exploring ways of visualizing functional connectivity · this thesis the design of an interactive...

54
Exploring Ways of Visualizing Functional Connectivity Jan Nyl´ en October 26, 2017 Master’s Thesis in Interaction Technology and Design, 30 hp Supervisors at UmU: Anders W˚ ahlin and Kalle Prorok Examiner: Thomas Mejtoft Ume ˚ a University Department of Applied Physics and Electronic SE-901 87 UME ˚ A SWEDEN

Upload: others

Post on 21-Jun-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Exploring Ways of VisualizingFunctional Connectivity

Jan Nylen

October 26, 2017Master’s Thesis in Interaction Technology and Design, 30 hp

Supervisors at UmU: Anders Wahlin and Kalle ProrokExaminer: Thomas Mejtoft

Umea UniversityDepartment of Applied Physics and Electronic

SE-901 87 UMEASWEDEN

Page 2: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an
Page 3: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Abstract

Functional connectivity is a field within neuroscience where measurements of co-activationbetween brain regions are used to test various hypotheses or explore how the brain activatesdepending on a given situation or task. After analysis, the underlying data in the fieldconsists of a n by n adjacency matrix where each cell represents a correlation value betweentwo regions in the brain. Depending on the research question the number of regions andmatrices incorporated varies and new visualizations are needed in order to portray them. Inthis thesis the design of an interactive web based visualization tool for functional connectivitywas explored through an iterative design process. The design of the tool was based onexisting guidelines, interviews and best practices in data visualization as well as an analysisof current visualization solutions used in functional connectivity. The final concept andprototype uses a network plot for functional connectivity called the connectogram as wellas a grouped bar graph to provide an intuitive and accessible way of comparing functionalconnectivity data by interacting with and highlighting networks and specific network datathrough direct manipulation. Results of qualitative evaluations of a prototype using datafrom a concurrent scientific project is presented. The prototype was found to be useful,engaging, easily perceivable and offered an easy and quick way of exploring data sets.

Page 4: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

ii

Page 5: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Contents

List of Figures v

1 Introduction 1

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Aim of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.3 Umea Center for Functional Brain Imaging (UFBI) . . . . . . . . . . . . . . . 2

1.4 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Theoretical Framework 3

2.1 Data Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1.1 Key Areas in Data Visualizations . . . . . . . . . . . . . . . . . . . . . 3

2.1.2 Design Guidelines for Data Driven Visualizations . . . . . . . . . . . . 4

2.1.3 Data Types and Tasks in Data Visualizations . . . . . . . . . . . . . . 6

2.1.4 Color . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.1.5 Guidelines on Visual Hierarchy (colors, shapes, position etc) . . . . . . 7

2.1.6 Involving Users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Functional Connectivity and Current Visualization Solutions 11

3.1 Overview of Functional Connectivity . . . . . . . . . . . . . . . . . . . . . . . 11

3.1.1 Resting and Working State . . . . . . . . . . . . . . . . . . . . . . . . 12

3.1.2 Overview of the Brain’s Anatomy . . . . . . . . . . . . . . . . . . . . . 12

3.1.3 Data structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.1.4 Domain Specific Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.1.5 Existing Visualizations of Functional Connectivity . . . . . . . . . . . 15

4 Methodology 19

4.1 Design-Build-Test and a Data Visualization Process . . . . . . . . . . . . . . 19

4.1.1 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.1.2 Design - Establishing Goals and Requirements . . . . . . . . . . . . . 20

4.1.3 Build - Implementation of Prototypes . . . . . . . . . . . . . . . . . . 21

4.1.4 Test - Testing and Evaluating . . . . . . . . . . . . . . . . . . . . . . . 21

iii

Page 6: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

iv CONTENTS

5 Results 25

5.1 Final Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

5.2 Design Rationale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5.2.1 Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5.2.2 Choice of Representation . . . . . . . . . . . . . . . . . . . . . . . . . 28

5.2.3 Colors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.2.4 Annotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.3 User Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.3.1 Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.3.2 Interviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

5.3.3 Visualization Effectiveness Profiles . . . . . . . . . . . . . . . . . . . . 31

6 Discussion and Conclusions 33

6.1 User Test Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

6.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6.3 Methodology development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6.4 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

7 Acknowledgements 39

8 Bibliography 41

A A - Usability Test Instructions 45

A.1 A Visualization Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

A.1.1 Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Page 7: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

List of Figures

3.1 A sagital slice of the brain portraying a few anatomical regions. Acquired

from https://tl.wikipedia.org/wiki/Utak####/media/File:NIA_human_

brain_drawing.jpg Accessed 2017-10-25 . . . . . . . . . . . . . . . . . . . . 13

3.2 The 12 networks of interest for the visulization case. The names for each icon

can be found in the list under section 3.1.3 with the corresponding number. . 14

3.3 Illustrations for weighted and unweighted graphs as well as an adjacency matrix. 15

3.4 Image depciting the transition from anatomical to connectional imaging for

functional connectivity. Retrieved from Marguilies et al. [14] . . . . . . . . . 16

5.1 The Index page and one of the visualization views of the interactive prototype

used for the case study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.2 A connectogram where all edges but the memory retrieval’s have been re-

moved network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.3 A heatbar showing the correlation thresholds used for the edges ni the connnec-

togram and a menu for hiding and showing networks. . . . . . . . . . . . . . . 27

5.4 Two Connectograms adapted to resting and working state of mean connec-

tivity data used in the final concept. . . . . . . . . . . . . . . . . . . . . . . . 27

5.5 A bar chart showing the resting and working state of the ventral attention

network for a mean connectivity data set. The y-axis stands for the r-value

between the x-axis networks and the ventral attention network. . . . . . . . . 28

5.6 Mock up views for more than two datasets as well as one of the earlier versions. 29

5.7 The visualization profile results. Each of the colored shapes represents one re-

searcher’s critique. Base figure acquired with permission from https://www.

perceptualedge.com/articles/visual_business_intelligence/data_visualization_

effectiveness_profile.pdf Retrieved 2017-08-10. . . . . . . . . . . . . . . 32

6.1 Stephen Few’s recommendation on what visulizations should aim for in terms

of his seven factors of visualization effectiveness. Acquired with permission

from https://www.perceptualedge.com/articles/visual_business_intelligence/

data_visualization_effectiveness_profile.pdf Retrieved 2017-08-10 . 34

v

Page 8: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

vi LIST OF FIGURES

Page 9: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 1

Introduction

1.1 Background

Information processing in the brain can be analyzed with functional magnetic resonanceimaging (fMRI) [34], an approach by which many of the discoveries on how the brain workshave relied on. The technique is sensitized to local blood oxygenation levels that depend onlevel of neural activity - by taking continuous snapshots of the brain, it is possible to seehow a certain brain region responds to a task.

Beyond analyzing activation of different parts of the brain separately it is necessaryto explore so called functional connectivity [27], a measurement of the degree of temporalcorrelation in activity between two different brain regions. In the big data era, coordi-nated collection of larger study samples as well as improved hardware and software haveallowed exploring connectivity between all possible combinations of brain regions and howsuch connectivity may be important for various behavioral parameters. This explosion indimensionality makes reporting the final result a pedagogical challenge. Initially this haslead to the adoption of common graph and network analysis plots to present neuroscientificdiscoveries [2][14], a pragmatic choice that unfortunately loses the brain anatomy inter-pretability. Therefore, it is difficult for someone without years of experience to assess agraphical visualization of the results.

A possible solution to this pedagogical issue is to allow for interaction with the visualiza-tion. An interactive web based visualization can be used as a tool that can be accessed whenreading a research paper, during presentations or even analysis of data through exploration.With its inherent widespread access and support for various devices the web allows for aninteractive web based visualization to be accessible near to anywhere. In this thesis wepresent the process and results of designing a web based visualization tool with the aim ofintuitively portraying functional connectivity in a more engaging way. Furthermore, the re-sults of a set of qualitative evaluations of the tool conducted with neuroscientific researchersinvolved throughout the thesis are presented.

1.2 Aim of the Thesis

The aim of this thesis was to design a visualization tool that assisted in the presentationand exploration of functional connectivity data. Specifically we wanted to develop a conceptthat 1) was easy to use for neuroscientific researchers, 2) visualized the data sets and allowed

1

Page 10: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

2 Chapter 1. Introduction

for comparison of them and 3) could provide pedagogical illustrations for publications. Inorder to fulfill the aims of the thesis the following sub goals were set up:

1. Find appropriate literature and design guidelines to base the design of the tool on.

2. Map what kind of visualizations that are used in functional connectivity publications.Furthermore, evaluate the strengths and weaknesses of these visualizations.

3. Go through an iterative design process and produce and evaluate prototypes based onthe reviews in point 1) and 2) in varying fidelity.

4. Evaluate the final prototype’s functionality against data from a current study togetherwith neuroscience researchers (domain experts).

1.3 Umea Center for Functional Brain Imaging (UFBI)

The project was supervised by and performed in collaboration with researchers at UmeaCenter for Functional Brain Imaging (UFBI). UFBI is an inter-disciplinary research centreat Umea University where the brain’s structure and function is examined in relation to bothbasic and clinical questions. It was established in October 2001 and has since conductedstudies and activities ranging from examining age-related changes of functional and struc-tural brain responses and associated changes in the dopamine system to training-relatedchanges in functional networks, and clinical imaging projects on stress 1.

1.4 Thesis outline

This thesis is structured into six chapters. In the previous and first chapter the problem,thesis subject and aim is introduced. In chapter two an overview over guidelines and bestpractices in data visualization is given. Chapter three briefly describes important aspectsof functional connectivity, the data used as a basis for the design, some basic anatomicalproperties of the brain as well as current visualization solutions in the field of functionalconnectivity. Following after the theoretical chapters one, two and three the methodologyof the thesis is introduced in chapter four. Thereafter, the results are presented in chapterfive and further discussed in chapter six.

