Using CBR as Design Methodology for Developing Adaptable Decision Support Systems - Efficient Decision Support Systems: Practice and Challenges from Current to Future - page 132

Fig. 5. Life cycle of the geneCBR system working in diagnostic mode
The adaptation of previous cases in order to solve a new FMD is accomplished in the reuse
stage (left bottom square in Figure 5). A growing cell structures (GCS) network (Fritzke,
1993) is trained with the whole case base, only taking the existing cases represented by the
genes selected in the previous stage as input. Then, the new FMD is presented to the GCS
network and the patients most similar from a genetic point of view are retrieved. Based on
this grouping, a proportional weighted voting mechanism is applied that ponders the level
of similarity with the new FMD. An initial class is assigned by the geneCBR system from
among the existing pathologies (Glez-Peña et al. 2009b).
In the revision stage (right bottom square in Figure 5) the expert is provided with useful
data about the decision made by the system. This information contains the selected DFP
genes, the grouping made by the GCS network, the weighting assigned to each class and a
set of See5 (Quinlan, 2000) classification rules generated from the most similar patients. The
expert contrasts the initial prediction given by the system with other external information
such as patient karyotype or clinical history in order to ascertain a revised prediction and a
final diagnostic.
Every time a new FMD is solved, the internal structure of the geneCBR system is updated
(top-right square in Figure 5). The new FMD is associated to its corresponding class and
added to the case base. The affected FP is updated and the system marks the most similar
patients selected for future classifications. In this stage the geneCBR system changes to edit-
mode and the expert is permitted to update patient's classification taking into account the
new knowledge generated.
The geneCBR represents a new translational decision support system that can effectively
support the integrative work of programmers, biomedical researches and clinicians working