Super-resolution ultrasound technology has been developed by researchers at the Beckman Institute for Advanced Science and Technology using deep learning. Traditional super-resolution ultrasound techniques use microbubbles, which are considered a contrast agent and can be injected into a blood vessel to increase the clarity of an ultrasound image. However, these techniques require a low concentration of microbubbles and processing speeds are much slower than traditional methods. In response to user feedback from Dr. Daniel Llano, associate professor of molecular and integrative physiology and neurologist at Carle Foundation Hospital, the Song group went back to the drawing board and modernized their approach to super-resolution technology, foregoing microbubble localization and tracking entirely. Instead, they used an artificial intelligence network to evaluate the information spatiotemporally across space and time. The new method allows blood flow to be visualized in real-time with processing speeds reduced from minutes to seconds, making it potentially useful for clinicians in a clinical setting.
According to a report, this is the first paper that achieved direct calculation of blood flow velocity using raw ultrasound data without any explicit microbubble localization or tracking. Processing speeds have been significantly improved from previous methods as well as post-processing being done in real-time. The collaboration between two research groups was made possible by Beckman’s shared environment.
The researchers hope that speeding up higher-resolution technology will make it a useful option for clinicians in medical settings. “We believe this technique has the potential for super-resolution to eventually be used in a clinical setting,” said Song.
This breakthrough demonstrates how deep learning can help develop innovative solutions that could lead us towards better healthcare outcomes with faster diagnosis times and more accurate results.