (ERP) that is typically elicited in the process of decision
making [97]. It is most commonly evoked during "odd-ball" tasks, where the participant is asked to respond to
infrequent or unexpected stimuli amongst a set of standard stimuli. The P300 wave occurs roughly in the range 300-
600 milliseconds after the presentation of the stimulus and is understood to be correlated with the degree of attention
allocated to a stimulus [87]. Dogs exhibited a P300 response in the study by Balint (2022) [17], with a more positive
ERP response to human as compared to dog vocalizations between 250 and 650 ms, which may reflect a difference in
the motivational significance and allocated attention to human and dog vocalizations.
Lastly, the N400 component is a well-researched event-related potential (ERP) [62, 63] that is often associated with
semantic processing in language comprehension. It’s characterized by a negative peak occurring around 400ms post-
stimulus onset. Dogs exhibited a more positive deflection for words with high transitional probability than with low
transitional probability between 220-470 ms after word onset, with late ERP effects also observed between 590-790
ms, potentially representing a higher order recognition at the word level.
10
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Table 3: Overview of the 18 datasets created by non-invasive canine EEG studies.
Name
Papers
No. of dogs
Unique dogs?
Type
Conditions
Events OR sessions per dog
Duration of recording per dog (min)
Total events OR duration of data
Open source?
Faces1 [61, 99] 8 Yes Epoch, Continuous 2 240 80 976 No
Sleep_Exploratory1[56] 22 Yes Continuous 1 1 180 3960 No
Sleep_Exploratory2[56] 7 No Continuous 2 2 360 2520 No
Sleep_Learning1 [55] 15 Unsure Continuous 2 2 360 5400 No
Auto_Artefacts [66] 9 Yes Continuous 1 90 N/A N/A No
Sleep_Activity [16] 16 Unsure Continuous 2 2 540 8640 No
Sleep_Variation [44] 155 Unsure Continuous 1 1 180 27900 No
Sleep_Aging [45] 58 Unsure Continuous 2 2 N/A N/A No
Sleep_Individual [57] 27 Unsure Continuous 1 1 180 4860 No
Sleep_Adapt [92] 24 Unsure Continuous 3 2 180 4320 No
Faces2 [60] 8 Unsure Epoch 8 1200 N/A 8000 Partially A
Phoentics [70] 17 Unsure Epoch 3 240 N/A 4080 Partially B
Sleep_Development[91] 91 Partial Continuous 1 3 540 49140 No
Statistical_Learning[10] 19 Unsure Epoch 4 320 12 6080 Partially B
Vocalizations [17] 17 Unsure Epoch 4 192-384 N/A N/A Partially C
Pilot_Exploratory [54] 2 Yes Epoch 2 2 N/A N/A No
Sleep_Attachment [18] 43 Unsure Continuous 1 1 1-3h N/A Partially D
Sleep_Hyperactivity[19] 86 Unsure Continuous 1 1 1.5-3h N/A Partially D
Table 4: Overview of selected time-windows that had significant event-related potential activity.
Time-window (ms) Description Studies
30-40 Aggressive dog faces [60]
75-110 Difference between faces/objects and scrambled images [60,61]
220-470 Word segmentation (transitional probability) [10]
250-650 Species vocalization sensitivity [17]
360-400 Difference between dog and human faces [60]
590-790 Word frequency [10]
650-850 Known versus nonsense words [70]
800-900 Valence in vocalizations [17]
Table 5: Overview of selected associations between sleep macrostructure and patterns of sleep and physiological and
psychological states.
Sleep feature Associated physiological or psychology
state/trait
Studies
Duration in NREM and REM Pre-sleep intensive activity, time and location of
sleep, owner-rated hyperactivity [16,19]
Alpha frequency power Age, cooperation, attachment [18,57]
Beta and delta frequency power Age, learning rate, cooperation [55,57]
Gamma frequency power Age [92]
Frequency, density, and amplitude of spindles Age, sex, learning gain, owner-rated hyperactivity [43–46]
11
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Table 6: Overview of selected overlapping EEG components between humans and dogs.
Component Human Canine
N1 (100ms) First negative deflection post-
stimulus onset Occurs at 75ms in response to visual stimuli, can
differentiate between human and dog faces.
Face-sensitive
(170ms) Holistic representation of a face is
generated at 170 ms post stimulus
onset
Emotional expression-dependent effects at 127–170
ms post stimulus onset, suggesting face processing
in dogs may be connected to the processing of the
affective content
Word-familiarity
(200-500ms) Observable word-familiarity effect
in human infants ERP differences between WORDS and NON-
SENSE conditions appeared between 650 and 800
ms following word-onset
P300 and LPP
(300ms) Associated with attention and stim-
ulus evaluation Dogs show a differential ERP response depending
on the species of the caller between 250 and 650
ms, suggesting attentional differences to human and
dog vocalizations
N400 (400ms) Indicates speech segmentation of
candidate words Significant effect of transitional probability (220-
470 ms) and possible higher order recognition (590-
790 ms) after word onset
Avenues for future development
Unexplored questions
While reviewed studies explored the neural correlates of visual and auditory processing, olfactory processing has so
far been unexplored. Studies with humans indicate that relevant olfactory processing features can be extracted from
EEG data [27,41]. An understanding of the neural correlates of olfaction would be vital to a greater understanding of
a dog’s perception and cognition. Early studies that measured EEG of sedated dogs showed promise in decipherable
differences between evoked potentials between stimuli [40] and multiple studies using fMRI have observed meaningful
neural correlates from olfactory tasks in dogs [50,89, 90].
Similarly, studies exploring questions of cognitive control or executive function, defined by Gazzaniga and colleagues
as the "set of psychological processes that enable us to use our perceptions, knowledge, and goals to bias the selection
of action and thoughts from a multitude of possibilities" [30], have remained relatively underexplored. Cognitive
control encompasses a wide range of processes, including working memory, attentional control, cognitive flexibility,
and inhibitory control, and some precedent for into these domains is provided by prior canine cognitive research using
fMRI [24,64]. As with olfaction, questions of cognitive control not only inform us about the neural underpinnings of
canine cognition but also have significant practical value in the way humans communicate with and train dogs.
