Kl. Comparison between Using Linear and Non-linear Features to Classify Uterine Electromyography Signals of Term and Preterm Deliveries

The main objective of this paper is to predict preterm deliveries at an early gestation period using uterine electromyography signals (EMG). Detecting such uterine signals can yield a promising approach to detennine and take actions to prevent this potential risk. Previous classification studies use only linear methods as classic spectral analysis to classify the uterine EMG that does not give clinically useful results. On another hand some studies make linear and non-linear analysis for the uterine EMG and find that the non-linear parameters can distinguish the preterm delivery uterine EMG from the term one. In this research, two ways will be taken combining the two previousideas;the first way is to take some uterine EMG linear parameters as features to a suitable neural network and the second one is to take some uterine EMG non-linear parameters as features to the same neural network. Then, the two ways' results are compared using ROC analysis which provesthat the chance of correctly classification increases markedly when applying the non-linear methods.