The quantum correlations of photon pairs inspired the development of ghost imaging, where only a single detector is placed after the object. This method offers fundamental advantages compared to conventional imaging, including the operation at ultra-low photon flux and the potential for higher resolution [1]. Over the last year, it was demonstrated experimentally that nonlinear metasurfaces with a thickness of only a few hundred nanometers can facilitate strongly enhanced generation of photon pairs through spontaneous parametric down-conversion. Furthermore, the quantum photon state can be tailored to feature spatial [2] and spectral [3] entanglement. However, the potential of metasurface photon-pair sources for quantum imaging remained untapped.

Here, we formulate and experimentally demonstrate, for the first time to our knowledge, the unique benefits and practical potential of quantum imaging based on nonlinear metasurfaces, facilitating an efficient protocol combining ghost and all-optical scanning imaging. The metasurface incorporates a subwavelength-scale silica metagrating on a lithium niobite thin film. A distinguishing feature of the metasurface is the capability to all-optically scan the emission angle in the direction across the grating simply by tuning the pump beam wavelength. Simultaneously, the photon emission is broad and anti-correlated along the grating direction, allowing for ghost imaging. Thereby, a 2D object can be reconstructed using just a 1D detector array in the idler and a bucket detector in the signal paths, by recording the dependencies of coincidences on the pump wavelength. Our results reveal new possibilities for quantum imaging and pave the way for advancements of quantum technologies using ultra-compact nanostructured metasurfaces.

References
[1] M. J. Padgett and R. W. Boyd, Philos. Trans. R. Soc. A 375, 20160233 (2017).
[2] J. H. Zhang et al, Sci. Adv. 8, eabq4240 (2022).
[3] T. Santiago-Cruz et al, Science 377, 991 (2022).

—

Entanglement quantification is paramount to fundamental research and many cutting-edge applications. Various approaches to entanglement detection have been proposed, but they usually provide only a witness or require the interference of multiple copies of the system under test. It was shown that quantum tomography is necessary to determine the entanglement in an unknown quantum state exactly. The tomography yields complete information with the drawback of unfeasible scaling with the system complexity.

Artificial neural networks were recently exploited for tomography by approximating the state wavefunction and for entanglement witnessing. Despite these achievements, the question of how precisely the entanglement can be estimated directly from the incomplete measured data remains open. We approach this problem using deep neural networks [1]. We focus on characterizing two- and three-qubit entanglement sources with imminent applications in quantum communications. We demonstrate significantly lower errors of quantum concurrence estimation from heavily undersampled local measurements compared to state-of-the-art quantum tomography.

To further verify and support our results, we applied our deep neural network methods to experimental data obtained from two entangled photon sources: spontaneous parametric down-conversion and a quantum dot. In both cases, deep neural network approaches outperform the tomographic approach. We further trained deep neural networks that infer four- and five-qubit mutual information matrix from local and informationally incomplete measurements. Our results indicate that we can omit most of all projectors to still reach an accurate prediction of quantum entanglement.

[1] D. Koutný, et al., Deep learning of quantum entanglement from incomplete measurements, Sci. Adv. 9, eadd7131 (2023).

—

Harnessing localized heating through efficient light absorption has opened up exciting possibilities for precise control of nanoscale systems. Such optothermal effects alter the environment locally through the induction of fluid flows enabling the active manipulation of systems out of equilibrium [1]. Here, we explore how the control of light-induced fluid flows facilitates complex, yet steerable, motion of micro- and nanoparticles [2-4]. We demonstrate that the rapid rotation of micromotors is solely powered by the generation of local temperature gradients, whose direction can be predetermined through light polarization. Furthermore, by regulating heat-mediated fluid flows, we have achieved the self-assembly of individual particles into colloidal structures that resemble molecular structures [5]. The inherent activity of these colloidal molecules provides valuable insights into the organization of fundamental building blocks into functional materials. Finally, we introduce digital holographic microscopy (DHM) as a powerful tool for investigating the onset of fluid flows in microfluidic chips through highly accurate 3D tracer tracking [6]. We combine this technique with optical diffraction tomography (ODT) to precisely determine the temperature profile. Our findings highlight the potential of optothermal manipulation in micro- and nanoscale systems as a viable alternative approach for nanotechnologies with enhanced functionality and control.

