An artificial neural network for classification of forced expired volume signals

An artificial neural network was developed for the classification of respiratory spirometric curves. A feedforward network utilizing the generalized delta rule learning algorithm was trained to recognize spirometric curves representing patients with normal, restricted, or obstructed pulmonary function. A set of 137 spirograms which had been previously classified into those categories was used to evaluate the performance of the neural net classifier. Five spirograms randomly selected from each group were used as a training set. After training, the network correctly classified 72% of the remaining 122 spirograms. The ability of the neural net to learn automatically patterns of abnormality in biological signals makes it a potentially powerful screening tool.<>