3D Biomedical Imaging
X-ray tomography and machine learning for tooth microcrack analysis (VULRC, Lithuania)
The combination of X-ray micro-computed tomography (μCT) with convolutional neural network (CNN) assisted voxel classification and volume segmentation has, for the first time, made it possible to take a look at the network of microcracks (MCs) inside a tooth and reveal a 3D image without destroying the sample.
This experimental work was performed by an interdisciplinary team including a dentist, an astronomer, and a laser physicist from Vilnius University, together with colleagues from Stanford and Swinburne Universities.
Figure 1: X-ray μCT data-cube projected density maps along three major axes indicate CNN-segmented voxels belonging to the tooth, enamel, dentine, and cracks
Figure from I. Dumbryte et al., Sci. Rep. 12: 22489 (2022), used under Creative Commons license http://creativecommons.org/licenses/by/4.0/
Four extracted, undamaged human premolars were scanned using a μCT instrument and segmented with a newly trained CNN image segmentation model to identify enamel, dentine and cracks. This technique allowed 3D characterisation of all the MCs in a tooth, regardless of where in the tooth they began and extended (Figure 1), along with evaluation of the arrangement of cracks and their structural features (Figure 2). Moreover, the morphological characteristics of the different tooth surfaces, such as the degree of convexity, surface roughness and enamel layer width, did not interfere with the MC assessment procedure.
Figure 2: X-ray μCT data-cube projected density maps of CNN-segmented enamel (green), dentine (blue), and cracks (red) reveal an intricate inner structure in the four healthy teeth used in the study
Figure from I. Dumbryte et al., Sci. Rep. 12: 22489 (2022), used under Creative Commons license http://creativecommons.org/licenses/by/4.0/
The work revealed an intricate starshaped network of MCs covering most of the inner tooth, suggesting that MCs could be considered as structural and possibly
functional elements of the tooth, offering a protective role (by redistributing forces), rather than a damaging one.
From a clinical point of view, there is a need to revise the definition of MCs, to reevaluate their role and impact on tooth integrity and longevity, and to develop new
algorithms for the monitoring and treat ment of teeth with MCs in daily clinical practice. More widely, this detailed volumetric imaging will expand understanding of the cracking pattern in natural hard materials, and provide a greater insight into how to design biologically-inspired solid structures and predict the propagation of cracks within them.
Irma Dumbrytė, Donatas Narbutis and Mangirdas Malinauskas (VU, Lithuania)
I. Dumbryte et al., Sci. Rep. 12: 22489 (2022)
Imaging of cancer cell adhesion (MUT-IOE, Poland)
High-resolution monitoring of cell adhesion structures is essential for the study of cell movement mechanisms. It determines the metastasis of cancer cells, which is of clinical importance in determining treatment. A soft X-ray contact microscopy (SXCM) technique was used by biologists at the Centre of Biomedical Engineering (CIBIO) at the Military University of Technology (MUT) in Warsaw, Poland, to study the metastasis of breast cancer cells.
A team at the Institute of Optoelectronics at MUT (MUT-IOE) developed a compact SXCM system, based on a laser plasma soft X-ray source with a gas puff target. This was then combined with a commercially available confocal laser scanning microscope, which allowed the structures of Focal Adhesion (FA) complexes responsible for cell attachment to the substrate to be uncovered.
Imaging of breast cancer cell adhesion structures using soft X-ray contact microscopy (SXCM) and confocal laser scanning microscopy (CLSM) techniques. Focal adhesion structures are visible.
Observing FA structures and their reorganisation in living cancer cells, without additional modifications, enabled scientists to elaborate on cell cancer movement. The figure illustrates cell adhesive structures (FAs and actin fibres) seen under the confocal laser scanning microscope, and the recorded image of the cell’s lamellipodium imprint on the photoresist surface using SXCM. Photoresist PMMA may be degraded during exposure, depending on the efficiency of radiation passage through cellular structures.
After removing the cell residue and degraded fragments of the PMMA, atomic force microscopy (AFM) was used to scan the imprint of the cells. AFM 3D images represent surface features of imprints with nanometre resolution.
SXCM, performed with a table-top system, helped to obtain a resolution showing the substructure of the FA protein complex without a cell fixation step, with the
technique used revealing the size, area and cross-sectional plots of FA inside HCC38 breast cancer cells. The detailed knowledge of cell adhesion and native FA structures could bring about new therapeutic options to combat cancer.
Paulina Osuchowska and Elżbieta Trafny (MUT-IOE)
P.N. Osuchowska et al., Int. J. Mol. Sci. 22: 7279 (2021)
XPulse laser plasma X-ray source for phase contrast mammography (CELIA, France)
left: XPulse 3D mammography Prototype, right: Laser plasma X-ray conversion system
Mammography is the most widely used imaging technique for breast cancer screening and diagnosis. Although studies have been conducted on its effectiveness for the detection of breast cancer and the reduction of associated mortality, the technology is still constantly evolving, in particular to improve the generation and detection of X-rays, and to exploit new methods and techniques of image acquisition and reconstruction.
