An Introduction to Visual Analytics in Healthcare

An Introduction to Visual Analytics in HealthcareA tutorial sponsored by theVisual Analytics Working Group

An Introduction to Visual Analytics in HealthcareDisclosure:David Gotz discloses that he has received grant funding from Amazon and the National Consortium for Data Science, an industry-academic partnership that receives funding from Cisco, Deloitte, EMC, GE, and IBM.Jesus Caban discloses that he has no relationships with commercial interests.Adam Perer discloses that he is employed by IBM.Josua Krause discloses that he has no relationships with commercial interests.

Tutorial OrganizersDavid Gotz

Jesus Caban

Adam Perer

Josua Krause

NYU

3

Tutorial AgendaSession 1: Introduction8:30 8:45: Welcome8:45 9:30: Introduction to Visual Analytics

Session 2: Application Domain (10:00 Coffee Break)9:30 10:00: Introduction to Healthcare Applications10:30 11:00: More Healthcare Applications

Session 3: Hands-on Experience11:00 11:50: Hands-on Activities11:50 12:00: Conclusion

Content LevelBasicIntermediateAdvancedBasicIntermediateAdvancedBasicIntermediateAdvanced

The Context: Setting the Stage for Visual AnalyticsMarch 2015 Issue of JAMIA

AMIA VIS Working GroupAMIAs newest working group

Are you an AMIA member?Sign up to get involvedhttp://communities.amia.org/vis-wgVIS-WG@lists.amia.org

Not an AMIA member?Become one then join working groupMailing list maintained by www.visualanalyticshealthcare.org/

6

Synergistic ActivitiesThe 6th Annual Visual Analytics inHealthcare (VAHC) Workshop

Chicago on October 25, 2015Part of IEEE VIS

Papers archived in ACM Digital Libraryhttp://dl.acm.org/Demos, posters, etc. archived on workshop websitehttp://www.visualanalyticshealthcare.org/

Past VAHC Workshops; Annually since 20102010-2012 at IEEE VIS; 2013-2014 events held at AMIAProceedings from previous years available on workshop websitehttp://www.visualanalyticshealthcare.org/proceedings.htmlFuture: AMIA in 2016?

Goal to alternate between IEEE VIS and AMIA to help bridge communities. Also a recent special issue of JAMIA; past panel at AMIA; was well attended; a popular topic.7

Get Involved!Vibrant communities depend on volunteersParticipate in eventsHelp generate ideas (events, resources, etc.)Donate time to help organize

How to get involved?

Join the VIS Working Group

Attend the VIS Working Group meeting

Contact us if you want to volunteer to help lead

8pmMonday Nov 16Franciscan B

Youre here because you have at least a passing interest in the topic. If, after the tutorial, you want to get involved, there are many ways to do so.8

What is Visual Analytics?

Why use it?

What should we know?

Lets start by lookingat a table of data.

First, A Test.

10

First, A Test.Raise your hand when you have found the single highest sales figure.How about the top 3? Or bottom 3?

All right. So with that example, showing how changes in the visual presentation can convert a task from hard to easy, lets talk about visualization.11

What is Visualization?Visualization is the communication of information using graphical representations.Ward et al., Interactive Data Visualization

Multiple Views of the Same Data

Lookup values; Identify OutliersTrends over time in each regionQuarterly patterns?

Why choose one view vs. another?

It depends what TASK you intend to support! Lookup specific values? TABLE.

See trend over time? Line charts.

See if there is a quarterly pattern? A different view makes sense.13

More Complex Examples

Can also integrate less familiar graphical representations, high levels of interactivity, connect with sophisticated analytics algorithms, to produce very powerful visual exploratory analysis platforms.14

John Snow, 1854http://en.wikipedia.org/wiki/1854_Broad_Street_cholera_outbreak

Map of Cholera fatalitiesby location in 1854outbreakLeading hypothesis wasbad air (germ theorywas not yet known)

Map showed connectionto water pumpMap persuaded the localauthorities to disable thepump

John Snow is sometimes called the Father of Modern Epidemiology

15

Why Does Visualization Work?Why is a good visualization easier to see than tables of numbers?

Our visual systems have tremendous power to:See patternsIdentify TrendsLocate Outliers and anomaliesMuch of that power is precognitiveFastEfficient

16

The Visualization Pipeline

Rendering is largely solvede.g., Canvas, SVG, OpenGL, DirectX, Java 2D

Creating a visualization includes designing foranalysis, filtering, mapping, and interaction

From Data to GraphicImage from http://www.infovis-wiki.net/index.php/Visualization_Pipeline, with modifications.User Interactions

17

Higher Level Model for Visual Analytics

From Sacha et al., Knowledge Generation Model for Visual Analytics (IEEE VAST 2014)

18

The Visualization Pipeline

Once data is prepared and filtered, it must be mapped to a graphic representationA process often called Visual Encoding

From Data to GraphicImage from http://www.infovis-wiki.net/index.php/Visualization_Pipeline, with modifications.

