I will share how my experiences in Gabi Popescu's lab as a graduate student shaped me as a person and as a scientist. I will also discuss my research in his lab and how it has led me to my current work on using high resolution imaging technologies to study embryonic development.

Quantitative phase imaging (QPI) has recently emerged as a potentially valuable label-free approach which, due to its high resolution and sensitivity, has enabled a broad range of new applications. However, being a label-free technique, structures in a QPI image of a cell cannot always be readily identified because the image lacks specificity. To address this, machine learning methods have been deployed to map an acquired QPI image to a different image type, such as a simulated fluorescence image or an image representing a labeled segmentation mask. In this talk, I review several productive collaborations I had with Prof. Gabi Popescu on this topic. The applications surveyed include live-dead assay on unlabeled cells, cell stage classification and artificial confocal microscopy for deep label-free imaging.

Imaging with second harmonic generation (SHG) and third harmonic generation (SHG) are powerful methods biological samples but usually requires slow laser scanning. We demonstrate epi and transmission holographic SHG and THG imaging for aberration-free imaging by estimating and correcting phase distortions. Tomography and super resolution imaging with nonlinear wide field microscopy will also be discussed.

We utilize Spatial Light Interference Microscopy (SLIM) with enhanced computational analysis to quantify bioactive molecules, dynamic cellular topography, and tissue-level architectural tomography in various models of disease.

The histological optical imaging is a gold standard method to observe the biological tissues, which follows routine process such as dissection, embedding, sectioning, staining, visualization and interpretation of specimens. This technique has a long history of development, and is used ubiquitously in pathology, despite being highly time and labour-intensive. Staining-free optical imaging offers the unique advantage of bypassing the cumbersome staining process during specimen preparation and distinguishing the structure of tissue slides based solely on optical contrast. Here, we introduce the potential of quantitative phase imaging as a new digital histopathologic tool. Optical imaging based on phase contrast in histopathology could build fast feedback of anatomy of tissues or organs due to its simplicity, efficiency, robustness, and high-throughput capabilities. This presentation covers the latest work of large-scale and fast tissue imaging using quantitative phase imaging. Specifically, the talk will highlight comparison study over the conventional method in histopahology and its adaptation with artificial intelligence such as the virtual staining and resolution enhancement.

Quantitative phase imaging (QPI) is essential in biomedicine and surface inspection for its 3D high-resolution imaging. Conventional QPI methods, hindered by large optical setups, face challenges in small spaces like in-vivo imaging of cancer tissues. To combat this, our study introduces a novel, ultra-thin, lensless microendoscope using multicore fiber, enabling effective QPI even in limited spaces. This device captures real-time holograms, reconstructing complex light fields with automatic digital refocusing, achieving up to 15 nm axial and 1-micron spatial resolution. Its compact 0.5 mm design significantly enhances access to restricted areas, revolutionizing surface inspection and in-vivo clinical diagnostic imaging.

One of the directions of development in quantitative phase imaging is to provide the capability to reconstruct the phase or preferably refractive index (RI) distribution within thick, highly scattering samples. This direction coincides with current trends in biology, where three-dimensional (3D) organoids are currently replacing standard 2D cultures as more physiological models for tissue growth and organ formation in a dish. The biological complexity of these 3D structures makes the imaging and RI reconstruction particularly challenging, and thus calibration as well as validation structures are important and sought-after tools in instrumentation development. For this reason, in this work, we present the full preparation and measurement procedure for organoid phantoms printed with two-photon polymerization along with the method to obtain the ground truth of the object structure independently of RI reconstruction errors and artifacts.

Nowadays, fluorescence microscopy has made the leap from 2D to 3D or even 4D (xyz+t) imaging. On the other hand, Zernike phase contrast microscopy, which was awarded the Nobel Prize in Physics in 1953, has become the standard feature for modern biological microscopy, but is still limited to 2D imaging. Currently, life science research urgently needs a new "label-free 3D microscopy" mode that complements confocal/two-photon/super-resolution 3D fluorescence microscopy technology to meet the needs of rapid, high-resolution, long-term imaging of live cells in 3D. In this talk, we will present some of our research progress in "noninterferometic" intensity diffraction tomography, including: quantitative phase imaging and diffraction tomography based on transport of intensity and Fourier ptychography. Our results highlight a new era in which strict coherence and interferometry are no longer prerequisites for quantitative phase imaging and diffraction tomography, paving the way toward new generation label-free three-dimensional microscopy, with applications in all branches of biomedicine.

