Analysis of cardiotocography signals and classifying with machine learning techniques

Cardiotocography (CTG) is a fetal surveillance technique which is widely used to determine fetal well-being during pregnancy and delivery. A CTG test consists of two biophysical signals which are fetal heart rate (FHR) and uterine contraction (UC) recorded simultaneously. As a surveillance technique, CTG has various problems, such as poor positive prediction value, high false positive rate, subjective of interpretation, inconsistency in reproducibility value and high level of disagreement between intra- and inter-observers. Although CTG has been practiced routinely in clinics since end of the 1960s, the debates on validity and usefulness of this diagnosis test have been carried on still in the literature. The computerized CTG analysis is seen as the most promising method to overcome these drawbacks. In the thesis, the traditionally used morphological features were taken into consideration primarily. At the same time, the linear and nonlinear features of FHR that are prominent indices in terms of diagnosis were extracted. Besides, several studies have been carried out in time and frequency domain to determine FHR variability that is regarded as indispensable part of clinic assessment. Nonlinear analysis based on entropy estimators was performed. In time-frequency field, a novel feature extraction method focusing on different frequency intervals and based on texture descriptors has been proposed. In this context, gray level co-occurrence matrix, local binary pattern, and segmentation fractal texture analysis methods have been added to the literature. As a result of the studies conducted in the thesis process, a software has been developed which enables the analysis of CTG signals in morphological, linear, nonlinear, frequency, and time-frequency fields. Various machine learning techniques, such as artificial neural network, extreme learning machine, support vector machine, radial basis network were employed to provide the classification of CTG signals as normal and hypoxic as well as the performance comparison.