1http://www.ufbi.umu.se/english/about-ufbi/ Website of UFBI. Accessed 2017-05-03

Page 11: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 2

Theoretical Framework

As a theoretical framework for this thesis literature surrounding key areas in data visualiza-tion, functional connectivity and visualizations of functional connectivity data are reviewed.Chapter two brings up literature regarding data visualization, while chapter three shortlyintroduces functional connectivity and a review on existing visualizations that are beingused in publications regarding functional connectivity.

2.1 Data Visualization

Data visualization has been around since the first time a picture representing data wasdrawn. However, as a formal scientific field it is relatively new. It is an interdisciplinaryfield combining knowledge from cognition, color theory, data types, interactivity paradigms,Human Computer Interaction (HCI) and art [3]. Exactly what areas that are used forvarious sub-fields and how the field, in a scientific context, should be defined varies. Forexample, Tory et al. [28] proposes a change in definition from a dichotomy consisting ofinformation visualization and scientific visualization to instead consist of continuous modelvisualization and discrete model visualization. In order to clarify what data refers to in thecontext of this thesis it is the visualization of functional brain connectivity data. A moreexplicit description of what functional connectivity is and what the data looks like is givenin section 3.2.

As young as the field is there is not a set approach on how to solve a visualizationproblem and it is probable that there never will be due to its creative nature. While thereare cognitive and perceptual knowledge that can be directly applied to problems, such aswhat color that is the most salient in a given context, current literature that addresses datavisualizations as a whole are combinations of various versions of the creative process, expertguidelines on how to achieve better visualizations, and samples on previously successfulvisualizations.

2.1.1 Key Areas in Data Visualizations

In this thesis the key areas of data visualization are regarded as representation, presenta-tion, exploiting visual perception abilities and to amplify cognition as presented in DataVisualization a Successful Design process by author Kirk Andy [12]. With representationKirk means what form a visualization should take, be it a bar chart, circle chart or anyother visual variable. Presentation is how it is integrated into the overall work. E.g what

3

Page 12: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

4 Chapter 2. Theoretical Framework

colors to choose, what annotations to include and what interactive properties to use. WithVisual perception abilities he emphasize the importance of understanding how our eyesand brain processes information. He exemplifies this with our abilities of spatial reasoning,pattern recognition and big-picture thinking. Finally, amplifying cognition concerns themaximization of how efficiently and effectively we can process the information into thoughts,insights and knowledge.

2.1.2 Design Guidelines for Data Driven Visualizations

Other than what can be derived from examples, specific guidelines for visualizations offunctional connectivity are scarce. Allen et. al [1] brings up a few guidelines regardingvisualization in neuroscience as a whole. Although, they are mostly centered on how onecan keep the integrity of ones’ results by showing information regarding uncertainties. Forexample, if a given number in a visualization has been statistically generated it is recom-mended to show the standard deviation of it in its graphic representation. However, despitewhat seems to be a lack of specific guidelines, general guidelines in the field of informationand data visualization are applicable when creating an interactive tool for visualizing func-tional connectivity. A famous name in information visualization is Edward Tufte. He haswritten four books on the subject in which he exemplifies what he considers as good or badvisualization practices: Beautiful Evidence, The Visual Display of Quantitative Information,Envisioning Information, and Visual Explanations [29, 32, 30, 31]. In the Visual Display ofQuantitative Information Edward Tufte presents a set of nine guidelines for what he callsgraphical excellence:

– 1) Show the data

– 2) Induce the viewer to think about the substance rather than about methodology,graphic design, the technology of graphic production or something else

– 3) Avoid distorting what the data has to say

– 4) Present many numbers in a small space

– 5) Make large data sets coherent

– 6) Encourage the eye to compare different pieces of data

– 7) Reveal the data at several levels of detail, from a broad overview to the fine structure

– 8) Serve a reasonably clear purpose: description, exploration, tabulation or decoration

– 9) Be closely integrated with the statistical and verbal descriptions of a data set.

”These guidelines are somewhat vague and are in many cases end goals of a visualization.

However, as Tufte comments, the designer has to filter through and choose what guidelinesthat are the most appropriate for their case. ”Most principles of design should be greetedwith some skepticism... we may come to see only through the lenses of word authority ratherthan with our own eyes” - Tufte [29].

In addition to the guidelines above, Tufte marks on the importance of avoiding chartjunk [30][33]. Meaning that unnecessary aesthetic or decorative graphics that does not holdrelevant information should be avoided in order to keep the data-to-ink ratio low and thereby

Page 13: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

2.1. Data Visualization 5

lessen the chance of visual overload. Although, claiming that aesthetics are irrelevant is notfair in all situations. Donald Norman has in his work[17] shown that the usability andmemorability of products can be affected by irrelevant aesthetic decorations. Norman callsthis emotional design and it can be summed up with: If we like something we are morelikely to remember it and use it again. Therefore, there might be value in considering notjust practical and pragmatic aspects of a visualization but also its aesthetic values.

In the article Human factors in visualization research Tory et al. [28] brings up anotherset of perception and Cognition-Based Design guidelines for visualization research. In com-parison to Tufte’s guidelines these are more directly applicable to the project of this thesisas they are specified for interactive visualization tools and not information visualization ina general sense, be it static, dynamic or interactive. The guidelines are defined below andany references within them can be found in Tory et al.’s article [28]:

– ”Since users’ information needs are domain and task dependent, design must either

• be domain and task specific or

• look at domain-independent sub tasks such as those defined by Shneiderman[71], Chapter 15]: overview, zoom, filter, details-on-demand, relate, history, andextract.

– To support users with different tasks and requirements, multiple visual representationsof the data should be available. Several representations may be visible at once usingmultiple view windows. If it is not possible to render a global view of the data set inwhich every element is precisely represented, it is possible to combine detailed, partialrepresentation(s) with vague, global representation(s). For example, in a medicalimaging data set, slices and subvolumes of specific areas could be combined with avolume rendered overview of the entire volume.

• Changing between representations and views should be easy.

• Using multiple views is not always appropriate. Baldonado et al. describe a setof guidelines for when and how to utilize multiple views for visualization tasks[2].

• Continuity should be maintained so the user does not get lost when switchingbetween representations. Woods provides several design guidelines to help pro-vide such continuity or “visual momentum” [82]. (For example, use gracefultransitions such as animation, maintain formatting consistency across views, andprovide features that are easily discernible in all views and thus act as perceptuallandmarks.)

– The following variables should always be visible:

• The set of data elements (an overview). With volume or fluid flow data, theoverview contains the entire object or space being visualized.

• Relationships between data elements. Relationships may be either explicit (e.g.,links between web pages) or implicit (e.g., relative positions of objects in a scene).

• Method of locomotion. In other words, cues should be present to help the userunderstand how to navigate through the display and modify display parameters.

• Details at the current location (e.g., the value of a voxel in volume data).

Page 14: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

6 Chapter 2. Theoretical Framework

• Details of the local neighborhood.

• Navigation history. In other words, a list of previously explored display param-eters, such as transfer functions in direct volume rendering (for details of thisexample, see Section 3.4).

• Data at the focus of interaction should be undistorted and represented at thehighest possible resolution.

• Navigation tools should be reused to maintain consistent interaction metaphorsthroughout the system.

2.1.3 Data Types and Tasks in Data Visualizations

In the article The Eyes Have It: A Task by Data Type Taxonomy for Information Visualiza-tions [23], Shneiderman presents a task by data type taxonomy for information visualizationas well as a design paradigm which is described with the mantra ”Overview first, zoom andfilter, then details on demand”. What kind of data types that are used for visualizationsvaries and depending on what the data looks like in a project the way that key areas shouldbe tackled and how guidelines can be applied differs. Shneiderman presents a list of thevarious forms that data can take as: 1-dimensional, 2-dimensional, 3-dimensional, multi-dimensional, temporal, tree and network. It is further stated that these data types can inmany cases be combined and the list is an abstraction of reality and not serve as an endall definition. In addition to the data type taxonomy, Shneiderman lists seven high levelabstraction tasks that are recommended to be incorporated in a visualization tool:

– ”Overview: Provide an overview of all the data

– Zoom: Zoom in on items of interest

– Filter: Filter out uninteresting items.

– Details-on-demand: Select an item or group and get details when needed

– Relate: View relationships among items

– History: Keep a History of actions to support undo, replay and progressive refinement

– extract: Allow extraction of sub-collections and of the query parameter”

Albeit, as mentioned in Tory et al.s guidelines [28] these tasks does not cover all possibletasks that a visualization tool can have and they further state that domain specific taskscan often further be defined.

2.1.4 Color

While color often serves as an aesthetic tool to set an appropriate tone and feel of a design itcan also be used as a tool to categorize or emphasize data in visualizations or interfaces. Inthe later case, especially when the data is comprised of multiple categories, it is important tobe consistent and make sure that the choice of colors enhances our visual perception. Whata color represents should not vary as it can confuse viewers and as color can emphasizeelements of a visualization it is important to consider what parts that are more or less

Page 15: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

2.1. Data Visualization 7

important [25]. When it comes to mapping with color, the choice of a color or combinationof colors depends on the type of data. If the data is nominal the choice of colors differs fromwhen it is ordinal, interval or ratio [4]. There are typically three types of color schemasdefined for the visualization of data: 1) Qualitative color schemas - Color sets used torepresent data of a categorical nature, 2) Divergent colors - color schemas that both showfor example a negative and a positive side of data and ”diverges” from the middle just likea temperature scale and 3) Sequential color schemas which are arranged from high to low.Essentially the same as if one would remove one side of a divergent one. Colorbrewer 2.0 isa tool that creates various kinds of color scales based on color theory and can be used tocreate efficient color scales for visualizations 1.

