1. Introduction
There is an increasing interest in “objectivization” of behavior assessment methods. As Karen Overall noted: “A review of behavioral data over the past decade supports a serious shift to crisper definitions of terms and quantifiable assessment of behaviors. Veterinary behavior and veterinary behavioral medicine are coming later to this approach, but the change is welcome” [
1].
One particularly relevant challenge in this context is the problem of assessment of dog ADHD-like (attention deficit hyperactivity disorder) behaviors. They are often expressed in the form of inattention, impulsivity, and aggressivity, greatly compromising life quality of the dog and its owner [
2,
3,
4]. In addition, recently its links to human ADHD were explored; in particular, a pilot study by Puruunen et al. [
5] identified associations between canine ADHD-like behaviors and metabolites involved in lipid and tryptophan metabolisms, which share similarity with earlier findings in human ADHD models.
Several assessment instruments relevant to ADHD-like behavior and its expressions such as impulsivity were considered in the literature. These include C-BARQ [
6], the dog ADHD rating scale [
3], Dog Impulsivity Assessment Scale (DIAS) [
7], and other questionnaires and scales [
2,
4]. However, all of these scales are owner administered and thus may be affected by non-objectivity and inaccuracy (e.g., there is evidence that owners may misinterpret the behavior of their pets [
8]).
Computational behavior analysis may offer new tools for more objective assessment of behavior in clinical settings. It is an emerging field encapsulating the use of modern methods from computer science and engineering to quantitatively measure animal behavior [
9]. In this paper we use computational analysis in the context of clinical assessment of ADHD-like behavior as part of an ongoing study for developing a decision support system for behavioral clinicians (some first results already appeared in [
10,
11,
12,
13]).
Our hypothesis in this paper is that dogs with ADHD-like behavior may exhibit distinguishable movement patterns during consultation. To investigate the hypothesis, we compare a set of movement parameters (e.g., average speed, number of turns, and other trajectory characteristics) automatically extracted from video footage of behavioral consultations of 12 dogs treated due to ADHD-like behavior (we henceforth refer to this group as H-group) and of a control group (C-group) of 12 dogs with no known behavioral problems.
2. Materials and Methods
2.1. Participants
Twenty-four dogs were selected for the study. The C-group (N = 12, 5 males, 7 females) had a median age of 2.75 years (±2.15, max 7 years, min 8 months), median weight of 22.00 kg (±6.70, max 36, min 13), while the H-group (N = 12, 7 males, 5 females) had a median age of 1.50 years (±2.06, max 7 years, min 6 months), median weight of 20.58 kg (±8.16, max 34, min 12) (see
Table 1 for details).
To rule out confounding effects of demographic factors on the experimental variables we ensured that the C-group was matched to H-group in terms of demographic factors. To verify this, we checked that there was no statistically significant difference between the two groups for any demographic factor: age did not differ significantly between the two groups (Mann–Whitney U test, U = 62, n1 = n2 = 24, p = 0.58 (two-tailed)), nor did weight (Mann–Whitney U test, U = 62, n1 = n2 = 24, p = 58 (two-tailed)). Categorical factors of sex (Fisher’s exact test, p = 0.56 (two tailed), φ = 0.31) and neutered state (Fisher’s exact test, p = 0.32 (two tailed), φ = −0.40) also did not differ significantly between the groups.
All subjects in the H-group were clinical patients of two behavioral veterinary clinics “Veterinar Toran Tel-Aviv” and “Veterinar Toran Petach Tikva” in Israel and were diagnosed with ADHD-like behavior. They were recorded at the time of their first visit to the clinic of the first author, a practicing behavioral veterinarian. Thus, the inclusion criteria for the H-group were the following:
This was the subjects’ first visit to the behavioral expert veterinarian.
During that visit the expert diagnosed the subject with ADHD-like behavior.
During that visit the expert prescribed a medical treatment for the above condition.
All subjects in the control group were dogs with no reported behavioral problems or issues. They were recruited in the two clinics “Veterinar Toran Tel-Aviv” and “Veterinar Toran Petach Tikva” in Israel during standard checkup and vaccination, or through the authors’ professional connections. The behavioral veterinarian ruled out any behavioral disorder during a consultation.
For all dogs from both groups it was their first visit to the consultation room. The owner-dog dyad entered the room and the dog was immediately released, at which point the recording was started.
2.2. Location
Dogs were observed in the veterinary behavioral clinics at the Petah Tikva and Tel Aviv Animal Hospitals. The testing space captured by a camera fixed on the ceiling was 100 × 65 cm and 300 × 160 cm respectively. (To rule out confounding effects from the difference in floor size in the different clinics, we verified whether there was any significant difference in any measured variable between recording from the Tel Aviv (N = 18) and Petah Tikva (N = 6) clinic. A two-tailed Mann–Whitney U test found no significant difference for any of the variables (
p > 0.05).) (see
Figure 1). The dog owners sat at a fixed place outside of the captured frame.
