Prediction of forced expiratory volume in normal and restrictive respiratory functions using spirometry and self-organizing map.

This work presents a prediction of forced expiratory volume in pulmonary function testing, using spirometry and neural networks. The pulmonary function data were recorded (n = 110) from volunteers using flow-volume spirometer with a standard acquisition protocol. From the recorded flow-volume curves, the acquired data are then used to predict forced expiratory volume in one second (FEV1) using a self-organizing map (SOM) and radial basis function neural networks. The SOM is used to determine the cluster centres of the hidden layer of radial basis function neural networks. The optimal widths of the Gaussian function of radial basis function neural networks were obtained from these centres and this network is then used to predict FEV1. The performance of the neural network model was evaluated by computing their prediction error statistics of average value, standard deviation, root mean square and their correlation with the true data for normal and abnormal cases. The correlation between measured and predicted values of FEV1 for normal subjects was found to be 0.9. The prediction error for normal subjects is lower than that of restrictive subjects. Results show that the adopted neural networks are capable of predicting FEV1 in both normal and abnormal cases.