|
|
ViStA: Visual Statistical Analyzer
K. Mueller and W. Zhu
It is well known that once the data become too large in size and/or
dimensionality, automated data mining approaches begin to fail. To address
these shortcomings, our current efforts target a classify-refine mechanism
that inserts the scientist into a tight loop in the data mining process.
Interaction with the system is very natural and intuitive, since the user
may reorganize, restructure, or fine-tune certain components of the model or
portions of the data directly within the visual display.
Viewing in the native domain. We have developed a 3D visualization
interface that displays correlational data in the native domain. This tool
is currently mostly used within the BrainMiner project, where the data and
the functional relationships implied by the data are displayed within the
brain anatomy. The correlation matrix is a 6-D object and a workable tool
for its visualization is a challenge. With user selection of a single row or
column, the problem reduces to three-dimensional visualization. For the
brain function correlation matrix, these data are presented, along with an
MRI volume and a digitized version of the Talairach brain atlas. Both can be
sliced in three orthogonal directions and can be overlaid on each other. A
basic view with a few regions of interest (ROIs) is shown in Figure 25,
which shows the Graphical User Interface (GUI) of our newly developed 3D
brain visualization software, along with a basic view of a small number of
ROIs embedded into a cut-out area of a normalized/standardized MRI brain.
Similar to the 2D viewer, the colors of the ROIs denote the strength of the
correlational relationship, on a rainbow scale. The root ROI is colored in
yellow. The GUI allows the user to slide the cutting planes up and down and
back and forth, to rotate the volume, and to select certain brain surfaces,
such as white matter, gray matter, and skull, to be semi-transparently
superimposed. The correlation thresholds can also be selected, and many more
features are available. The number of ROIs to be displayed, however, can
become quite large (about 120-140), which poses challenging problems in the
visualization task: in a space too crowded with statistically significant
ROIs it becomes very hard, if not impossible, for the user to tell the 3D
positions of the individual ROIs. To overcome these difficulties, a number
of techniques [1-2] were devised, some of which are illustrated and
described in Figure 1.
Viewing in an abstracted feature domain. An alternative way to view
and edit causal and hierarchical relationships is in a complementary
abstracted feature-centric display. We illustrated this type of display by
way of our SpectrumMiner domain application, but similar strategies will
also be available to edit causal models for our BrainMiner domain
applications. A particular challenge is imposed by time-varying data. Our
interactive dendrogram viewer provides the following two mechanisms to cope
with this kind of data: (i) the 4D time-slice selector and (ii) the 3D
ThemeRiver. We shall describe these two displays in turn.
Figure 1. (Left) The present 3D visual interface of BrainMiner, along
with a basic view of a small number of ROIs embedded into a cut-out
area of a normalized/standardized MRI brain. The display can be switched between three
correlation matrices: those of the two studies as well as the
difference correlation matrix. The latter display shows the ROIs
that have changed, under the influence of the drug, their
statistical relationship with respect to the selected root ROI;
(center column top:) superimposing a Talairach atlas slice (or an
MRI slice) that can be slid up and down the volume;
(right column
top:) enhancing the ROIs by colored halos or coasters, where the
colors code their height and depth on a rainbow color scheme, to aid
the perception of 3D depth relations; (center column bottom:) near
cortex ROI-correlations projected onto the cortex surface;
(right
column bottom:) depth cues provided drop lines.
The 4D time-slice selector. The interactive dendrogram
represents time-varying data as a cylindrical shape composed of a
stack of circular time-slice dendrograms. The time-slice selector
shown above the dendrogram in Figure 2 shows the unwrapped outer
surface of this cylinder, with each horizontal slice capturing the
leaf nodes of one time-slice dendrogram. Patterns in the data
distribution over time are clearly visible. Stepping across the time
slices will animate the data-related coloring of the dendrogram’s
arcs.
The 3D ThemeRiver. To show the fluctuations of different variables
(i.e., nodes in the time-varying dendrogram) over time, we use the
ThemeRiver paradigm [3] that visually illustrates multiple variables as
parallel streams in which the width of the stream maps to node magnitude. A
limitation of the existing ThemeRiver scheme is that only one attribute can
be displayed per theme. We therefore have devised a 3D extension, which
enables us to display two attributes of each variable in the data stream.
The new 3D ThemeRiver [6] can display any ternary relationships within the
data, and not just as a function of time. The surface itself is modeled as a
smooth 3D.
Figure 2. Left: The dendrogram with a user-drawn closed freeform line
specifying the time axis-centric surface to be visualized with 3D ThemeRiver.
Right: A 3D ThemeRiver visualization of 17 clusters of organic aerosols.
Width encodes overall cluster distribution and the height encodes incidence
of zinc. The horizontal axis represents the increasing concentrations
of ozone in the atmosphere.
References
-
[1] Mueller, K., Welsh, T., Zhu, W., Meade, J., and Volkow, N. Brain
Miner: A Visualization Tool for ROI-Based Discovery of Functional
Relationships in the Human Brain, New Paradigms in Information
Visualization and Manipulation (NPIVM) 2000, Washington DC, November
2000.
-
[2] Welsh, T., Mueller, K., Zhu, W., Volkow, N., and Meade, J. Graphical
strategies to convey functional relationships in the human brain: A case
study. Visualization '01, pp. 481-485, San Diego, October 2001.
-
[3] Havre, S., Hetzler, E., Whitney, P., and Nowell, L. ThemeRiver:
Visualizing thematic changes in large document collections. IEEE Trans.
Visualization and Computer Graphics 8(1): 9-20 (2002).
-
[4] Imrich, P., Mugno, R., Mueller, K., Imre, D., Zelenyuk, A., and Zhu,
W. Interactive poster: Visual data mining with the interactive
dendrogram. IEEE Information Visualization Symposium’02, October 2002.
-
[5] Volkow, N.D., Zhu, W., Felder, C., Mueller, K., Welsh, T., Wang,
G-J., and De Leon, M. Changes in brain functional homogeneity in
subjects with Alzheimer’s disease. Psychiatry Research Neuroimaging
114(1): 39-50 (2002).
-
[6] Imrich, P., Mueller, K., Imre, D., Zelenyuk, A., and Zhu, W.
Interactive Poster: 3D ThemeRiver. IEEE Information Visualization
Symposium ‘03, October 2003.
-
[7] Imrich, P., Mueller, K., Imre, D., Zelenyuk, A., and Zhu, W.
Interactive Poster: 3D ThemeRiver. IEEE Information Visualization
Symposium ‘03, October 2003.
-
[8] Zhu, W., Volkow, N.D., Ma, Y., Fowler, J.S., and Wang, G-J.
Relationships of ethanol induced changes in brain regional metabolism
and motor, behavioral and cognitive functions. Alcohol and Alcoholism.
39(1), 53-58 (2004).
-
[9] Kim, J., Zhu, W., Chang, L., Benter, P., and Ernst, T. A unified
structural equation modeling approach for the analysis of multi-subject,
multivariate functional MRI data. Human Brain Mapping. May 22; [Epub
ahead of print] (2006).
Last Modified: January 31, 2008
Please forward all questions about this site to:
Claire Lamberti
|
