Enhancing time-frequency parameters estimation for Doppler Ultrasound blood-flow signals

Abstract

Doppler Ultrasound (DU) blood flow signals, particularly when collected under intra-operative conditions are noisy; accurate extraction of clinical parameters from their spectra becomes a difficult task. The spectral center frequency and bandwidth were estimated using two estimators with alternative time-frequency resolutions: a fixed resolution method, the Short-Time Fourier Transform (STFT) and the multi-resolution Continuous Wavelet Transform (CWT). Their performance was also assessed when the DU signals were pre-processed by a recently proposed Noise Cancellation Technique (NCTech). The NCTech algorithm enables quantification of the magnitude of the canceled noise in the form of percentage, called Cancellation Level (CL). Quantitative comparisons have been performed in terms of bias of the estimators when four signal-to-noise (SNRs) on DU simulated signals are employed: infinity, 20 dB, 10 dB and 5 dB. Results prove that CWT produced spectral parameters estimates with less bias than STFT; however these estimates were less consistent than the STFT ones. When NCTech is primarily applied to the signal, the STFT is the method to benefit most from this pre-processing technique. The CWT combined with NCTech produced estimates of both spectral parameters with better accuracy over the majority of the cardiac cycle, except where the frequency varies within a small range of frequencies during a short period of time.