User Interactions

19

What is Visual Encoding?Mapping of data entities, attributes, and relationships to a geometric representation that facilitates visual interpretation.

102530

The Designers RoleYour job as a visualization designerDesign an interpretable visual representationDefine the mapping function to algorithmically convert data to geometry

The algorithmic requirement is importantNot a one time designRepeatable for a defined class of dataWhat types of data? What prerequisites are there?What are the edge cases that need to work?How would the appearance of outliers impact the design?How will it scale to larger volumes of data?

This is what makes mapping challenging (and fun!)

The Visual VariablesEight visual variables that can be controlled during the mapping processPositionMarkSizeBrightnessColorOrientationTextureMotion

"Interactive Data Visualizationby Matthew Ward, Georges Grinstein and Daniel Keim

PositionThe location of a visual object1D

2D3D (use rarely)

Age

MarksA mark is an atomic graphical primitive Often called a glyph or symbolEmbodied by the shape of a graphical objectA distinct composition of lines, areas, volumes Scale, orientation, color/shade are NOT considered

Example marks:

.24

Required: Position and MarksBoth position and marks are required to define a visualization.This is the minimum: a mark drawn at a particular spotWithout either, there is nothing to see

b

Age

Height

25

SizeMarks can be drawn with varied size1D: length2D: area 3D: volume

3rd visual variable26

Beware Perception of SizeOur ability to judge size is easily confused:

Suggestions For EffectiveSize ComparisonWhen possible

Attempt to limit differences in size to 1D

Use position to align shapes

28

BrightnessLike size, brightness (aka luminance) can be used to distinguish marksPerception of brightness less precise than size.Hard to estimate magnitude of differencesSorting objects easier than magnitude of differencesSmall differences may be imperceptible

Compare brightness to size:Line length with 5% difference

The right square has 5% less brightness

29

Beware Fancy Shading With Gradients

The Gradient Illusion

ColorBrightness is part of colorMaps to the lightness (or darkness) of a color

Other aspects of colorHue is the primary wavelength (color)Saturation is amount of color vs. gray

This gives us the HSL color space:

Hue Saturation Lightness31

Color SpacesHSL is one example of a color space

Most common color space in software is RGB (Red-Green-Blue)

RGB is the standard colormodel for the WWWThree common notations:whitergb(255,255,255)#ffffff

We will use RGB32

ColormapsColormaps provide mapping between a variables value and color

Can be discrete or continuousGradients for continuous ratio values

Discrete, ordinal, or categorical data use palettes

Ratio values: gradientsDiscrete, ordinal, categorical: use defined palettes.

Q: WHICH PALETE IS FOR ORDINALS?33

Suggestions for the EffectiveUse of Color

Avoid rainbow color maps

Be aware of color blindness1 in 12 men (8%)1 in 200 women (0.5%)

Color theory has much to say about designing good colormaps. Seek advicehttp://colorbrewer2.org/

Source for color blindness rates: http://www.colourblindawareness.org/colour-blindness/

34

OrientationMarks can have an orientationMap attribute value to angle of rotation

35

Orientation ExampleVisualization of wind spead from NOAAPosition shows time of predictionOrientation shows forecast wind direction

Question: Why L-shaped marks?

36

Orientation and Mark SymmetryMarks can have an orientationMap attribute value to angle of rotationRange of angle values depends on mark symmetry

37

TextureTextureColor gradientsHatchingMarks within a markNot common. Most often in black-and-white graphics where color is not an option

http://www.indezine.com/products/powerpoint/ppezine/048.htmlhttp://www.archblocks.com/archblocks-cad-blocks-and-products-previews/autocad-hatch-patterns

Why rarely used?Lots of high frequency visual variation.DISTRACTING! Attracts attention. Also, HARD TO DISTINGUISH!

