Motion-resistant vascular ultrasound imaging based on real-time eigen-filtering

Doppler flow imaging has become a standard clinical modality for vascular diagnostics. Nevertheless, it remains challenging to perform vascular ultrasound in more complicated diagnostic scenarios, because significant flashing artifacts may appear due to inadequate suppression of Doppler clutter arising from moving tissues. For one decade, we have been striving to achieve motion-resistant vascular ultrasound by designing advanced eigen-filtering algorithms whose attenuation response is adapted to clutter characteristics. Using a receiver operating characteristics analysis approach, we showed that in the presence of vessel pulsation and tissue vibration, our eigen-based motion-resistant signal processing chain yielded a significantly higher true positive rate (>90%) in depicting flow in comparison to non-adaptive signal processing chains. Another engineering challenge that we have overcome is the high computational demand of eigen-processing algorithms. We have successfully devised real-time implementations of eigen-based motion-resistant signal processing through designing parallel computing kernels that are executed on a graphical processing unit (GTX Titan X). In particular, we achieved real-time video-range throughput for full-view Doppler frames, up to a scan depth of 5 cm for slow-time ensemble length of 16 samples (i.e., beyond the typical requirement for carotid scans). These findings serve well to substantiate the practical feasibility of performing motion-resistant vascular ultrasound.Doppler flow imaging has become a standard clinical modality for vascular diagnostics. Nevertheless, it remains challenging to perform vascular ultrasound in more complicated diagnostic scenarios, because significant flashing artifacts may appear due to inadequate suppression of Doppler clutter arising from moving tissues. For one decade, we have been striving to achieve motion-resistant vascular ultrasound by designing advanced eigen-filtering algorithms whose attenuation response is adapted to clutter characteristics. Using a receiver operating characteristics analysis approach, we showed that in the presence of vessel pulsation and tissue vibration, our eigen-based motion-resistant signal processing chain yielded a significantly higher true positive rate (>90%) in depicting flow in comparison to non-adaptive signal processing chains. Another engineering challenge that we have overcome is the high computational demand of eigen-processing algorithms. We have successfully devised real-time implementations...