Applying AI to improve ultrasound accuracy

Prof. Yonina Eldar of the Department of Computer Science and Applied Mathematics, devises data processing methods with applications in fields as diverse as medicine, radar and communication systems. She aims to incorporate artificial intelligence (AI) tools to better acquire and extract information, laying the foundations for new technologies that can see, hear and communicate beyond existing limits.

To process analog signals – speech, music, or images – we first have to sample and transform them into bits of computer data. Once they are in digital form, sophisticated algorithms can be applied to improve the information. A good example of signal processing is hearing aids, which record environmental sounds, process them, and then output an improved sound. However, signal acquisition and processing are limited by the sampling rate, a primary boundary that the Eldar lab set out to break – by introducing a new paradigm of reduced sampling.

One intriguing application for the Eldar group’s method pertains to ultrasound imaging, a common diagnostic tool. Since ultrasound is a wave-based imaging technology, its precision is limited by physical limitations. To get past these limitations, a contrast agent may be used. The contrast agent contains microbubbles that can be injected to the blood vessels, where they act as reflectors of ultrasound beams. The use of models and AI algorithms can help pinpoint the locations of these individual microbubbles as they travel through the bloodstream, creating a map of small blood vessels that were not visible before. The Eldar lab has recently employed these methods in a clinical study that was conducted in collaboration with radiologists from Beilinson Hospital in Israel under Dr. Ahuva Grubstein.

During this study, 21 female patients with breast tumors, both benign and malignant, were scanned after they were injected with a contrast agent containing microbubbles. The researchers used AI deep learning methods to extract and demonstrate the pattern of the small blood vessels inside the different kinds of breast tumors. Their findings were presented at the 2021 conference of the Medical Image Computing and Computer Assisted Interventions Society.

“Our team showed that this method is able to map out small blood vessels inside tumors,” says Dr. Keren Peri-Hanania, a member of the Eldar lab. “The implication is that using these super-resolution methods could be a viable way to better characterize tumors using ultrasound.”

While these early results may end up assisting physicians with breast cancer diagnosis, the Eldar team is working hard on various other ways to improve clinical diagnosis using ultrasound. This includes hardware improvements of the ultrasound device, with the goal of producing a high-quality, small and portable ultrasound device that may be also worn as a patch over a patient’s body – enabling continuous monitoring; using AI to combine the input of ultrasound with that of other imaging modalities or to guide novice ultrasound users while conducting scans; and extracting important properties of body tissues from the pattern of ultrasound transmission.

Prof. Yonina Eldar’s research is supported by Miel de Botton; Jean and Terry de Gunzburg; the Henry Chanoch Krenter Institute for Biomedical Imaging and Genomics; Marcos Lederman; and the Sagol Weizmann-MIT Bridge Program. Prof. Eldar is Head of the Manya Igel Centre for Biomedical Engineering and Signal Processing, and the incumbent of the Dorothy and Patrick Gorman Professorial Chair.