Intertwined with these questions on cognitive control are questions on the onset and processing of emotions. Both
Kujala et. al. [60] and Balint et. al. [17] investigated the effect of positive and negative stimuli, and found observable
differences in the respective evoked potentials. A broader and deeper investigation into the realm of canine emotions
would likely elucidate further such evidence, as work with fMRI on emotions such as jealousy shows [23,53]. Re-
latedly, it is worth investigating whether fundamental physiological states, such as hunger, stress, and the need of
elimination, are also represented by neural correlates that can be be consistently identified by non-invasive EEG. A
fundamental constraint on such questions is the poor spatial depth of EEG, as emotional and physiological processing
involve structures located deeper in the brain, to which EEG is not the adequate technique to map.
Another underlying challenge with some of these questions is the problem of ascertaining ground truth. This is
especially made apparent by the question of identifying emotions and to create ethical experimental conditions where
specific emotions can be consistently engendered in a non-human animal. However, there is potential in cross-modality
methods being used to triangulate such ground truth, such as the use of machine vision [9] and FACS (Facial Action
Coding System) [102] as well as physiological data on respiration and heart-rate from wearable sensors [13,28].
Finally, the majority of reviewed studies deployed group-level analysis frameworks. Such frameworks have dominated
canine cognition research, indeed Arden and colleagues find that from 1911 to 2016, only three studies took an explicit
individual-differences approach to exploring canine cognition [6]. Given the large inter-species variation amongst dog
breeds, especially in head and brain shape, a greater number of individual-differences analyses is well-warranted.
12
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Standard Setups for Dog EEG
A challenge downstream of canine variance is the difficulty in standardizing electrode montages. The convention used
in human research is the 10-20 system that allows for a consistent placement of electrodes across individuals [49].
Reviewed studies borrowed from the human 10-20 system in the placement of electrodes, although they differed in
their labels for similar electrode montages. An important question is if the 10-20 system is capable of transferring over
to canines in a useful way given the marked and asymmetrical difference in head shapes between dolichocephalic,
mesocephalic, and brachycephalic dogs [96]. For instance, the distance of a 10% posterior increment between a
greyhound and a bulldog would likely lead to stark differences in the brain regions that are recorded by the same
electrode position [37].
A related issue is the lack of a standard reference electrode. Three different electrode position labels were used by
the six different setups, and it is not clear that these labels refer to the same anatomical position. As mentioned prior,
the position of the reference electrode has a strong and irrevocable influence on the EEG recording, and it is unlikely
that meaningful comparisons can be made across subjects and labs, if the same anatomical reference is not used. Four
studies used the ear as a reference position, which runs contrary to the 2020 recommendations for reproducible human
EEG research issued by the Organization for Human Brain Mapping (OHBM) [85], which recommended against
physically linked earlobe or mastoid electrodes as they are not a neutral reference and can introduce distortions in
the data that make modelling intractable. This is likely the case for canine EEG as well, especially given the natural
tendency for dogs to move their ears to attune to stimuli, although it is important to note the influence of breed type,
as Beagles arguably display less ear movement given their floppy nature. The other four setups either use Cz or Pz
as a reference. It is worth noting, as the authors themselves do [17], that the choice of Pz could lead to an attenuated
recording from Cz relative to Fz, given the relative distance between the two. Future research could incorporate
other reference systems such as bipolar montages, where each channel represents the potential difference between two
adjacent electrodes, or the laplacian montage, where the reference is averaged signal of neighboring electrodes [1].
The development of a standard montage and validation of different reference montage systems, centered on the speci-
ficity of canine anatomy, and equipped to deal with the large variance amongst canine individuals would be greatly
beneficial to future progress in the field.
A Standard Data Structure for Dog EEG
As the volume and complexity of cognitive neuroscience methods has grown, several challenges emerged in the orga-
nization, dissemination, and analysis of data, leading to the creation of new standards and protocols for neuroscience
data structure and management. The brain imaging data structure (BIDS), first proposed in 2016 for magnetic reso-
nance imaging [33], is an exemplar of such a standard that embodies the FAIR principles of findability, accessibility,
interoperability, and reusability [104]. Recently, a BIDS standard for EEG data - EEG-BIDS - was proposed to address
the same concerns [86]. BIDS allows for the friction-less sharing of data within and between laboratories, as well as
enabling the automation of analysis scripts, that all serve to address the replicability of findings.
As noted prior, none of the reviewed studies had datasets that were fully accessible to external researchers. Three
studies that did attempt to make data open-source lacked the necessary information to replicate analysis, either omitting
raw data or lacking vital meta-data. While researchers should be continued to be encouraged to open-source their data,
in the vein of suggested guidelines by the Organization for Human Brain Mapping (OHBM), it is also vital that such
shared data is interoperable and accessible for researchers across labs and time. Adapting EEG-BIDS for canines
would be a crucial step to ensuring replicable and robust canine EEG research, and can readily be done, with the
primary challenge being the adoption of a standard anatomical coordination system to serve as the necessary metadata
file. As such a data structure would require adoption by researchers in the field and it should be a priority to ensure
consensus on its creation and use. Additional canine specific metadata, such as head size measurements, might also
be included in such a protocol to allow for flexibility and refinement in analysis. Upon the adoption of canine EEG-
BIDS, data can be made accessible on a neuroscience-specific repository such as OpenNeuro, which would allow for
large datasets. Furthermore, common analysis pipelines, such as developed by the BCI2000 open source system for
humans [94], could be readily adapted to canine data, opening up the world of canine brain-computer interfaces.
Improving Signal-to-Noise Ratio
The presence of a furry and muscular scalp makes improving the Signal-to-Noise Ratio (SNR) an important challenge
to overcome for non-invasive canine EEG experiments. One approach to improving SNR would be using electromyo-
graphy (EMG), alongside EEG, to quantify the contribution of scalp muscle activity. With enough measurements from
individuals and across dog breeds, it could be possible to regress out muscle activity from neural activity, and allow
robust recordings from further electrode positions.
