Software that assists in making a diagnosis of disease must be efficient (it must not demand additional time of the clinician), comprehensive (users must have confidence that it is authoritative within its application domain), and must be able to explain how it arrived at a conclusion. 

MYCIN was an early example of an expert system that could deal effectively with incomplete or ambiguous information about a patient’s condition.  It dealt only with a very specific scenario: it was designed to help diagnose and treat patients with certain kinds of bacterial infections.  MYCIN consisted essentially of a set of IF-THEN rules, built upon the knowledge base of clinical data, and certainty weights.  At its core are Bayesian Belief Networks (BBNs), a method of describing complex probabilistic reasoning using weighted network elements.

In pseudocode, the program’s operation followed (for instance):

IF the infection is pimary-bacteremia

AND the site of the culture is one of the sterile sites

AND the suspected portal of entry is the gastrointestinal tract

THEN there is suggestive evidence (0.7) that infection is bacteroid.

ANNs

The most widely-used approach in medical applications is the artificial neural network.  It, too, uses weighted networks, but can in contrast to BNN-based expert systems learn from new information and evidence.  ANNs are often used in pattern recognition and data mining applications such as EEG analysis, image recognition, ECG monitoring (actively used in fetal monitoring), and modeling of gene signaling networks.

Agents

In their comprehensive textbook of AI, Norvig and Russell present tools of artificial intelligence -- pattern recognition, fuzzy logic, neural networks, and so on (see diagram on “History”) -- in the context of the creation of intelligent software agents.  They define agents loosely as semiautonomous software entities capable of acquiring and assimilating new information, reasoning and acting in ways suited to a task at hand.

In OO-lingo, an agent has state and behavior within an environment, and is able to modify both to suit a changing environment.  A recent example of agent technology in practice, in which software agents “roam” a network and provide continuously-updated patient information, is from critical care (Dafonte et al, 2007):