Movement Entropy in a Gardening Design Task as a Diagnostic

Marker for Mental Disorders: Results of a Pilot Study

Sebastian Unger

, Sebastian Appelbaum

, Thomas Ostermann

and Christina Niedermann

2,3

Didactics and Educational Research in Health Science, Witten/Herdecke University, Germany

Methods and Statistics in Psychology, Faculty of Health, Witten/Herdecke University, Germany

Fine Arts, University of Applied Sciences and Arts Ottersberg, Germany

Keywords: Mental Health, Garden Design, Movement Analysis, Heatmap Analysis, Entropy, Homography.

Abstract: Movement, actions, and intentions are important psychological skills in human behavior. Studies have shown

correlations between movement activity and a variety of mental disorders. In this context, planning and

designing of gardens and outdoor spaces as an intentional activity might play an important role as a marker

for mental health. Thus, in this study, 16 subjects (8 female) aged between 19 and 60 were asked to do a

gardening task in an experimentally constructed environment while their movement activity was recorded

with a camera from a fixed viewpoint. Movement heatmaps and entropy then was calculated and correlated

with mental state measured via the Global Severity Index (GSI) of the Brief Symptom Inventory (BSI-18)

questionnaire. After finding an optimal grid size of the heatmaps, we were able to find a moderate negative

correlation of r = -0.463 between these quantities in an overall of both genders, explaining 21.4 % of variance.

After considering the gender of the test group, a noticeable gender effect could be revealed. We found a

significant interaction effect of entropy with gender meaning that a lower movement entropy in a gardening

task correlates with a higher mental distress for men, but lower for women. Multivariate regression found that

this model explained 77.44 % of variance (R = 0.88). Despite of these promising results, further investigations

in this area should overcome some limitations in this pilot study in the field of position tracking and movement

feature extraction.

1 INTRODUCTION

Movement, actions, and intentions are important

psychological skills in human behavior. A recent

systematic review of the relationships between motor

proficiency physical abilities and academic

performance in mathematics and reading tasks

showed that motor proficiency was able to predict the

academic performances of children and adolescents,

in particular in the early years of school (Macdonald

et al., 2018). Other reviews from the field of clinical

psychology showed correlations between

pathological movement features and a variety of

mental disorders. A recent review of Rohani et al.

(2018) found significant correlations between

behavioral features and depressive mood symptoms

in adults, while the review of Zhu et al. (2019) found

that physical activity of children and adolescents is

associated with a reduced risk of experiencing

anxiety. Other studies in patients suffering from

dementia reported in (Collier et al., 2018) suggest that

human movement analysis can be used as a diagnostic

marker for early dementia. In this respect, the

intentional meaning of movement seems to play an

important role. According to (Clark et al., 2015; p.1)

mindful movement “may improve the functional

quality of rehearsed procedures, cultivating a

transferrable skill of attention”.

Hence and in agreement with the findings of

(O’Brien et al., 2017), continuous and everyday

monitoring of activity and motion could be a

promising real-world biomarker for early detection of

mental disorders.

With respect to the definition of “real world”,

studies in elderly populations found that the planning

and designing of gardens and outdoor spaces was

attributed as an intentional activity with importance

for active daily living (Kim and Ohara, 2010; Milke

et al., 2009; Wang and Glicksman, 2013; Yen et al.,

2014). Moreover, gardening seems to have a low

threshold with respect to participation and a high

degree of prior experience or interest for a broad

range of participants (Bleasdale et al., 2011).

Unger, S., Appelbaum, S., Ostermann, T. and Niedermann, C.

Movement Entropy in a Gardening Design Task as a Diagnostic Marker for Mental Disorders: Results of a Pilot Study.

DOI: 10.5220/0010227203370343

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 5: HEALTHINF, pages 337-343

ISBN: 978-989-758-490-9

337

Our pilot study thus aimed at investigating, how

gardening activity could be assessed and used as a

diagnostic marker in an experimental setting. We

therefore used human movement entropy. As

described in (O’Brien et al., 2017), entropy is

interpreted as a measure that describes how “vital” a

movement is. Low entropy thus represents stationary

and slow movement, while a high entropy is

associated with expansive and faster movement

behavior.

