RESUMO
Ultrafast ultrasound imaging enables the visualization of rapidly changing blood flow dynamics in the chambers of the heart. Singular value decomposition (SVD) filters outperform conventional high pass clutter rejection filters for ultrafast blood flow imaging of small and shallow fields of view (e.g., functional imaging of brain activity). However, implementing SVD filters can be challenging in cardiac imaging due to the complex spatially and temporally varying tissue characteristics. To address this challenge, we describe a method that involves excluding the proximal portion of the image (near the chest wall) and divides the reduced field of view into overlapped segments, within which tissue signals are expected to be spatially and temporally coherent. SVD filtering with automatic selection of cut-off singular vector orders to remove tissue and noise signals is implemented for each segment. Auto-thresholding is based on the coherence of spatial singular vectors, delineating tissue, blood, and noise subspaces within a spatial similarity matrix calculated for each segment. Filtered blood flow signals from the segments are reconstructed and then combined and Doppler processing is used to form a set of blood flow images. Preliminary experimental results suggest that the spatially segmented approach improves the separation of the tissue and blood subsets in the spatial similarity matrix so that automatic thresholding is significantly improved, and tissue clutter can then be rejected more effectively in cardiac ultrafast imaging, compared to using the full field of view. In the case studied, spatially segmented SVD improved the rate of correct automatic selection of thresholds from 78% to 98.7% for the investigated cases and improved the post-filter power of blood signals by an average of more than 10 dB during a cardiac cycle.