Isolation of signal from stationary microbubbles adhered to vessel walls using an adaptive regression filtering technique

Abstract

The singular value filter (SVF) is proposed for isolation of adherent microbubble signal in ultrasound-based targeted molecular imaging. The SVF method involves signal decomposition of complex echo data such that ensembles are re-expressed along a new basis determined from principal component analysis (PCA) using the singular value decomposition (SVD) method. In contrast to many previously proposed PCA-based approaches, filter coefficients in SVF are dictated by a weighting function allowing for non-binary coefficients and based upon a signal model of the underlying source signals. The weighting function allows for filter coefficients to be determined adaptively from the shape of the singular value spectra of local regions of echo data, which is quantified using a parameter called the normalized singular spectrum area (NSSA). Simulations in FIELD II are performed to quantify the effects of acoustic scatterer motion characteristics, such as motion and decorrelation, on NSSA. Results confirm that the singular value spectrum flattens, and thus NSSA increases, monotonically with increased axial shift of scatterers between A-lines and increased differential motion. The SVF filter is validated experimentally in an ex vivo porcine artery with adherent microbubbles collecting on the lower wall due to application of acoustic radiation force. SVF was compared to a low-pass infinite impulse response (IIR) filter operating on pulse inversion (PI) data. Results from our ex vivo experiments indicate that signal from adherent microbubbles exhibits higher dimensionality and thus higher NSSA than signal from vessel wall and free microbubbles. SVF provided > 40dB contrast of adherent microbubble signal over vessel wall and > 32dB contrast of adherent microbubble over free microbubble signal.