Traditionally, ultrasound beamforming is done with an assumed constant sound speed across the entire image. However, in actual imaging scenarios, especially in vivo, areas have varying sound speeds, which can cause distortions (phase aberration) in the received signal wavefront. These distortions, in turn, lead to increased noise in the final image and decreased visibility and resolution in regions of interest. This image degradation can especially be prevalent in cases such as abdominal imaging for high BMI patients. In order to improve image quality, estimated heterogeneous sound speed maps can be used in the beamforming process, but current estimation methods for pulse-echo ultrasound are applicable in only specific acquisition situations, may only resolve well in one direction, or do not easily work in real time. To improve the sound speed estimation process, we are working on a data driven approach for estimation by training a neural network to estimate sound speed. Deep learning has been relatively less explored for sound speed estimation and can potentially better link the complex relations between received echoes and sound speeds in the field.  

In addition, we are exploring ways to make beamforming faster on ultrasound devices with limited GPU capability, such as point of care ultrasound devices, by using frequency-domain beamforming methods that take fewer computational steps to form an image compared to traditional time-domain beamforming. With faster algorithms, we can expedite the image formation process in a variety of patient care scenarios, including emergency rooms and rural areas. Currently, we are working on a method that avoids interpolation during the beamforming process and performs purely phase multiplications in the frequency domain to transform received signals into a final image in the spatial domain. We are also exploring the use of chirp transmits and their effect on frequency-domain beamforming and image quality for high-depth images.

L. Zhuang, W. Simson, O. Ostras, D. Hyun, G. Pinton and J. Dahl, "Abdominal Sound Speed Estimation Using Neural Networks Trained on Wave Propagation Physics," 2023 IEEE International Ultrasonics Symposium (IUS), Montreal, QC, Canada, 2023, pp. 1-4, doi: 10.1109/IUS51837.2023.10308076.