FETAL HEART RATE EXTRACTION AND DETECTION USING EMPIRICAL MODE DECOMPOSITION

• Main Author: SUJATMIKO, WAHYU
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/35350
• Institution: Institut Teknologi Bandung

During pregnancy, frequent fetal monitoring is a necessary. One of the parameters that is used to measure fetal health condition is fetal heart rate (FHR). Current method that can be safely used to monitor long-term FHR is phonocardiography. There are two main problems to develop this method. The first problem is how to extract fetal phonocardiography (fPCG) signal from other acoustic signal such as maternal phonocardiography, respiration, uterine and digestive signals, fetal movement signal, signals produced by sensor movement during recording, and ambient noise. The second problem is how to detect FHR using fPCG signal. This study aims to extract and detect FHR signal using empirical mode decomposition (EMD). This method needs to be supported by other method such as fast fourier transform (FFT), selector, high pass filter (HPF), gaussian smoothing filter (GSF), wavelet denoising, envelope extraction, signal normalization, and peaks detection. This study is divided into two systems, extraction and detection. Method used in extraction system includes EMD, FFT, selector, and HPF, while method used in detection system includes GSF, wavelet denoising, envelope extraction, normalization and peaks detection. Empirical mode decomposition is used to separate maternal audio signal into several parts. Those signals are changed from time to frequency domain using FFT. After changed into frequency domain, selector choses output signals from empirical mode decomposition, which have dominant peak more than 65 Hz, to be candidates for fPCG signals. The output signals from selector are then passed through HPF to suppress the dominant low frequency component with maternal phonocardiography signal. After successfully extracting the signals, FHR detection system begins with performing baseline removal using GSF. The output signals are then passed through wavelet denoising to eliminate up to the smallest noise so that the fPCG signals can be clearly seen. After that, the envelopes of the wavelet denoising output signals are taken using envelope extraction method. Signal normalization is then used so that the signals amplitude ranges are limited from zero to one. Results of the signal normalization are processed using peaks detection method to obtain FHR signals. There are 115 fPCG and mPCG signals that are being used in this study. Those data were obtained from Hafez Hospital, Shiraz University of Medical Science. Empirical mode decomposition can be used to extract fPCG signal with 96.1% accuracy and detect FHR with 90.9% accuracy.