The diagnostic case base is a data base of patient cases with completed diagnoses organized as a discrimination net by the findings, to use in case-based diagnostic reasoning. This case base uses a self-organizing memory system implemented by Koton[10] and patterned after the memory system developed by Kolodner[8]. It consists of two kinds of data structures gens, which are generalizations representing two or more cases, and cases, which contain the identification, description, input findings, and accepted diagnosis of an individual case. These structures are organized as a network with the case structures as terminal nodes. Each gen has a feature list with the features that occur in at least two-thirds of the cases represented by the gen, a diffs list with features used to differentiate among the cases, a causal list with the parameter states common to all of the cases, and a node count.
Whenever the correct diagnosis is determined for a case, it is added to the discrimination net. A case structure is generated to represent it and an appropriate location is found for it by following the diffs list paths through the network. At each point in the network, if a gen is found the appropriate features on the diffs list of that gen are followed, if a case is found a new gen is produced from the two cases, if nothing is found the case becomes the new terminal node. Thus, the network is built reflecting the features found in the cases diagnosed by the program. The purpose of this structure is to allow the case-based reasoner to rapidly find the cases that are similar to a new case as a starting point for a diagnosis.