2.3. Analysis Software
We used a self-developed software tool, K9-Blyzer (first presented in Amir et al. [
10] and later extended in [
11,
12,
13]) for analyzing the video footage recorded at the clinics and producing the data on the movement variables discussed below. The system takes as input raw video of a dog freely moving in a room and produces as output the dog’s location (x,y) in each frame. Due to different distances of the camera from floor in the two clinics, calibration specific to clinic was applied to convert distances in pixels to centimeters. The system is based on a convolutional neural network which was trained on 6000 annotated frames of dogs in the two clinics. For frames where the dog is clearly visible, the system reaches 98% detection accuracy.
Figure 2 shows some example frames where the tool tracks the dog.
2.4. Processed Videos
For each participant, the first three minutes of the consultation were taken for the computational analysis. Videos where the dog was not clearly identified in more than 30% of the frames were excluded from the analysis (the number of excluded dogs was 5).
2.5. Analyzed Parameters
In addition to basic movement related parameters: distance and average speed (that were previously used in the context of a vet consultation room [
11]), we used five metrics known as animal movement indices in the literature: Alamedia et al. [
14] mention intensity of use, straightness, sinuosity, mean squared displacement, and fractal dimension.
Table 2 below provides explanations for each metric. We further added two more metrics intuitively related to “erratic” movement of an impulsive dog: number of turns the dog made, divided into four groups according to angle sharpness and number of approximation points on the curve of the dog’s trajectory, obtained by applying a variant of Ramer–Douglas–Pecker curve approximation algorithm [
15].
3. Results
Analysis of the difference in measured variables between H-group and C-group using Mann–Whitney U test (given small sample sizes and non-normally distributed range of variables) revealed several statistically significant differences, as summarized in
Table 3. In order to account for confounding effects of dog size on movement (e.g., larger dog breeds may cover more absolute distance than smaller dog breeds by nature of their size) we used an approximated normalization for the relevant parameters affected here (distance, average_speed) by dividing them by weight. We calculated the statistical power of each test to account for low sample size, disregarding any identified distinctions with a statistical power < 0.80. In particular, distance (normalized) and average_speed (normalized) revealed a significant difference between groups (
p < 0.01, resp. Cohen’s d = 1.28, 1.45), as did most of the turn variables (30_60, 60_90, and 90_120;
p < 0.01, Cohen’s d resp. 1.42, 1.64 and 1.28), number of points (
p < 0.01, Cohen’s d 1.43), straightness (
p < 0.05, Cohen’s d −1.24), and fractal dimension (
p < 0.01, Cohen’s d 1.96).
4. Discussion
The parameters found significant by our analysis can be related to three dimensions for further exploration in the context of quantification and objectivization of assessment of ADHD-like behavior: (i) speed (as induced by average_speed parameter), (ii) coverage of room space (as induced by distance and FD), and (iii) reorientation in space (as induced by num_of_points, straightness, and the turn parameters).
It seems that the parameter of intensity of use is also strongly relevant to the second dimension. Indeed, while they were not found significant using the widely accepted minimal threshold of 0.8 for statistical power, they were close to it, and further experimentation with additional samples may highlight a different picture.
The identified dimensions of distance and coverage of space support the suggested links of ADHD-like behavior to increased motor activity [
4] and lack of stop signal in the behavioral sequence and low threshold of sensorial homeostasis [
2]. High frequency of re-orientation may also be explained by the link to inattentiveness [
4] and strong reaction to the lightest stimuli [
2].
These dimensions can form the basis for further developing computer-based instruments for quantified assessment of ADHD-like behavior. Eventually their envisioned use is for decision support system for behavioral clinicians in diagnosis and treatment of behavioral problems and disorders. It should be noted, however, that the current study considered only the differences between dogs with pure ADHD-like behavior to normal dogs, so no conclusions about the specificity of the system can be made at this stage, hence dogs with other behavioral problems, such as anxiety, may also exhibit similar patterns. This remains to be tested and is one immediate direction for future research.
Moreover, the above-mentioned dimensions are not expected to replace owner-reported information but complement it in new objective ways. Another future research direction we are already involved in is integrating a web-based decision support system into the behavioral vet’s workflow, which analyzes video footage in real-time, providing the different parameters values and visualizing them. An additional research direction will be further validation of the video-tracking system using the various owner-reported questionnaires mentioned above, which continue to provide the common practice for an assessment of canine ADHD-like behavior. Finally, further data collection for a more in-depth comparison with respect to groups of dogs diagnosed with other behavioral disorders (such as anxiety or depression) is another direction we plan to explore.
