Application of Echo-PTV and Deep Learning for the Visualization and Quantification of the Cerebral Microcirculation

Ultrasound Localization Microscopy involves tracking of microbubbles (<5 μm) in contrast enhanced ultrasound (CEUS) images for reconstructing the microvasculature and microcirculation in organs. The current work introduces several methods for enhancing the localization and tracking of microbubbles in clinical CEUS images for measuring the cerebral microcirculation in piglets. A U-net based deep learning method is used for enhancing noisy raw CEUS images, improving the precision of microbubble detection. Bubble tracking uses Kalman filtering along with a series of criteria to insure the spatio-temporal consistency in flow direction, velocity magnitude, and bubble image morphology. Trajectory assignments are then globally optimized. Based on synthetic data, the U-net enhancement significantly improves the processing speed and localization accuracy over conventional methods. Heatmaps of the bubble trajectories depict the complex cerebral micro-vessel network, where neighboring vessels separated by 40 μm can be distinguished. Based on the current framerate (44 fps), blood flow speeds in the 0.1 to 12 cm/s range can be well captured, but non-clinical systems with much higher framerates can capture higher speeds.