Nowadays, prediction of preterm birth is still uncertain and numerous scientists are searching for methods to improve its predictability. Some of the current researches are based on the analysis of electrical activity of uterus which contains rich information about it's electrophysical properties.
We developed algorithms for automatic detection, characterization and automatic classification of the electrical activity of uterus. We characterized them on the basis of power spectrums and evaluated the results with the one-way analysis of the variance ANOVA. The efficiency of detector has been assessed by special rate of efficiency. With detector’s optimization we achieved 80.2 % sensitivity and 69.8 % positive predictability. The efficiency of the classifier has been assessed by the standard rate of efficiency. The best distinguishing of uterine activity we achieved with the nearest neighbor classifier. We achieved 99.1 % sensitivity and 97.3 % specificity.
|