13
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Alongside, impedance benchmarks for different breeds, as well as for different electrode types (e.g. wet vs dry),
would inform optimal electrode design to increase SNR. An example of such a study was performed by Luca and
colleagues for wired vs wireless sub-dermal electrodes with canines [67]. The use of custom canine phantoms to
measure impedance for different systems could also be productive, as seen for human EEG studies [35,58, 81]. It is
worth noting that the majority of reviewed studies used wired systems and only one study used a canine-specific cap
to hold electrodes in place. The use and development of canine-specific caps coupled with wireless modern systems
would greatly reduce the noise from electrode slippage and wire movement [79] and lead to an increase in SNR.
Moreover, such systems would allow experiments in freely moving dogs in naturalistic settings.
Along with custom wireless canine systems, the use of active electrodes, over the passive one used by all studies, have
the potential to greatly boost SNR. Active electrodes consist of electrodes with a mini-amplifier system loaded onto
the electrode itself, boosting signal quality at the source, and thus increasing SNR overall [34]. Another avenue could
be the creation of ground-truth standardization tests based on steady-state visual-evoked potentials (SSVEPs) [4,80].
A SSVEP is the brain’s evoked potential in response to the presentation of periodically flashing visual stimuli, and this
has been observed to form a stable EEG wave with a frequency that matches the presented stimulus. Thus, a SSVEP
framework can provide a grounding upon which the SNR of a particular system can be measured. Auditory steady-
state responses (ASSRs) are similar to SSVEPs but with specific frequencies embedded in audio. ASSR paradigms are
potentially easier to do with dogs than SSVEPs as dogs don’t have to be trained to stare at a flashing stimulus. Similar
to SSVEPs, ASSRs can provide a metric to ascertain the SNR of a system [77].
Finally, the detection of artifacts is crucial to ensuring high SNR. All awake experimental reviewed studies deployed
manual or algorithmic preprocessing pipelines to clean their recorded canine EEG data. However, as shown by Levitt
and colleagues [66], machine learning models are capable of eye-blink and muscular artifact identification and removal
for non-invasive canine EEG data, although it is important to point out the limited accuracy of 80%. Moreover, the
same models used for human EEG data performed well when trained on canine EEG data, suggesting the potential for
transferring models with weights from human EEG data to canine EEG data. While the use of ML artifact removal
pipelines continues to be debated in the field [47], and further larger studies should be conducted before making any
strong conclusions, it seems possible for ML preprocessing pipelines to be a valuable addition to the field, especially
as more than half the data from some awake dog studies had to be manually thrown out because of artifacts [70]. As
such pipelines could feasibly allow the identification and removal of artifacts without losing out on the data from the
entire trial, this could significantly increase the quantity of data available for analysis, boosting SNR, especially as the
number and quantity of EEG recordings increases.
A search for better models for canine EEG
We observed in the reviewed studies a shift in focus from explanatory models to prediction models, a distinction
articulated by Yarkoni and Westfall in their 2017 paper [107]. The distinction is raised with the pertinent criticism that
psychology often prioritizes explanatory models, which are prone to overfitting and rarely tested for out-of-sample
accuracy. They further argue that this emphasis on explanatory models has contributed significantly to the replication
crisis in psychology, as these models often fail to generalize to different data sets. In contrast, a predictive approach
aims to construct models that can effectively predict out-of-sample data by leveraging established machine learning
techniques such as leave one-out cross-validation [12,107]. This shift toward prediction models offers a promising
direction for improving the robustness and generalizability of canine EEG research.
A particular pitfall to be addressed, is the observed limitation in some reviewed studies, where multiple comparisons
went uncorrected or under-corrected. As the nature of EEG studies involves many degrees of freedom, including mul-
tiple channel locations, time windows, and subjects, spurious significance values are likely. Some studies attempted to
adjust for this using methods such as the Bonferroni correction, which involves adjusting the αvalue by the number of
hypotheses (m) by the following formulae — actual alpha = desired α/m. However, some studies omitted to consider
the true range of hypotheses being tested, e.g. not considering time-window choice as a relevant hypothesis. This lack
of adequate correction for multiple comparisons, coupled with the relatively small sample size of studies raises serious
questions of replicability.
One challenge with the use of predictive models is the need for large amounts of data to train models [26]. One
tractable approach could be to combine a predictive model framework with an individual level analysis. That is, large
amounts of EEG data can be collected from a few dogs for a specific task, and machine learning models can be trained
on data from these individual dogs. Comparing the performance of such models across dogs and tasks, as well as
correlation between model performance with other behavioral and cognitive traits should provide insight on questions
of canine cognition, whilst providing an alternative model to solve the problem of multiple comparisons.
14
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
Conclusion
The rise of non-invasive canine EEG can be traced to lying at the intersection of three trends — the increasing ma-
turity of cognitive neuroscience, the rejuvenation of canine science, and the increasing sophistication in portable and
accessible neuroimaging methods. The latter is primarily due to the increasing interest in brain-computer interfaces
(BCIs) [101]. We may be at the cusp of the emerging field of canine brain-computer interfaces, where wearable and
non-invasive systems could allow dogs to interact with objects, their environment, and humans through cognitive pro-
cesses alone. Such systems could be beneficial to researchers in the field of canine science, as it extends the field of
possibilities in experimental design, as well as potentially reducing the time needed for operant training. The devel-
opment of canine BCIs would be benefited by the framings of the field of Animal-Computer Interaction (ACI), which
works to design technologies for non-human users using a development model which incorporates iterative prototyp-
ing, animal welfare as a central value, and a direct involvement of animal experts at all stages of the development
process [100].
It would be vital to remember that neuroscience needs behavior [59], and that further work with canine EEG continues
to incorporate behavioral measures alongside neural data. One useful approach could be the lens of embodied and 4E
cognition, where cognition is seen to extend beyond the confines of the brain [78,82]. An embodied approach to canine
EEG would emphasize embodied data, such as from wearable heart and respiration sensors, as well as acknowledge
the influence of the environment, as well as human and con-specifics, on neural patterns.
In conclusion, the utility of non-invasive EEG encompasses the diverse and expansive roles that dogs have come to
occupy in our societies, providing a portable, accessible, and ethical method to derive quantitative data on canine
cognition. Non-invasive EEG can lead to insights on the shared neurological conditions as well as cognitive processes
in humans and canines, and provide a data-driven amplifier to the training and deployment of working dogs.
Author contributions statement
A.K. and A.Z. conceived the study, A.K. and D.L. conducted the study, A.K, D.L., A.Z. and L.F. analysed the results.