References:
[1] P. S. Kollipara, et al., ACS Nano. 2023, 17(8), pp- 7051-7063.
[2] D. B. Ciriza, et al., ArXiv 2023, 2305.06688.
[3] F. Schmidt, et al., Phys. Rev. Lett. 2018. 120(6), pp- 068004.
[4] F. Schmidt, et al. Nat. Commun. 2021, 12(1), pp- 1902.
[5] F. Schmidt, et al., J. Chem. Phys. 2019, 150(9), pp- 094905.
[6] B. Ciraulo, et al. Nat. Commun 2021, 12(1), pp- 2001.

—

Speckle, the granular interference pattern produced when coherent light is scattered by a rough surface, is commonly thought to be detrimental to optical systems. However, the processes which form the speckle are entirely linear, and speckle can therefore be used as a surprisingly sensitive probe of the properties of the light, the environment and the scatterer.

Here, I will show how recent innovations in speckle metrology have pushed the sensitivities forward by multiple orders of magnitude. This has allowed the measurement of the wavelength of light to attometre precision [Laser Photonics Rev. 14, 2000120, 2020], detected changes in the refractive index of the environment at parts per billion [ACS Photonics 9, 830−836, 2022]; and enabled displacement of the scattering surface by tens of picometres [arXiv:2110.15939v1, 2021].

—

Kerr-instability amplification (KIA) is a process by which two pump photons amplify a signal photon to create an idler photon. KIA is like four-wave optical parametric amplification (FWOPA), but high intensities allow for a modified definition of phase matching. Contrary to FWOPA, which has a strict phase-matching condition, we can utilize one angle to amplify multiple frequencies at one time, thus amplifying supercontinuum spectra.

It has previously been shown that cascaded four-wave mixing (CFWM) can be used as a tunable Raman probe for broadband stimulated Raman spectroscopy (SRS) experiments. Given the broadband amplification that is possible with KIA, we can create a tunable probe by directly amplifying the wavelength of interest from a supercontinuum seed or use a cascaded beamlet created from KIA above saturation. If we amplify directly, we can achieve better power stability than CFWM. Furthermore, KIA imparts less dispersion on the pulse than the material itself, so the amplified pulse is not only broadband but also near transform limited.

Given the tunability and stability of our method, we expect our new approach to become a desirable way to perform broadband tunable stimulated Raman spectroscopy experiments. We are applying this method, which we have coined Kerr amplification broadband stimulated Raman spectroscopy (KAB-SRS) to measure trace amounts of nutrients and contaminants in the Laurentian Great Lakes.

—

Widefield fluorescence imaging has significant challenges in visualizing neuronal fibers near cell bodies. Specifically, out-of-focus and scattered light from the bright cellbody often obscure nearby dim fibers and degrade their signal-to-background ratio. Scanning techniques can solve this problem but are limited by reduced imaging speed and increased cost. We greatly reduce stray light by modulating the illumination intensity to different structures. We use a digital micromirror device (DMD) in the illumination channel of a common widefield microscope to target illumination to dim fibers. We identify fibers by real-time image processing comprised of several techniques. With the setup, we illuminate bright cell bodies with minimal low light intensity, and illuminate in-focus fiber-like structures with high light intensity to reveal dim fibers. Thus, we minimize the background and enhance the visibility of fibers in the final image. This target illumination significantly improves fiber contrast while maintaining a fast-imaging speed and low cost. Using a target illumination setup in a widefield microscope, we demonstrate confocal quality imaging of complex neurons in live anesthetized C. elegans, restrained zebrafish larva, and in vitro mouse brain slice.