2.1.5 Guidelines on Visual Hierarchy (colors, shapes, position etc)

In their article Graphical Perception: The Visual Decoding of Quantitative Informationon Graphical Displays of Data McGill and Cleveland [6] ranked how easily different visualaspects are to perceive. The rankings are listed below:

1. Position along a common scale

2. Positions along nonaligned scales

3. Length, direction, angle

4. Area

5. Volume, curvature

6. Shading, color saturation

The implications of these results are not transferable or applicable to all visualizationsbut in essence it can be concluded that visualizations in first hand should consider shapesand positions rather than color when encoding data.

2.1.6 Involving Users

Conducting user studies is an established way to evaluate the usability of interfaces inhuman computer interaction (HCI). Tory et al. [28] states that data visualization andHCI are related and several of the factors that are important when designing an interfacerelated to HCI are also important when designing an interactive visualization. One of themain concerns in both fields are the users. Involving users in the design process, be it avisualization or a user interface, is a way to control for quality and to lower the chance of theend result not fulfilling its goal [20, 13]. While HCI has worked with user centered design fora long time, the field of data visualization has just started. Tory et al. reviewed articles in thedata visualization journal Transactions on Visualization and Computer Graphics (TVCG)journal IEEE and concluded that only 23 percent of them had considered human factorsduring the design process. As of yet there is no conclusive research for how users effectivelycan be involved when designing visualizations [28]. However, despite the lack of knowledgeor tried methodologies in visualization design Tory et al. implores that inspiration shouldbe taken from methods used in HCI.

There are a variety of ways of conducting user tests. Ranging from qualitative tests, suchas ethnological studies, interviews and Thinking Aloud to quantitative tests like A/B testing

1http://www.colorbrewer2.org/ Website of Color Brewer. Accesed 2017-08-10

Page 16: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

8 Chapter 2. Theoretical Framework

or heuristic evaluations like the GOMS model [20]. The choice of which test to use dependson the context of the project and product. For example, what the underlying question is,which phase the project is in, the scope of the project, type of users as well as to whatdegree one has access to test participants [20]. The test subjects of this thesis were alreadypredefined at the start of the thesis as the UFBI researchers. However, Tufte[30] claims thatboth readers as well as writers should be considered as the users of a visualization used ina text. The writer is the one that uses the visualization to enhance their text and get areference point to talk about. Meanwhile, the reader consumes the end result of both textand visualization.

While visualization projects can make use of HCI evaluation techniques as it always isimportant to make tools intuitive and easy to use. It is still debated how and what kindof tests that should be used in the field. Typical HCI evaluations evaluate the usabilityand ease of use while visualizations for domain experts might require years of experienceto understand and therefore be lacking in how easy they are to use or understand, anddespite this they could still be considered useful. Long term case studies has been proposedas a way to measure the success of visualization tools [24] by evaluating the success interms of number of discoveries throughout research projects. An obvious negative sideor problem with long term approaches is the time requirement. Besides evaluating thesuccess in terms of number of discoveries it has also been proposed by Bang Wong, in hisarticle Points of view: Visualizing biological data 2 that visualization designers working withscientific data should incorporate domain experts as much as possible throughout the designprocess in order to make sure that the visualization stays true to the intent of the data,and research questions as a whole. However, there is no set method for how they should beincorporated or how evaluations of visualization with domain experts should be conducted.Stephen Few has attempted to create a visualization evaluation profile which he calls ”DataVisualization Effectiveness Profile”. In it he proposes seven factors for which a visualizationshould be evaluated and critiqued against and further exemplifies what he considers as anin general good rating to compare an evaluation against. The seven factors are dividedinto one informative category and one emotive category: The Informative factors consists ofusefulness, completeness, perceptibility, truthfulness and intuitiveness and Emotive factorsare aesthetics and engagement.

2.2 Conclusions

As presented in this chapter there exists several guidelines for data visualization. The onesthat has been brought up are not an exhaustive list. They are in many cases based on expertknowledge and experience, rather than empirical science. Albeit, it is likely that they havebeen produced and are derived with empirical knowledge regarding cognition and perceptionin mind. An implication of the lack of empirical knowledge is that the guidelines should notbe taken as de facto solutions or that they always have to be followed. Designing is not anexact science and while every part of the guidelines mentioned above is not applicable toall kinds of visualization there is still value in regarding them and making use of them asinspiration.

All of the visualization guidelines brought up two points repeatedly. These can beparaphrased as:

1) Provide an overview of the data in order to allow for the user or reader to understand

2http://www.nature.com.proxy.ub.umu.se/nmeth/journal/v9/n12/full/nmeth.2258.html An Articlediscussing the involvement of domain experts in visualization projects. Accessed 2017-03-24

Page 17: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

2.2. Conclusions 9

the context of any sub-views and thus make it easier for them to refer back to where in thedata they currently are located or exploring.

2) Allow for the user to filter and put more focus on a select subset of a data that isthough to be of more interest.

As all of the guidelines incorporate a version of these points they have been emphasizedthroughout the project.

Page 18: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

10 Chapter 2. Theoretical Framework

Page 19: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 3

Functional Connectivity andCurrent Visualization Solutions

Scientific data visualizations tend to be multi-dimensional [30] and have a spatial contextwhich on its own carries valuable information for understanding scientific results. In thisthesis the spatial context is the brain and the underlying data is a matrix where each cellin the matrix contains information regarding the temporal correlation between two regionsin the brain. In this chapter the structure of the data set to be visualized, its context, aswell as what kinds of representations there are for this kind of data is introduced.

3.1 Overview of Functional Connectivity

Functional connectivity is derived from an indirect measurement of neural activity calledthe blood-oxygenation-level-dependent-signal (BOLD-signal) which is a measurement of aprocess that happens in the brain when blood streams more to active neurons than inactiveneurons effectively heightening the oxygen level in that area. This variation in oxygenationthat arises has proven to be measurable with recent techniques and is what the BOLD-signalrepresents [33].

The BOLD-signal is further used to indicate the degree of functional connectivity be-tween different brain regions. Functional connectivity refers to the covariance of BOLDactivity between different regions in the brain usually measured during an experimentallypre-defined task or during resting state [34]. The covariance of two regions is extracted bycorrelating the time series of two regions as measured through fMRI.

Through further graph and statistical analysis of this correlation data it has been discov-ered that there exists functional networks in the brain, consisting of areas in the brain thatco-activate during task or resting state. An example of such a network is the Default ModeNetwork (DMN) which is hypothesized to be intrinsic in its functioning[8]. Meaning that itactivates when we reminisce or think about the future. The DMN and other networks areto a varying degree accepted definitions which are spatially spread out and/or clustered inthe brain.

This mapping and analysis of functional connections between brain networks and sub-regions is a way of studying the human functional connectome[27]. Functional connectome isan umbrella term for how all brain networks are connected with each other. It is importantto note that on the one side the brain consists of thousands upon thousands of networks,

11

Page 20: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

12 Chapter 3. Functional Connectivity and Current Visualization Solutions

but in order to practically conduct research in this area the number of networks researchersuse in their definitions varies depending on the objective of the neuroscientific research.

3.1.1 Resting and Working State

As briefly mentioned above functional connectivity experiments usually measures the ac-tivity of the brain during either resting state, working state or both. During resting stateexperiments, the test-subject is instructed to ”rest” by lying down in a brain scanner, think-ing of nothing in specific, while looking up at a cross above them attached to the scanner.Working state is when brain activity is measured during various behavioural conditions.An example of such a task is the n-back test which the UFBI researchers has used whencollecting the data that this thesis has been exploring [15].

The n-back test has been developed to target the working memory and is used to measurewhat networks that are functionally connected during working memory. The n-back taskis defined by that a test subject is presented with a sequence of stimuli, and whenever thecurrent stimuli is the same as n-steps back the subject is instructed to note it. The difficultyof the task is regulated by varying the number of steps n [11].

3.1.2 Overview of the Brain’s Anatomy

As a part of this thesis is to explore how functional connectivity data can more pedagogicallybe visualized together with how it is anatomically situated it is important to know the largerstructures that exists in the brain. The brain is immensely complex and fully presenting itis out scope of this thesis. However, a brief overview of the brain’s anatomy is given.

The brain has been found to have anatomically separated structures with separate mainfunctions[18]. However, functional connectivity research [34] has shown that these struc-tures and brain networks have emergent properties and that different components of thebrain interact with each other creating networks that are spatially separated. During ourdevelopment, from our early years to adulthood, our brain develops both functionally as wellas anatomically. It is during our early years that most anatomical development happens.In its fully developed state the brain consists of five divisions called the Telencephalon, Di-encephalon, Mesencephalon, Metencephalon and Myelencephalon. All of these divisions arefurther subdivided into anatomical regions such as the cerebral cortex, cerebellum, hippo-campus and more. Fig 3.1 is a sagital slice view of the brain and portrays a few of theseanatomical regions and gives a sense of how the brain is structured.

3.1.3 Data structure

Powers et al. [19] describes the brain as a graph with sub graphs in varying hierarchies andhas developed a technique to extract a graph version of the brain consisting of 264 nodesdivided into 14 sub graph network classifications as shown in the list below. Fig 3.2 portraystwelve of these networks. The network icons have been created using an imaging tool calledMricron 1.

1. ’Sensory/somatomotor Hand’

2. ’Sensory/somatomotor Mouth’

3. ’Cingulo-opercular Task Control’

1https://www.nitrc.org/projects/mricron The website for Mricron. Accessed 2017-08-07

Page 21: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

3.1. Overview of Functional Connectivity 13

Figure 3.1: A sagital slice of the brain portraying a few anatomical regions. Acquiredfrom https://tl.wikipedia.org/wiki/Utak#/media/File:NIA_human_brain_drawing.

jpg Accessed 2017-10-25

4. ’Auditory’

5. ’Default mode’

6. ’Memory retrieval?’