All authors wrote the main manuscript text and reviewed the manuscript.
References
[1] Jayant N. Acharya and Vinita J. Acharya. Overview of EEG montages and principles of localization. Journal
of Clinical Neurophysiology, 36(5):325, 2019.
[2] Mary D Salter Ainsworth. Object relations, dependency, and attachment: A theoretical review of the infant-
mother relationship. Child development, pages 969–1025, 1969.
[3] Thomas D Albright, Eric R Kandel, and Michael I Posner. Cognitive neuroscience. Current Opinion in Neuro-
biology, 10(5):612–624, October 2000.
[4] Brendan Z. Allison, Dennis J. McFarland, Gerwin Schalk, Shi Dong Zheng, Melody Moore Jackson, and
Jonathan R. Wolpaw. Towards an independent brain–computer interface using steady state visual evoked po-
tentials. Clinical Neurophysiology, 119(2):399–408, 2008.
[5] Attila Andics, Márta Gácsi, Tamás Faragó, Anna Kis, and Ádám Miklósi. Voice-Sensitive Regions in the Dog
and Human Brain Are Revealed by Comparative fMRI. Current Biology, 24(5):574–578, 2014.
[6] Rosalind Arden and Mark James Adams. A general intelligence factor in dogs. Intelligence, 55:79–85, 2016.
[7] Massimo Aria, Alessandra Alterisio, Anna Scandurra, Claudia Pinelli, and Biagio D’Aniello. The scholar’s
best friend: Research trends in dog cognitive and behavioral studies. Animal Cognition, 24(3):541–553, 2021.
[8] Gregory S Berns, Andrew M Brooks, and Mark Spivak. Functional mri in awake unrestrained dogs. PloS one,
7(5):e38027, 2012.
[9] Tali Boneh-Shitrit, Marcelo Feighelstein, Annika Bremhorst, Shir Amir, Tomer Distelfeld, Yaniv Dassa, Sharon
Yaroshetsky, Stefanie Riemer, Ilan Shimshoni, Daniel S. Mills, and Anna Zamansky. Explainable automated
recognition of emotional states from canine facial expressions: the case of positive anticipation and frustration.
Scientific Reports, 12(1):22611, 2022. Number: 1 Publisher: Nature Publishing Group.
[10] Marianna Boros, Lilla Magyari, Dávid Török, Anett Bozsik, Andrea Deme, and Attila Andics. Neural processes
underlying statistical learning for speech segmentation in dogs. Current Biology, 31(24):5512–5521.e5, 2021.
15
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
[11] Michael E Brandt, Ben H Jansen, and Joseph P Carbonari. Pre-stimulus spectral eeg patterns and the visual
evoked response. Electroencephalography and Clinical Neurophysiology/Evoked Potentials Section, 80(1):16–
20, 1991.
[12] Michael W Browne. Cross-validation methods. Journal of Mathematical Psychology, 44(1):108–132, 2000.
[13] Rita Brugarolas, Sherrie Yuschak, Darcy Adin, David L. Roberts, Barbara L. Sherman, and Alper Bozkurt. Si-
multaneous monitoring of canine heart rate and respiratory patterns during scent detection tasks. IEEE Sensors
Journal, 19(4):1454–1462, 2019. Conference Name: IEEE Sensors Journal.
[14] Caleb M Bryce, Michael S Davis, Matthew E Gompper, Aimee Hurt, Jeremy M Koster, Greger Larson, Elaine A
Ostrander, Monique AR Udell, Silvan Urfer, Aaron J Wirsing, et al. Biology’s best friend: bridging disciplinary
gaps to advance canine science, 2021.
[15] Nóra Bunford, Attila Andics, Anna Kis, Ádám Miklósi, and Márta Gácsi. Canis familiaris As a Model for
Non-Invasive Comparative Neuroscience. Trends in Neurosciences, 40(7):438–452, 2017.
[16] Nóra Bunford, Vivien Reicher, Anna Kis, Ákos Pogány, Ferenc Gombos, Róbert Bódizs, and Márta Gácsi.
Differences in pre-sleep activity and sleep location are associated with variability in daytime/nighttime sleep
electrophysiology in the domestic dog. Sci Rep, 8(1):7109, 2018.
[17] Anna Bálint, Huba Ele˝
od, Lilla Magyari, Anna Kis, and Márta Gácsi. Differences in dogs’ event-related
potentials in response to human and dog vocal stimuli; a non-invasive study. R. Soc. open sci., 9(4):211769,
2022.
[18] Cecília Carreiro, Vivien Reicher, Anna Kis, and Márta Gácsi. Attachment towards the owner is associated
with spontaneous sleep EEG parameters in family dogs. Animals, 12(7):895, 2022. Number: 7 Publisher:
Multidisciplinary Digital Publishing Institute.
[19] Cecília Carreiro, Vivien Reicher, Anna Kis, and Márta Gácsi. Owner-rated hyperactivity/impulsivity is associ-
ated with sleep efficiency in family dogs: A non-invasive EEG study. Sci Rep, 13(1):1291, 2023.
[20] Frances L. Chen, Madeline Zimmermann, Jessica P. Hekman, Kathryn A. Lord, Brittney Logan, Jane Russen-
berger, Eldin A. Leighton, and Elinor K. Karlsson. Advancing genetic selection and behavioral genomics of
working dogs through collaborative science. Frontiers in Veterinary Science, 8, 2021.
[21] Brittany C Clawson, Jaclyn Durkin, Sara J Aton, et al. Form and function of sleep spindles across the lifespan.
Neural plasticity, 2016, 2016.
[22] Michael X Cohen. Where does eeg come from and what does it mean? Trends in neurosciences, 40(4):208–218,
2017.
[23] Peter Cook, Ashley Prichard, Mark Spivak, and Gregory S. Berns. Jealousy in dogs? Evidence from brain
imaging. Animal Sentience, 3(22), 2018.
[24] Peter F. Cook, Mark Spivak, and Gregory Berns. Neurobehavioral evidence for individual differences in canine
cognitive control: An awake fMRI study. Anim Cogn, 19(5):867–878, 2016.