—

Phthalocyanines absorb intensely at far-red and near-infrared wavelengths, in the phototherapeutic window. Their photoproperties can be modulated by playing with the substitution and metalation pattern. Phthalocyanines for anti-cancer photodynamic therapy should be water-soluble or embedded into DDS. Our different strategies will be developed: fluorination, encapsulation into liposomes, PVP and polymeric micellar formulations, covalent incorporation into silesquioxane or poly-L-glutamic acid nanoparticles. Their respective advantages and subsequent molecular design will be discussed.

Some references:
Chem. Soc. Rev., 2020, 49, 1041-1056
Coord. Chem. Rev. 2010, 254, 2792-2847
J. Porphyrins Phthalocyanines 2019, 23, 611-618
Eur. J. Med. Chem. 2016, 124, 284-298
J. Porphyrins Phthalocyanines 2019, 23, 1587-1591
Mol. Pharmaceutics 2018, 15, 2594-2605.
Macromolecular Bioscience 2022, 22, 2200130
Chem. Commun. 2019, 55, 11619-11622.
Nanoscale, 2017, 9, 16622-16626
Mol. Pharmaceutics 2023, doi.org/10.1021/acs.molpharmaceut.3c00306

—

The use of scientific techniques to examine heritage materials in situ is an established methodology for many decades [1] The stained-glass technique consists of assembling small flat pieces of cut whitish and colored glass to form a picture. A commonly applied coloring technique from the Middle Ages through the present times is silver staining. With this method an elemental silver-containing film is applied to the glass surface. Depending on the fabrication process parameters a wide variety of colors could be obtained. Driven by the need to have better insights in the unravelling of the fabrication conditions via non-destructive means, a study was performed to explore the benefits of using absorption spectroscopy. The basis of these analyses was bringing the optical characteristics of the recorded transmission spectra (surface plasmon resonance profile and color) in relation to the applied technology, i.e., the firing temperature and type of silver compound, the silver concentration, the type of ochre, the application of an Ag/Cu mixture, the presence of a metal tin layer at the top surface and the iron composition of the base matrix. In this research we investigated samples with known fabrication conditions. A benchmarking between the selected optical based non-destructive and destructive chemical-based methods, allowed us identifying the optical measures that match with the material properties. The latter include the penetration depth, the generated particle sizes, and distributions. The research resulted in a flowchart scheme allowing us to give clues on the silver-staining technology via non-destructive means [2].

[1] W. Meulebroeck et al., Authenticity screening of stained glass windows using photonics, Sci Rep. (2016) 1–10.
[2] W. Meulebroeck et al., Using absorption spectroscopy as a non-destructive tool for the study of silver-staining in glass: An operational flowchart to assign the technology parameters, J Non Cryst Solids. 602 (2023).

—

New generations of laser sources, with high pulse energy, wide continuous tunable range, and a compact form, are in high demand to advance two-photon and three-photon microscopy to its full potential for deep-tissue imaging. Multimode fibers have emerged as an early-stage but promising candidate. Effective control of nonlinear processes at high power levels in multimode fibers (MMFs) would unlock new possibilities for diverse applications including high-power fiber lasers, which potentially address the aforementioned need. This talk will discuss our recently developed approach that exploits the spatial and temporal degrees of control of nonlinear effects in step-index MMFs using a 3D-printed programmable fiber piano. By leveraging the rich spatiotemporal degrees of freedom and the high spectral brilliance in SI MMF, We have achieved broadband high-peak-power spanning nm, resulting from combined spectral energy reallocation and temporal shortening uniquely enabled by the fiber shaper. Its potential as a nonlinear imaging source is further demonstrated by applying the MMF source to multiphoton microscopy, where multi-fold signal enhancement is achieved for label-free tissue imaging with adaptive optimization.

