Announcing a new publication from Opto-Electronic Sciences ; DOI 10.29026/oes.2026.260019 .
Researchers have developed a physics-informed neural-field framework for partially coherent quantitative phase imaging. Termed the universal neural-field solver for differential phase contrast microscopy (USDPC), it embeds a rigorous bilinear image-formation model into neural-field optimization to jointly recover nonlinear sample phase and unknown pupil aberrations without ground-truth labels or additional calibration measurements. Experiments on standard targets, histological sections, and live cells demonstrate accurate recovery of large phase variations and improved spatial resolution under aberrated conditions, providing a unified approach to challenging inverse problems in computational imaging.
Transparent biological specimens contain rich structural and dynamic information that remains invisible to conventional intensity-based microscopy. From subtle changes in living cells to complex micro- and nanoscale structures, visualizing these features is essential for understanding biological processes, investigating disease mechanisms, and advancing biomedical research. Quantitative phase imaging (QPI) provides a powerful solution by recovering specimen-induced phase variations, enabling label-free and quantitative visualization. Among QPI techniques, differential phase contrast (DPC) microscopy is particularly attractive for its stable, speckle-free phase measurements, resolution beyond the coherent diffraction limit, and compatibility with conventional bright-field microscopes.
However, accurately recovering phase information from complex specimens and nonideal optical systems remains challenging. Conventional DPC reconstruction relies on linear models based on the weak-object approximation and an ideal pupil. These assumptions break down for large phase variations, substantial absorption, or unknown optical aberrations, leading to phase underestimation, structural distortion, and reduced spatial resolution. Existing nonlinear, aberration-correction, and learning-based methods address parts of this problem but often require separate processing, labeled data, or additional calibration. A unified framework that can recover nonlinear specimen information while correcting unknown aberrations directly from standard DPC measurements is therefore highly desirable.
To address this challenge, the authors of this article developed a new computational framework called USDPC. Rather than linearizing the imaging process, USDPC incorporates the physics of partially coherent imaging directly into reconstruction, enabling the joint recovery of specimen information and unknown optical aberrations.
USDPC uses two implicit neural fields to represent the specimen’s complex transmittance and the system’s pupil aberrations. A differentiable bilinear forward model then jointly optimizes these representations directly from the measured intensity images. This allows USDPC to recover phase and absorption while correcting unknown aberrations, without relying on the weak-object approximation or an ideal pupil.
Comprehensive simulations and experiments demonstrated that USDPC accurately recovered large phase variations and effectively corrected unknown optical aberrations, achieving improved reconstruction accuracy and spatial resolution compared with conventional DPC. In biological imaging, USDPC produced clearer tissue boundaries and finer structural details in stained kidney tissue through correction of spatially varying aberrations. Long-term imaging of living HeLa cells further demonstrated stable quantitative phase reconstruction and captured dynamic cellular processes, including contraction, migration, division, and fusion.
By combining physical image-formation models with neural fields, USDPC could offer a flexible strategy for tackling other nonlinear inverse problems. The framework may ultimately extend beyond DPC microscopy to broader inverse scattering and computational imaging applications, opening new possibilities for accurate imaging under challenging experimental conditions.
Keywords: phase retrieval, nonlinear inverse problem, aberration correction, neural fields, differential phase contrast microscopy
# # # # # #
The Smart Computational Imaging Laboratory (SCILab: www.scilaboratory.com ) at Nanjing University of Science and Technology is affiliated with the Ministry of Education's "Spectral Imaging and Information Processing" Changjiang Scholars Innovation Team and an inaugural National Huang Danian-Style Faculty Team , led by Professor Qian Chen. Its academic leader, Professor Chao Zuo , is a Changjiang Distinguished Professor and a Fellow of SPIE, Optica, and IOP .
Supported by eight high-level research platforms, SCILab advances smart computational imaging and instrumentation for life sciences, biomedicine, intelligent manufacturing, remote sensing, and national defense. It has published more than 300 SCI-indexed papers , including more than 50 cover papers and 30 ESI Highly Cited/Hot Papers, with more than 24,000 citations. Its students have earned major dissertation, scholarship, and innovation awards, including the National Outstanding Doctoral Dissertation Award, Wang Daheng Optical Award, and overall championship in the "Internet+" Competition . The talent-development achievements have been featured by CCTV's Focus Report, People's Daily, and Optics & Photonics News.
Opto-Electronic Science (OES) is a peer-reviewed, open access, interdisciplinary and international journal. OES has been indexed in ESCI, Scopus, DOAJ, and CAS databases, with a Scopus CiteScore Tracker of 30.7 for 2025 and a JCR Immediate Impact Factor of 26.
More information: https://www.oejournal.org/oes
Editorial Board: https://www.oejournal.org/oes/editorialboard/list
All issues available in the online archive ( https://www.oejournal.org/oes/archive )
Submission of OES may be made using ScholarOne ( https://mc03.manuscriptcentral.com/oes )
CN 51-1800/O4
ISSN 2097-0382
Contact Us: oes@ioe.ac.cn
Twitter: @OptoElectronAdv ( https://twitter.com/OptoElectronAdv?lang=en )
WeChat: OE_Journal
Zhou S, Huang XY, Lu LP et al. Physics-informed neural fields enable blind aberration correction for partially coherent quantitative phase imaging. Opto-Electron Sci 5 , 260019 (2026). DOI: 10.29026/oes.2026.260019
Opto-Electronic Science