[25] Daniel D. Dilks, Peter Cook, Samuel K. Weiller, Helen P. Berns, Mark Spivak, and Gregory S. Berns. Awake
fMRI reveals a specialized region in dog temporal cortex for face processing. PeerJ, 3:e1115, 2015.
[26] Hauke Dose, Jakob S Møller, Helle K Iversen, and Sadasivan Puthusserypady. An end-to-end deep learning
approach to mi-eeg signal classification for bcis. Expert Systems with Applications, 114:532–542, 2018.
[27] Kiana Ezzatdoost, Hadi Hojjati, and Hamid Aghajan. Decoding olfactory stimuli in EEG data using nonlinear
features: A pilot study. Journal of Neuroscience Methods, 341:108780, 2020.
[28] Marc Foster, Jianxun Wang, Evan Williams, David Roberts, and Alper Bozkurt. ECG and respiration signal
reconstruction from an IMU at various orientations during rest or sleep for dog welfare monitoring. In Eight
International Conference on Animal-Computer Interaction, pages 1–9. ACM, 2021.
[29] K. J. Friston, A. P. Holmes, K. J. Worsley, J.-P. Poline, C. D. Frith, and R. S. J. Frackowiak. Statistical parametric
maps in functional imaging: A general linear approach. Human Brain Mapping, 2(4):189–210, 1994.
[30] Michael S. Gazzaniga, Richard B. Ivry, and G. R. Mangun. Cognitive Neuroscience: The Biology of the Mind.
W.W. Norton & Company, fifth edition edition, 2019.
[31] Anna Gergely, Orsolya Kiss, Vivien Reicher, Ivaylo Iotchev, Enik˝
o Kovács, Ferenc Gombos, András Benczúr,
Ágoston Galambos, József Topál, and Anna Kis. Reliability of family dogs’ sleep structure scoring based on
manual and automated sleep stage identification. Animals, 10(6):927, 2020. Number: 6 Publisher: Multidisci-
plinary Digital Publishing Institute.
16
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
[32] Gitanjali E. Gnanadesikan, Brian Hare, Noah Snyder-Mackler, and Evan L. MacLean. Estimating the heritabil-
ity of cognitive traits across dog breeds reveals highly heritable inhibitory control and communication factors.
Animal Cognition, 23(5):953–964, 2020.
[33] Krzysztof J. Gorgolewski, Tibor Auer, Vince D. Calhoun, R. Cameron Craddock, Samir Das, Eugene P. Duff,
Guillaume Flandin, Satrajit S. Ghosh, Tristan Glatard, Yaroslav O. Halchenko, Daniel A. Handwerker, Michael
Hanke, David Keator, Xiangrui Li, Zachary Michael, Camille Maumet, B. Nolan Nichols, Thomas E. Nichols,
John Pellman, Jean-Baptiste Poline, Ariel Rokem, Gunnar Schaefer, Vanessa Sochat, William Triplett, Jes-
sica A. Turner, Gaël Varoquaux, and Russell A. Poldrack. The brain imaging data structure, a format for
organizing and describing outputs of neuroimaging experiments. Scientific Data, 3(1):160044, 2016. Number:
1 Publisher: Nature Publishing Group.
[34] Erwin Habibzadeh Tonekabony Shad, Marta Molinas, and Trond Ytterdal. Impedance and Noise of Passive and
Active Dry EEG Electrodes: A Review. IEEE Sensors J., 20(24):14565–14577, 2020.
[35] W. David Hairston, Geoffrey A. Slipher, and Alfred B. Yu. Ballistic gelatin as a putative substrate for EEG
phantom devices, 2016.
[36] E. E. Hecht, I. Zapata, C. E. Alvarez, D. A. Gutman, T. M. Preuss, M. Kent, and J. A. Serpell. Neurodevelop-
mental scaling is a major driver of brain–behavior differences in temperament across dog breeds. Brain Struct
Funct, 226(8):2725–2739, 2021.
[37] Erin E. Hecht, Jeroen B. Smaers, William D. Dunn, Marc Kent, Todd M. Preuss, and David A. Gutman. Sig-
nificant Neuroanatomical Variation Among Domestic Dog Breeds. J. Neurosci., 39(39):7748–7758, 2019.
[38] Suzana Herculano-Houzel. The human brain in numbers: a linearly scaled-up primate brain. Frontiers in
Human Neuroscience, 3, 2009.
[39] Hermann Hinrichs, Michael Scholz, Anne Katrin Baum, Julia W. Y. Kam, Robert T. Knight, and Hans-Jochen
Heinze. Comparison between a wireless dry electrode EEG system with a conventional wired wet electrode
EEG system for clinical applications. Sci Rep, 10(1):5218, 2020.
[40] Y. Hirano, T. Oosawa, and K. Tonosaki. Electroencephalographic olfactometry (EEGO) analysis of odour
responses in dogs. Research in Veterinary Science, 69(3):263–265, 2000.
[41] Hui-Rang Hou, Qing-Hao Meng, and Biao Sun. A triangular hashing learning approach for olfactory EEG
signal recognition. Applied Soft Computing, 118:108471, 2022.
[42] Tiffani J. Howell, Russell Conduit, Samia Toukhsati, and Pauleen Bennett. Auditory stimulus discrimination
recorded in dogs, as indicated by mismatch negativity (MMN). Behavioural Processes, 89(1):8–13, 2012.
[43] Ivaylo Borislavov Iotchev, Anna Kis, Róbert Bódizs, Gilles van Luijtelaar, and Enik˝
o Kubinyi. EEG transients
in the sigma range during non-REM sleep predict learning in dogs. Scientific Reports, 7(1):12936, 2017.
Number: 1 Publisher: Nature Publishing Group.
[44] Ivaylo Borislavov Iotchev, Anna Kis, Borbála Turcsán, Daniel Rodrigo Tejeda Fernández de Lara, Vivien Re-
icher, and Enik˝
o Kubinyi. Age-related differences and sexual dimorphism in canine sleep spindles. Sci Rep,
9(1):10092, 2019.
[45] Ivaylo Borislavov Iotchev, Vivien Reicher, Enik˝
o Kovács, Tímea Kovács, Anna Kis, Márta Gácsi, and Enik ˝
o
Kubinyi. Averaging sleep spindle occurrence in dogs predicts learning performance better than single measures.