According to a study of Rohani et al. (2018)

movement entropy was found to have a negative

correlation with mental burden. Thus, a high entropy

would indicate a better mood of the participants,

which is the underlying hypothesis of this pilot study.

2 METHODS

2.1 Setting and Participants

The study was announced in the Witten/Herdecke

University and University staff members, their

relatives and friends and students were invited to

participate in the study on a voluntary basis.

Prior to the gardening task, all participants had to

sign an informed consent form and were asked to

complete a questionnaire, including demographic

items such as age, gender, self reported gardening

experience, creativity as well as the Brief Symptom

Inventory (BSI-18) to assess the mental state

expressed by the global severity index (GSI) of the

BSI-18 (Spitzer et al., 2011).

The main task for the participants consisted in

creating a landscape in a 2.5 x 2.5 m squared sandpit

area (Figure 1) in a maximum of 30 min (minutes).

Material to be used included plants, flowers,

branches and stones. They were allowed to use the full

space of the gardening area and were not limited in

their spatial movement or in times of resting.

Participants started from an identical point and were

videotaped with a user-independent camera installed

on a tripod at a height of 1.50 m and a 30-degree angle.

Ethical approval for this study was obtained from

the ethical committee of Arts Therapies of the

University of Applied Science, Nürtingen. The

complete study description is provided in

(Niedermann and Ostermann, 2019).

2.2 Data Acquisition

To analyze movement behavior, all videos of the test

group were played by a self-developed windows

application to manually track the positions of the

Figure 1: Setting of the sandpit from the camera perspective

(a) before and (b) after the gardening task.

participants. For this task, five psychology students

(3females and 2 males between the ages 21 and 24)

from the Department of Psychology and

Psychotherapy of the Witten/Herdecke University

volunteered. They were asked to always follow the

movements of the subjects by holding the mouse

pointer as near as possible to the subjects’ ankle.

Since some of the videos were quite long and the

student’s ability to concentrate should not be

overstrained, they could pause the movement tracking

process at any time. During this period, the tracking

automatically stopped.

In order to compensate distortions and other

sources of error that could arise from the tracking, the

whole set of the videos was fed through the analysis

software of each individual volunteer as shown in

Figure 2. While the students followed the movements,

Figure 2: Experimental setup.

HEALTHINF 2021 - 14th International Conference on Health Informatics

338

a background process tracked the position of the

mouse pointer. The interval between two time points

was set at 100 ms (milliseconds). Due to the 30-

degree angle between camera and ground, the

collected coordinates (x

, y

) were transformed by a

homography approach.

This approach is based on the Direct Linear

Transform (DLT), which uses a 2D (two-

dimensional) transformation matrix H (Hartley and

Zissermann, 2003). For the calculation of this 3 x 3

normalized matrix, a special case was considered, i.e.

there were 8 degrees of freedom and the position h

was set to one. The remaining eight 2D points

... P

and P’

... P’

) were obtained by measuring

the corners of the projected sandpit using the

following formula:

𝐴



𝐴







𝐴



𝐵



 

ℎ



⋮

ℎ





(1)

For each point pair P

and P’

(j = 1…4) two rows

of the matrix A were calculated as follows:

⎝

⎜

⎛

𝑃





𝑃





𝑃





𝑃′





𝑃





𝑃′





𝑃





𝑃





𝑃





𝑃′





𝑃





𝑃′





⎠

⎟

⎞



 

𝑎

 







 

𝑎

 







 



(2)

The same algorithm was applied to matrix B,

except that only the projection points were needed:



𝑃′





𝑃′









 

𝑏

 







 

𝑏

 







 



(3)

Figure 3 visualizes the viewpoints, one (O) that is

related to the location of the camera and is the state

before the transformation and one (O’) that represents

the state after the homography was applied. Both

point in the direction of the sandpit. The two 2D

planes that are located in between show the resulting

sandpit from each view. Whereas it is formed as an

isosceles trapezoid near the bottom, looking from

viewpoint O, from viewpoint O’ it is drawn as square,

which is in the middle of its plane.

The whole data, including the calculated

coordinates, video length, video identifier and date,

was exported as a comma-separated values file (CSV-

file) for further statistical analysis.

Figure 3: Viewpoints toward the sandpit.

