Case-Based Reasoning in the Health Sciences: Foundations and Research Directions - Computational Intelligence in Healthcare: Advanced Methodologies - page 142

infrequently mined to discover more general knowledge. Such knowledge mining
would not only improve the performance of CBR systems, but could turn case
bases into valuable assets for clinical research.
The integrated synergy between knowledge discovery, data mining, and CBR
is exemplified by Janichen and Perner [47] who present a memory organization
for ecient retrieval of images based on incremental conceptual clustering for
case-based object recognition. These authors explain that case-based image in-
terpretation in a biological domain, such as fungi spore detection, requires storing
a series of cases corresponding to different variants of the object to be recognized
in the case base. The conceptual clustering approach provides an answer to the
question of how to group similar object shapes together and how to speed up the
search through memory. Their system learns a hierarchy of prototypical cases
representing structural shapes and measures the degree of generalization of each
prototype [47].
Bichindaritz [51] explores automatically learning prototypical cases from the
biomedical literature. The topic of case mining is an important recent trend
in CBR, enabling CBR to capitalize on clinical databases, electronic patient
records, and biomedical literature databases. This author explores how mined
prototypical cases can guide the case-based reasoning of case-based decision-
support systems and the different roles prototypical cases can play, particularly
in aligning systems with recent biomedical findings [51].
7.3 Multimodal Reasoning Architectures
CBR is frequently used in conjunction with other AI approaches in order to cap-
italize on complementary strengths and synergies [76,77]. The first multimodal
reasoning system in the health sciences was CASEY [23], which diagnosed heart
failures,asdescribedinSection4.Itintegrated CBR with model-based reason-
ing, which it used when it did not have a suciently similar case in its case
base to reach a satisfactory diagnosis. Rule-based reasoning (RBR) is the AI
technique most commonly integrated with CBR. One example is the Auguste
project, which combines CBR and RBR to recommend neuroleptic drugs for
Alzheimer's patients with behavioral problems [35]. Here, CBR is used to de-
termine whether or not a patient would benefit from a neuroleptic drug, and if
so, RBR is used to determine the best choice of available neuroleptic drugs for
the individual. Another example is the 4 Diabetes Support System [78], which
helps to achieve and maintain good blood glucose control in patients with type
1 diabetes on insulin pump therapy. This system uses RBR to detect glucose
control problems requiring changes in therapy and CBR to suggest the appro-
priate therapeutic changes. Researchers on the 4 Diabetes Support System have
recently integrated a naive Bayes classifier to detect excessive glucose variability
and are currently exploring the integration of other machine learning techniques
for problem detection in continuous blood glucose data.
Several other AI methodologies and techniques have also been combined with
CBR to advantage. Dıaz et al. demonstrate the applicability of CBR to the
classification of DNA microarray data and show that CBR can be applied