NEURAL NETWORKS - WHAT ARE THEY

Neural networks are computer systems that mimic biologic nervous systems in that they sum inhibitory and excitatory inputs to produce a single output, which in turn synapses with one or more similar processors. They learn to recognize patterns by training on a set of sample inputs that correspond to desired outputs. Unlike other computer-derived clinical algorithms, they do not explicitly describe the relative contribution of each data element to the desired output. A report by Baxt in this issue of Annals describes how he applied neural networks to the problem of identifying patients who truly had myocardial infarction from among a set of patients presenting with chest pain. His study had remarkably high sensitivity and specificity. Clinicians may well be wary of this "black box" technology. However, its value in clinical medicine should be assessed objectively. We look forward to future evaluations of neural networks applied to clinical problems.