—

Optical imaging through scattering media [1] is an important challenge in a variety of fields ranging from microscopy to telecommunications. While advanced wavefront shaping techniques have offered significant breakthroughs in the past decade [1–3], current techniques still require a known guide-star and a high-resolution spatial-light-modulator (SLM) [4], or a very large number of measurements, and are limited in their correction field-of-view.
We introduce a guide-star free non-invasive approach that can correct more than 350,000 scattered modes using just 100 holographically-measured scattered random light fields. This is achieved by computationally emulating an image-guided wavefront-shaping experiment [5], where several ‘virtual SLMs’ can be simultaneously optimized to maximize the reconstructed image quality. Our method shifts the burden from the physical hardware to a digital, naturally parallelizable computation, leveraging state-of-the-art automatic-differentiation optimization tools used for the training of neural-networks [6]. We demonstrate the flexibility and generality of this framework by applying it to imaging through various complex samples and imaging modalities, including anisoplanatic multi-conjugate correction of highly scattering layers, lensless-endoscopy in multicore fibers, and acousto-optic tomography. The versatility, effectiveness, and generality of the presented approach have great potential for rapid noninvasive imaging in diverse applications.
References:
1. J. Bertolotti and O. Katz, “Imaging in complex media,” Nat. Phys. 18, 1008–1017 (2022).
2. I. M. Vellekoop and A. P. Mosk, “Focusing coherent light through opaque strongly scattering media,” Opt. Lett., OL 32, 2309–2311 (2007).
3. A. P. Mosk, A. Lagendijk, G. Lerosey, and M. Fink, “Controlling waves in space and time for imaging and focusing in complex media,” Nature Photon 6, 283–292 (2012).
4. R. Horstmeyer, H. Ruan, and C. Yang, “Guidestar-assisted wavefront-shaping methods

—

The main theme of the present research is to open a new avenue to further development of suitable devices for medical imaging that can promise to improve diagnostic capabilities and patient comfort.

Capsule endoscopy plays a crucial role in non-invasive medical diagnostics, enabling visualization of the gastrointestinal tract and detection of abnormalities. However, optical aberrations and large optical sizes because of multiple traditional lenses significantly limit the applications of capsule endoscopy.

In this work, we propose the design of a novel capsule endoscopy camera based on metalens technology. Our design methodology, which integrates three techniques—Zemax ray-tracing, Zemax physical propagation, and finite-difference time-domain simulation—represents an innovative approach to design and optimization.

The results of this research are important in several respects. The capsule endoscopy camera designed in this work achieved a remarkable 160° field of view and high image quality, as evidenced by a modulation transfer function value of 45% at 300 lp/mm and relative illumination of higher than 98%. Achieving a remarkably high field of view is one of the requirements for a comprehensive examination of internal organs and areas of interest. Furthermore, we have achieved a compact design with an optical diameter of around 1.6 mm and a total track length of 1.4 mm by incorporating metalens technology. The integration of metalens in capsule endoscopy represents a breakthrough in compact optical design.

To the best of our knowledge, it is the first time that a capsule endoscope camera simultaneously has such a high FOV, compact design and superior image quality. This innovative approach shows great potential in improving diagnostic capabilities and patient comfort.

—

We present the development and application of a beam-scanning Synchronous Red Green Blue – Fluorescence lifetime imaging microscope (SyncRGB-FLIM) system, optimized for real-time monitoring of metabolic activity of cells in 2D and 3D live cell models. The patented SyncRGB-FLIM method can be seen as an extension of conventional multiphoton MP-FLIM, while deploying a sub-10 fs ultra-broadband (>400 nm bandwidth) laser capable of simultaneous excitation of spectrally well-separated dyes or autofluorescence molecules [1]. The SyncRGB-FLIM method was used to retrieve metabolic imaging of both 2D and 3D live cell models. The specific fluorescence lifetime properties of endogenous metabolites of e. g. NADP/NADPH and FAD allow mapping the cell bioenergetics [2], while additional labeling can be used to identify the localization of selected cell compartments. Results will be presented on A549 lung cancer cell based-spheroids subjected to anticancer treatment with the drug paclitaxel. The treatment was studied and quantified through the mapping of lifetime ratios of the bound and unbound state of the metabolites that are spectrally separated. The effect of the drug was differentiated on the outer layer (proliferating cells) compared to the inner layer (quiescent cells). In summary, we present a novel bioimaging technique, with the potential to track the efficiency of cancer treatments through metabolic imaging in 3D cell samples.

