Computationally Efficient Implementation of Aperture Domain Model Image Reconstruction

• Published in: IEEE transactions on ultrasonics, ferroelectrics, and frequency control ferroelectrics, and frequency control, 2019-10, Vol.66 (10), p.1546-1559
• Main Authors: Dei, Kazuyuki, Schlunk, Siegfried, Byram, Brett
• Format: Article
• Language: English
• Subjects: <italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">in vivo; Acoustic noise; Acoustics; Algorithms; Aperture domain signal; Apertures; beamforming; Clutter; Computational efficiency; Computational modeling; Computer simulation; Data models; dimensionality reduction; Fourier transforms; Image quality; Image reconstruction; In vivo methods and tests; Independent component analysis; Liver; medical ultrasound; Microsoft Windows; Noise levels; physical model; Predictive models; reverberation clutter; Run time (computers); signal processing; simulation; Singular value decomposition
• ISSN: 0885-3010; 1525-8955
• DOI: 10.1109/TUFFC.2019.2924824

Aperture domain model image reconstruction (ADMIRE) is a useful tool to mitigate ultrasound imaging artifacts caused by acoustic clutter. However, its lengthy run-time impairs its usefulness. To overcome this drawback, we evaluated the reduced model methods with otherwise similar performance to ADMIRE. We also assessed other approaches to speed up ADMIRE, including the use of different levels of short-time Fourier transform (STFT) window overlap and examining the degrees of freedom of the model fit. In this study, we conducted an analysis of the reduced models, including those using Gram-Schmidt orthonormalization (GSO), singular value decomposition (SVD), and independent component analysis (ICA). We evaluated these reduced models using the data from simulations, experimental phantoms, and in vivo liver scans. We then tested ADMIRE's performance using full, GSO, SVD, and ICA-fourth-order blind identification (ICA-FOBI) models. The results from simulations, experimental phantoms, and in vivo data indicate that a model reduced using the ICA-FOBI method is the most promising for accelerating ADMIRE implementation. In the in vivo liver data, the improvements in contrast relative to delay-and-sum (DAS) using a full model, GSO, SVD, and ICA-FOBI models are 6.29 ± 0.25 dB, 11.88 ± 0.90 dB, 9.01 ± 0.67 dB, and 6.36 ± 0.27 dB, respectively; whereas, the contrast-to-noise ratio (CNR) improvement values in the same order are 0.04 ± 0.06 dB, -1.70 ± 0.17 dB, -1.51 ± 0.19 dB, and 0.12 ± 0.07 dB, respectively. The implementation of ADMIRE using the reduced model methods can decrease ADMIRE's computational complexity over three orders of magnitude. The use of a 50% STFT window overlap reduces ADMIRE's serial run time by more than one order of magnitude without any remarkable loss of image quality, when compared to the use of a 90% window overlap used previously. Based on these findings, a combination of the ICA-FOBI model and the use of a 50% STFT window overlap makes the ADMIRE algorithm more computationally efficient.