7. ’Visual’

8. ’Fronto-parietal Task Control’

9. ’Salience’

10. ’Subcortical’

11. ’Ventral attention’

12. ’Dorsal attention’

13. ’Cerebellar’

14. ’Uncertain’

As mentioned in section 3.1 the number of networks that are defined during a fMRIproject varies and therefore a visualization tool should be general in how many nodes andedges it can visualize. However, in order to make the project feasible there has to be alimitation on the number of data points as a visualization of 100 data points vs 10000 datapoints varies greatly in how it can and should be portrayed.

The data sets that were used as a basis for the design and user evaluation in this thesisfollows Power et al.’s [19] classification. Albeit, the UFBI research team has altered the

Page 22: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

14 Chapter 3. Functional Connectivity and Current Visualization Solutions

Figure 3.2: The 12 networks of interest for the visulization case. The names for each iconcan be found in the list under section 3.1.3 with the corresponding number.

classifications and averaged all nodes belonging to the same network into one value. The datasets therefore consists of 14 networks, each with a correlation with each other, with a total of196 connections. These connections can be classified into intra and inter correlations. Whereinter correlations are the correlation of one network to another and intra correlations is thecorrelation of a network with itself. As the correlation values are derived from statisticalformulas all correlation values also have a corresponding significance value.

Each data set has been collected from 181 healthy subjects during resting state andn-back working state from a UFBI research project called Cobra. More details about thedata collected for this study is available in Nevalainen et al. [15] The format of the dataused for the final representation is a graph representation of brain connectivity consistingof 12 pre-defined networks (the cerebellar and uncertain categories were further left out asthey were not of interest for the researchers). The edges between two different networks(inter-correlation) and one and the same pre-defined networks (intra-correlation) have beenaveraged respectively.

3.1.4 Domain Specific Tasks

In order to design the interactive visualization there was a need to know what kind of tasksthat neuroscientists, and especially the scientists at UFBI, typically wanted to achieve whenvisualizing their results. For example, what kind of hypothesis did they want to answer andconsequently, how and what information should be portrayed. Alper et al. [2] conducteda task analysis by interviewing neuroscientists and based on the interviews defines a set oftypical tasks when viewing functional connectivity data. The task analysis resulted in sevendifferent tasks which shows that comparison of different data sets is the most frequent andpossibly most important task in the current state of the field. Based on this task analysisa discussion was held with the researchers whereas it was concluded that two main cases of

Page 23: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

3.1. Overview of Functional Connectivity 15

(a) Weighted graph(lower) and un-weighted graph(upper).

(b) Adjacency matrix

Figure 3.3: Illustrations for weighted and unweighted graphs as well as an adjacency matrix.

comparison were of interest. One being the comparison of resting state vs working state andthe other the comparison of different age groups. While working and resting state comparestwo data sets, the comparison of age groups can involve two or more data sets.

3.1.5 Existing Visualizations of Functional Connectivity

Networks and the visualization of networks have been studied extensively and consists of twomain parts. Nodes and edges. In functional connectivity, brain networks corresponds to thenodes and the correlation value between two networks represents the edges. Conventionalgraph plots of networks are visualized as nodes with edge-lines between them or as adjacencymatrices [23]. A problem with network visualizations is that their legibility relies on theamount of data to be presented (in terms of number of nodes, edges as well as the densityof a graph). Otherwise, they can quickly become what is popularly called unintelligiblehairballs.

Allen et al. reviewed the visualization practices in neuroscience in their article Datavisualization in the neurosciences: overcoming the curse of dimensionality [1]. Their mainfindings were that 3D visualizations has more information loss than 2D. This aligns withwhat Edward Tufte [29] writes regarding how 3D visualizations distorts data and makes itharder to portray. Although, in the end both the 2D and 3D visualizations were deemedto lack information and thereby also data integrity. They furthermore include a table ofproblem areas as well as suggested solutions that can be used to evaluate neuroscientificgraphs.

Similar to Allen et al. [1], Marguilies et al. [14] reviews and exemplifies the techniquesused to visualize functional connectivity between 2005 and 2012 and discusses implicationsbetween the transition and combination of the functional connectome with the anatomicaland spatial properties of the brain. It is shown that researchers in the field of connectivityhave collectively gone from portraying functional networks as graphs inside of a volumetricvisualization or two dimensional slice of the brain to a more connection centric perspective(see fig 3.4). No clear conclusion as for what the best visualizations are given. Rather, it is

Page 24: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

16 Chapter 3. Functional Connectivity and Current Visualization Solutions

Figure 3.4: Image depciting the transition from anatomical to connectional imaging forfunctional connectivity. Retrieved from Marguilies et al. [14]

described how various visualizations has trended and been used throughout the time spanof the reviewed articles. In an attempt to add on to Marguilies et al.’s list and look at whichtechniques that has been used in recent years, as well as earlier ones, a list of visualizationsin addition to the ones that Marguilies et al. discusses is presented below. The review isnot exhaustive. However, it covers recent visualizations gathered from scientific databases.Together with Marguilies et al.’s findings six main groups of visualizations were identified.Both used in combination with each other as well as standalone.

1. Presenting connectivity matrices as they are with the correlation values in the cellscolor coded according to a defined color scale. Additionally, the various brain ar-eas/regions of interest that the cells correspond to are presented with a textual repre-sentation or categorization of multiple nodes. [2, 5].

2. Plotting the functional network as an undirected weighted or unweighted graph. Wherethe nodes in the graph represents the networks of interest that measurements have

Page 25: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

3.1. Overview of Functional Connectivity 17

been taken from. In weighted graphs the edges not only represent a connection be-tween nodes but also the strength of it [2, 14].

3. Using the volume of the brain and plotting networks as a weighted or unweightedgraph inside of the brain and placing nodes in the form of dots in accordance to thexyz-coordinates of the regions of interest and then drawing lines between them todepict correlation values [14].

4. Plotting functional networks as a graph with a 2D slice of the brain as spatial con-straints. For example, using a sagital or coronal plane slice as a silhouette to commu-nicate a sense of the position that nodes have in the brain [14].

5. Through a variant of a circledigram, called a connectogram as presented by Irimiaet al.[10]. Where sections of the outer edge of a circle represents a brain area andthe middle part of the circle have been taken out and lines are drawn between thesections to represent connectivity values. Furthermore, in some articles anatomicalreference images have been added to the connectogram next to each section [5, 26].The images used are fMRI images that highlight the spatial locations of certain brainareas/networks.

6. Using gray scale fMRI images directly and coloring the activity which corresponds tonetworks of interest [26] (see fig 3.2 for an example).

None of these visualization categories have been found to, or can be viewed as the oneperfect solution when visualizing functional connectivity. Rather, they complement eachother and each of them can be valuable as a visualization. Just as it in Shneiderman’sand Tufte’s [23, 29] guidelines is proposed to offer multiple views of a data set to allow fordifferent kinds of insights.

At the end of Irimia et al.’s article[10], where the connectogram is presented, they com-pare previous visualizations and discuss their strengths and weaknesses. They argue thatVisualizations of smaller networks are inherently more legible as visual clutter is reducedand reading for example the edge strength gets easier. This, can intuitively be understoodby considering that more data points requires more space and therefore would more easilycreate a visually cluttered view. In for example fig 3.4 the Botger graph depicts a graphdrawn within a lateral slice of the brain. It is possible to visually appreciate where thenodes are located in the brain. Atleast, in a horizontal (X,Y) perspective. However, as thenumber of vertices grows the spatial environment becomes cluttered and intricate detailsgets lost. Albeit, even a cluttered view can allow for a visual appreciation of what nodesthat are more or less connected and, as Elmqvist et al. discusses [7], can have value as avisualization depending on what context that the information is to be presented in. Forexample, if it is to get a sense of what nodes that have the largest degree of centrality inthe network, portraying all connections might be useful.

3D solutions offers the possibility to show functional data directly inside of the volume ofthe brain, thereby allowing for the interpretation of where the data is situated in the brain.The limitation is that, as mentioned, the data gets distorted [29] due to how 3D perspectivesare perceived. And when the number of visualized nodes increases, the visualization becomesmore and more cluttered and information that is to be presented gets lost behind informationcloser to the viewer. Making it a questionable format when publishing in journals.

In graph visualizations of functional connectivity, both when they are plotted on paperas 3D volumes as well as in 2D graphs, there are limitations as for how we can detectdifferences in edge strength when they are encoded by size. In 3D, when data is encoded

Page 26: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

18 Chapter 3. Functional Connectivity and Current Visualization Solutions

in terms of size as a volume, the eye can easily get tricked [30] and in the end the attemptto visualize edge strength distorts the data. In 2D it is easier to appreciate the strength ofa connection when it is encoded through the size with for example the thickness of a line.However, the problem of visual cluttering resides in 2D visualizations as well.

Page 27: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 4

Methodology

In this chapter the design methodology of the thesis, a combination of the iterative Design-Build-Test cycle[35] and the data visualization design process from Data Visualization: asuccessful design process [12], is described. First, the overarching procedure is introducedand thereafter how the methods have been combined as well as how each step of the iterativecycles has been performed.

4.1 Design-Build-Test and a Data Visualization Process

Historically designing and designers in general has had, and still have, very fluid and in-dividual methodologies. The process has in many cases been viewed as an enigma wherecreativity, inspiration and intuition is what drives it rather than a fixed pattern. However,even though it can be hard to pinpoint in detail what it is that allows an individual orgroup of individuals to produce successful designs. Iterative processes such as the engineer-ing design cycle called Design-Build-Test (DBT) as introduced by Steven C. Wheelwrightand Kim B. Clark [35] have proven to more reliably assist in producing successful results.

