Comparison of methods for detection of contrast agents in ultrasound signals

Detection and segmentation of events in noisy random signals, which are non-stationary or stationary by segments, are used when differentiation between tissues or between tissue and Ultrasound Contrast Agent (UCA) is required. Here 4 different detection algorithms were studied, for detecting small sections of UCA within tissue - 'transients'. The algorithms are based on linear time-frequency transforms: Autoregressive (AR) and the Short Time Fourier Transform (STFT). The processed signals were clustered and classified into 'Tissue' or 'Transient', by the Newman Pierson Decision Principle, the STFT with Smooth Threshold and the Novelty Detection algorithm with Kernel transform. The detection ability of the 4 methods were compared, using simulated signals and signals generated experimentally. The simulated signals include signals with different Transient-to-Tissue Energy Ratio (from -25 dB to 5dB) and different durations (20 and 150 samples in length). 'In-vitro' experiments were carried out with UCA (Optison) flowing through 2-6 mm Latex tubes inserted into real and artificial tissues. The STFT with Smooth Threshold method and Novelty Detection with Kernel Function method performed best. Thus, the series of laboratory experiments verified the simulation results that under similar conditions, flow of UCA in 2 mm tubes/arteries can be successfully detected.