Providing Case-Based Retrieval as a Decision Support Strategy in Time Dependent Medical Domains - Computational Intelligence in Healthcare: Advanced Methodologies - page 200

Chapter 10
Providing Case-Based Retrieval as a Decision
Support Strategy in Time Dependent Medical
Domains

Stefania Montani

Dipartimento di Informatica, Universita del Piemonte Orientale, Alessandria, Italy

Abstract. Case-based Reasoning (CBR), and more specifically case-
based retrieval, is recently being recognized as a valuable decision
support methodology in “time dependent” medical domains, i.e. in all
domains in which the observed phenomenon dynamics have to be dealt
with. However, adopting CBR in these applications is non trivial, since
the need for describing the process dynamics impacts both on case rep-
resentation and on the retrieval activity itself.
The aim of this chapter is the one of analysing the different methodolo-
gies introduced in the literature in order to implement time dependent
medical CBR applications, with a particular emphasis on time series

representation and retrieval.
Among the others, a novel approach, which relies on Temporal Ab-
stractions for time series dimensionality reduction, is analysed in depth,
and illustrated by means of a case study in haemodialysis.

1 Introduction

Several real world applications require to capture the evolution of the observed
phenomenon over time, in order to describe its behaviour, and to exploit this
information for future problem solving.
This issue is particularly relevant in medical applications, where the physician
typically needs to recall the clinical history that led the patient to the current
condition, before prescribing a therapy; actually, the pattern of the patient's
changes is often more important than her/his final state [24]. The need for cap-
turing the phenomenon's temporal evolution emerges even more strongly when
a continuous monitoring of the patient's health indicators is required, such as in
chronic diseases [5], or when control instruments that automatically sample and
record biological signals are adopted, such as in the Intensive Care Unit [39],
haemodialysis, or instrumental diagnostic procedures [33]. In these applications,
(many) process features are naturally collected in the form of time series ,ei-
ther automatically generated and stored by the control instruments (as e.g. in
Intensive Care Unit monitoring), or obtained by listing single values extracted
from temporally consecutive situations (as e.g. the series of glycated haemoglobin
values, measured on a diabetic patient once every two months).