Conclusion - Data Warehouse Systems: Design and Implementation - page 574

of their images) and incremental update of Visual Cube (given new images,
e ciently update the cube).
In the medical domain (as is also the case, e.g., in bioinformatics),
multimedia data constitute valuable information for the decision-making
process. Arigon et al. [ 8 , 9 ] have applied the idea of multimedia data
warehouses for analyzing electrocardiograms (ECGs). This work aims at
extending clinical decision-support systems with multimedia information.
This requirement poses many challenges. On the one hand, advanced
modeling features such as those studied in this topic are needed. Examples
include the support of complex hierarchies (a pathology could belong to many
classes), many-to-many relationships (a patient may have many pathologies
and vice versa), and so on. On the other hand, complex multimedia data,
and probably also textual data, must be supported. To deal with these
data, first users need to develop e cient algorithms (e.g., based on signal
or image processing, pattern recognition, statistical methodologies, among
other ones) in order to transform the initial raw data (e.g., an ECG or an
X-ray) into data descriptors. Selecting an appropriate set of descriptors is
a challenge and depends on the domain under analysis. In the work we are
commenting, the authors use the star and snowflake schemas as modeling
tools. In these schemas, the dimensions are the descriptors of these data.
There are three dimensions related to the patient: principal pathology, age,
and gender (description-based descriptors) and two dimensions related to
ECG acquisition: time and technology. Finally, two dimensions are related to
the content of the ECGs: the QT duration (the time after the ventricles are
repolarized) and the noise level (content-based descriptors). The fact table is
composed of the foreign key of such descriptors and the ECG signal. Thus, we
can, for example, count the number of ECGs that have a given characteristics,
or compute an average over a list of ECGs that have a given characteristics
in order to obtain a “medium ECG”. Note the similarity to the Visual Cube
approach described above.
Along similar lines, music data warehouses are starting to attract the
attention of researchers and the industry, arising from the interest in so-
called music information retrieval. The work by Deliege and Pedersen [ 37 , 38 ]
envisions an extension of data warehouse technologies to music warehouses
that integrate a large variety of music-related information, including both
low-level features and high-level musical information. The authors define
a music warehouse as a dedicated data warehouse optimized for storing
and analyzing music content. The work analyzes the features that a music
warehouse must support and the dimensions that a music cube must contain,
including a classification of music metadata in four categories: (a) editorial,
which covers administrative and historical information; (b) cultural, defined
as knowledge produced by the environment (like reviews, for instance); (c)
acoustic, which refers to acoustic features, like spectral analysis, or wavelets,
which describe the music content; and (d) physical, which refers to the storage
medium. Music data warehouses can be built based on these characteristics.