2.3 Data Processing

For data processing, the images from the total of five

data sets were combined into a single one for each of

the 16 videos, using a special algorithm implemented

in R, which summarized the pixel coordinates of the

related images in disjoint intervals with a span of 0.5

s (seconds). Every interval contained around 20

measurement points, from which the median was

taken to set the pixel of the final image. Compared to

averaging with the mean, the advantage of this

method is that the original trace of the subjects could

be retained.

Based on the approach of (Riungu et al., 2018)

who analyzed park visitors’ spatial behavior,

heatmaps were created as a graphical aid to analyze

place and frequency of the participants’ movement in

the work environment. In this heatmaps, any point

that was ever visited from the participant were

highlighted in color. The range of the colors starts

with dark red, meaning that the pixel coordinate was

tracked once. If there is more than one visit, a lighter

color is indicated, up to a pure white, which denotes

the highest visiting frequency. All other pixels of the

output image, i.e. coordinates that never occurred at

least once, remained in a dark black.

Based on the distribution of the movement,

Shannon’s entropy E was calculated. It is based on a

system of mutually exclusive and exhaustive spatial

clusters A

, A

, …, A

and a set of probabilities p

p(A

), p

:= p(A

), …, p

:= p(A

), describing the

probability of having visited a respective cluster

(Ostermann and Schuster., 2015), and has been

applied in a variety of settings in health services

research. Then, the entropy is given by

𝐸



𝑝



,…,𝑝





 𝑝



𝑙𝑜𝑔 𝑝







(4)

where 0  log 0 = 0 is assumed.

Movement Entropy in a Gardening Design Task as a Diagnostic Marker for Mental Disorders: Results of a Pilot Study

339

Because of an unknown optimal grid size for

producing the maps, several densities were used in

order to find the optimal resolution, regarding to the

cluster dimension. For that reason, Pearson’s

correlation coefficient of the global severity index

GSI and the movement entropy was calculated and

plotted against the grid size g of the clusters. In a final

step, a curve was fitted through the points using an

ordinary least square (OLS) approach. We assumed a

convergence of the correlation from a certain grid size

and thus used the exponential approach

𝑟𝑎𝑒



𝑐

(5)

to fit the curve. After the parameters a, b, and c were

found, the optimal grid was determined for the grid

size g where the tangent had a slope m = -1. This

specific point divides the e-function into two

symmetrically equal parts and was determined using

the first derivation:

𝑟



𝑏𝑎𝑒



1

(6)

To detect for further associations, we finally fitted

a multivariate regression model with the GSI as

dependent variable and entropy, gender and two self-

assessed items (gardening experience and creativity)

as independent variables.

3 RESULTS

Sixteen individuals (8 female) aged between 19 and

60 years (x

age

= 28.67 years) participated in this pilot

study. Scores of the GSI showed a range of 2 to 27

points indicating a low to moderate mental burden.

The length of the videos ranged between 4.00 min to

30.00 min (x

length

= 23.00 min).

An example of the trajectory lines is given in

Figure 4. This example shows two dissimilar

Figure 4: Trajectory lines of a participant with a high BSI

on the left and a participant with a low BSI on the right. The

white lines exemplify a given grid.

movement styles. The tracing lines on the left belongs

to a participant with an extraordinarily high GSI

value. Other than the trajectories on the right, which

visualizes the active movement style of a participant

with a very low GSI value, these lines visually seem

less condensed and thin while the trajectories on the

right are bold indicating a higher pixel density and

thus a more active movement in the same area.

Figure 5 displays four heatmaps, corresponding to

the right participant of Figure 4. The maps were

created by using different grid sizes. From left to right

and from top to bottom, the raster increases in width

and height. A closer look reveals that this increases

the image resolution, too. If these heatmaps were

compared to those from the left participant of Figure

4, the latter would show a much lower distribution of

track points, resulting in an inferior entropy.

Figure 5: Heatmaps using different raster sizes. Top left ≙

50 x 50. Top right ≙ 100 x 100. Bottom left ≙ 200 x 200.

Bottom right ≙ 300 x 300.

Pearson’s correlation coefficient of the global

severity index GSI and the movement entropy plotted

against the grid size g of the clusters is displayed in

Figure 6. The red line denotes the curve found by

OLS-regression and is given by the following

exponential function:

𝑟0.44 𝑒

.

