Classification of fetal abnormalities based on CTG signal

number: 
2413
English
Degree: 
Author: 
Asraa Razak Swadie
Supervisor: 
Dr. Safa'a S. Mahdi
year: 
2010
Abstract:

This thesis presents an approach for fetal heart rate (FHR) signal processing based on Artificial Neural Networks ,Fuzzy Logic and frequency domain Discrete Wavelet Transform in order to perform automatic analysis using personal computers. Cardiotocography (CTG) is a primary biophysical method of fetal monitoring. The assessment of the printed Cardiotocography traces is based on the visual analysis of patterns describing the variability of fetal heart rate signal. Fetal heart rate data of 55 normal fetuses and 15 abnormal fetuses from pregnant women with pregnancy between 38 and 40 weeks of gestation were studied. The data of fetal heart rate were taken from the Maternal/Fetal Monitoring: Clinical Applications ,operator manual P/N 15457AA REV.C, GE Medical Systems in Bagdad Teaching Hospital . The work in this thesis describes the analysis of fetal heart rate signals by using MATLAB version 7.7 R2008b program . The first stage in the system is to convert the cardiotocograghy tracing into digital series so that the system can be analyzed .While,in the second stage the FHR time series was transformed using Discrete Wavelet Transform in order to obtain the system features .At the last stage the approximate coefficients resulted from the Discrete Wavelet Transform are fed to the Artificial Neural Networks with Backpropaction Learning Algorithm and to the Fuzzy Logic. The accuracy of the two systems was evaluated for different frequencies. The Most of signal power are located at the lower frequencies and both algorithms gave the optimal recognition of low frequency spectrum .The artificial neural network algorithm recognized all 14 sample data based on low frequency components, knowing that 56 samples were used for training and 14 samples were used for training and 14 samples were used for test input data. While fuzzy logic system recognized 63samples out of 70 samples based on low frequency components.