ABSTRACT
Many techniques have been raised for improving the survival of preterm infants. However, premature delivery remains as a common problem for the new born deaths. The complications due to preterm deliveries may even lead to significant health problems of the new born babies and also many other economical problems. There exists a strong evidence that the analysis of uterine electrical signals could provide an easy path towards preterm birth detection and even prediction. Exploring this idea, the paper focus on the Electrohysterography (EHG) signal processing and efficient machine learning algorithms. Features based on the one dimensional Local Binary Patterns (1D-LBP), a powerful texture classification feature was proposed here. This study uses an open dataset, Term-Preterm Electro Hystero Gram (TPEHG) database, containing 300 delivery records (38 preterm and 262 term) for data acquisition. This paper provides a simple, but efficient way of approach for predicting premature deliveries. An accuracy of about 60% is evaluated as the performance of classifying term and preterm records.
