Neural fields correct aberrations in phase imaging

Researchers published a physics-informed neural-field method that can recover phase and optical aberrations from partially coherent quantitative phase imaging without ground-truth labels or extra calibration. The approach improved reconstruction quality on simulations, tissue sections and live cells, pointing to a broader path for harder computational imaging problems.

Why it matters: - Transparent biological specimens often reveal important structure only through phase changes, not intensity alone. - Quantitative phase imaging can measure those phase shifts without labels, which is valuable for biology and biomedical research. - Conventional differential phase contrast microscopy struggles when specimens have large phase changes or the optical system has unknown aberrations. - Those failures can cause phase underestimation, distorted structure and lower spatial resolution.

What happened: - Researchers published a new physics-informed neural-field framework in Opto-Electronic Sciences on September 14, 2026. - The method is called the universal neural-field solver for differential phase contrast microscopy, or USDPC. - USDPC jointly recovers nonlinear sample phase and unknown pupil aberrations from standard DPC measurements. - The framework does this without ground-truth labels or extra calibration measurements.

The details: - USDPC embeds a rigorous bilinear image-formation model into neural-field optimization. - Two implicit neural fields represent the specimen’s complex transmittance and the system’s pupil aberrations. - A differentiable bilinear forward model optimizes both representations directly from measured intensity images. - The approach avoids the weak-object approximation and does not assume an ideal pupil. - Simulations and experiments showed accurate recovery of large phase variations and better spatial resolution under aberrated conditions. - Tests included standard targets, histological sections and live cells. - In stained kidney tissue, USDPC produced clearer tissue boundaries and finer structural detail after correcting spatially varying aberrations. - In long-term HeLa cell imaging, the method delivered stable quantitative phase reconstruction and captured contraction, migration, division and fusion. - The publication appears in Opto-Electronic Sciences under DOI 10.29026/oes.2026.260019.

Between the lines: - The work targets a core limitation in computational imaging: reconstruction methods often separate phase recovery from aberration correction. - By folding physics into neural fields, USDPC points to a single framework that can handle nonlinear specimens and imperfect optics at the same time. - That combination may be especially useful where calibration is difficult or where the sample changes over time. - The researchers say the same strategy could extend beyond DPC microscopy to inverse scattering and other computational imaging tasks.

What's next: - Future work may adapt the USDPC framework to other nonlinear inverse problems in imaging. - The method could also be tested in broader inverse scattering settings and additional challenging experimental conditions. - If those extensions hold up, physics-informed neural fields could become a more general tool for blind correction in computational imaging.

The bottom line: - USDPC shows that neural fields plus a physics-based forward model can recover phase and correct aberrations at the same time, without extra calibration or labels.