Scientific Reports, 10(1):22461, 2020. Number: 1 Publisher: Nature Publishing Group.
[46] Ivaylo Borislavov Iotchev, Dóra Szabó, Anna Kis, and Enik˝
o Kubinyi. Possible association between spindle
frequency and reversal-learning in aged family dogs. Sci Rep, 10:6505, 2020.
[47] Md Kafiul Islam, Amir Rastegarnia, and Zhi Yang. Methods for artifact detection and removal from scalp EEG:
A review. Neurophysiologie Clinique/Clinical Neurophysiology, 46(4):287–305, 2016.
[48] Débora Jardim-Messeder, Kelly Lambert, Stephen Noctor, Fernanda M. Pestana, Maria E. de Castro Leal,
Mads F. Bertelsen, Abdulaziz N. Alagaili, Osama B. Mohammad, Paul R. Manger, and Suzana Herculano-
Houzel. Dogs have the most neurons, though not the largest brain: Trade-off between body mass and number
of neurons in the cerebral cortex of large carnivoran species. Frontiers in Neuroanatomy, 11, 2017.
[49] Herbert H Jasper. Ten-twenty electrode system of the international federation. Electroencephalogr. Clin. Neu-
rophysiol., 10:371–375, 1958.
[50] Hao Jia, Oleg M. Pustovyy, Paul Waggoner, Ronald J. Beyers, John Schumacher, Chester Wildey, Jay Bar-
rett, Edward Morrison, Nouha Salibi, Thomas S. Denney, Vitaly J. Vodyanoy, and Gopikrishna Deshpande.
Functional MRI of the Olfactory System in Conscious Dogs. PLOS ONE, 9(1):e86362, 2014.
17
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
[51] Matt Kaeberlein, Kate E. Creevy, and Daniel E. L. Promislow. The dog aging project: Translational geroscience
in companion animals. Mammalian Genome, 27(7):279–288, 2016.
[52] Emily S. Kappenman and Steven J. Luck. The Effects of Electrode Impedance on Data Quality and Statistical
Significance in ERP Recordings. Psychophysiology, 47(5):888–904, 2010.
[53] Sabrina Karl, Ronald Sladky, Claus Lamm, and Ludwig Huber. Neural Responses of Pet Dogs Witnessing
Their Caregiver’s Positive Interactions with a Conspecific: An fMRI Study. Cerebral Cortex Communications,
2(3):tgab047, 2021.
[54] Mirosław Karpi´
nski, Sławomir Kotyra, Małgorzata Goleman, Justyna Wojta´
s, Aleksandra Garbiec, Wanda
Krupa, Damian Zieli´
nski, and Piotr Czy˙
zowski. EEG readings in dogs depending on their living condition
and training. Preliminary study. Journal of Veterinary Behavior, 50:53–59, 2022.
[55] Anna Kis, Sára Szakadát, Márta Gácsi, Enik˝
o Kovács, Péter Simor, Csenge Török, Ferenc Gombos, Róbert
Bódizs, and József Topál. The interrelated effect of sleep and learning in dogs (Canis familiaris); an EEG and
behavioural study. Sci Rep, 7(1):41873, 2017.
[56] Anna Kis, Sára Szakadát, Enik ˝
o Kovács, Márta Gácsi, Péter Simor, Ferenc Gombos, József Topál, Ádám Mik-
lósi, and Róbert Bódizs. Development of a non-invasive polysomnography technique for dogs (Canis familiaris).
Physiology & Behavior, 130:149–156, 2014.
[57] Orsolya Kiss, Anna Kis, Katalin Scheiling, and József Topál. Behavioral and neurophysiological correlates
of dogs’ individual sensitivities to being observed by their owners while performing a repetitive fetching task.
Frontiers in Psychology, 11, 2020.
[58] S. Kohli, S. Krachunov, and A. J. Casson. Towards closed-loop transcranial electrical stimulation: A comparison
of methods for real time tES-EEG artefact removal using a phantom head model. Brain Stimulation: Basic,
Translational, and Clinical Research in Neuromodulation, 10(2):467–468, 2017.
[59] John W. Krakauer, Asif A. Ghazanfar, Alex Gomez-Marin, Malcolm A. MacIver, and David Poeppel. Neuro-
science Needs Behavior: Correcting a Reductionist Bias. Neuron, 93(3):480–490, 2017.
[60] Miiamaaria V. Kujala, Jukka-Pekka Kauppi, Heini Törnqvist, Liisa Helle, Outi Vainio, Jan Kujala, and Lauri
Parkkonen. Time-resolved classification of dog brain signals reveals early processing of faces, species and
emotion. Sci Rep, 10(1):19846, 2020.
[61] Miiamaaria V. Kujala, Heini Törnqvist, Sanni Somppi, Laura Hänninen, Christina M. Krause, Outi Vainio, and
Jan Kujala. Reactivity of Dogs’ Brain Oscillations to Visual Stimuli Measured with Non-Invasive Electroen-
cephalography. PLoS ONE, 8(5):e61818, 2013.
[62] Marta Kutas and Kara D Federmeier. Thirty years and counting: finding meaning in the n400 component of the
event-related brain potential (erp). Annual review of psychology, 62:621–647, 2011.
[63] Marta Kutas and Steven A Hillyard. Reading senseless sentences: Brain potentials reflect semantic incongruity.
Science, 207(4427):203–205, 1980.
[64] Sreenath P. Kyathanahally, Hao Jia, Oleg M. Pustovyy, Paul Waggoner, Ronald Beyers, John Schumacher, Jay
Barrett, Edward E. Morrison, Nouha Salibi, Thomas S. Denney, Vitaly J. Vodyanoy, and Gopikrishna Desh-
pande. Anterior–posterior dissociation of the default mode network in dogs. Brain Struct Funct, 220(2):1063–
1076, 2015.
[65] Troy M. Lau, Joseph T. Gwin, and Daniel P. Ferris. How many electrodes are really needed for EEG-based
mobile brain imaging? Journal of Behavioral and Brain Science, 02(3):387–393, 2012.
[66] Joshua Levitt, Adam Nitenson, Suguru Koyama, Lonne Heijmans, James Curry, Jason T. Ross, Steven Kamer-
ling, and Carl Y. Saab. Automated detection of electroencephalography artifacts in human, rodent and canine
subjects using machine learning. Journal of Neuroscience Methods, 307:53–59, 2018.