[1] C. Maibohm et al., “SyncRGB-FLIM: synchronous fluorescence imaging of red, green and blue dyes enabled by ultra-broadband few-cycle laser excitation and fluorescence lifetime detection,” Biomed. Opt. Express, vol. 10, no. 4, 2019, doi: 10.1364/boe.10.001891.
[2] A. R. Faria, O. F. Silvestre, C. Maibohm, R. M. R. Adão, B. F. B. Silva, and J. B. Nieder, Nano Res., vol. 12, no. 5, 2019, doi: 10.1007/s12274-018-2231-5.

Acknowledgments: This research is funded by ANI via the ExtreMED project #NORTE-01-0247-FEDER-045932.

—

A multi-photon 3D micro-/nano-lithography technique will be introduced by explaining its principles, techniques, applications as a tool for rapid prototyping and technology for advanced additive manufacturing.
A possibility to use any color of spectrum from 500-nm-to-1200-nm with controlled pulse widths of 100-fs will be demonstrated revealing a delicate interplay of photo-physical mechanisms more than just two-photon absorption inducing localized photo-polymerization. An evolution of the polymerised volume during direct laser writing (DLW) via different energy delivery mechanisms will be discussed: one-/two-/three-photon absorption, avalanche ionization, and thermal diffusion leading to controlled photo-polymerization are revealed. The results can be used to tailor polymerized volume for increasing the 3D nano-printing performance. A non-trivial energy deposition by X-photon absorption with an onset of a strong lateral size increase at the higher pulse energy at longer wavelengths and can be understood as due to reaching epsilon-near-zero conditions. Such recent findings are valuable for further developing MPP technology to reduce the footprint size and increase its efficiency. Understanding mechanisms and appearance of λ-tunable commercial lasers are benefiting broad applications in advanced optical additive manufacturing areas of micro-optics, nano-photonic devices, meta-materials, and integrated-chips, and tissue engineering.
Finally controlled refractive index, high transparency and resilient as well as active micro-optical components will be showcased as their production route is enabled X-photon lithography in combination with calcination and atomic layer deposition. The achievements have immediate applications in sensing under harsh conditions, open space, and unmanned aerial vehicles (UAV).

—

Introducing multi-material properties for additively manufactured parts is currently a largely unmet challenge and most solutions rely on labor-intensive engineering during the manufacturing process. While grey-tone lithography has achieved significant progress to alter material behavior in a spatially resolved manner through the use of a single photoresist, such solutions mostly focus on mechanical alteration of the resist. We herein pioneer a two-in-one photoresist, from which stable microstructures and degradable structures can be printed by simple alteration of the laser power during manufacturing (Figure 1 a). [1]

The photoresist is based on a light-stabilized material, which utilizes naphthalene-BisTAD dynamic bonds (Figure 1 b) that are stable under green light irradiation but degrade in darkness. [2] We found that alteration of laser intensity during direct laser writing can modify the crosslinking density of resulting microstructures so that full degradation can be achieved at low intensity within five days, while structures written at high laser powers remain stable (Figure 1 c-d). Chemical alterations due to varying writing parameters were excluded with TOF-SIMS analysis. AFM-studies revealed unprecedentedly detailed elucidation of the degradation mechanism. Furthermore, degradation rates are tunable by the solvent in which the structures are immersed.

References
[1] Gauci S. C., Ehrmann K, Gernhardt M., Tuten B., Blasco E., Frisch H., Houck H. A., Jayalatharachchi V., Blinco J. P., Barner-Kowollik C. Adv. Mater. 2023, in press.
[2] Gauci S. C., Gernhardt M., Frisch H., Houck H. A., Blinco J. P., Blasco E., Tuten B., Barner-Kowollik C. Adv. Funct Mater 2022, 2206303

—

Optical fibers are key devices of modern photonics and have been successfully applied in a variety of fields, including telecommunication and life science. While the actual light transport capabilities are excellent, the main problem arises at the beginning of the fiber where the light is to be collected. However, the disadvantage of commercial step-index single-mode fibers is the low refractive index contrast between core and cladding leading to the low numerical aperture. As a consequence, light may be efficiently collected into the fiber up to a maximum angle of incidence of 20 degrees [1].