In-line Digital holographic microscopy is easy to implement because it only requires a coherent illumination source. An out-of-focus image (hologram) of the sample is recorded and numerically reconstructed to retrieve phase shift information due to the sample. Effects of the optical setup, such as aberrations, can distort the hologram signal, and ignoring these effects results in biased reconstructions. In this context, we study the effects of misalignment and wrong cover-slip thickness by statistically analysing the reconstructions of beads distributed in the whole field of view. Our aberration-wise reconstruction approach produces debiased estimations and allows precise and realistic estimation of optical aberrations from a single hologram.

In the context of multi-wavelength in-line holographic microscopy of micrometer-sized samples, we propose a 2-step methodology to estimate geometric and chromatic aberrations of the optical system and then use this calibration step to better reconstruct the complex transmittance at focus. The first step uses an aberration-wise Lorenz-Mie model to jointly estimate the parameters of calibration beads spread in the sample and 14 Zernike coefficients at each wavelength. Then, the reconstruction step is performed using a regularized inverse problems approach reconstruction of the whole multi-wavelength data set with a colocalization hypothesis. This general methodology is applied to the case of Gram-stained bacteria on blood smears. On these samples, in addition to providing a new information (phase), we show interesting improvements on the image quality, which promises better discrimination between bacteria types and enhanced repeatability.

To unravel dynamic processes underpinning key functions in cell biology, it is essential to develop imaging technologies able to track the movement of individual bio-nano-objects under physiologically relevant conditions, hence at high speed (ms) and in 3D. We demonstrate interferometric gated off-axis reflectometry (iGOR) which detects the back-scattered light of the structure of interest using an external off-axis reference, enabling label-free high-speed tracking of nanoparticles and suspended membranes in 3D volumes. Employing coherence time-gating by femtosecond pulses, the axial extension of the detected volume is controlled. We show tracking of single nanoparticles down to 10nm size freely diffusing in volume, which allows us to determine their geometrical and hydrodynamic radius as well as non-sphericity. We also show the spatiotemporal dynamics of suspended lipid bilayers, and the influence of lipid phase transitions on these dynamics with sub-nm thickness precision.

Wavefront shaping allows focusing and imaging at depth in disordered media such as biological tissues, exploiting the ability to control multiply scattered light. Many so-called guide-star mechanisms have been investigated to deliver light and image non-invasively, among which incoherent processes such as non-linear fluorescence feedback. However, the most common microscopy contrast mechanism, linear fluorescence, remains extremely challenging. I will discuss some of our recent works, exploiting signal processing and machine learning frameworks, to recover images behind scattering layers exploiting linear fluorescence.

In optical imaging, light propagation is affected by the heterogeneities of the refractive index. At shallow depths, these fluctuations induce wave-front distortions that degrade the image resolution and contrast. Beyond a few scattering mean free paths, multiple scattering starts to predominate and gives rise to a random speckle image without any connection with the medium reflectivity. To overcome these detrimental phenomena, we develop a general matrix approach of optical imaging and demonstrate its benefit by means of numerical simulations. By stacking a set of random phase screens, we model forward multiple scattering and its short-range memory effect. A computational multi-conjugate adaptive optics strategy is then proposed to exploit these snake photons ad optimize the focusing process at any point inside the medium.

Holographic retinal imaging can be affected by optical distortions from the eye and lenses. A digital Shack-Hartmann algorithm corrects this by splitting the Fourier plane into sub-apertures to measure local wavefront gradients via cross-correlation of sub-images. We examine wavefront regularization by Zernike polynomials for better aberration correction, and introduce a new method for calculating retinal image shifts. Using the entire computed image as a reference, rather than just the central sub-image, minimizes bias. Furthermore, we use a direct wavefront reconstruction approach, using overlapping sub-apertures and a 2D gradient integration algorithm to estimate the wavefront without regularization. Our findings show that this direct wavefront estimation enhances image resolution and contrast for Doppler holography of the eye fundus compared to wavefront regularization by Zernike polynomials.

Imaging and measuring moving objects behind translucide layers is of high interest in numerous applications, in various fields. We developed in this study a method using digital holographic microscopy for that purpose, with a partial spatial coherence illumination. The presented study includes numeric simulations for the assessment of the measurement accuracy, as a function of both the object and the translucide layer parameters. Experiments are performed with the in vitro analysis of human red blood cells moving behind an edothelial cell layer. Such a configuration mimics the real blood vessel, which is of crucial importance for a better understanding of the behaviour of the blood cells in blood circulation.

