A mapping network approach for classifying the respiratory forced expired volume signal is presented. Using reconstructed spirograms, the development and application of a backpropagation mapping network simulator to two pulmonary function classification problems is described. In the first problem, the mapping network correctly classified 95% of previously unseen volume-time curves as being indicative of normal, restricted, or obstructed pulmonary function. In the second problem, the mapping network performed at a level equivalent to a discriminant function based on standard spirometric parameters in differentiating between spirograms indicative of normal and diseased subjects. The ability of the neural network to automatically learn patterns of abnormality in biological signals makes it a potentially powerful screening tool. >