In this work, we show both theoretically and experimentally that nanoprinted all-dielectric nanostructures implemented on fiber end faces allow to significantly enhance light in-coupling into fibers up to the unprecedently high levels. The polymer axial-symmetric nanostructures are fabricated at fiber end face using the direct laser writing technique. Taking advantage of the unique properties of the nanoprinting technology, we outperform the functionality of bare fibers by orders of magnitude [1], in particular, at almost grazing incidence [2]. The various types of axially symmetric structures (e.g., single-pitch and double-pitch gratings or aperiodic arrays) were placed on the facet of SMF-28, resulting in structures of exceptional quality and coupling efficiency beyond all previously known devices [1]. We demonstrated the light in-coupling enhancement at the multiple selected angles and across the large angular intervals.

Applications can be found in a variety of cutting-edge fields that require highly efficient light collection over selected angular intervals, including endoscopy or quantum technologies.

[1] O. Yermakov et al., Applied Physics Reviews, 10(1), 011401 (2023).
[2] O. Yermakov et al., ACS Photonics, 7(10), 2834 (2020).

—

Nanoscale 3D printing enables a wide variety of applications from photonics to biology and medicine. Specifically, using volumetric 3D printing and a digital micromirror device (DMD) to project light into a resin can allow rapid light-assisted printing by polymerizing many 3D pixels (voxels) simultaneously. However, this requires a way to select a single plane at a time. To do so, we use triplet fusion upconversion, a quadratic photonic process, to print, driving polymerization with high spatial resolution deep within a resin vat. The intensities required are orders of magnitude smaller than those needed in two-photon polymerization, which scans a tightly focused high-intensity laser spot to print at small scales. Here, we study how these low power requirements can allow for DMD-based printing. As the voxels are projected into the resin, however, the z-resolution and precision of the printing system can be limited by substantial light intensity above and below the focal plane. Using a home-built setup, we study the behavior of light at and around the focal plane, and how we can manipulate the images uploaded to the DMD to optimize the resolution of our prints. We have designed both optical and software solutions to optimize the images projected into our resin, and we will present our recent findings towards high-resolution nanoscale volumetric 3D printing.

—

Light microscopy is a powerful single-cell technique that allows for quantitative spatial information at subcellular resolution. However, unlike flow cytometry and single-cell sequencing techniques, microscopy has issues achieving high-quality population-wide sample characterization while maintaining high resolution resulting in a compromise between resolution and population context. It is especially challenging to know the contextual relevance of data being acquired for high-resolution live imaging applications where the field of view limits cell population analysis. I will present a general framework, data-driven microscopy (DDM), that uses real-time population-wide object characterization to enable data-driven high-fidelity imaging of relevant phenotypes based on the population context. DDM combines data-independent and data-dependent steps to synergistically enhance data acquired using different imaging modalities. We have developed plugins for improved high-content screening and live adaptive microscopy for cell migration and infection studies that capture events of interest, rare or common, with high precision and resolution. We believe DDM will be a valuable approach for reducing human bias, increasing reproducibility, and placing single-cell characteristics in the context of the sample population when interpreting microscopy data, leading to an overall increase in data fidelity.

—

Light waves suffer from multiple scattering in turbid media, which limits the imaging depth. Wavefront shaping can overcome scattering but is limited by the isoplanatic patch and typically requires repeated updates of hardware as well as feedback from invasive detectors or guidestars, restricting its use in large-volume 3D imaging. Here, we present scattering matrix tomography (SMT) that employs virtual spatial and temporal wavefront control to realize guidestar-free hardware-free wavefront optimization for every isoplanatic patch, enabling noninvasive volume imaging deep inside scattering media.