[67] Julia Luca, Michal Hazenfratz, Gabrielle Monteith, Andrea Sanchez, Luis Gaitero, and Fiona James. Elec-
trode scalp impedance differences between electroencephalography machines in healthy dogs. Can J Vet Res,
85(4):309–311, 2021.
[68] Wolfgang Löscher. Dogs as a Natural Animal Model of Epilepsy. Front Vet Sci, 9:928009, 2022.
[69] Evan L. MacLean, Aubrey Fine, Harold Herzog, Eric Strauss, and Mia L. Cobb. The New Era of Canine
Science: Reshaping Our Relationships With Dogs. Frontiers in Veterinary Science, 8, 2021.
[70] L. Magyari, given-i=Zs family=Huszár, given=Zs., A. Turzó, and A. Andics. Event-related potentials reveal
limited readiness to access phonetic details during word processing in dogs. R. Soc. open sci., 7(12):200851,
2020.
18
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
[71] ManyDogs Project ManyDogs Project, Daniela Alberghina, Emily E. Bray, Daphna Buchsbaum, Sarah-
Elizabeth Byosiere, Julia Espinosa, Gitanjali E. Gnanadesikan, C.-N. Alexandrina Guran, Elizabeth Hare,
Daniel J. Horschler, Ludwig Huber, Valerie A. Kuhlmeier, Evan L. MacLean, Madeline H. Pelgrim, Bryan
Perez, Dana Ravid-Schurr, Liza Rothkoff, Courtney L. Sexton, Zachary A. Silver, and Jeffrey R. Stevens.
ManyDogs project: A big team science approach to investigating canine behavior and cognition. Comparative
Cognition & Behavior Reviews, 18:59–77, 2023.
[72] Gian Gastone Mascetti. Unihemispheric sleep and asymmetrical sleep: behavioral, neurophysiological, and
functional perspectives. Nature and Science of Sleep, pages 221–238, 2016.
[73] Sandra McCune and Daniel Promislow. Healthy, Active Aging for People and Dogs. Frontiers in Veterinary
Science, 8, 2021.
[74] Ádám Miklósi. Dog behaviour, evolution, and cognition. oUp Oxford, 2014.
[75] Debra L Mills, Chantel Prat, Renate Zangl, Christine L Stager, Helen J Neville, and Janet F Werker. Language
experience and the organization of brain activity to phonetically similar words: Erp evidence from 14-and
20-month-olds. Journal of cognitive neuroscience, 16(8):1452–1464, 2004.
[76] Wajid Mumtaz, Suleman Rasheed, and Alina Irfan. Review of challenges associated with the EEG artifact
removal methods. Biomedical Signal Processing and Control, 68:102741, 2021.
[77] T Nakamura, H Namba, and T Matsumoto. Classification of auditory steady-state responses to speech data. In
2013 6th International IEEE/EMBS Conference on Neural Engineering (NER), pages 1025–1028. IEEE, 2013.
[78] Albert Newen, Leon De Bruin, and Shaun Gallagher, editors. 4E Cognition: Historical Roots, Key Concepts,
and Central Issues, pages 2–16. Oxford University Press, 2018.
[79] Guiomar Niso, Elena Romero, Jeremy T. Moreau, Alvaro Araujo, and Laurens R. Krol. Wireless EEG: A survey
of systems and studies. NeuroImage, 269:119774, 2023.
[80] Anthony M. Norcia, L. Gregory Appelbaum, Justin M. Ales, Benoit R. Cottereau, and Bruno Rossion. The
steady-state visual evoked potential in vision research: A review. Journal of Vision, 15(6):4, 2015.
[81] Amani Yousef Owda and Alexander J. Casson. Electrical properties, accuracy, and multi-day performance of
gelatine phantoms for electrophysiology, 2020. Pages: 2020.05.30.125070 Section: New Results.
[82] Francisco J. Parada and Alejandra Rossi. Perfect timing: Mobile brain/body imaging scaffolds the 4e-
cognition research program. European Journal of Neuroscience, 54(12):8081–8091, 2021. _eprint:
https://onlinelibrary.wiley.com/doi/pdf/10.1111/ejn.14783.
[83] Josef Parvizi and Sabine Kastner. Promises and limitations of human intracranial electroencephalography.
Nature Neuroscience, 21(4):474–483, 2018.
[84] Fernando C Pellegrino and Roberto E. P Sica. Canine electroencephalographic recording technique: Findings
in normal and epileptic dogs. Clinical Neurophysiology, 115(2):477–487, 2004.
[85] Cyril Pernet, Marta I. Garrido, Alexandre Gramfort, Natasha Maurits, Christoph M. Michel, Elizabeth Pang,
Riitta Salmelin, Jan Mathijs Schoffelen, Pedro A. Valdes-Sosa, and Aina Puce. Issues and recommendations
from the OHBM COBIDAS MEEG committee for reproducible EEG and MEG research. Nature Neuroscience,
23(12):1473–1483, 2020.
[86] Cyril R. Pernet, Stefan Appelhoff, Krzysztof J. Gorgolewski, Guillaume Flandin, Christophe Phillips, Arnaud
Delorme, and Robert Oostenveld. EEG-BIDS, an extension to the brain imaging data structure for electroen-
cephalography. Sci Data, 6(1):103, 2019.
[87] John Polich. Updating p300: an integrative theory of p3a and p3b. Clinical neurophysiology, 118(10):2128–
2148, 2007.
[88] Ashley Prichard, Raveena Chhibber, Kate Athanassiades, Mark Spivak, and Gregory S. Berns. Fast neural
learning in dogs: A multimodal sensory fMRI study. Sci Rep, 8(1):14614, 2018.
[89] Ashley Prichard, Raveena Chhibber, Jon King, Kate Athanassiades, Mark Spivak, and Gregory S Berns. De-
coding Odor Mixtures in the Dog Brain: An Awake fMRI Study. Chemical Senses, 45(9):833–844, 2020.
