Classification of the normal and abnormal electrogastrogram using back-propagation neural networks

The cutaneous recording of the myoelectrical activity of the stomach is called the electrogastrogram (EGG). The percentages of the normal activity and dysrhythmia in an EGG recording are important in clinical diagnosis of patients. The aim of this study was to develop an optimal backpropagation network (BPN) for automated assessment of the normality and abnormality of the EGG. Raw EGG data, running spectral data and autoregressive moving average (ARMA) modeling parameters of the EGG were, respectively, used as input data to single-hidden-layer BPNs with different number of hidden nodes and were compared with each other. Experimental results showed that the best type of input data to the network for this specific application was ARMA modeling parameters of the EGG and the optimal network configuration was 22:15:2 (the number of input nodes: hidden nodes: output nodes), which provided a recall of 0.95.