Mining Spatial Data Using an Interactive Rule-based Approach

1
Geovisualization and Spatial Analysis of Cancer
Data Developing Visual-Computational Spatial
Tools for Cancer Data Research
Dr. Alan M. MacEachren, Dr. Mark N. Gahegan
(GeoVISTA), Dr. Eugene Lengerich (ACN), Dr. Luc
Anselin, CSISS. Dr. James Macgill (GeoVISTA), Ann
Ward (ACN)

• Challenges for Spatial Data Analysis in Cancer
  Research
• Data with high dimensionality (e.g. screening
  rates, mortality rates, risk factors, demographic
  covariates, etc.) are not easy to analyze using
  traditional techniques.
• Geographic variations in potential and
  hypothesized relationships can be hard to
  identify and verify.
• Overall Project Goal
• To develop, implement, assess, and disseminate
  the next generation of cross-platform,
  visually-enabled geospatial analysis methods and
  tools to support cancer-related public health
  research and policy.
• Methods and tools under development will
  facilitate the integration of epidemiological,
  demographic, and health-policy data, enabling
  researchers and analysts to take a holistic view
  of communities, their health with respect to
  cancer, and relationships to health policy (e.g.
  screening, accessibility).
• Specific Project Objectives
• Develop dynamic visual analysis tools, e.g.
  interactive maps and other data visualizations,
  linked with each other.
• Develop computational methods for assisting
  users to navigate and analyze spatial and
  non-spatial patterns and relationships
• Integrate the above methods and tools into a
  flexible software environment for cancer
  specialists.
• For more details, see www.geovista.psu.edu/grants
  /nci-esda/index.html

INTRODUCTION
PROJECT FRAMEWORK

• Collaborating Researchers
• Researchers in geographic information science
  spatial statistics
• -GeoVISTA Center (Penn State)
• -Center for Spatially Integrated Social Sciences
  (CSISS Univ. of Illinois)
• Cancer epidemiology control specialists Dept.
  of Health Evaluation Sciences Appalachia Cancer
  Network (ACN)
• Methods
• Proof-of-Concept Case Studies
• to address specific cancer research questions
  relevant to the ACN
• to demonstrate and assess the methods and tools
  developed
• Usability Assessments
• applied throughout software design,
  implementation and deployment
• to ensure methods and tools are accessible and
  usable by cancer researchers analysts
• Dissemination
• to provide software / training in cancer research
  policy communities
• modeled on existing outreach efforts by ACN
  CSISS

PROTOTYPE CASE STUDY
SOFTWARE DEVELOPMENT
GeoVISTA Studio A Programming-Free
Environment Studio is an open software
development environment for geospatial data. Our
intention is to allow non-expert users to build
applications for integrating visual, statistical
and computational methods, quickly, and from a
growing range of interoperable components. We are
using Studio as a the platform for developing
tools for cancer data research. See
www.geovistastudio.psu.edu for further
information Component Based Implementation Tools
as Collections of Java Beans Tools are being
implemented using a component-based approach.
Major functions of tools are encapsulated in
different components (separate Java Beans). Each
component functions independently. Components
can be combined flexibly, by the user, tailored
to the needs of different analyses. These
combinations of components do not require
modifying the source code therefore building
tools does not require programming expertise.

Left Designing a Tool Visual programming in
Studio. Icons show individual Java Beans, linked
into a tool Right The user interface Showing the
tool in action for a specific data analysis
We can see that cervical and lung cancer co-vary
positively, because we see more grays and whites
in this map. Similarly, we can tell that cervical
and colon cancer vary inversely, because we can
see more cyans and reds in this map. There is
also a selection shown in all maps and
scatterplots the user interactively selected a
set of observations in one scatterplot and the
counties represented are highlighted in all
views. The observations selected were those that
have high rates of breast cancer but low rates of
cervical cancer. We can see in the maps that
there is a high concentration of these cases in
Pennsylvania and that these observations tend to
be low in lung cancer and high in colon cancer.
Poster prepared by Alistair Geddes, Diansheng
Guo, Frank Hardisty, and Dr. Alan M. MacEachren,
GeoVISTA Center