Researchers at the Urbana, Ill.-based Beckman Institute for Advanced Science and Technology have made a breakthrough in ultrasound technology by developing a new framework for super-resolution ultrasound using deep learning techniques. This cutting-edge development aims to address the challenges posed by conventional super-resolution ultrasound, which offers a clearer picture than traditional methods but suffers from slow processing speeds.
Super-resolution ultrasound techniques traditionally rely on the use of microbubbles, tiny gas spheres encased in a protein or lipid shell, as contrast agents to enhance image clarity. However, the need for microbubble localization and tracking slows down the processing time, making it impractical for real-time imaging in a clinical setting.
Pengfei Song, PhD, an assistant professor of electrical and computer engineering at the University of Illinois Urbana-Champaign, collaborated with Daniel Llano, MD, an associate professor of molecular and integrative physiology and a neurologist at Carle Foundation Hospital, to explore a new approach to super-resolution ultrasound technology. Their research paper, published in IEEE Transactions on Medical Imaging, presents their framework.
The conventional super-resolution ultrasound method falls short when differentiating vessels that are closely located. With the new super-resolution imaging, the distinction between single and multiple vessels becomes possible, providing valuable diagnostic information, particularly for conditions like cancerous tumors. However, the processing time required for these high-resolution images has hindered its practical implementation.
To overcome this challenge, the researchers decided to forgo microbubble localization and tracking and adopt a holistic approach. They evaluated the information spatiotemporally, considering both space and time, using an artificial intelligence network. This approach allowed them to determine the speed of blood flow and convert blurred images into high-resolution, clear ones.
The team achieved remarkable results by avoiding the laborious process of explicitly identifying microbubbles. Matt Lowerison, PhD, a Beckman Institute Postdoctoral Fellow and co-author of the paper, explained that the conventional method accumulates localized dots frame by frame to construct a super-resolved image, but this process is time-consuming. The deep learning network employed in the new framework eliminates the need for explicit microbubble identification, resulting in a faster production of super-resolution images.
With this groundbreaking technology, real-time visualization of blood flow becomes possible, as opposed to the still images generated by the slow conventional method. The processing speeds have been significantly reduced from minutes to seconds, allowing for real-time post-processing. The researchers anticipate that these advancements will make the higher-resolution super-resolution ultrasound a valuable tool for clinicians.
Song emphasizes the potential of this technique in a clinical setting. He highlights the significance of the collaboration between researchers from different fields, made possible by the shared lab space at Beckman Institute. The successful fusion of electrical and computer engineering with molecular and integrative physiology led to this breakthrough in ultrasound technology.
The framework developed by the Beckman Institute researchers holds immense promise for improving diagnostics and patient care, experts say. By providing real-time, high-resolution imaging, this super-resolution ultrasound technology has the potential to revolutionize medical imaging in a clinical setting, advancing the field of ultrasound and benefiting patients worldwide.