Case-Based Reasoning in the Health Sciences: Foundations and Research Directions - Computational Intelligence in Healthcare: Advanced Methodologies - page 128

Additional complexities arise in the medical domain when dealing with safety
critical constraints, assisting the elderly and disabled [52], and providing useful
explanations [53].
Recently, a major trend seems to be the broadening of CBR applications
beyond the traditional diagnosis, treatment, or quality control types toward the
applicability of CBR to novel reasoning tasks . An interesting example studies
how cases can represent instances of non-compliance with clinical guidelines,
eventually leading to expert refinement of these guidelines [54]. Another paper
demonstrates the usefulness of CBR in configuring parameters for hemodialysis
[50]. These papers open new fields of application for CBR, which will foster the
spread of CBR in biomedical domains.
CBR in the Health Sciences papers address all aspects of the CBR methodol-
ogy and attempt to advance basic research in CBR. For example, some research
addresses retrieval questions [55], while others address adaptation [56,57]. Bichin-
daritz [55] shows how memory organization for CBR can bridge the gap between
CBR and information retrieval systems. This article surveys the different mem-
ory organizations implemented in CBR systems, and the different approaches
used by these systems to tackle the problem of ecient reasoning with large case
bases. The author then proposes a memory organization to support case-based
retrieval similar to the memory organization of information retrieval systems
and particularly Internet search engines. This memory organization is based on
an inverted index mapping case features with the cases in memory.
D'Aquin et al. provide principles and examples of adaptation knowledge acqui-
sition in the context of their KASIMIR system for breast cancer decision support
[56]. These authors have identified some key adaptation patterns, such as adap-
tation of an inapplicable decision, and adaptation based on the consequences
of a decision. In addition, KASIMIR has also acquired retrieval knowledge to
take into account missing data and the threshold effect. The paper broadens
the discussion by proposing a sketch of a methodology for adaptation knowl-
edge acquisition from experts. Several authors have focused on the importance
of prototype-based knowledge representation for CBR in the Health Sciences
[51,58], which encourages further research in this direction.
The impact on CBR is also seen in multimodal reasoning and synergism with
other AI approaches and methodologies. The complementarity of CBR with AI
as a whole is examined in depth in Section 7.

5.2 Impact on the Health Sciences

Several CBR in the Health Sciences systems have been tested successfully in
clinical settings. However, it is a misperception that only clinical applications
are pertinent to biomedical domains. Biomedical research support also lies within
the purview of AI and CBR in the Health Sciences. For example, bioinformatics
systemsoftenaimtoanalyzedata,asdodataminingsystems,whichismoreof
value to biomedical researchers. Clinical CBR systems may remain in the realm
of pilot testing or clinical trial rather than daily clinical use. However, the shift
from CBR driven systems to application domain driven systems is currently