Unsupervised dealiasing and denoising of color-Doppler data

Abstract

Color Doppler imaging (CDI) is the premiere modality to analyze blood flow in clinical practice. In the prospect of producing new CDI-based tools, we developed a fast unsupervised denoiser and dealiaser (DeAN) algorithm for color Doppler raw data. The proposed technique uses robust and automated image post-processing techniques that make the DeAN clinically compliant. The DeAN includes three consecutive advanced and hands-off numerical tools: (1) statistical region merging segmentation, (2) recursive dealiasing process, and (3) regularized robust smoothing. The performance of the DeAN was evaluated using Monte-Carlo simulations on mock Doppler data corrupted by aliasing and inhomogeneous noise. Fifty aliased Doppler images of the left ventricle acquired with a clinical ultrasound scanner were also analyzed. The analytical study demonstrated that color Doppler data can be reconstructed with high accuracy despite the presence of strong corruption. The normalized RMS error on the numerical data was less than 8% even with signal-to-noise ratio as low as 10dB. The algorithm also allowed us to recover highly reliable Doppler flows in clinical data. The DeAN is fast, accurate and not observer-dependent. Preliminary results showed that it is also directly applicable to 3-D data. This will offer the possibility of developing new tools to better decipher the blood flow dynamics in cardiovascular diseases.