High speed terahertz imaging based on optimized galvanometric illumination Yuzhe Zhang, Ran Ning, Jie Zhao, Shufeng Lin, Lu Rong, Dayong Wang The pursuit of high-resolution, high-fidelity, real-time imaging is receiving significant attention in terahertz community. In this study, a versatile illumination approach based on a dual-mirror galvanometer is proposed and optimized for terahertz full-field imaging and computed terahertz tomography. We analyzed the mechanism of galvanometric illumination and elucidated three main factors affecting its homogeneity properties. In this illumination module, the terahertz beam is deflected rapidly by the galvanometer which is driven by triangular voltage signals, and then focused by a self-designed aspherical f-θ lens to illuminate the object at an equal lateral scanning velocity. A homogeneous illumination field with a speckle contrast of 0.11 and isotropic imaging resolution is recorded by an array detector in the form of non-correlated accumulation in a single integration time. By virtue of leading illumination homogeneity and parallelism, a compact imaging system is built for 2D and 3D terahertz imaging with high imaging speed and fidelity.

THz digital holography with camera sensor in off-axis configuration suffers from difficult recording conditions.In order to obtain a high resolution on the object, the latter must be close to the sensor, yielding high recording angles. It was already shown that iterative reconstruction methods can be used to limit the impact of spurious fringes obtained in such schemes. In this paper, we propose an inverse problem-based reconstruction technique that jointly reconstructs the object field and the amplitude of the reference field. Regularization in the wavelet domain promotes a sparse object solution. A single objective function combining the data-fidelity and regularization terms is optimized with a dedicated algorithm based on an ADMM framework. Each iteration alternates between two consecutive optimizations using projections operating on each solution and one soft thresholding operator applying to the object solution. The method is preceded by a windowing process to alleviate artifacts due to the mismatch between camera frame truncation and periodic boundary conditions assumed to implement convolution operators. Experiments demonstrate the effectiveness of the method.

Digital holography is an imaging technique that enables a 3-dimensional reconstruction of both the amplitude and phase of an electromagnetic field after its interaction with an object. Second harmonic generation being a coherent process, it can also be used to generate interferences and holograms. Since collagen molecules exhibit a significant second order response, we apply harmonic holography to porcine cornea samples to obtain single-shot measurements of the 3D spatial distribution of the collagen fibers. In addition, we propose polarization multiplexing to study the polarization dependence of the sample response.

This research presents the WUTScope, a novel interferometric microscope developed by the Quantitative Computational Imaging group at Warsaw University of Technology. This system, leveraging Quantitative Phase Microscopy and Optical Diffraction Tomography, provides insightful three-dimensional reconstructions of the refractive index distribution in semi-transparent objects. The WUTScope is distinguished by its compact design and capability to operate under partially coherent illumination, using polarization diffraction gratings for beam splitting and recombination. This approach allows for efficient phase shifting and reduces speckle noise, enhancing image signal-to-noise ratio. The system's achromatic nature, due to the identical optical paths of the diffraction orders, facilitates the use of less coherent light sources, a distinct advantage over traditional holographic methods. Its effectiveness is demonstrated through tomographic reconstruction of a 3D-printed brain sample and analysis of refractive index changes in HeLa cells' lipid droplets, revealing the impact of cholesterol accumulation.

Our ability to ask questions about living systems has been limited by our ability to measure their holistic, real-time structure, function and dynamics. When slow imaging speeds and challenging sample geometries necessitate complex preparations, perturbations and manipulations, or repeated induction of phenomena, we are not observing the system in its natural, evolving state. Recent developments in high-speed, multi-spectral, 3D single-objective light sheet microscopy have made it possible to image large regions, whole brains and even whole organisms in real-time, from 3D cell cultures to C. elegans worms, fruit flies and zebrafish larvae and the brains of living mice. Fluorescent reporters can provide real-time read-outs of cellular function, while tracking algorithms can simultaneously extract complex real-time behaviors. The next challenge is then exploring and interpreting the resulting TB-scale real-time data for scientific discovery. This talk will summarize recent high-speed 3D imaging advances combined with data-driven approaches to analyzing and interpreting physiological phenomena in a range of living systems.

