AI Gives Ultrasound Imaging a Boost
The preferred tool for imaging cells is the optical microscope, as it has the resolution needed to capture a cell’s main components. The technique works well if the cells only need to be observed for a few hours, but it breaks down for longer times, as photons from the light source eventually damage the cells. Ultrasound imaging allows for longer viewing times—24 hours or more—but its long wavelength, which can’t be shorter than a few hundred micrometers, leads to a resolution that is typically too low to be useful. Now Hirotsugu Ogi of the University of Osaka, Japan, and colleagues have demonstrated an artificial-intelligence (AI)-based method that can give low-resolution images of cells a high-resolution makeover [1]. Ogi expects that their method could improve the resolution of other types of microscopes.
For their demonstration the researchers collected more than 100,000 ultrasound and optical images of the same live stem cells. The images were then fed into an AI algorithm, which was trained to learn the relationship between the features of the two sets of images. The team then tested the algorithm with a new set of ultrasound images of cells and found that the output images matched the optical images of those cells. Ogi says that the key to the success was including both low- and high-frequency acoustic signals in the images. The low-frequency data capture mechanical resonances of the cell nucleus, which turned out to be needed to train the AI algorithm correctly.
–Katherine Wright
Katherine Wright is the Deputy Editor of Physics Magazine.
References
- N. Fujiwara et al., “Deep-learning generation of high-resolution images of live cells in culture using tri-frequency acoustic images,” Phys. Rev. X 15, 021015 (2025).