An Absolute Limit to Precision

“Imagine observing a small object behind an irregular, cloudy pane of glass,” explains Prof. Stefan Rotter from TU Wien’s Institute of Theoretical Physics. “Instead of a clear image, we see a complex light pattern with varying bright and dark patches. The question is: how precisely can we determine the object’s true position from this pattern—and what is the fundamental limit of that precision?”

Figure 1. AI Pushes Optical Imaging to Its Physical Limits.

This problem is highly relevant in fields like biophysics and medical imaging. When light scatters through biological tissue, information about deeper structures seems to be lost. But physics sets strict boundaries on how much of this information can, in principle, be recovered. Figure 1 shows AI Pushes Optical Imaging to Its Physical Limits.

To quantify this, the team used a theoretical measure called Fisher information, which describes how much an optical signal tells us about an unknown parameter, such as the object’s position. If the Fisher information is low, no method—no matter how sophisticated—can achieve high precision. Using this approach, the researchers calculated the theoretical upper limit of precision in various experimental scenarios.

Neural Networks Decode Chaotic Light

While TU Wien’s team handled the theoretical side, an experiment was carried out by Dorian Bouchet (University of Grenoble) with Ilya Starshynov and Daniele Faccio (University of Glasgow). They directed a laser at a small reflective object behind a turbid liquid, producing images that appeared as highly distorted light patterns. The turbidity was varied, making precise position detection more or less challenging.

“To the human eye, these images seem random,” says Maximilian Weimar (TU Wien). “But if we train a neural network on many such images with known object positions, it learns the relationship between patterns and positions.” After training, the network could predict the object’s location with remarkable accuracy—even for previously unseen patterns.

Approaching the Physical Limit

Crucially, the network’s precision was only slightly below the theoretical maximum calculated via Fisher information. “Our AI algorithm is not just effective—it is nearly optimal,” says Rotter. “It achieves almost exactly the precision that physics allows.”

This breakthrough opens possibilities for significantly enhancing optical measurement techniques across fields ranging from medical diagnostics to materials research and quantum technologies. The team now plans to collaborate with applied physics and medical partners to explore real-world applications of these AI-supported methods.

Source: SciTECHDaily

Cite this article:

Priyadharshini S (2025), AI Takes Imaging Right to the Edge of Physical Limits, AnaTechMaz, pp.851