[90] Bhavitha Ramaihgari, Oleg M. Pustovyy, Paul Waggoner, Ronald J. Beyers, Chester Wildey, Edward Morrison,
Nouha Salibi, Jeffrey S. Katz, Thomas S. Denney, Vitaly J. Vodyanoy, and Gopikrishna Deshpande. Zinc
Nanoparticles Enhance Brain Connectivity in the Canine Olfactory Network: Evidence From an fMRI Study in
Unrestrained Awake Dogs. Frontiers in Veterinary Science, 5, 2018.
[91] Vivien Reicher, Nóra Bunford, Anna Kis, Cecília Carreiro, Barbara Csibra, Lorraine Kratz, and Márta Gácsi.
Developmental features of sleep electrophysiology in family dogs. Scientific Reports, 11(1):22760, 2021. Num-
ber: 1 Publisher: Nature Publishing Group.
19
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
Non-invasive canine electroencephalography (EEG): a systematic review PREPRINT
[92] Vivien Reicher, Anna Kis, Péter Simor, Róbert Bódizs, Ferenc Gombos, and Márta Gácsi. Repeated afternoon
sleep recordings indicate first-night-effect-like adaptation process in family dogs. Journal of Sleep Research,
29(6):e12998, 2020. _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/jsr.12998.
[93] Vivien Reicher, Anna Kis, Péter Simor, Róbert Bódizs, and Márta Gácsi. Interhemispheric asymmetry during
NREM sleep in the dog. Sci Rep, 11(1):18817, 2021.
[94] G. Schalk, D.J. McFarland, T. Hinterberger, N. Birbaumer, and J.R. Wolpaw. BCI2000: a general-purpose
brain-computer interface (BCI) system. IEEE Transactions on Biomedical Engineering, 51(6):1034–1043,
2004. Conference Name: IEEE Transactions on Biomedical Engineering.
[95] Trine Schütt, Jan T. Pedersen, and Mette Berendt. Chapter 15 - The Domestic Dog as a Model for Human Brain
Aging and Alzheimer’s Disease. In Jeffrey L. Ram and P. Michael Conn, editors, Conn’s Handbook of Models
for Human Aging (Second Edition), pages 177–194. Academic Press, 2018.
[96] Andy Shores and Brigitte A Brisson. Current techniques in canine and feline neurosurgery. John Wiley &
Sons, 2017.
[97] Samuel Sutton, Margery Braren, Joseph Zubin, and ER John. Evoked-potential correlates of stimulus uncer-
tainty. Science, 150(3700):1187–1188, 1965.
[98] Mario Tudor, Lorainne Tudor, and Katarina Ivana Tudor. Hans Berger (1873-1941)–the history of electroen-
cephalography. Acta medica Croatica: casopis Hravatske akademije medicinskih znanosti, 59(4):307–313,
2005.
[99] Heini Törnqvist, Miiamaaria V. Kujala, Sanni Somppi, Laura Hänninen, Matti Pastell, Christina M. Krause, Jan
Kujala, and Outi Vainio. Visual event-related potentials of dogs: A non-invasive electroencephalography study.
Anim Cogn, 16(6):973–982, 2013.
[100] Dirk van der Linden and Anna Zamansky. Agile with animals: Towards a development method. In 2017 IEEE
25th international requirements engineering conference workshops (REW), pages 423–426. IEEE, 2017.
[101] Marcel van Gerven, , Jason Farquhar, Rebecca Schaefer, Rutger Vlek, Jeroen Geuze, Anton Nijholt, Nick
Ramsey, Pim Haselager, Louis Vuurpijl, Stan Gielen, and Desain. The brain–computer interface cycle. Journal
of Neural Engineering, 6(4):041001, 2009.
[102] B Waller, CC Caeiro, K Peirce, A Burrows, J Kaminski, et al. DogFACS: the dog facial action coding system,
2013.
[103] Marco Weiergräber, Anna Papazoglou, Karl Broich, and Ralf Müller. Sampling rate, signal bandwidth and
related pitfalls in EEG analysis. Journal of Neuroscience Methods, 268:53–55, 2016.
[104] Mark D. Wilkinson, Michel Dumontier, IJsbrand Jan Aalbersberg, Gabrielle Appleton, Myles Axton, Arie
Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E. Bourne, Jildau Bouwman,
Anthony J. Brookes, Tim Clark, Mercè Crosas, Ingrid Dillo, Olivier Dumon, Scott Edmunds, Chris T. Evelo,
Richard Finkers, Alejandra Gonzalez-Beltran, Alasdair J. G. Gray, Paul Groth, Carole Goble, Jeffrey S. Grethe,
Jaap Heringa, Peter A. C. ’t Hoen, Rob Hooft, Tobias Kuhn, Ruben Kok, Joost Kok, Scott J. Lusher, Maryann E.
Martone, Albert Mons, Abel L. Packer, Bengt Persson, Philippe Rocca-Serra, Marco Roos, Rene van Schaik,
Susanna-Assunta Sansone, Erik Schultes, Thierry Sengstag, Ted Slater, George Strawn, Morris A. Swertz, Mark
Thompson, Johan van der Lei, Erik van Mulligen, Jan Velterop, Andra Waagmeester, Peter Wittenburg, Kather-
ine Wolstencroft, Jun Zhao, and Barend Mons. The FAIR guiding principles for scientific data management
and stewardship. Scientific Data, 3(1):160018, 2016. Number: 1 Publisher: Nature Publishing Group.
[105] K. J. Worsley and K. J. Friston. Analysis of fMRI Time-Series Revisited—Again. NeuroImage, 2(3):173–181,
1995.
[106] Jiawei Xu, Srinjoy Mitra, Chris Van Hoof, Refet Firat Yazicioglu, and Kofi A. A. Makinwa. Active Electrodes
for Wearable EEG Acquisition: Review and Electronics Design Methodology. IEEE Rev Biomed Eng, 10:187–
198, 2017.
[107] Tal Yarkoni and Jacob Westfall. Choosing Prediction Over Explanation in Psychology: Lessons From Machine
Learning. Perspectives on Psychological Science, 12(6):1100–1122, 2017.
[108] Galit Yovel. Neural and cognitive face-selective markers: An integrative review. Neuropsychologia, 83:5–13,
2016.
20
.CC-BY 4.0 International licenseavailable under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprintthis version posted August 16, 2023. ; https://doi.org/10.1101/2023.08.14.552730doi: bioRxiv preprint
