1
00:00:00,880 --> 00:00:05,440
Aardvark uses composite visualizations 
of trees time series and images to help

2
00:00:05,440 --> 00:00:09,560
researchers understand how cancer 
cells grow divide and spread.

3
00:00:09,560 --> 00:00:12,080
Our collaborators collect time series microscopy

4
00:00:12,080 --> 00:00:21,160
images where cells are tracked over time
derived attributes such as cell mass

5
00:00:21,160 --> 00:00:26,320
and cell location are measured 
resulting in time series data

6
00:00:26,320 --> 00:00:37,240
and when cells divide the relationship is 
recorded to produce a tree or cell lineage

7
00:00:37,240 --> 00:00:41,480
each of these data types represent 
a different aspect of the data

8
00:00:41,480 --> 00:00:46,200
and visualizing them independently 
cannot tell the whole story

9
00:00:46,200 --> 00:00:51,080
for instance, simply plotting the mass 
over time may show one cell growing

10
00:00:51,080 --> 00:00:58,200
dividing into two, then four
but what if the line chart shows an outlier

11
00:00:58,200 --> 00:01:01,800
in this chart alone we cannot verif 
ver ify if this is a cell growing

12
00:01:01,800 --> 00:01:06,760
abnormally large or an Imaging error
to answer this we need to find the

13
00:01:06,760 --> 00:01:12,280
corresponding cell in the images
in another example we see a

14
00:01:12,280 --> 00:01:15,960
divergence in the cell growth
but again we cannot tell in

15
00:01:15,960 --> 00:01:22,000
this view if the faster growing cell 
divides into the faster growing offspring

16
00:01:22,000 --> 00:01:26,960
to do this we need to find the 
corresponding cells in a tree view

17
00:01:26,960 --> 00:01:31,680
our collaborators are frequently 
crosschecking between visualization views

18
00:01:31,680 --> 00:01:37,440
but this requires mentally linking 
separate visual data elements together

19
00:01:37,440 --> 00:01:41,280
therefore our work focuses on 
easing this burden by showing

20
00:01:41,280 --> 00:01:45,360
all three data types together 
in a composite visualization

21
00:01:45,360 --> 00:01:51,000
but how do we actually combine these 
disparate types into a seamless visualization

22
00:01:51,000 --> 00:01:56,240
let's look at our first composite 
visualization to explain our design principles

23
00:01:56,240 --> 00:02:02,560
here we select trees as the primary data type 
which we show in a node-link representation

24
00:02:02,560 --> 00:02:09,120
this visualization will then 
host our secondary data types

25
00:02:09,120 --> 00:02:14,800
we can then choose a composition technique 
to embed the client visualization

26
00:02:14,800 --> 00:02:19,600
in this case we nest the time 
series data inside each node

27
00:02:19,600 --> 00:02:29,840
alternatively for images we superimpose them 
along the nodes automatically and on demand

28
00:02:31,920 --> 00:02:36,240
in Aardvark we use horizon charts 
to represent time series data

29
00:02:36,240 --> 00:02:40,280
image snippets of cells are 
superimposed above the horizon charts

30
00:02:40,280 --> 00:02:44,800
we place images on the tree 
topology near division points

31
00:02:44,800 --> 00:02:51,160
and where large changes occur in cell attributes
as the size of the tree changes additional

32
00:02:51,160 --> 00:02:58,360
cells are added or removed
mousing over the horizon chart

33
00:02:58,360 --> 00:03:04,280
shows thumbnails with a full cell 
boundary beyond the clipped region

34
00:03:04,280 --> 00:03:09,960
hovering over a cell image shows 
the previous and next frames

35
00:03:09,960 --> 00:03:19,800
and selecting it pins those images
in this case the images reveal a missed division

36
00:03:19,800 --> 00:03:29,120
the cell snippets show that the cell divided 
but one of the daughter cells is tracked

37
00:03:29,120 --> 00:03:34,840
the second composite visualization uses 
the time series data as a primary data type

38
00:03:34,840 --> 00:03:39,640
a line chart is well suited for comparing 
the development of multiple cells

39
00:03:39,640 --> 00:03:46,280
the tree data is shown on demand 
through connecting lines and styling

40
00:03:46,280 --> 00:03:53,760
images of cells are shown on demand 
above their corresponding data points

41
00:03:53,760 --> 00:03:58,080
here each line represents the 
mass of a single cell over time

42
00:03:58,080 --> 00:04:02,200
hovering over a line will show 
the cell at that point in time

43
00:04:02,200 --> 00:04:07,240
selecting a cell will display connections 
explicitly with a dashed line and use this

44
00:04:07,240 --> 00:04:10,360
color to differentiate cells in the two branches

45
00:04:10,360 --> 00:04:17,200
here we can see a divergence in the growth 
between the green and violet branches

46
00:04:17,200 --> 00:04:21,960
the final composite visualization 
uses images as the primary data type

47
00:04:21,960 --> 00:04:27,280
showing the images directly helps reveal 
the spatial relationships between cells

48
00:04:27,280 --> 00:04:32,680
cell positions over time are superimposed 
directly in this coordinate space

49
00:04:32,680 --> 00:04:36,480
cell relationships are shown 
with a node-link representation

50
00:04:36,480 --> 00:04:47,480
leaf nodes coincide with the cells in the image 
and internal nodes represent cell ancestors

51
00:04:47,480 --> 00:04:55,320
in Aardvark we can see how this looks as one 
cell divides into two, then four, then eight

52
00:04:55,320 --> 00:05:00,080
as the number of cells increases we 
can see when they begin to overlap,

53
00:05:00,080 --> 00:05:04,440
resulting in issues with tracking

54
00:05:04,440 --> 00:05:07,320
together these three views comprise Aardvark,

55
00:05:07,320 --> 00:05:17,240
an open source visualization system 
for trees, time-series, and images

56
00:05:17,240 --> 00:05:19,960
thanks for watching and remember 
with Aardvark the answer is just

57
00:05:19,960 --> 00:05:23,480
a nose length away so keep on digging!