Case-Based Reasoning in the Health Sciences: Foundations and Research Directions - Computational Intelligence in Healthcare: Advanced Methodologies - page 120

contracts tick-borne encephalitis in the Vienna woods, an Austrian physician
will more readily be able to recognize and treat the disease than the traveler's
own physician in a tick-free part of the world.
CBR systems derive their power from capturing and recalling past experi-
ences in much the same way as people do. Knowledge is stored in the form of
individual cases, each of which represents some past problem or experience. For
example, a case might contain a map from point A to point B, or the description
of a patient's signs and symptoms with the treatment prescribed and the clinical
outcome. The precise contents of a case depends upon the task that the system is
meant to perform. To be useful for automated reasoning, a case must contain in-
formation pertinent to the task at hand. Often, this is the same information that
a person would need to perform the task manually. If, for example, a physician
was deciding whether or not to prescribe a neuroleptic drug for a patient with
Alzheimer's disease, important factors would include medical history, physical
state, emotional state, observed behaviors, cognitive status, caregiver support,
and safety concerns. Some of this data might be available in the patient's chart
or electronic health record, and some might be acquired during the oce visit
at which the decision is made. Each case in a CBR system built to support this
decision making process would contain the specific details of these factors for an
individual patient, along with a record of whether or not a neuroleptic drug was
actually prescribed.
The knowledge base, then, is a collection of cases, known as a case library or
case base. The case base is organized to facilitate retrieval of the most similar, or
most useful, experiences when a new problem arises. When the system is given
a new problem to solve, it searches its case base for applicable past cases. The
solutions stored in past cases provide a starting point for solving the problem
at hand. The reasoning process employed may be envisioned as a CBR cycle, as
first described in the classic CBR overview by Aamodt and Plaza [6], and shown
in Figure 1. Next, let us consider a simple example to illustrate the workings of
this CBR cycle.
Suppose that a real estate agent needs to determine a fair selling price for
a house. If the price is set too high, the house will not sell, but if the price is
set too low, the agent will lose money. The agent, whether human or software,
can use recent selling prices of similar past homes to assist with this task. Each
past sale would constitute a case. A case would contain the selling price, plus
a description of the house sold, including its address, school district, year of
construction, living area, lot size, number of bedrooms, number of bathrooms,
type of garage, and other amenities.
Reasoning begins when a new house comes on the market. The agent can
ascertain its descriptive properties, but has not yet determined a selling price.
The new house becomes a new case at the beginning of the CBR cycle. The
new house is compared to the cases in the case base during the Retrieve phase.
The most similar past case is retrieved, becoming the retrieved case shown in
Figure 1. This retrieved case contains a selling price that can be considered for
the new house. However, before finalizing the price, the agent would examine