AI enhances phase imaging, promises clearer views of cells

04 Mar 2025

HZDR team exploits chromatic aberrations as route to clinically useful images.

Quantitative phase imaging (QPI), in which changes in parameters such as phase shift or path length reveal key visual information about targets, is potentially a valuable technique in biomedical and other imaging applications.

A project involving the Center for Advanced Systems Understanding (CASUS), part of the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) research center, plus partners Imperial College London and University College London, has now applied a generative AI model to the imaging technique.

Discussed at the recent 39th AAAI Conference on Artificial Intelligence (and in a preprint paper via HZDR), the AI approach exploits the usually undesirable chromatic aberrations in optical data, and allows one single exposure to produce the image quality needed for QPI to be more readily applied in biomedical applications.

Computational QPI has some inherent advantages, commented the researchers, since it requires less expensive equipment that conventional QPI. One of the most prominent computational QPI approaches is solving the transport-of-intensity equation (TIE), a differential equation calculating an image of the sample based on the recorded phase changes.

This is easy to integrate into an existing optical microscope set-up, but often requires multiple acquisitions with different focus distances to deal with image artifacts.

"Our approach relies on similar principles to TIE, but only needs one image because of a clever combination of physics and generative AI," noted Artur Yakimovich from CASUS.

Computational techniques used in clinical settings

Chromatic aberrations occur when lens systems cannot bring all wavelengths of polychromatic white light to a single converging point perfectly. Unless specialized corrective lenses are in place, the different wavelengths can have slightly different focus distances.

The new project's breakthrough was to record the phase shifts of the red, green and blue wavelengths separately using a conventional RGB detector, and use that information to build a "through-focus stack" of various focal planes. This stack then facilitates computational QPI to create an enhanced image.

"Using chromatic aberrations to realize QPI poses one challenge: the distance between the red light focus and the blue light focus is very small," commented CASUS researcher Gabriel della Maggiora.

"Solving the TIE the standard way just does not give meaningful results. So we reasoned that we could use artificial intelligence. After training a generative AI model with an open-access data set consisting of 1.2 million images, the model was able to retrieve phase information even when relying on just the very limited data input from the recording."

In trials the generative AI-based approach was applied to the analysis of red blood cells in a human urine sample, and was able to unveil the shape of these cells where an established TIE-based approach was not. An additional advantage was the virtual absence of cloud artifacts in the images calculated with the new generative AI-based QPI, said the project.

The CASUS Machine Learning for Infection and Disease group under Yakimovich is developing novel computational techniques for microscopy that could be immediately applied in clinical settings, and believes that the potential for the generative-AI approach is huge.

"Generative AI is prone to produce hallucinations, and incorporating physics-based elements is a key approach to reducing them," commented CASUS. "As the AI-based quantitative phase imaging example shows, this route is very promising."