DBT is an iterative design method where the whole process of designing, implementingand testing consists of several cycles. Where the aim is to go through each step repeatedly,and produce a set of prototypes in various degree of fidelity in order to explore alternativesand improve the quality of the end result [21]. Thereby, avoiding the case where only thefirst good idea gets implemented and instead being more exhaustive to cover as many ideasas possible (within the time frame of the project).

In order to be time efficient and exhaustive during idea generation DBT advocates thatthe designer filters through ideas by producing lo-fi sketches and prototypes to quicklyevaluate, discuss, test and refine ideas [21]. Thereafter, the ideas that are deemed to bebetter gets implemented in the form of a hi-fi prototype where more intricate details of theidea can be implemented and further tested.

DBT is a general design methodology that is often applied in human computer interface(HCI) design projects. In order to better adapt the DBT methodology to this thesis, inspi-ration has also been taken from the design process that is introduced in Data Visualization:a successful design process [12]. Both of the methodologies are similar as they are bothiterative. They were combined as we previously had experience of the DBT process andwere therefore confident in it as an iterative structure while the second one was used toguide the design project’s visualization aspect in more detail.

In Data Visualization: a successful design process, five crucial steps are outlined.

19

Page 28: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

20 Chapter 4. Methodology

1. Setting the purpose and identifying key factors

2. Demonstrating editorial focus and learning about your data

3. Conceiving and reasoning visualization design options

4. The visualization anatomy – data representation

5. Creating interactivity

6. Annotation

The first step, defining a purpose, was already defined at the start of the thesis and willnot be touched on any further. The second step has been an ongoing procedure throughoutthe thesis but was mainly approached during the literature review. The last four steps wereaddressed in an iterative fashion as part of the iterative cycles.

4.1.1 Literature Review

While a literature review is important in any project and subject to get an understandingof the theoretical underpinnings, one of the most important aspects in the field of datavisualization has been argued [12][30] to be to get to know ones data and thereby betterunderstand its possibilities and limitations. While the results of the literature review inthis thesis are presented in chapters two and three a short description of how it was done isgiven below.

As a first step in understanding the subject area and thus the data an informal pre-studywas performed to get a better appreciation of what the field of functional connectivity con-sists of as well as to better understand the underlying data to be used for the visualization.After getting a basic understanding the informal study proceeded into a more formal litera-ture study during which relevant research papers in the area of functional connectivity wereidentified and a mapping of existing visualizations was conducted.

Furthermore, informal meetings were repeatedly held with the researchers to better un-derstand the data, analysis process and general aims with the research. Working closelywith the researchers allowed for a better understanding of the data and therefore betterclarity on how to design. Tufte warns [[30], p. 80] that the reason that graphical represen-tation often distorts the underlying data is due to a lack of knowledge from the designer.Therefore, working closely with the researchers was done in an effort to lessen the risk ofnot keeping the integrity of the data.

4.1.2 Design - Establishing Goals and Requirements

The first step in establishing the goals was described under the Literature review sectionabove. After that the network representations were set out to be chosen. The set of alter-natives to choose between were based on the types of visualizations found in the reviewedfunctional connectivity publications in section 3. However, as the choice of representationsalso depended on the functionality of the tool, semi-structured interviews and brain storm-ing sessions [20] were held together with UFBI researchers in order to explore what functionsthat were needed. The following questions were used as a basis:

1. What questions does the researchers want to explore/present with the visualization?

2. At what resolution or combination of resolutions should (needs) the data be portrayed?

Page 29: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

4.1. Design-Build-Test and a Data Visualization Process 21

3. What aspects of the resulting data are the most important? What data is relevant?

4. What functions should the tool have?

4.1.3 Build - Implementation of Prototypes

The prototyping was divided into three phases. The phases portrays how the prototyping hasbeen conducted throughout the thesis but have not been performed strictly chronologically.

1. Lo-fi prototyping

2. Mid-fi prototyping

3. Hi-fi prototyping

During the first phase brainstorming sessions were held and various lo-fi prototypes wereconceptualized with paper-sketches to explore concepts and quickly evaluate ideas based onthe interviews and literature study. The various concepts were discussed together with theresearchers. No formal evaluation techniques were used during these evaluations. Rather,the meetings were kept short to quickly gauge the usefulness that they as domain expertssaw in the various concepts.

In the second phase the concepts deemed as more promising during the lo-fi prototypingwere further explored in the form of mid-fi prototypes. The mid-fi prototypes were concep-tualized in the prototyping tool Axure 1 where color coding and interactivity of the datarepresentations were further explored.

After exploring concepts at a lo-fi and mid-fi level a vertical interactive hi-fi prototypewas implemented. The hi-fi prototype was also implemented with Axure with additionalfunctionality added through the JavaScript library Google charts 2 in combination withGoogle Sheets 3 in order to allow for an evaluation of the visualizations against real data. Thedata driven prototype was developed to explore the interaction, filtering and functionalityconcerning the data sets in more depth. The concept was further horizontally exploredwithout interactivity for cases with more data sets than two. Albeit, this part of the conceptwas in the end never tested with users.

4.1.4 Test - Testing and Evaluating

Two kinds of qualitative tests were used for evaluation during this thesis. Continuous testswith lo-fi and mid-fi prototypes and concepts and a final set of test where the think aloudmethod [20] was used to evaluate and test a vertical hi-fi prototype. Quantitative user tests,such as A/B testing [20], were left out as they were deemed too demanding for the size ofthis project and more rapid testing were needed.

Continuous Evaluation - As the UFBI researchers were both considered as users andcollaborators they were involved in the design process throughout the project by continu-ously evaluating ideas, designs and prototypes as the work progressed. Bang Wong arguesin his article 4 that visualization designers working with scientific data should incorporate

1https://www.axure.com/ Website of axure prototyping tool. Acessed 2017-08-152https://developers.google.com/chart/ Website of Google Charts. Accessed 2017-08-073https://www.google.com/sheets/about/ Website of Google Sheets. Accessed 2017-08-074http://www.nature.com.proxy.ub.umu.se/nmeth/journal/v9/n12/full/nmeth.2258.html An article

discussing the use of engaging researchers in the development of visualization tools Accessed 2017-08-07

Page 30: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

22 Chapter 4. Methodology

the knowledge of the domain experts (scientists) in order to make sure that the visualizationstays true to the intent of the data, and further on the research questions as a whole. Theseevaluations were held at a week by week basis depending on how the work with the designand prototype went. The tests did not follow a strict layout but were inspired from Nielsen’spaper on discount usability studies [16]. The article discusses the use of user studies thattakes a short amount of time. Exemplifying with narrowed down prototypes and heuristicevaluations against guidelines (In this case the guidelines evaluated against were the onesmentioned in the theoretical framework).

Think Aloud - Evaluating with Real Data In order to get higher ecological validity[20] for the final test of the visualization concept a ”Think Aloud” 5 [20] case study withtwo data sets from a concurrent UFBI research project was performed on an interactivehi-fi prototype. Three test participants from the UFBI research team were recruited. TheNielsen Norman group recommends that around five users 6 should be tested in order to getas much value as possible out of each user test. Kruger [13] further argues that the threefirst users will find the most significant problems.

A set of seven tasks based on how the functionality, compare and perceive the visualiza-tion, of the tool could be explored were defined for the test case of each data set. The sametasks were used for both data sets. The participants were informed beforehand regardingthe nature of the data and were instructed to say out loud what they were thinking as well asexpecting to happen when performing the tasks. Besides a short introduction to what kindof data that was being visualized no further guidance were given for the tool. Each sessionwas audio recorded and an observer took notes throughout. For a complete description ofthe user instructions see appendix A.

After all tasks had been performed with both data sets an open-ended interview [20](p.228) was held with each participant. Where problems noted by the observer as well asthe participant’s’ overall sense of the tool were discussed. Following after the interviewsthe participants were instructed to rate the tool against the seven factors of Stephen Few’sData visualization Effectiveness Profile. 7.

The tasks were as follows:

1. Enter the visualization from the first page.

2. Find the exact value of the subcortical network’s intra correlation.

3. Give an approximation of the difference between resting state and active state of thatnetwork’s intra correlation.

4. Find the top three networks with the strongest inter-correlations.

5. Change the visualization so that only two network’s inter-correlations are shown inthe circular representations.

6. Save the visualization.

5https://www.nngroup.com/articles/thinking-aloud-the-1-usability-tool An article describingthe use and process of performing Think Aloud usability tests. Acessed 2017-08-06/

6https://www.nngroup.com/articles/how-many-test-users/ An article discussing the number of usersto test when running user tests. Accessed 2017-08-04

7https://www.perceptualedge.com/articles/visual_business_intelligence/data_visualization_

effectiveness_profile.pdf A newsletter introducing and describing a visualization critique profile.Accessed 2017-08-04

Page 31: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

4.1. Design-Build-Test and a Data Visualization Process 23

7. This isn’t a particular task. Instead you are free to explore the visualization as youwish.

In the second trial, with mean connectivity data, you will start out in the visualizationview.

Page 32: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

24 Chapter 4. Methodology

Page 33: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 5

Results

This chapter is divided into three main parts. The first one describes the final prototypeand concept that was tested with real data, the second one is a design rationale for the finalconcept and the last section describes the results from the final case study user tests.

5.1 Final Concept

The final concept consisted of two pages as shown in fig 5.1. The first page 5.1a was theindex page where the tool was introduced and data in some file format (the file formathas not been defined in this thesis) can be uploaded. The second page 5.1b was wherethe data was visualized and consisted of two main representations: a variable number ofconnectograms as well as a grouped bar chart adapted to the number of connectograms.The visualization view worked as a single page application, meaning that a user stayed onthe same page and never had to move between two different pages when exploring a dataset.