The XPulse project – led by ALPhANOV, Amplitude Laser Group, CELIA Laboratory, Imagine Optic and Institut Bergonié and funded by the Region Nouvelle Aquitaine
and FEDER – aims to develop an innovative X-ray phase contrast mammographic system, based on the use of a laser plasma X ray source. This system will improve the contrast and resolution of the images, reduce the deposited dose, and improve the patient’s comfort by realising 3D full tomographic images without breast compression.
Phase contrast breast imaging applications often require the use of high power and high brightness sources. While such conditions are easily available on synchrotron sources, laboratory-scale sources remain a technological challenge. Laser-driven x-ray sources represent a promising solution, since they are mostly limited by the available power from short pulse laser systems: with rapidly evolving laser technology, available power is increasing each year. For mammographic applications, the laser driver will need to be a kW-class, short pulse (ps or sub-ps) laser system. The XPulse collaboration has prototyped a laser plasma conversion system combined with a kHz, 100 W-class laser driver. This system will soon be integrated into the phase contrast mammography imaging prototype and used for preclinical tests on breast phantoms and biological samples.
Fabien Dorchies (CELIA)
Real-time molecular imaging of near-surface tissue using Raman spectroscopy (Leibniz IPHT, Germany)
Current medical imaging techniques mostly provide information based on morphological or anatomical differences in the tissue, disregarding the underlying molecular composition. Raman spectroscopy has the technological potential to provide the intrinsic molecular fingerprint of a sample label-free, non-invasively, and non-destructively. Such molecular sensitive methods could be used for clinical in vivo diagnostics, to detect, and potentially delineate, cancer from healthy tissues; however, current fibre optic probe systems do not support image acquisition or provide instantaneous data analysis. To overcome these challenges, a team of scientists has been working to develop a fibre optic probe-based imaging system that can exploit the whole potential of Raman spectroscopy. [1]
a. Raman-based system for the acquisition of MVR images b. Data accessing the topology of a hemisphere phantom c. Brightfield information mapped on the topology information d. Molecular information combined with AR and the topological information and e. information directly projected on the sample
f and g. Visualisation of pharmaceutical and lipid-rich compounds on a brain tissue sample
The proposed system enables handheld imaging acquisition using a fibre optic probe, as well as real-time data processing and reconstruction of molecular information. It combines Raman spectra measurements, simultaneous computer vision-based positional tracking with real-time data processing, and real-time formation of molecular virtual reality (MVR) images. These images can be observed as augmented chemical reality (AR) on a computer screen or can be projected onto the tissue itself, creating mixed reality (MR) information that can be viewed in real-time with the naked eye. Sample topologies can be added to the acquired data, enabling mapping of the molecular information onto a 3D sample surface.
The proposed system will facilitate real-time molecularly specific clinical diagnostics and molecular boundary demarcation, with smart and intuitive visualisation of the data using AR and MR, opening future clinical applications. It offers easy, direct access to the patient site and can provide biochemical distributions from the region of interest, for disease tissue differentiation during surgical resection. With a spatial resolution of 0.5 mm in the transverse plane, a topology resolution of 0.6 mm, and a spectral sampling frequency of 10 Hz, large tissue areas can be sampled in a few minutes, making the system highly suitable for clinical tissue-boundary demarcation. It also has non-medical applications, for example in manufacturing and quality control.
Wei Yang, Florian Knorr, Ines Latka, Matthias Vogt, Gunther O. Hofmann, Jürgen Popp and Iwan W. Schie (Leibniz IPHT, Germany)
[1] W. Yang et al., Light-Sci. Appl. 11: 90 (2022)
SILMAS - 3D microscopy for the quantification of pathology in whole mouse brains (LLC, Sweden)
Structured Illumination Light-sheet Microscopy with Axial Sweeping (SILMAS) is a volumetric imaging method for cleared tissue being developed by laser diagnostics researchers and neuroscientists at Lund University, Sweden.
Comparison between standard sectioning in a) and SILMAS sectioning in b). Below are examples of fluorescent pathology at different penetration depths in cleared tissue. In a) the z-resolution is deteriorating with depth, while in b) the z-resolution is uniform throughout.
Tissue clearing has become increasingly popular with the neuroimaging community over the last decade, as it allows researchers to image uncut brain tissue and visualise fluorescently-tagged neural structures using visible light. SILMAS is specifically designed to image such samples at a high, uniform, and isotropic resolution, offering improved contrast and more quantitative intensity values than standard light-sheet microscopy.
Light-sheet volumetric imaging is performed by stacking optically-sectioned 2D planes into 3D volumes. Despite tissue clearing, the sectioning quality inevitably deteriorates with penetration depth, due to light scattering. Additionally, most microscope designs have limited field-of-view (FOV), so multiple images are stitched together into one plane to achieve a larger FOV. As a result, the optical resolution suffers in the third dimension, and is non-uniform within each plane.
