Biomedical Engineering Reference
In-Depth Information
Recommenders/collaborative fi ltering - methods to identify likelihood
of one group to share common traits with other groups who exhibit
certain traits. Example: potential for understanding reactions to drugs
based on the reactions of other similar users.
But how should analysts bring these algorithms and the massive
amounts of fl exibly structured data together? This is an exciting fi eld and
quite immature, but below we review how the industry has begun to
piece together the constituent parts to make this happen. There are
several options and the Business Intelligence (BI) stack companies,
Pentaho [26] and JasperSoft [27], have begun to assemble SQL and
NoSQL connectors, statistics packages, analysis packages, ETL packages,
and Cubing and OLAP packages. Table 20.2 shows several options for
different capabilities in different stacks and platforms.
Pentaho Open BI Suite is a complete packaged suite for business-
enabled reporting solutions. The community edition is open source and
the breadth of functionality is very thorough. The solution is very
business-oriented, and its target use-cases are guided by a philosophy of
actionable reporting through complete round-trip business processes that
supports alerts and scheduling. From a visualization perspective, Pentaho
delivers a portal-based product that allows administrators, reports
designers, and reports users to defi ne, design, and deliver reports and
charts. It has a robust plug-in capability that allows others to replace or
add to many of the components currently embedded. A central meta-data
repository maps the physical database from the logical visualization tier.
The portal and the meta-data model work together to allow users to
defi ne reports and charts via JFreeReports and JFreeCharts.
To handle extreme scales of future clinical data, many different slices
of the data will be produced, based on the clinical problem being solved.
For example, a database whose data model is only focused on the factor
of heart disease may be created and loaded with information only from
medical records from patients with this condition. Data would be
extracted from the fl exible data store, subjected to semantic and clinical
normalization, and transformed into a view that can more easily be
navigated. As discussed above, the desire is to create an architecture that
allows the raw CDA to be re-processed based on the current rule set of
interest. In the study on acute heart disease, analysts might later fi nd that
a new factor exists and might want to re-build the data mart to include
this. Thus, the best path is a tradeoff between processing power and
fi nding the perfect data model, and in our experience, the quest for the
latter only delays projects.
￿ ￿ ￿ ￿ ￿
 
Search WWH ::




Custom Search