Any uploaded data was represented at full with one or several connectograms as por-trayed in fig 5.4 and 5.6. The nodes in the connectograms were represented with a superiorslice view of the brain where the coordinates of nodes in that network had been enlargedaround the pixel coordinate and qualitatively colored. Intra-correlation edge values wereshown as a split circle attached to the outside of each node while inter-correlations wererepresented by lines drawn between the nodes within the circular white area. Furthermore,the edge values were color coded with a divergent color scale adapted to the highest andlowest edge values in an uploaded data sets. For example, as shown in fig 5.3 the scale wasadapted to the data set to the typical range of connectivity values. There were five negativeand five positive steps for how the edges were color coded.

When a network node in one of the connectograms was clicked the exact edge valuesfor that node and corresponding nodes in other connectograms were shown in the groupedbar chart below the connectograms. Any interaction with one of the connectograms wasmirrored in the others. In other words, the groups included in the bar chart were decidedby the number of networks uploaded or currently showing. For example, in fig 5.1b the redgroup was resting state data and the blue group was working state data. If another dataset would be showing as a connectogram a third group would also show in the grouped barchart as in 5.6b. Thereby, only one group of network edges were shown at the same time inthe bar chart. Additionally, clicking a node would hide or show the edges of that network(see fig 5.2 for an example) making it possible to visually filter the connectograms.

25

Page 34: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

26 Chapter 5. Results

(a) Index page (b) Data view page adapted to a mean connectiv-ity data set.

Figure 5.1: The Index page and one of the visualization views of the interactive prototypeused for the case study.

Figure 5.2: A connectogram where all edges but the memory retrieval’s have been removednetwork.

Page 35: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

5.2. Design Rationale 27

Figure 5.3: A heatbar showing the correlation thresholds used for the edges ni the connnec-togram and a menu for hiding and showing networks.

Figure 5.4: Two Connectograms adapted to resting and working state of mean connectivitydata used in the final concept.

The white square with two circular upper icons in 5.3, with a one and a two in them,represented a menu for hiding and showing uploaded data sets. For example, if a user hasuploaded three data sets on the index page, as in 5.6b, and only wants to look at two ofthe networks then a user clicks either of the icons to hide that network. Finally, in 5.1b thebutton in the upper left corner was a link to the index page and the button in the upperright corner was a button to save the current visualization.

5.2 Design Rationale

At the start of the design phase various design concepts which combined different networkrepresentations in order to portray several sides of one and the same data set were explored.After discussing the first few prototypes it was concluded that it was the comparison oftwo or more data sets that was the most interesting high level task for the researchers.Therefore, focus was drawn away from combining different versions of representations forone set of data and instead put into how multiple networks could be represented togetherand how the comparison of them could be facilitated.

The next few design concepts as well as the final one were, besides the task change,inspired from the concept of multiple representations which Tufte writes about[29]. Hediscusses the use of having multiple smaller representations beside each other instead ofhaving all data in one and the same representation. He exemplifies this with sparkcharts(linecharts) by showing the efficacy of having several smaller charts with only one line

Page 36: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

28 Chapter 5. Results

Figure 5.5: A bar chart showing the resting and working state of the ventral attentionnetwork for a mean connectivity data set. The y-axis stands for the r-value between thex-axis networks and the ventral attention network.

in each chart next to each other instead of all lines within the same graph. However,both based on interviews with the UFBI researchers and the reviewed papers in section 3functional connectivity representations, such as the connectogram and adjacency matrices,were deemed to be lacking in legibility and therefore needed to be enhanced.

5.2.1 Layout

The concept of a single page application is partly based on Tory et al.s and Shneiderman’s[23, 28] guidelines which states that there should be consistency in the dataview as it makesit easier for users to understand the relations between data sets. In addition, the layout wasdesigned with Shneiderman’s ”Overview, zoom and details on demand” mantra in mind.First, an overview of current data sets that have been uploaded is presented. Thereafterthe representation of the whole data sets, and color codes for the data values are shown.Interactivity comes into play when details regarding the strength of correlation values areenhances by presenting them ”on demand” in a grouped bar chart when interacting withthe visualization.

5.2.2 Choice of Representation

The choice of the connectogram as representation is mainly based on two aspects. First, asopposed to matrix representations, it directly portrays the relationship between networkswith lines drawn between the nodes and therefore it arguably follows Tory et al.s [28] rec-ommendations on directly showing relationships. Second, during the prototyping it wasnoticed that it was easier to add graphical representations for the networks (see fig 3.2 forexamples) to the connectogram as compared to an adjacency matrix. The adjacency matri-ces effectively took up too much space as compared to the connectogram when combiningthem with icons for the networks.

A grouped bar chart was used to enhance the comparison task (established way of com-paring values) for the researchers and it also served as a filter by only showing a subset ofthe data. The transitions between what edges the bar chart showed was further animated

Page 37: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

5.2. Design Rationale 29

(a) Data view page for 5 networks.

(b) Data view page with three networks.

(c) Early data view prototype.

Figure 5.6: Mock up views for more than two datasets as well as one of the earlier versions.

Page 38: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

30 Chapter 5. Results

to provide a fluid feel when transitioning between networks and avoiding the case where twoarrays of data looks almost the same and its hard to know whether they change or not.

5.2.3 Colors

Grey is used as a base color to allow for other colors as well as interaction elements notconcerning the visualization to stand out from the background. Black and white was used forinteraction elements outside of the visualization in order to be consistent and have neutralcolors and thereby have free use of other colors in the visualization. (color was already aproblem with the amount of networks). The palettes for the qualitative and divergent colorsets for the heatbar and connectograms were derived from color brewer1.

Having a dynamically scaled color scale for the edge values was deemed necessary aftera first visualization trial of the data sets. As pretty much all data in the test data fell in thesame range of 0 to 0.2 the connectograms became bland, and in a sense non informative,as no differences could be seen. After discussing this with the researchers it was thoughtthat having a slider to change the intervals would be a good solution. This was due to thatdifferences between lower values were still of interest. Adding more colors was not thoughtas a feasible option as it would make it harder to visually distinguish the colors on the colorscale.

5.2.4 Annotations

The design of the annotations in 5.1b were based on that the size and font should be easilyreadable while still being able to fit inside of the visualization. Otherwise, the actual textcontent was adapted from the named categories and networks defined in the data by theresearchers. Considerations were taken on whether or not it was necessary to have both thenetwork image icons and textual representations. As what networks that are incorporatedand what the network icons looks like can vary for different data sets it was deemed necessaryto have both as it could not be assumed that all users would be familiar with all kinds ofnetworks.

5.3 User Tests

Here notable observations drawn during the user tests are first presented thereafter anintegration of all interviews and at last a summary of the resulting visualization profiles.

5.3.1 Observations

All participants successfully finished the test cases. However, during three of the tasksapparent problems or deviations from what was expected were noted and are thereforepresented.

First of, when searching for the three strongest correlations during the fourth task, noneof the participants seemed to use the connectogram to any obvious extent when filteringthrough the data. Instead they clicked through all nodes and memorized the bars that hadthe strongest correlation. Giving their answers when they seemingly were confident enoughin that they had the correct values.

1http://colorbrewer2.org/#type=sequential&scheme=BuGn&n=3 The website of the colorbrewer tool.Accessed 2017-08-07

Page 39: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

5.3. User Tests 31

Secondly, all test subjects were confused during the fifth task where they were instructedto change the visualization so that only two network’s edges were shown in the connec-tograms. They clicked around the prototype and tried to find how they could achieve thetask. Two of the participants managed to figure it out on their own after a while and athird one was guided to the correct interaction. However, this only happened when the firstdata set was tested and all users completed the task without hesitation when performingthe same task with the second data set.

Thirdly, when trying to find the three strongest correlations all participants refrainedfrom using the connectogram. While they acknowledged it as a possible solution they feltmore confident in looking up the exact values in the grouped bar charts.

Additionally, during the final task, where they could freely explore the data, two of theparticipants expressed a wish to be able to filter the data by what edges that were significantor not. Two of them also tried to drag the black bars on the divergent color scale in anattempt to filter the data. One participant wanted clearer indication for selected networksas the black squares were thought to be too inconspicuous. Furthermore all the participantstried to interact with the edges within the connectogram during atleast one of the taskswhere they had to compare or find a specific edge value.

5.3.2 Interviews

Overall all of the test participants regarded the tool as useful. After being asked aboutusing the connectogram as a first filtering device the users acknowledged the connectogramas a possible solution. However, two said that they felt more confident in looking up thevalues in the grouped bar chart as they could see the exact values. One further commentedthat it could be due to having more experience with bar charts and that if they had moreexperience with the tool they would probably use the connectogram more.

When asked about task number five, where the users had trouble hiding and showingcertain networks’ edges, they did not think it would be a problem in further use. Two ofthem felt that even though the interaction was not clear at a glance it was still useful andeasy to learn. Furthermore, they thought that if a short introduction or guide was availableat the first page regarding how it worked it would not be an issue.

When asked for further functionality all researchers wanted the ability to filter the edgesthrough the heatbar (level of correlation) as well as filter the edges by significance. Onealso noted on the use of different annotations for the same data in the bar graphs versusconnectograms. Preferably he would see the same annotations for both. Furthermore, onenoted that it would be good to have the ability to completely remove a network if it was notof interest and thereby have the option to have a more focused visualization. For example, ifthey would come to the conclusion that one of the networks were not of interest they wouldwant to remove it.

5.3.3 Visualization Effectiveness Profiles

In fig 5.7 the resulting profiles that the researchers filled in are presented.

