Case-Based Reasoning in the Health Sciences: Foundations and Research Directions - Computational Intelligence in Healthcare: Advanced Methodologies - page 141

One major step was to include AI systems in clinical practice. AI systems in
use today are numerous. One of the first was NeoGanesh, developed to regulate
the automatic ventilation system in the Intensive Care Unit (ICU), in use since
1995 [73]. Another example is Dxplain, a general expert system for the medical
field, associating 4,500 clinical findings, including laboratory test results, with
more than 2,000 diseases [74]. Some of these systems are available for routine
purchase in medical supplies catalogues.
Several studies have shown the effectiveness of systems in clinical practice
in terms of improving quality of care, safety, and eciency [72]. One example
is the study of a 1998 computer-based clinical reminder system that showed
a particular clinical act - discussing advance directives with a patient - was
performed significantly better with the clinical reminders than without them
[75]. More generally, prescription decision-support systems (PDSS) and clinical
reminder systems, often based on clinical guidelines implementation, have con-
sistently shown clinical benefit in several studies [75]. However, clinical outcomes
are rarely measured, while process variables and user satisfaction are more fre-
quently measured. Obviously, computer system intrinsic measures are always
reported.
The success of AI in the health sciences is explained by the shift of focus
from centering the system success on the computational performance to the
application domain performance. Indeed, successful systems provide a practical
solution to a specific healthcare or health research problem. The systems with
the largest impact, such as the clinical reminders, do not have to represent a
challenging AI diculty, but they do have to fit the clinical domain in which
they are embedded. They are application domain driven rather than AI driven.
7.2 Synergies with Data Mining and Knowledge Discovery
These synergies may arise in different ways. They may involve using data mining
as a separate pre-process for CBR, for example, to mine features from time series
data [49]. Data mining may also be used for prototype mining [51], or during
the CBR reasoning cycle, for example, to retrieve cases with temporal features
[50], or for memory organization [47].
In the decoupled synergy between knowledge discovery, data mining, and CBR ,
Funk and Xiong present a case-based decision-support system for diagnosing
stress related disorders [49]. This system deals with signal measurements such
as breathing and heart rate expressed as physiological time series. The main com-
ponents of the system are a signal-classifier and a pattern identifier. HR3Modul,
the signal-classifier, uses a feature mining technique called wavelet extraction to
learn features from the continuous signals. Being a case-based reasoning system,
HR3Modul classifies the signals based on retrieving similar patterns to deter-
mine whether a patient may be suffering from a stress related disorder as well
as the nature of the disorder. Advancing this research, Funk and Xiong argue
that medical CBR systems incorporating time series data and patterns of events
are fertile ground for knowledge discovery [49]. While CBR systems have tradi-
tionally learned from newly acquired individual cases, case bases as a whole are