SMT conducts a one-time measurement of the hyperspectral scattering matrix of the sample and digitally synthesizes confocal spatial gating with refractive index mismatch correction and time gating with pulse compression. Moreover, wavefront modulation and optimization are digitally performed for both input and output with noninvasive feedback from the reconstructed image. Such wavefront-optimized confocal spatiotemporal focus is digitally scanned in 3D without moving the sample, forming a high-resolution image across a large volume.

Experimentally, we use SMT to realize an ideal diffraction-limited resolution where the target signal is reduced by over ten million-fold due to the presence of multiple scattering. We achieve a depth-over-resolution ratio of 910 behind one-millimeter-thick ex vivo mouse brain tissue, which is the highest ratio reported in the literature (including all optical and non-optical label-free methods) to our knowledge. We also realize 3D volumetric imaging with ideal transverse and axial resolutions inside a dense colloid, where conventional imaging methods fail, across a large depth of field of over 70 times the Rayleigh range.

—

Wave-matter interactions are traditionally controlled by shaping the input wavefront or by inverse-designing a metamaterial. Currently, a new trend based on complex media with massive in situ tunability is emerging across scales and wave phenomena, spanning from nanophotonics and optics via acoustics to microwaves. Yet, to date, the full potential of a massively programmable complex medium (MPCM) remains inaccessible because no technique exists to accurately predict the transfer function of a specific MPCM as a function of the configuration of its degrees of freedom. The challenge lies in the unknown geometry and material properties of the MPCM, and in the non-linear dependence of the transfer function on the MPCM’s structural configuration. Here, we overcome this roadblock by experimentally demonstrating that a compact model derived from first physical principles can precisely predict how the transfer function of a complex scattering environment depends on the configuration of an embedded programmable metasurface. Our model is calibrated using a very small random subset of all possible metasurface configurations and without knowing the setup’s geometry. Our approach achieves two orders of magnitude higher precision than a deep learning-based digital-twin benchmark while involving hundred times fewer parameters. Strikingly, when only phaseless calibration data is available, our model can nonetheless retrieve the precise phase relations of the scattering matrix as well as their dependencies on the metasurface configuration. Thereby, we achieve coherent wave control (focusing or enhancing absorption) and phase-shift-keying backscatter communications without ever having measured phase information. Finally, our model is also capable of retrieving the essential properties of scattering coefficients for which no calibration data was ever provided. These unique generalization capabilities of our pure-physics model significantly alleviate the measurement complexity.

—

Deep learning has transformed computational imaging, but traditional pixel-based representations limit their ability to capture continuous, multiscale details of objects. Here we introduce a novel Local Conditional Neural Fields (LCNF) framework, leveraging a continuous implicit neural representation to address this limitation. LCNF enables flexible object representation and facilitates the reconstruction of multiscale information. We demonstrate the capabilities of LCNF in solving the highly ill-posed inverse problem in Fourier ptychographic microscopy (FPM) with multiplexed measurements, achieving robust, scalable, and generalizable large-scale phase retrieval. Unlike traditional neural fields frameworks, LCNF incorporates a local conditional representation that promotes model generalization, learning multiscale information, and efficient processing of large-scale imaging data. By combining an encoder and a decoder conditioned on a learned latent vector, LCNF achieves versatile continuous-domain super-resolution image reconstruction. We demonstrate accurate reconstruction of wide field-of-view, high-resolution phase images using only a few multiplexed measurements. LCNF robustly captures the continuous object priors and eliminates various phase artifacts, even when it is trained on imperfect datasets. The framework exhibits strong generalization, reconstructing diverse objects even with limited training data. Furthermore, LCNF can be trained on a physics simulator using natural images and successfully applied to experimental measurements on biological samples. Our results highlight the potential of LCNF for solving large-scale inverse problems in computational imaging, with broad applicability in various deep-learning-based techniques.

—