Researchers Use AI to Improve Ultrasound Image Quality

Ultrasound scans, also known as sonograms, are among the most common medical imaging procedures. They are used to track the development of a fetus during pregnancy and also to visualize the heart, liver, and other organs. However, ultrasound images are often of poor quality, not very detailed, so that it may be hard for a doctor to see relevant diagnostic information.

A team of researchers at the University of Virginia has an idea to make ultrasound scans better with the help of generative artificial intelligence (AI). Their method allows for improving the quality of images without the need for extensive data sets of patient scans, which could make diagnosis more accurate and broader applications of AI in medicine possible.

A Different Approach to AI Training

The research was carried out by Soumee Guha, a Ph.D. student at the University of Virginia’s Department of Electrical and Computer Engineering. The main goal of the project is to create new generative AI models that enhance image quality. Unlike traditional AI, such technology does not require a large data set of annotated images to train the algorithm.

It is hard to get enough medical images for training because of privacy issues. That is why new algorithms that enable working with less data are groundbreaking. The researcher stated that the developed AI could make some “visual ‘noise’ in ultrasound images,” thus, improving the quality of scans.

Potential Benefits Across Multiple Medical Fields

Although the study primarily focuses on prenatal ultrasound, the technology can be applied to different areas of medicine. Enhanced imaging capabilities can contribute to diagnosing liver disease, heart problems, and other conditions requiring an ultrasound examination, according to the researchers.

The technology allows a more comprehensive view of the body’s interior in order to obtain more information about a patient’s condition without causing harm or additional discomfort. Moreover, the study revealed the potential of using generative AI in healthcare when large-scale annotated datasets are unavailable.

“I love when math meets the real world,” said Guha. “I find it exciting to be able to take abstract ideas and turn them into algorithms that will directly benefit patients.” Her doctoral dissertation is titled Learning Under Multiplicative Noise: Principled Image Enhancement Frameworks for Coherent Imaging.

The development of such technology will benefit patients, enabling doctors to get accurate results faster and easier, improving treatment outcomes. The study and the work of many other scientists also demonstrate how AI can make medical imaging more accessible and valuable in clinical practice.