Vector Flow Velocity Imaging of Human Placenta using Angle-resolved Ultrasound and Deep Learning - This proposal describes a five-year research and career development program to prepare Dr. You Li for a career as an independent investigator. The program will build upon Dr. Li’s multidisciplinary background as a biomedical engineer, trained in medical ultrasound imaging, by providing expertise in obstetrics, the application of machine learning in medical imaging, and translational research. The PI will be mentored at Stanford University by Drs. Jeremy Dahl (primary mentor, medical ultrasound), Virginia Winn (co-mentor, obstetrics and gynecology), and Matthew Lungren (co-mentor, radiology and artificial intelligence). Human placenta plays a vital role in human development, and its abnormalities may cause significant consequences to both the mother and the fetus. Preeclampsia, in particular, is a common disorder that affects approximately 1 in 33 pregnancies in the United States and accounts for 18% of pregnancy-associated maternal death. Many placental abnormalities, including preeclampsia, are related to the hemodynamics and growth of vessels in placenta. Despite the severe consequences of placental abnormalities, our understanding in placenta and placental abnormalities is lacking. One primary reason for the gap of knowledge is the inability to observe the hemodynamics of placenta in vivo. Currently, B-mode and Doppler ultrasound are the primary imaging modalities in imaging the placenta and its vasculature. However, significant limitations exist in the ability of Doppler ultrasound to visualize and measure detailed flow velocities in placental vasculature. It has low sensitivity to small vessels in the placenta, including spiral arteries and chorionic villi, and can only measure flow velocity along the ultrasound beam direction, requiring tedious manual angle correction for flow along any of the visible vessels if quantitative information is desired. These limitations make conventional Doppler ultrasound poorly suited for imaging the hemodynamics of the complex vasculature of human placenta. To provide full and detailed characterization of placental hemodynamics, we propose to develop a vector flow velocity imaging technique using deep neural network models and multiple angle plane wave ultrasound transmits. This technique will be able to quantitatively image both the flow velocity magnitudes and flow directions of millimeter-diameter vessels in human placenta over a large field of view. Aim 1 and 2 will be focused on the technical development of the technique to provide a semi- real-time vector flow imaging system based on a research ultrasound scanner, paving way for Aim 3, which will be focused on validating the clinical value of the technique on a pilot clinical study to image the hemodynamics in spiral arteries and chorionic villi of pregnant women. Successful completion of the project will provide a novel technique for the scientific and early clinical assessment of placental hemodynamics, development, and abnormalities including the development of the villous tree structure, placenta accreta, preeclampsia, and placental insufficiency. During the project, the PI will receive training in machine learning, obstetrics, translational research, and career development skills, which will transition the PI into an independent faculty.