Page 40: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

32 Chapter 5. Results

Figure 5.7: The visualization profile results. Each of the colored shapes rep-resents one researcher’s critique. Base figure acquired with permission fromhttps://www.perceptualedge.com/articles/visual_business_intelligence/data_

visualization_effectiveness_profile.pdf Retrieved 2017-08-10.

Page 41: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 6

Discussion and Conclusions

In this thesis the design of an interactive web based visualization for functional connectivityhave been explored. In specific, we have looked at how existing visualizations of functionalconnectivity data sets can more easily be interpreted as well as compared with the aid ofinteractivity. In this chapter the resulting concept, users tests and limitations are discussedin relation to the reviewed literature in chapter two and three. Furthermore, based on thediscussion, conclusions regarding the results are drawn.

6.1 User Test Analysis

In the article where Stephen Few proposes his visualization effectiveness profile he providesrecommendations for what evaluation level visualizations in general should aim for, in termsof seven factors. When comparing these recommendations (see fig 5.1) with the resultingprofiles in fig 5.7 it can be seen that completeness, truthfulness and intuitiveness were allscored below or on the edge of what Few recommends. Based on the user interviews themain issue with the intuitiveness of the concept in fig 5.1bwas the double interaction ofboth showing values in the bar chart and hiding and showing a network’s edge. The userssimply did not notice that a network was removed as they clicked through and the barchart slid up at the same time. When starting in the original view 5.1b and all edges wereportrayed, removing the first one to three networks was hard to notice or pay attention to.Since all networks were connected to each other removing the first network did not removeits connections as the other ones were also connected to it. However, as this interaction wasnot a problem during the second test case and that the users regarded it as easy to learn itcould be argued to be a non problem. Although, a possible improvement would be to add avisual indication for nodes that became ”inactivated” by e.g greying them out at the sametime as their respective edges were removed. Additionally, the problem could be attributedto functionality. As the prototype was set up in a way so that each node had their ownconnection to all other nodes removing the first one, as mentioned, still left the connectionsthat belonged to the other nodes. Having it so that all edges connected to a node wouldbe hidden when clicked could arguably be a more natural interaction. However, this wouldmake it harder to create visualizations that shows edges for a few networks.

When it comes to the lackings in the factors of truthfullness and completeness the faultscould be attributed to limited functionality of the prototype. An integral part of the dataare the significance values and not showing them lessens the truthfullness and completenessof the visualization. During the interviews all of the researchers expressed that they wanted

33

Page 42: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

34 Chapter 6. Discussion and Conclusions

Figure 6.1: Stephen Few’s recommendation on what visulizations should aim for interms of his seven factors of visualization effectiveness. Acquired with permission fromhttps://www.perceptualedge.com/articles/visual_business_intelligence/data_

visualization_effectiveness_profile.pdf Retrieved 2017-08-10

to filter the data by significance values as well as to threshold what edge values that wereshown. These functions had already been defined before the tests, although, there wasnot enough time to implement them. Therefore, the researchers were instead interviewedregarding how they would like the function to work and what kind of interaction that theyexpected. One of the researchers commented that it was of value to see the raw data at firstand then being able to filter by significance. Thus, it can be argued that the tool shouldnot filter the values automatically. Instead it should be possible to explore the data withthree settings. (1) all of the raw data is shown as it is, (2) only the significant values areshown and (3) all of the data is shown and an indication is given for significant edges.

The tested researchers rated and expressed that they felt engaged in the data when usingthe tool. It could be hypothesized that this lessens the risk of missing interesting detailsor making faulty conclusions and aesthetic values can increase retainability as Norman [17]has shown, which arguably is an integral part when exploring data and trying to synthesizeinformation in order to draw conclusions.

Page 43: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

6.2. Limitations 35

6.2 Limitations

There are a number of limitations with this project due to its limited time span and otherfactors. First off, the evaluation of existing visualizations did not have a strict methodologyand was loosely based on the reviewed guidelines. As is discussed in section 3.2 differentrepresentations have different strengths and it could be useful to be able to switch between,for example, having matrices and connectograms. To reach a decisive conclusion as forwhether any representation is the best in this scenario a more valid method for choosinga representation would be to conduct comparative tests of the representations, such as theone in Alper et al. [2].

Another limitation was the number of users tested in the final test. As user tests theygave useful insight. However, due to the low amount of users tested, the generalizability ofthe results is hard to argue for. All the while, an argument can be made that as UFBI isat the forefront of fMRI research their critique and opinions gives extra value to the usertests. However, as this is a case of call to authority it does not make the results decisive.Both evaluating the experience of a possible reader (non domain expert) and the pedagogicalcontent creator was challenging and in the end it was skipped as we could not find a suitablemethod.

Furthermore, the visualization effectiveness profile has not been validated in a contextsuch as this thesis before. While we argue it to have been a useful resource it is still anuntested methodology and there is a risk for bias in the results as the designer was familiarwith the tested users.

Another limitation or factor that have not been discussed to any length in this projectis the use or role of screen real estate. What quality and size of screen that is used plays alarge part in visualizations as degraded quality or too small visual elements can effectivelymake it useless.

6.3 Methodology development

There is a lack of tested methodologies for conducting user tests with visualizations in ashort term scope. The chosen methodology of working close with the researchers, conductingthink aloud case user studies with real data and then using Stephen Few’s visualizationprofile as a further evaluation tool was successful for this project. For example, in additionto allowing for a bench mark as Few proposes the profile also got the domain experts tothink and evaluate in visualization terms, leading to a better discussion that cleared outproblems with the tool. It could be argued that the profile is to be used as a critique toolfor visualization experts. Albeit, it further seems viable as a tool to allow non-visualizationexperts to evaluate with as well. Even though this project does not give a definitive answer asfor whether or not the approach used would be successful for other projects it can be used asa basis for future similar projects. What could be further worth to consider is long term casestudies such as the one proposed in [24]. As it might be necessary in order to reliably evaluatethe effectiveness in terms of how successful researchers are in getting insights regarding theirdata. However, as methods and research questions within neuroscience are constantly proneto changes long term case studies might be too slow of an approach.

”Currently, many researchers evaluate visualizations based on the time it takes to com-plete a task and the response accuracy. However these metrics are not relevant for un-derstanding stories. Meaningful story metrics include engagement and interest, ability toremember key points, information provided to make more informed decisions, and so on.”-Segal et al [22]

Page 44: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

36 Chapter 6. Discussion and Conclusions

In regards to the qoute above, using real data for the evaluations seemed to naturallyengage the researchers during the user tests. Thereby, the visualization and exploration con-sequently was of interest for them. Possibly leading to more reliable evaluations. However,this might at the same time bias the evaluation of the tool.

6.4 Future Work

When it comes to taking the concept further, a first step would be to iterate the conceptonce more and adapt it against the feedback given during the evaluation. After that the con-cept could be used as a basis for a programatical implementation with the help of javascriptvisualization libraries such as d3.js 1 and google charts so that the visualizations automat-ically adapts to a given data set. A recommendation for an intermediate step would be todefine a file format according to the standards that functional connectivity researchers useso that their data can be automatically parsed. Furthermore, as it was the visualizationthat was in focus in this project the concept as a whole was not explored. Work regardingfile format, how to save, possible log in and database management for files and users couldhave a significant impact on the layout of elements outside of the visualization view, such asthe index page. The current version of the Index page was not tested fully. However, as perthe request of two of the participants a guide on how the visualization worked at the startwould be good. Furthermore it could be good to separate the act of uploading data fromthe first page into a separate one so that the user flow would have a better structure. Forexample: Enter the first page, read the guide, press an arrow or a start button and enteran upload page to upload data. Finally, press the visualize button to enter the visualizationpage.

As this work only tested and covered the case of having two data sets in the same viewfurther work is also needed to determine where the limitations for how many and how largedata sets that are feasible in one and the same view. The size of chosen representationsis arguably a big issue for visualizations. How many data sets and number of nodes andedges that can be incorporated in a visualization while still remaining visually discernableis something that is left to be answered. A possible way to enable larger data sets couldbe to combine this design concept with visual sorting algorithms, such as hierarchical edgebundling [9].

As is discussed in [14] a future solution to problems with static visualization couldlie in the use of interactive visualizations that allows researchers to show more of theirdata. Visualizations that are ranked low in Mcgill et al’s[6] ranking can be enhanced bycombining them with visualizations that are ranked higher. For example, in this thesisthe connectogram which uses the low ranked color coding for edge values was enhanced byadding a barchart which uses positions along a common scale that are lacking in a certainarea. However, this offers new challenges in terms of how the data should be narrated[22]. One approach could be to change the format of the traditional article by involvingthe interactive visualization directly inside of it. Another approach would be to adapt thetool so that it can have a more detailed narrative built into it. The resulting concept inthis thesis could be used as a basis for such a study or project. For example, by adaptingit or using it in conjunction with an article published in an online journal or by having forexample a QR-code in an article which can be accessed through a smart phone when readingit in paper format.

1https://d3js.org/. Website of the javascript visualization library d3.js. Accessed 2017-08-14

Page 45: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

6.5. Conclusions 37

6.5 Conclusions

In this thesis a concept of a functional connectivity visualization tool was developed. Basedon high scores on several factors on a visualization evaluation test with domain experts andas by interpretation of user test interviews it can be concluded that combining full scalerepresentations such as the connectogram with dynamically adapted grouped bar graphsto enhance decoding allowed for an easy interpretation and comparison of tested data sets.Additionally, according to all tests by domain experts the resulting concept was perceived asan engaging and useful tool for exploring functional connectivity data sets. In regards to theconclusions above the aims listed in the introduction, of creating an easy to use concept fora visualization tool that allowed for comparison of data sets for researchers, were achieved.While further work is still needed to finalize it, the concept could be implemented as is andprovide a valuable way of interacting with and exploring functional connectivity data sets.

