Exploring Non-Contrast Power Doppler Imaging: Impacts of Advanced Acquisition Techniques, Filtering Architectures, and Signal Component Selection

Abstract

Hepatocellular carcinoma (HCC) is the most prevalent form of liver cancer and is the third leading cause of cancer death. Transarterial chemoembolization (TACE) is the most common first-pass treatment for HCC. TACE endpoint assessment is difficult intra-operatively, and post-operative CT assessment requires a 1-2 month latency period to allow the embolic agent to dissipate. Non-contrast power Doppler ultrasound is a potential solution for more robust intra-operative endpoint assessment for TACE, providing visualization of blood flow, or lack thereof, in the tumor. Such a modality would enable immediate retreatment if endpoints are not initially met. However, physiological and sonographer hand motion and poor imaging conditions due to co-morbidities limit the effectiveness of non-contrast power Doppler ultrasound. In this dissertation, we propose three solutions to enable visualization of slow blood flow with non-contrast power Doppler ultrasound. First, we explore block-wise filtering architectures to mitigate depth-dependent noise. We analyze differing spatial block sizes across varying noise gradients in simulation; we propose utilizing physically square blocks and realize contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) gains of 2.5 ± 0.56 dB and 2.2 ± 0.36, respectively, compared to standard filtering architectures. Second, we propose utilizing normalized cross-correlation as a new parameter for selecting singular values. Our method approaches performance of optimal manual selection of singular values, at a deficit of 0.03 dB in CNR and 0.02 dB in blood power error and outperforms adaptive singular value selection by 1.1 dB and 4.4 dB for those same metrics. Finally, we evaluate the effects of SNR – and by extension, coded excitation transmits – on singular value subspace delineation. As SNR increases, we observe an average increase of 14.0 ± 3.5 singular value components from 0 dB SNR to 50 dB SNR in the blood subspace, and a 35.7 ± 5.2 increase between matched in vivo uncoded and coded transmit data from healthy liver datasets. These findings suggest that filtering cutoff selection should be modified significantly as imaging scenarios, and thus SNR, change. The three techniques explored here can be implemented independently, or in concert to improve slow blood flow visualization and ultimately improve TACE outcomes.