 0.476

(7)

Using equation (6), an optimal grid was found at

g = 198.8, which resulted in a rounded grid size of

200 x 200 considered as optimum.

HEALTHINF 2021 - 14th International Conference on Health Informatics

340

Figure 6: Correlation between the size of the raster and

Pearsons’s r (approximated as red curve) as well as the

corresponding p values (blue curve).

Looking at the optimum point, the Pearson’s

correlation value of r

= -0.463 nearly reached the

maximum of all the values, which ranged between

min

= -0.19 and r

max

= -0.487. Statistical significance

with a value of p

= 0.071 was slightly missed and

ranged between p

min

= 0.48 and p

max

= 0.056 for all

grid sizes (see the blue line in Figure 6).

Thus as a first result, it could be assumed that a

more vital and spacious movement can be associated

with lower mental burden which is displayed in the

bottom graph of Figure 7, in which the GSI is plotted

against the movement entropy for all participants for

the optimal grid size.

However, a closer examination of the dataset by

dividing it into two gender-based subgroups revealed

a clear gender specific effect, similar as found by Van

Tuyckom et al. (2012). The assumption made above

only applies to males (top graph of Figure 7), while

females participants behaved in the complete

opposite way (middle graph of Figure 7), presumably

to compensate their depressive mood with an

increased activity in order to achieve a feeling of

satisfaction by looking at their work of art afterwards

on the gardening task (Milligan et. al., 2004).

In addition to the visual illustration of the gender

effect, it was also statistically verified: whereas the

correlation between the GSI and the entropy strongly

increased in the females’ group (r

females

= 0.661),

there was an extremely sharp decrease in the

correlation line for the males (r

males

= -0.922).

Finally, using a multiple regression model with

the independent variables entropy, gender and the

product of both, a high significance could be obtained

(p < 0.001 for all predictors; r = 0.88). Other

Figure 7: GSI plotted against the movement entropy for all

participants for the optimal grid size of g = 200 x 200

separated into three gender groups.

variables such as self reported gardening experience

or creativity did not have any influence on the

outcome as it is shown in Table 1 below.

Table 1: Results of the multivariate regression model.

Variable Standardized β Significance

Entropy -2.544 0.001

Gender -13.086 0.001

Entropy × Gender 13.841 0.001

Experience 0.001 0.995

Creativity 0.117 0.518

4 CONCLUSIONS

In the field of garden and landscape design,

therapeutic studies have already shown effectiveness

(Clatworthy et al., 2013). In the field of diagnostics,

however, there were so far only rudimentary

approaches to record corresponding movement and

behavior patterns.

In this pilot study we thus aimed at investigating,

whether a reduced spatial movement in a gardening

task was related to higher value of mental distress. By

using a movement entropy approach and after finding

an optimal grid size, we finally were able to find a

highly significant association between these

quantities, which however were moderated by gender

and explained 77.4 % of variance. We thus can

assume that depending on the gender, movement

Movement Entropy in a Gardening Design Task as a Diagnostic Marker for Mental Disorders: Results of a Pilot Study

341

entropy in a gardening task is significantly associated

with mental distress.

This result is in accordance with other findings in

the field of human movement analysis. In the

Systematic Review of Rohani et al. (2018), entropy

amongst other features was the most promising

candidate to predict mental distress in 6 of 46

included studies. Moreover, the amount of explained

variance was also comparable with the results of an

exploratory study (Saeb et al, 2015) on depressive

symptom severity in daily-life behavior and

normalized entropy as an indicator of mobility

between favorite locations (r

= 33.64 %, p = 0.012).

Other approaches which examined the movement

patterns in drawing tests also found associations of

cognitive impairment and drawing entropy (Robens

et al., 2019). And in contrast to potentially similar

measures like pixel density (Unger et al., 2020), the

association between entropy and the GSI was more

pronounced.

From a methodological point of view, a further

important effect could be discovered. Changing the

size of the raster that ran over the input images had a

clear impact on the correlation of entropy and GSI:

The larger the dimensions was adjusted, the higher

was the amount of correlation. This went down until

a raster size of around 300 x 300 pixels after which

the amount of coefficient seemed to stagnate.

