Classification of labour contractions using KNN classifier

Uterine contractions of desired strength and frequency are one of the important requirements for the smooth progress of cervical dilation and delivery of baby. Presently, Tocograph is used to monitor the strength, duration and frequency of uterine contractions. One of the major drawbacks of Tocography is subjectivity in interpretation. Therefore, there is a need for an objective method for classification of uterine contractions. In this paper, a K Nearest Neighbor based classification method is presented for automated classification of different types of uterine contractions during labour. For the study, CTG signals from Physionet database were used. The signals were annotated by an expert Doctor in three categories: mild (n=33), moderate (n=64) and strong (n=96). After processing of signals, eight features were extracted, followed by implementation of an appropriate classifier. K Nearest Neighbor and Rule based K Nearest Neighbor classifiers were used to classify the uterine contractions into mild, moderate and strong. We achieved an accuracy of 90.91%, 85% and 85.71% for classification using Rule based K Nearest Neighbor classification method.