Page 46: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

38 Chapter 6. Discussion and Conclusions

Page 47: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 7

Acknowledgements

I want to thank all of the researchers, personnel and my supervisors, Lars Nyberg andAnders Wahlin, at UFBI that has helped me throughout the thesis with concept evaluation,idea generation and much more. A special thank you to Anders, my main supervisor atUFBI, for always showing up with a smile and being enthusiastic and helpful with invaluablefeedback whenever needed. I would like to thank my peer reviewers Joakim Ljungren andMattias Edin for their healthy perspective, feedback and interesting discussions. Also, thankyou to my supervisor at Umea University, Kalle Prorok, for your encouraging words andenlightening insights.

A final thank you too my family and friends for encouragement and for just being there.

39

Page 48: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

40 Chapter 7. Acknowledgements

Page 49: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Chapter 8

Bibliography

[1] Allen, E. A., Erhardt, E. B., and Calhoun, V. D. Data visualization in theneurosciences: overcoming the curse of dimensionality. Neuron 74, 4 (2012), 603–608.

[2] Alper, B., Bach, B., Henry Riche, N., Isenberg, T., and Fekete, J.-D.Weighted graph comparison techniques for brain connectivity analysis. In Proceed-ings of the SIGCHI Conference on Human Factors in Computing Systems (New York,NY, USA, 2013), CHI ’13, ACM, pp. 483–492.

[3] Aparicio, M., and Costa, C. J. Data visualization. Commun. Des. Q. Rev 3, 1(Jan. 2015), 7–11.

[4] Brewer, C. A. Color use guidelines for mapping. Visualization in Modern Cartogra-phy (1994), 123–148.

[5] Buckner, R. L., Krienen, F. M., and Yeo, B. T. Opportunities and limitationsof intrinsic functional connectivity mri. Nature Neuroscience 16, 7 (2013), 832–837.

[6] Cleveland, W. S., and McGill, R. Graphical perception: The visual decoding ofquantitative information on graphical displays of data. Journal of the Royal StatisticalSociety. Series A (General) (1987), 192–229.

[7] Elmqvist, N., Do, T.-N., Goodell, H., Henry, N., and Fekete, J.-D. Zame:Interactive large-scale graph visualization. In Visualization Symposium, 2008. Paci-ficVIS’08. IEEE Pacific (2008), IEEE, pp. 215–222.

[8] Habeck, C., and Moeller, J. R. Intrinsic functional-connectivity networks fordiagnosis: just beautiful pictures? Brain connectivity 1, 2 (2011), 99–103.

[9] Holten, D. Hierarchical edge bundles: Visualization of adjacency relations in hierar-chical data. IEEE Transactions on Visualization and Computer Graphics 12, 5 (2006),741–748.

[10] Irimia, A., Chambers, M. C., Torgerson, C. M., and Van Horn, J. D. Circularrepresentation of human cortical networks for subject and population-level connectomicvisualization. Neuroimage 60, 2 (2012), 1340–1351.

[11] Kane, M. J., Conway, A. R., Miura, T. K., and Colflesh, G. J. Workingmemory, attention control, and the n-back task: a question of construct validity. Journalof Experimental Psychology: Learning, Memory, and Cognition 33, 3 (2007), 615.

41

Page 50: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

42 Chapter 8. Bibliography

[12] Kirk, A. Data Visualization: a successful design process. Packt Publishing Ltd, 2012.

[13] Krug, S. Don’t make me think!: a common sense approach to Web usability. PearsonEducation India, 2000.

[14] Margulies, D. S., Bottger, J., Watanabe, A., and Gorgolewski, K. J. Vi-sualizing the human connectome. NeuroImage 80 (2013), 445–461.

[15] Nevalainen, N., Riklund, K., Andersson, M., Axelsson, J., Ogren, M.,Lovden, M., Lindenberger, U., Backman, L., and Nyberg, L. Cobra: Aprospective multimodal imaging study of dopamine, brain structure and function, andcognition. Brain research 1612 (2015), 83–103.

[16] Nielsen, J. Usability engineering at a discount. In Proceedings of the third inter-national conference on human-computer interaction on Designing and using human-computer interfaces and knowledge based systems (2nd ed.) (1989), Elsevier ScienceInc., pp. 394–401.

[17] Norman, D. A. Emotional design: Why we love (or hate) everyday things. BasicCivitas Books, 2004.

[18] Pinel, J. P. Biopsychology. Pearson Higher Ed, 2015.

[19] Power, J. D., Cohen, A. L., Nelson, S. M., Wig, G. S., Barnes, K. A.,Church, J. A., Vogel, A. C., Laumann, T. O., Miezin, F. M., Schlaggar,B. L., et al. Functional network organization of the human brain. Neuron 72, 4(2011), 665–678.

[20] Rogers, Y., Sharp, H., and Preece, J. Interaction design: beyond human-computer interaction. John Wiley & Sons, 2011.

[21] Sefelin, R., Tscheligi, M., and Giller, V. Paper prototyping-what is it goodfor?: a comparison of paper-and computer-based low-fidelity prototyping. In CHI’03extended abstracts on Human factors in computing systems (2003), ACM, pp. 778–779.

[22] Segel, E., and Heer, J. Narrative visualization: Telling stories with data. IEEETrans. Visualization & Comp. Graphics (Proc. InfoVis) (2010).

[23] Shneiderman, B. The eyes have it: A task by data type taxonomy for informationvisualizations. In Visual Languages, 1996. Proceedings., IEEE Symposium on (1996),IEEE, pp. 336–343.

[24] Shneiderman, B., and Plaisant, C. Strategies for evaluating information visu-alization tools: Multi-dimensional in-depth long-term case studies. In Proceedings ofthe 2006 AVI Workshop on BEyond Time and Errors: Novel Evaluation Methods forInformation Visualization (New York, NY, USA, 2006), BELIV ’06, ACM, pp. 1–7.

[25] Silva, S., Santos, B. S., and Madeira, J. Using color in visualization: A sur-vey. Computers Graphics 35, 2 (2011), 320 – 333. Virtual Reality in Brazil VisualComputing in Biology and MedicineSemantic 3D media and contentCultural Heritage.

[26] Smith, S. M., Nichols, T. E., Vidaurre, D., Winkler, A. M., Behrens, T. E.,Glasser, M. F., Ugurbil, K., Barch, D. M., Van Essen, D. C., and Miller,K. L. A positive-negative mode of population covariation links brain connectivity,demographics and behavior. Nature Neuroscience 18, 11 (2015), 1565–1567.

Page 51: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

43

[27] Sporns, O. The human connectome: Origins and challenges. NeuroImage 80 (2013),53 – 61. Mapping the Connectome.

[28] Tory, M., and Moller, T. Human factors in visualization research. IEEE Trans-actions on Visualization and Computer Graphics 10, 1 (Jan 2004), 72–84.

[29] Tufte, E. R. The visual display of quantitative informations 2nd ed. Graphics Press,Cheshire, Conn., 2001.

[30] Tufte, E. R. Beautiful Evidence. Graphics Press, 2006.

[31] Tufte, E. R. Visual explanations: images and quantities, evidence and narrative.Graphics Press, 2012.

[32] Tufte, E. R. Envisioning Information. Graphics Press, 2013.

[33] Turk-Browne, N. B. Functional interactions as big data in the human brain. Science342, 6158 (2013), 580–584.

[34] van den Heuvel, M. P., and Pol, H. E. H. Exploring the brain network: A reviewon resting-state fmri functional connectivity. European Neuropsychopharmacology 20,8 (2010), 519 – 534.

[35] Wheelwright, S. C., and Clark, K. B. Accelerating the design-build-test cyclefor effective product development. International Marketing Review 11, 1 (1994), 32–46.

Page 52: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

44 Chapter 8. Bibliography

Page 53: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

Appendix A

A - Usability Test Instructions

A.1 A Visualization Case Study

A.1.1 Instructions

This is a concept and prototype of a web based visualization tool that is supposed to allowyou as a user to upload connectome data and thereafter visually, as well as quantitatively,compare values between two to five data sets interactively. Specifically we want to visualizehow one can correlate connectivity against an external variable e.g. behaviour or dopamine.

This test is a case study on a prototype based on two sets of data. The first set ofdata that will be loaded and evaluated with is mean connectivity data and the secondone is dopamine covariation data. Each set has been collected from 181 healthy cobrasubjects during resting state and n-back working state. The format of the data used for therepresentation is a graph representation of brain connectivity consisting of 12 pre-definednetworks. The edges between two different networks (inter-correlation) and one and thesame pre-defined networks (intra-correlation) have been averaged respectively.

Below you have a set of tasks to perform on each of the data sets. Note, as this is acase study on predefined data sets you will start of the test when the datasets have alreadybeen uploaded. Try to think out loud while performing the tasks. That is - verbalize yourthoughts as you move through the interactions. For example, when you are clicking onsomething, say what you expect is going to happen.

Tasks:

1. Enter the visualization from the first page.

2. Find the exact value of the subcortical network’s intra correlation.

3. Give an approximation of the difference between resting state and active state of thatnetwork’s intra correlation.

4. Find the top three networks with the strongest inter-correlations.

5. Change the visualization so that only two network’s inter-correlations are shown inthe circular representations.

6. Save the visualization.

45

Page 54: Exploring Ways of Visualizing Functional Connectivity · this thesis the design of an interactive web based visualization tool for functional connectivity was explored through an

46 Chapter A. A - Usability Test Instructions

7. This isn’t a particular task. Instead you are free to explore the visualization as youwish.

In the second trial, with the Dopamine correlation data, you will start out in the visu-alization view and thereafter redo the tasks listed above.

Thank you for your participation!