Information Technology Reference
In-Depth Information
Despite these advantages, the approach in [26] is not as simple as TA, which
allow a very clear interpretation of the qualitative description of the data pro-
vided by the abstraction process itself. As a matter of fact, such description is
closer to the language of clinicians [48], and easily adapts to a domain where
data values that are considered as normal at one time, or in a given context,
may become dangerously abnormal in a different situation (e.g. due to disease
progression or to treatments obsolescence). The ease at which knowledge can be
adapted and understood by experts is an aspect that impacts upon the suitabil-
ity and the usefulness of intelligent data analysis systems in clinical practice.
Due to these characteristics, TA have been largely exploited to support intelli-
gent data analysis in different medical application areas (from diabetes mellitus
[47,5,45], to artificial ventilation of intensive care units patients [28,4,12]; see also
the survey in [51]). However, TA have been typically adopted with the aim to
solve a data interpretation task [46], and not as a retrieval support facility. For
instance, TA have been adopted to study the co-occurrence of certain episodes
in a set of clinical time series, which may justify a given diagnosis. Indeed, in
[44] TA have been exploited in for time series retrieval, but basically for data
pre-processing as a noise filtering tool.
Therefore, our proposal to rely on TA for time series dimensionality reduc-
tion and retrieval appears to be significantly innovative in the recent medical
informatics literature panorama.
In the next section, we will introduce a flexible TA-based time series retrieval
approach, in which orthogonal index structures optimize the response time. An
example application to the field of haemodyalisis will be provided as well.
It is also worth noting that TA series can be considered as an input to a fur-
ther knowledge discovery process, able to extract key symbol sequences in the
symbolic series themselves. These key sequences might, for instance, highlight
significant transitional patterns between symbols, which can be more important
than static symbols within single intervals, especially in medical diagnosis sup-
port. Such key sequences are also able to further reduce the data dimensionality,
Fig. 3. Detection of key sequences in symbolic (e.g. TA) series in [14] (the picture is
taken from [14]).
 
Search WWH ::




Custom Search