Hidden Markov Models for the identification of fetal phases in CTG recordings

Cardiotocography (CTG) is the most widely used diagnostic examination to check the status of the fetus during pregnancy. The complex nature of CTG signals strongly affects the interpretation of the latter, causing a large intra and inter-observer variability in the analysis. A very challenging task in this context is the automatic identification of fetal activity and quiet periods within the tracings. The importance of distinguishing these periods resides in the fact that different neural mechanism are involved in the regulation of fetal heart, depending on quiet or activity fetal stages. This would also allow the evaluation of any differences present in diverse fetal stages between signals from healthy and pathological fetuses, increasing the diagnostic capability of CTG analysis. With the present work we address the problem of fetal state clustering in a totally unsupervised manner, resorting to the use of Hidden Markov Models. The obtained results have been brought to the attention of an experienced clinician, who expressed a 93% level of agreement between his interpretation and the fetal state clustering obtained with the model.