What kind of data? CATEGORICAL. Used carefully, perhaps ORDINAL. 38

MotionA change to any of the other seven propertiesAnimation can be used used to interpolate between valuesTypically associated with eitherInteractionDynamic data

Use judiciously!Show correspondingdatapoints Across a transitionAcross views

39

The Visual VariablesEight visual variablesPositionMarkSizeBrightnessColorOrientationTextureMotionDuring mapping, we convert attribute values to these visual properties

Relative Interpretation Not all visual variables are equalStudy by Cleveland and McGill examined accuracy of human perception and produced a rankingPosition along a common scaleScatter plot, Points on a mapPosition along an identical but non-aligned scaleScatter plot matrixLengthBar chartHistogramAngle and slopePie chartGradient linesAreaTreemapBubble chartVolume, density, and color saturation3D visualizationHeat mapColor hueColor scales

Position along a common scalePosition along an identical by non-aligned scaleLengthAngle or SlopeAreaVolumeDensityColor SaturationColor HueGraphical Perception: Theory, Experimentation, and Application of the Development of Graphical Methods. William S. Cleveland and Robert McGill. JSTOR. 1984.

41

An Example: Room for ImprovementFrom a meeting at NIH last week

1. 3D perspective, distorting our ability to compare size. REMEMBER THE THREE CARS?2. Brightness gradients making colors less distinctive (NOT SUCH A BIG DEAL with categorical data, but still unnecessary)3. Pie chart uses angles. Why not use a sorted bar chart? Length much easier to compare than angle.

This chart isnt such a big deal. No decision is being made, just some light information. But what if it was patient data, and you had to make a treatment decision of some kind? Now much more important.42

Moving Beyond Individual MarksThese eight variables apply to individual marks

Graphical elements are not interpreted in isolation.

Relationships between visual elements also have perceptual power

Patterns

Gestalt LawsProximitySimilarityConnectednessContinuitySymmetryClosureFigure and Ground

ProximityItems positioned near each other are perceptually grouped together.

Implication:Marks representing related information should be positioned close together.

Is seen as rows because dots are closer together horizontally.Is seen as columns because dots are closer verticallyWe perceive two groups because of spatial proximity.45

SimilarityItems with a similar appearance are perceptually grouped together.

Implication:Use similar graphics define rows, columns or other groupings of marks.

ConnectednessConnecting marks also definegroupsTypically more powerful thanproximity or similarityNot part of original Gestaltprinciples

Implication:Use connectors to link grouped marksCaveat:Adds ink to the screen, making it messier than proximity and similarity (visual complexity)

ContinuityOur minds more naturally interpolate smooth shapesWhich pathsare easier foryou to trace?

Implication:Avoid discontinuities or abrupt changes in shapee.g., curves instead of Manhattan-style lines

SymmetryWe seek balanced, symmetric intepretations of shape

In isolation, we usehorizontal and vertical axes

Implication:Use axes or other frames of reference to support the intended interpretation of your design

Larger patterns can provide alternativeframes of reference

Is a squareIs a diamond

49

ClosureWe tend to perceive closed contours.

Our minds attempt to complete a shape, guessing what is behind an occluding object

Implications:Occluding shapes can produce incorrect assumptionsBackground contours (and other containing boundaries) can effectively denote groups even if partially obscured

Our minds see a complete circle in (a); not a broken arc.50

Figure and GroundSmaller parts of a pattern are perceived as in the foreground (the figure)Larger parts appear in the background (the ground)

Implication:Smaller areas within larger boundaries will be the objects which users first attempt to interpret for meaning

Black elements appear to float in part A.Part B has green objectsPart C has 1 white object.

Often encountered in maps. Regions cut off by zoom can seem to disappear into the background.51

What Else To ConsiderAxesUses axes to define the meaning of positionLegendsUse to define your other mappings:Size, Color, Mark, etc.LabelsPowerful, but cognitively demanding (must be read)Make your labelsInformative. Direct users attention to interesting elementsConcise. Save long text for other UI elements (e.g., sidebar)

Those are the Gestalt Laws. We also talked about visual variables.

What else should we consider?52

The Mapping ProcessVisual Variables and Gestalt Laws give us ground rules for designWhat can be controlled?How are those things perceived?

Based on rules

Define mapping function to convert data to a geometric representation

Last hands on: HYBRID.

X axis was LOCAL.Y axis was GLOBAL.

53

The Basis for Wide Range of Examples

Combine those mappings, with interaction to control parameters of the mappings, and to link across views

Many powerful views54

Developing the Right Design

FundamentalConceptsData

Task

Whats Next?Session 1: Introduction8:30 8:45: Welcome8:45 9:30: Introduction to Visual Analytics

Session 2: Application Domain (10:00 Coffee Break)9:30 10:00: Introduction to Healthcare Applications10:30 11:00: More Healthcare Applications

Session 3: Hands-on Experience11:00 11:50: Hands-on Activities11:50 12:00: Conclusion

Content LevelBasicIntermediateAdvancedBasicIntermediateAdvancedBasicIntermediateAdvanced