Spectral analysis of coronary bypass doppler blood flow signals

Zabihian, B.

http://hdl.handle.net/10400.1/3773

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
thesis_2011_04_01_FINAL.pdf		1.54 MB	Adobe PDF	Download

Send Feedback

Authors

Zabihian, B.

Advisor(s)

Ruano, M. Graca

Abstract(s)

The pulsed Doppler ultrasound (DU) is one of the important tools in the study of vessel diseases and the investigation of flow conditions. Due to its non-invasive nature, it has been increasingly used in medicine in the last few decades. Accurate estimation of DU spectral center frequency and bandwidth parameters are extremely important for blood flow diagnostic purposes. Under real-time data acquisition conditions the DU signal is generally corrupted with different types of noise. In these situations the identification of signal components solely belonging to the blood flow signal is a difficult task. This thesis was aimed to study spectral techniques to enhance spectral parameter estimation, in particular the center frequency. Spectral estimates were obtained using the Short Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT). STFT was applied to short duration data segments, respecting signals’ stationary properties. Two CWT functions have been studied: varying bandwidth filter and fixed bandwidth filter. Since different filter bandwidth values yield different results, bandwidths for fixed bandwidth filter were investigate and the most proper one has been used on the performance comparative studies. To enhance the blood flow signal content of noise-embedded clinical Doppler signals, a STFT-based technique was proposed to reduce the signals’ noise components. Quantitative evaluation of the spectral methods was primarily performed on simulated signals with deterministic center frequency and bandwidth. Different signal to noise ratio signals were simulated. It has been observed that STFT spectral center frequency and bandwidth estimators were less biased than the CWT ones, although the last ones were less sensitive to the center frequency variations. Applying the proposed noise cancellation technique to simulated signals reduces the spectral estimators’ errors. As an example, a typical noisy signal with 10dbSNR, a reduction of 88% and 97% was obtained on the RMS bias of the estimation of the center frequency and bandwidth estimators respectively.