Author Contributions
Conceptualization, A.Z., S.B.-E. and D.v.d.L.; methodology, A.Z., S.M. and S.B.-E.; software, S.R. and A.S.; validation, S.R. and A.F.; formal analysis, A.S.; investigation, A.Z., D.v.d.L. and S.M.; resources, S.B.-E. and S.M.; data curation, S.B.-E. and S.M.; writing—original draft preparation, A.Z. and A.F.; writing—review and editing, D.K.; visualization, A.S. and A.F.; supervision, A.Z. and D.K.; project administration, A.Z.
Funding
This research received no external funding.
Acknowledgments
Anna Zamansky was supported by the NVIDIA GPU grant program.
Conflicts of Interest
The authors declare no conflict of interest.
References
- Overall, K.L. The ethogram project. J. Vet. Behav. Clin. Appl. Res. 2014, 9, 1–5. [Google Scholar] [CrossRef]
- Masson, S.; Gaultier, E. Retrospecive study on hypersensitivity-hyperactivity syndrome in dogs: Long-term outcome of high dose fluoxetine treatment and proposal of a clinical score. Dog Behav. 2018, 4, 15–35. [Google Scholar]
- Vas, J.; Topál, J.; Péch, E.; Miklósi, A. Measuring attention deficit and activity in dogs: A new application and validation of a human ADHD questionnaire. Appl. Anim. Behav. Sci. 2007, 103, 105–117. [Google Scholar] [CrossRef]
- Hoppe, N.; Bininda-Emonds, O.R.P.; Gansloßer, U. Correlates of attention deficit hyperactivity disorder (ADHD)-like behavior in domestic dogs: First results from a questionnaire-based study. Vet. Med. Open J. 2017, 2, 95–131. [Google Scholar] [CrossRef]
- Puurunen, J.; Sulkama, S.; Tiira, K.; Araujo, C.; Lehtonen, M.; Hanhineva, K.; Lohi, H. A non-targeted metabolite profiling pilot study suggests that tryptophan and lipid metabolisms are linked with ADHD-like behaviours in dogs. Behav. Brain Funct. 2016, 12, 27. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Hsu, Y.; Serpell, J.A. Development and validation of a questionnaire for measuring behavior and temperament traits in pet dogs. J. Am. Vet. Med Assoc. 2003, 223, 1293–1300. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Wright, H.F.; Mills, D.S.; Pollux, P.M. Development and validation of a psychometric tool for assessing impulsivity in the domestic dog (Canis familiaris). Int. J. Comp. Psychol. 2011, 24, 210–225. [Google Scholar]
- Lit, L.; Schweitzer, J.B.; Iosif, A.M.; Oberbauer, A.M. Owner reports of attention, activity, and impulsivity in dogs: A replication study. Behav. Brain Funct. 2010, 6, 1. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Egnor, S.R.; Branson, K. Computational analysis of behavior. Annu. Rev. Neurosci. 2016, 39, 217–236. [Google Scholar] [CrossRef] [PubMed]
- Amir, S.; Zamansky, A.; van der Linden, D. K9-Blyzer: Towards Video-Based Automatic Analysis of Canine Behavior. In Proceedings of the Fourth International Conference on Animal-Computer Interaction, Milton Keynes, UK, 21–23 November 2017; p. 9. [Google Scholar]
- Zamansky, S.; Bleuer-Elsner, S.; Masson, S.; Amir, O.; Magen, D.v.d.L. Effects of anxiety on canine movement in dog-robot interactions. Anim. Behav. Cogn. 2018, 5, 280–287. [Google Scholar] [CrossRef]
- Zamansky, A.; Sinitca, A.M.; Kaplun, D.I.; Plazner, M.; Schork, I.G.; Young, R.J.; de Azevedo, C.S. An Application of Convolutional Neural Networks for Analyzing Dogs’ Sleep Patterns. In Proceedings of the ICANN: International Conference on Artificial Neural Networks, Munich, Germany, 17–19 September 2019. [Google Scholar]
- Kaplun, D.A.; Sinitca, A.; Zamansky, M.; Plazner, S.; Bleuer-Elsner, D.v.d.L. Animal Health Informatics: Towards a Generic Framework for Automatic Behavior Analysis. In Proceedings of the 12th International Conference on Health Informatics, Prague, Czech Republic, 22–24 February 2019. [Google Scholar]
- Almeida, P.J.; Vieira, M.V.; Kajin, M.; Forero-Medina, G.; Cerqueira, R. Indices of movement behaviour: Conceptual background, effects of scale and location errors. Zoologia 2010, 27. [Google Scholar] [CrossRef]
- Saalfeld, A. Topologically consistent line simplification with the Douglas-Peucker algorithm. Cartogr. Geogr. Inf. Sci. 1999, 26, 7–18. [Google Scholar] [CrossRef]
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).