Fig. 1.

Block diagram of a human observer and channelized numerical observer in a lesion-detection task. A human observer reports score Y describing his confidence that a lesion is present. Development of a numerical observer can be viewed as a supervised-learning problem, in which the goal is to identify the unknown human system. In other words, the goal is to train the numerical observer to produce a value Ŷ = f(x) that closely match the human-observer score Y.