As part of the DeepQuadrature project, a team of researchers from the Warsaw University of Technology is working on simplifying and accelerating optical measurements: from metal microstructures to biological cells. DeepQuadrature is an idea focused on “seeing more with less.” The goal is to extract the maximum amount of information from a single image taken using a measurement technique that employs light as a marker. Normally, to extract this information (e.g., layer thickness, surface shape, internal structure of a biological cell), multiple images with complementary data are required to enable precise reconstruction. DeepQuadrature, a deep learning model, makes it possible to achieve all this using just a single photograph thanks to artificial intelligence. Instead of constructing complex hardware, the researchers aim to provide a software tool that “adds” missing information. They are working to ensure that this system functions not only in a general sense but is also tailored to specific applications, such as studying certain types of cells.

Measurement techniques based on fringe pattern imaging, such as holographic microscopy and interferometry, are widely used in biology, optics, and materials engineering due to their high precision and non-invasive nature. A key limitation of their accuracy is information throughput, known as the space-bandwidth product (SBP), which determines the quantity and quality of information that can be extracted from an image. DeepQuadrature increases SBP purely numerically, without the need to modify the optical system. Thanks to training on synthetic data with diverse geometries and fringe frequencies, it can adaptively analyse both technical and biological objects—ranging from surface microstructures of metal alloys to images of HeLa cells. In experimental studies, researchers at QCI Lab, in international collaboration, achieved accuracy comparable to multi-frame methods while maintaining the simplicity of a single-frame approach.

As part of the project, our researchers are training the neural network to "fill in" missing data based on a single image. They are evaluating how well it performs on real-life data, testing its potential applications in biological imaging and technical material analysis, and developing datasets and tools that will later be made available to other researchers, particularly microbiologists and doctors.

The team from WUT consists of four researchers from the Quantum Imaging Laboratory (QCI Lab) at the Faculty of Mechatronics: Maria Cywińska, PhD, Prof. Michał Jóźwik, Prof. Krzysztof Patorski, and Prof. Maciej Trusiak. Each was responsible for different aspects of the project—from building the AI model to designing experiments and analysing data. Our researchers collaborated with UiT The Arctic University of Norway in Tromsø (Dr. Azeem Ahmad and Prof. Balpreet Ahluwalia) and the University of Münster (Prof. Bjorn Kemper). The result of this collaboration is a scientific publication in the prestigious Journal of Physics: Photonics. The publication can be accessed here. The lead author of the study is Maria Cywińska, PhD, who developed the neural network-based algorithm and conducted its testing. The neural network can “infer” what an image looks like when phase-shifted by 90 degrees. She was also involved in experimental testing and result analysis: the team examined whether their method works not only on computer-generated data but also in real-life measurements.

QCI Lab is part of the Institute of Micromechanics and Photonics at the Faculty of Mechatronics at the Warsaw University of Technology. The laboratory aims to create new frameworks for computational imaging by integrating advancements in numerical reconstruction algorithms with experimental optical systems. Its main research focus is on the development of optical microscopy – striving to create technologies that enable deeper, faster, and more reliable imaging without the use of markers, featuring an exceptionally high signal-to-noise ratio and a large space-time bandwidth product. The laboratory focuses on both coherent imaging techniques (interferometry, holography) and incoherent imaging techniques (Fourier ptychography, differential phase contrast). Particular emphasis is placed on the quantitative nature of optical measurements, providing efficient, modern tools for non-invasive and precise diagnostics. QCI Lab is led by Prof. Maciej Trusiak and Piotr Zdańkowski, PhD, with Błażej Żyliński responsible for coordinating the work.

Saving time, space, and money

Multi-frame methods require expensive equipment to precisely generate frames. DeepQuadrature allows costly hardware components to be replaced with software.

Maria Cywińska, PhD, highlights the savings enabled by the tool’s application: “The consequence is not just cost reduction. Single-frame measurements are faster and easier, allowing for mobile use beyond the optical laboratory. This presents a great opportunity for the widespread adoption of this technology.”

The greatest beneficiaries will be those who rely on advanced optical measurements: scientists, doctors, and materials engineers. With DeepQuadrature, measurements can be more affordable (as expensive equipment is no longer needed) and faster (since a single capture suffices). This makes the tool viable for real-life applications beyond the laboratory: in hospitals, in the field, or even in education. It’s worth emphasizing that the method delivers results as reliable as much more costly and complex techniques. Moreover, DeepQuadrature enables “replacing hardware with software”, which is essentially like carrying a laboratory in your pocket.

The tool paves the way for more accessible diagnostics and precise measurements across various fields. Our researchers have trained DeepQuadrature for a general case where the network is prepared for a wide range of tasks. “In the future, we aim to conduct training with a specific experimental task in mind, such as observing a particular type of biological samples. This will allow us to compare results between these two scenarios,” summarizes Maria Cywińska, PhD.

The research activities are funded by the National Science Centre under the PRELUDIUM programme (2021/41/N/ST7/04057, budget: 139,568 PLN) and the OPUS programme (2020/37/B/ST7/03629, budget approx. 2,000,000 PLN), as well as the Ministry of Education and Science (Polish Metrology PM/SP/0008/2021/1, budget: 572,000 PLN).

The entire DeepQuadrature model, along with the datasets used for its training and validation, is publicly available on GitHub, promoting open science.