An explanation for this effect could be that a raster

with a lower resolution only generates heatmaps, in

which the clusters for the calculation are too close

together. However, the constant course of the curve,

after reaching the optimal value at a grid size of 200,

could be an indicator that the true value of r is located

within this area.

Despite of these promising results, there are also

some limitations. First of all, the number of

participants is too low to draw any statistically robust

conclusions out of our analyses. As this study was a

pilot study, we did not primarily focus on a sufficient

sample size to detect significant correlations, but

rather tried to investigate the feasibility of the

approach. Thus, our conclusions have to be

interpreted with care.

In addition, other specific factors would be

expected to influence movement entropy such as

creativity or previous gardening experience.

Although this was only a pilot study we were able to

show, that these variables did not influence our

results, indicating that no prior gardening experience

or a special amount of creativity is needed. However,

further studies should try to replicate our results with

a higher sample size to determine whether the results

remain stable.

Moreover, our result could be improved by

looking at the intersection lines that would result from

connecting two geographic tracing points. On the

other hand, this method is much more complex and

would rather appreciate the precise position of the

subjects.

From a technical point, the automatic object

detection and electronic tracking methods as

summarized in (Hatwar et al., 2018) might be used to

get a more precise picture of movement patterns. In

addition, the markerless measurement and evaluation

of kinetic features as proposed in (Trujillo et al.,

2019) might also contribute to a more differentiated

movement analysis without deterioration of the

original setup and might produce more reliable data.

REFERENCES

Bleasdale, T., Crouch, C., & Harlan, S. L. (2011).

Community gardening in disadvantaged neighborhoods

in Phoenix, Arizona: Aligning programs with

perceptions. Journal of Agriculture, Food Systems, and

Community Development, 1(3), 99-114.

Clark, D., Schumann, F. & Mostofsky, S.H. (2015).

Mindful movement and skilled attention. Frontiers in

Human Neuroscience, 9, 297.

Clatworthy, J., Hinds, J. & Camic, M.P. (2013). Gardening

as a mental health intervention: a review. Mental Health

Review Journal, 18(4), 214–225.

Collier, S., Monette, P., Hobbs, K., Tabasky, E.,

Forester, B. P. & Vahia, I. V. (2018). Mapping

Movement: Applying Motion Measurement

Technologies to the Psychiatric Care of Older Adults.

Current Psychiatry Reports, 20(8), 64.

Ghosh, S. & Ghosh, S.K. (2018). Exploring the association

between mobility behaviours and academic

performances of students: a context-aware traj-graph

(CTG) analysis. Progress in Artificial Intelligence,

7(4), 307–326.

Hartley, R. & Zisserman, A. (2003). Multiple view

geometry in computer vision (2nd ed.). Cambridge

University Press.

Hatwar, R.B., Kamble, S.D., Thakur, N.V. & Kakde, S.

(2018). A Review on Moving Object Detectionand

Tracking Methods in Video. International Journal of

Pure and Applied Mathematics, 118(16), 511–526.

Kim, D. & Ohara, K. (2010). A Study on the Role of

Gardening and Planning of Green Environments for

Daily Use by Residents in Senior Housing. Journal of

Asian Architecture and Building Engineering, 9(1), 55–

61.

Macdonald, K., Milne, N., Orr, R., & Pope, R. (2018).

Relationships between motor proficiency and academic

performance in mathematics and reading in school-

aged children and adolescents: a systematic review.

International journal of environmental research and

public health, 15(8), 1603.

HEALTHINF 2021 - 14th International Conference on Health Informatics

342

Milke, D.L., Beck, C.H.M., Danes, S. & Leask, J. (2009).

Behavioral Mapping of Residents’ Activity in Five

Residential Style Care Centers for Elderly Persons

Diagnosed with Dementia: Small Differences in Sites

Can Affect Behaviors. Journal of Housing for the

Elderly, 23(4), 335–367.

Milligan, C., Gatrell, A. & Bingley, A. (2004). Cultivating

health: therapeutic landscapes and older people in

northern England. Social Science & Medicine, 58,

1781–1793.