We present a diffraction-phase and fluorescence 3D microscope as novel bimodal imaging technique, which provides simultaneous phase and multi-color epi-fluorescence acquisitions of living multicellular samples. The instrument consists of an LED-array to acquire intensity images at different illumination angles and an epifluorescence setup for fluorescence excitation. The 3D sample’s optical properties are reconstructed using the beam propagation method embedded inside a deep learning framework. To obtain the fluorescence reconstructions, we developed a novel incoherent model that takes into account the heterogeneous refractive indexes of the scattering sample. We validated the technique on long-term acquisitions of mouse embryos and 3D liver organoids under physiological conditions.

We present a computational approach for hyperspectral computational Selective Plane Illumination Microscopy (SPIM), offering fast 3D imaging with reduced photobleaching. Inspired by Hadamard spectroscopy, our method employs structured light sheets via a digital micromirror device. A data-driven reconstruction strategy, implemented through an end-to-end trained neural network, demonstrates robust performance under varying noise levels. Leveraging non-negative least squares minimization, we obtain component maps, exemplifying applications such as autofluorescence removal in transgenic zebrafish and discrimination of closely matched red proteins. Our findings showcase the potential of computational strategies to advance hyperspectral SPIM in photonic research.

Hyperspectral imaging has been explored for clinical applications in various medical disciplines. Based on our experiences, the potential of this versatile imaging method as well as pitfalls and limitations of current approaches are discussed. The use of reference samples and simple image modeling strategies are suggested to avoid misinterpretation and achieve a more conclusive image interpretation.

Dynamic metabolic reprogramming among neurons and glial cells can be characterized by label-free two photon excited fluorescence intensity and lifetime (FLIM) imaging of engineered brain tissue models consisting of human neurons, astrocytes, and microglia tri-cultures. Lipofuscin is a significant contributor to the overall fluorescence detected. Its removal is important for accurate recovery of metabolic function metrics. Results reveal the important function of glial cells to reduce oxidative stress in neurons over fourteen weeks of co-culture. Distinct metabolic profiles and dynamic changes are also evident for neurons and glial cells. Our study highlights that engineered brain tissue models in combination with label-free two photon imaging offer an excellent platform for understanding cell-cell interactions and are well-suited to improve understanding of the role of metabolic and mitochondrial dysfunction in neurodegenerative diseases.

The combination of femtosecond pulses with microscopes resolves processes at ultrashort time scales with spatial resolution. However, an integration into three-dimensional imaging methods, allowing to retrieve ultrashort processes as a function of three-dimensional space and time in complex and crowded environments, is lacking. We have achieved the implementation of broadband optical pulses with more than 100 nm bandwidth into a holographic optical diffraction tomography (ODT) setup. Besides having overcome a critical step towards ultrafast three-dimensional imaging we realized spectrally resolved ODT, retrieving the specimen’s refractive index as a function of 3D space and wavelength over the entire bandwidth.

Minimally invasive endoscopic imaging is indispensable for several important biomedical imaging applications such as in neurology. However, traditional fiber-optic endoscopes require bulky lens systems that typically enable 2D imaging only. Unconventional imaging using a diffuser for encoding 3D objects into 2D speckle patterns is presented. A multicore fiber transmits the speckle intensity information of fluorescent light from biological tissue. The decoding of the speckle pattern is accomplished by neural networks with a combination of U-Net and single-layer perceptron. It turns out 3D image reconstruction at almost video rate. The diffuser fiber endoscope is promising for in vivo deep brain diagnostics with cellular resolution and keyhole access. Advances in physics-informed neural networks and quantum imaging with entangled photons are also outlined. In conclusion, unconventional lensless imaging using multicore fibers, deep learning and quantum imaging is highlighted as paradigm shift for biomedicine.

We present a learning-enabled lens-free microscope for quantitative analysis of cell cultures. Leveraging the advances of recent years in learning algorithms, we developed a suite of neural networks that detect, quantify and track the cells. The detection algorithm locates the cells. The quantification algorithm, measures different cell metrics directly from cell phase image patches centred on the cells detections. Measured features include among others: cell morphology (dry mass, thickness, aspect ratio, ...) and local neighbourhood (density, contact surface, …). Finally, the tracking algorithm predicts the position of a given cell at next time point, making it possible to monitor a cell across time. To train these models we designed a semi-automated pipeline able to generate a supervised training datasets of up to millions of cells. The measurements obtained from the proposed method open up for modelling the cell cultures and providing biological insights.

