Recent developments in deep learning have created immense potential for ultrasound imaging research. This challenge is designed to explore the benefits of using deep learning to create images after:

  1. Plane wave ultrasound transmission, which has the potential to create new opportunities for ultrafast ultrasound imaging.
  2. Traditional focused ultrasound transmission, which is widely used in most clinical ultrasound systems available today.

Deep learning beamforming challenge with plane-wave transmissions: 

Ultrafast imaging is achieved by transmitting plane waves that span a wide region of interest (as opposed to transmitting focused beams line-by-line). Plane wave imaging increases frame rates by over 100-fold, when compared to focused transmissions, thereby enabling applications, such as real-time brain activity monitoring, Doppler imaging, or shear wave elastography.  Increasing frame rates for future tasks requires using only a few plane waves for beamforming and as a result, suffers from image quality degradation.  While previous work presented during the PICMUS challenge explored different beamforming methods for plane-wave images, including minimum variance beamforming and short-lag spatial coherence beamforming (among others), these methods are known to suffer from high computational complexity and complicate real-time applications.  The first and second tasks of this challenge will aim to balance both image quality and frame rates with novel deep learning approaches to plane wave imaging.

Deep learning beamforming challenge with focused ultrasound transmissions: 

Focused transmissions are typically used in most commercial ultrasound imaging scanners. One downside of focused transmissions is that only one focus is allowed per image acquisition. While synthetic aperture beamforming can be used to dynamically focus ultrasound transmissions, dynamic focusing generally involves a high computational load. The third task of the challenge explores deep learning approaches to achieve dynamic transmit focusing with the attractive possibility of faster and simpler computations than existing methods. 

Training data

A review of current literature on the topic of deep learning for ultrasound beamforming reveals that there are many different training approaches. For example, one approach does not use ultrasound images for training, but still arrives at good results. Another approach uses high-quality images during training. Other approaches operate directly on channel data prior to applying receive delays, use sub-aperture beamforming, or only replace portions of the beamforming process, etc. Given these multiple training approaches, the organizers have decided to keep the training open-ended and focus on challenging participants to produce a network that achieves specific tasks and meets specified requirements. Nonetheless, plane wave data that can be used for network training is available on the PICMUS website.

Literature References

The CUBDL organizers created a list of literature references that focus on applications of deep learning in ultrasound systems. Our goal is to provide researchers with an entry point to this growing field by collecting various literature demonstrating the impact of deep learning methodologies on many aspects of ultrasound imaging.

Updated July 5, 2021:

Our journal paper describing challenge outcomes was accepted for publication! 

D. Hyun, A. Wiacek, S. Goudarzi, S. Rothlübbers, A. Asif, K. Eickel, Y. C. Eldar, J. Huang, M. Mischi, H. Rivaz, D. Sinden, R.J.G. van Sloun, H. Strohm, M. A. L. Bell, Deep Learning for Ultrasound Image Formation: CUBDL Evaluation Framework & Open Datasets, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control (accepted July 1, 2021) [pdf]

Additional details surrounding challenge motivation and organization are available in our associated conference proceedings:

MAL Bell, J Huang, D Hyun, YC Eldar, R van Sloun, M Mischi, “Challenge on Ultrasound Beamforming with Deep Learning (CUBDL)”, Proceedings of the 2020 IEEE International Ultrasonics Symposium, 2020. [pdf]