Niedermann, C. & Ostermann, T. (2019). Zur Bedeutung

von Gartengestaltung für die Erfassung mentaler

Gesundheit: Konzeptentwurf eines Pilotprojekts. In C.

Schulze-Stampa & K. Schoch (Eds.), In/ter/ven/tion: in

mit durch Kunst (5th ed., pp. 58–63). Hochschule für

Künste im Sozialen Ottersberg.

O'Brien, J. T., Gallagher, P., Stow, D., Hammerla, N.,

Ploetz, T., Firbank, M., Ladha, C., Ladha, K., Jackson,

D., McNaney, R., & Ferrier, I. N. (2017). A study of

wrist-worn activity measurement as a potential real-

world biomarker for late-life depression. Psychological

medicine, 47(1), 93-102.

Ostermann, T., & Schuster, R. (2015). An Information-

theoretical Approach to Classify Hospitals with

Respect to Their Diagnostic Diversity using Shannon's

Entropy. Proceedings of the International Conference

on Health Informatics 2015: 325-329.

Riungu, G. K., Peterson, B. A., Beeco, J. A. & Brown, G.

(2018). Understanding visitors’ spatial behavior: a

review of spatial applications in parks. Tourism

Geographies, 20(5), 833–857.

Robens, S., Heymann, P., Gienger, R., Hett, A., Müller, S.,

Laske, C., Loy, R., Ostermann, T. & Elbing, U. (2019).

The Digital Tree Drawing Test for Screening of Early

Dementia: An Explorative Study Comparing Healthy

Controls, Patients with Mild Cognitive Impairment, and

Patients with Early Dementia of the Alzheimer Type.

Journal of Alzheimer's Disease: JAD, 68(4), 1561–

1574.

Rohani, D. A., Faurholt-Jepsen, M., Kessing, L. V., &

Bardram, J. E. (2018). Correlations between objective

behavioral features collected from mobile and wearable

devices and depressive mood symptoms in patients with

affective disorders: systematic review. Journal of

medical Internet research, mHealth and uHealth, 6(8),

e165.

Spitzer, C., Hammer, S., Löwe, B., Grabe, H. J., Barnow,

S., Rose, M., Wingenfeld, K., Freyberger, H. J. &

Franke, G. H. (2011). The short version of the Brief

Symptom Inventory (BSI -18): preliminary

psychometric properties of the German translation.

Fortschritte der Neurologie-Psychiatrie, 79(9), 517–

523.

Saeb, S., Zhang, M., Karr, C. J., Schueller, S. M., Corden,

M. E., Kording, K. P., & Mohr, D. C. (2015). Mobile

phone sensor correlates of depressive symptom severity

in daily-life behavior: an exploratory study. Journal of

medical Internet research, 17(7), e175.

Trujillo, J.P., Vaitonyte, J., Simanova, I. & Özyürek, A.

(2019). Toward the markerless and automatic analysis

of kinematic features: A toolkit for gesture and

movement research. Behavior Research Methods,

51(2), 769–777.

Unger, S., Ostermann, T., Appelbaum, S. & Niedermann,

C. (2021). Tracking Human Spatial Motion in the

Design of Gardens as a Potential Method to detect

Mental Disorders – A Pilot Study. Acta Horticulture,

Accepted for publication.

Van Tuyckom, C., Van de Velde, S., & Bracke, P. (2012).

Does country-context matter? A cross-national analysis

of gender and leisure time physical inactivity in Europe.

European Journal of Public Health, 23(3), 452–457.

Wang, D. & Glicksman, A. (2013). “Being Grounded”:

Benefits of Gardening for Older Adults in Low-Income

Housing. Journal of Housing for the Elderly, 27(1-2),

89–104.

Yen, I. H., Flood, F. J., Thompson, H., Anderson, L. A. &

Wong, G. (2014). How design of places promotes or

inhibits mobility of older adults: Realist synthesis of 20

years of research. Journal of Aging and Health, 26(8),

1340–1372.

Zhu, X., Haegele, J. A., & Healy, S. (2019). Movement and

mental health: Behavioral correlates of anxiety and

depression among children of 6–17 years old in the US.

Mental Health and Physical Activity, 16, 60-65.

Movement Entropy in a Gardening Design Task as a Diagnostic Marker for Mental Disorders: Results of a Pilot Study

343