Circulating tumor cell clusters (CTCCs) are associated with high metastatic potential and poor patient prognosis. However, they are difficult to detect and isolate because of their extremely low numbers. Here, we report on the use of machine learning and deep learning based analysis to achieve accurate detection of CTCCs in flowing whole blood samples relying on the confocal detection of endogenous light scattering signals. Our custom flow cytometer utilizes laser excitation at 405, 488, and 633 nm and confocal detection of the corresponding light scattering signals as well as fluorescence in the 510-530nm range. Samples consist of whole blood isolated from rats spiked with varying concentrations of CTCCs, flowed through the channels of a microfluidic device. The CTCCs express GFP, so that we can detect the strong GFP signal with the 520 nm detector and use it as the ground truth for assessing the performance of algorithms relying on the endogenous signals of the same peaks detected by the other detectors. We achieve a low false alarm rate of 0.78 events/min, a detection purity of 72%, and a sensitivity of 35.3%.

Lensless digital holographic microscopy (LDHM), as one of key computational microscopy techniques, performs high throughput in silico imaging. Numerical propagation of digitally recorded in-line Gabor holograms allows for accessing both amplitude (absorption) and phase (refraction) contrast, devoid of microscope objective limitations, e.g., in depth of field and field of view. The in-line coherent holographic framework induces inherent twin image errors and various coherent artifacts, however. The signal-to-noise ratio of reconstructed holograms additionally deteriorates due to low photon budget environment, favorable in terms of time-lapse photostimulation-free bioimaging of live cells. In this contribution, we discuss several techniques for minimization of LDHM reconstruction errors, with the emphasis on simultaneous validation of phase measurement fidelity via calibration target testing. Crafted using two-photon polymerization, our targets enable large field of view phase imaging verification and assess the efficacy of the 3D printing method itself. We also present bio-applications of enhanced LDHM in dynamic (migrating neural cells) and static (brain tissue slices) scenarios.

Cellular heterogeneity is the hallmark of many cancers, referring to the co-existence of different phenotypes with very distinct biological behaviours in single isolates. Automatically detecting single-cell heterogeneity is therefore critical, and can provide important information on cancer initiation. We present a clustering algorithm that allows identifying heterogeneity in cell culture from time-lapses of lensless microscopic images. A preliminary segmentation and tracking pipeline extract quantitative features (morphology, motility and reproduction cycle) for each cell. An unsupervised learning algorithm then clusters the time-series of the cell tracks measurements, in two steps. We validate our approach on co-cultures of mixed cells lines, and on murine fibroblasts isolated from genetically modified mice, where the modified genome promotes the establishment of cancers and heterogeneous cell morphologies and behaviours

Straylight (SL) characterization using ultrafast time of flight imaging (ToF) has been demonstrated for the testing of refractive telescopes, using a streak tube with a femtosecond laser. It was shown that individual SL contributors such as different ghost reflections and scattering features can be measured individually and identified by temporal discrimination due to the specific optical path length of each of them. This allows to analyze them individually for a better understanding of straylight properties in instruments. Recently, we have used the ToF approach to characterize a testing facility that was then used in the frame of the calibration campaign for the Narrow Angle Camera (NAC) of the Earth Return Orbiter mission. The facility itself could generate its own SL that has to be retrieved from that coming from the instrument. Due to the large facility dimensions, optical path lengths can be discriminated by using a low temporal resolution that is enabled by picosecond lasers associated to a SPAD detector. At the end, the SL coming from the facility can be reverse engineered to find its origin and either removed by facility adaptation or by processing.

We propose a novel design to perform quantitative phase confocal imaging with super-resolution imaging capabilities. Our approach, derived from image scanning microscopy, allows to perform quantitative label-free images up to 24kHz for 2D imaging. We demonstrate the capability to quantitatively image living adherent cells and biological tissue slices.

Photonic nanojets (PNJs) are highly localized optical probes that promise label-free measurements beyond the classical diffraction limit. We here demonstrate numerically the feasibility of label-free, self-calibrating, super-resolution optical detection and imaging using far-field scatterometry in conjunction with rapid scanning photonic nanojet excitation achieved with no opto-mechanical intervention. We realize PNJ scanning by computed structured illumination of refractive dielectric micro-elements such as micro-spheres and micro-cubes. Our far-field measurement data are phaseless. In proof-of-concept computations, we use our steerable optical probe to extract information on nanoparticles, aggregates of nanoparticles, and thin-film structures beyond the classical lateral and vertical resolution limits, in the presence of supporting structures such as substrates.