In SILMAS, a structured light-sheet is used to reject of out-of-plane signals from the scattered light that is deteriorating the image. The light-sheet has a high numerical aperture and is swept across the FOV to obtain thin and uniform optical sectioning. As a result, the method can reach uniform volumetric resolutions down to 1 μm3 in whole mouse brains. By rejecting out-of-plane signals a quantitative intensity signal, attenuated with tissue depth, is obtained at high contrast. This attenuation can be compensated for, which allows for accurate visualisation. These characteristics make SILMAS data uniquely fitted to the quantification of fluorescently-tagged pathology throughout whole brains from mouse models.
3D volume of a whole cleared mouse brain captured with a prototype SILMAS instrument at 2 μm isotropic resolution
In an ongoing study, a neural network is being developed for the quantification of Lewy pathology, which is key in the development of Parkinson’s disease. The study aims to determine to what extent the improved data uniformity can optimise the training of a neural network, thereby reducing the need for manually labelled data. This would significantly streamline the workflow for data processing in neuroscience, and contribute to more rapid progress in the field.
David Frantz and Edouard Berrocal (Lund University)
D. Frantz et al., Biomed. Opt. Express 13: 4907 (2022)
Introducing a modulated excitation to localise single molecules (ISMO, France)
Recent developments in fluorescence microscopy have enabled imaging beyond the diffraction limit, reaching previously inaccessible observation scales in biological samples. Single molecule localisation microscopy (SMLM) techniques (PALM, (d)STORM, PAINT, etc.) can achieve lateral localisation precisions of a few nanometres, but improvement in the axial direction remains a major challenge to the development of a nanoscope with isotropic 3D resolution, capable of imaging at several tens of microns depth.
In SMLM, the position of the fluorescent emitters is usually obtained by fitting the point spread function (PSF). Spatially-based localisation precision thus strongly relies on the PSF shape, which is degraded by defocusing or aberrations, and strongly affects 3D imaging. Researchers from ISMO, in collaboration with Institut Langevin, recently proposed a new localisation strategy, Modulated Localisation (ModLoc) [1,2], based on a time signature to retrieve the fluorophores’ position over the whole field-of-view.
With this strategy, wide-field uniform excitation of the sample is replaced with a shifting structured excitation, typically a moving fringe pattern (cf Figure a). This induces a time-modulated emission of the illuminated fluorophores, where the phase holds its position. The short on-time of a single molecule in dSTORM means that modulation frequencies typically over 500 Hz are required, which is too fast for sequential image demodulation for most cameras. An optical assembly, based on a Pockels cell or galvanometric mirrors, steers the photons in four sub-images that are then acquired in a single camera frame (cf Figure b), achieving demodulation without photon loss.
Results to date show an improvement in localisation precision by a factor of 2.4, and the unique capability to achieve uniform, sub 7 nm precision for in-depth imaging (cf Figures c, d). Furthermore, because it is resistant to optical aberrations, ModLoc allows imaging of complex samples, such as tissues or organoids, up to several tens of microns in depth.
a) Introduction of a tilted, structured illumination pattern to encode axial information; b) modulated fluorescence is sampled at ~kHz on four sub-arrays on the camera to retrieve the phase; c) lateral and axial precision within the first 7 μm; d) 3D imaging of mitochondria in COS7 in dSTORM (Alexa Fluor 647)
Work is ongoing to apply the ModLoc technique to identify multiple dyes associated with various proteins, with further developments planned within the framework
of the EIC Pathfinder Open project RT-SuperES beginning in July 2023, which aims to deliver a multiscale nanoscope.
Abigail Illand, Pierre Jouchet (ISMO, Université Paris Saclay), Emmanuel Fort (Institut Langevin), Sandrine Lévêque-Fort (ISMO)
[1] P. Jouchet et al., Nature Photonics 15(4): 1–8 (2021)
[2] P. Jouchet et al., Phil. Trans. R. Soc. A. 380: 20200299 (2022)
Adaptive optics fluorescence microscopy for high resolution in vivo imaging (ISMO, France)
Light-sheet fluorescence microscopy and two-photon excited fluorescence microscopy are popular techniques for 3D imaging of living samples, offering low phototoxicity, sectioning capability, and good spatio-temporal resolution. They are widely used in embryonic development imaging, cell tracking and vasculature studies, and have played a key role in deciphering brain functions in neuro-imaging. However, the refractive index inhomogeneity of biological tissues produces optical aberrations when targeting high-resolution, in-depth imaging using high numerical aperture objectives, significantly reducing contrast and resolution.
A novel adaptive optics (AO) method has been proposed to compensate for such aberrations and increase the image quality, based on direct wavefront sensing from an extended scene Shack-Hartmann wavefront sensor (SHWFS). This approach delivers fast, efficient and easier correction [1], as it neither requires the iterative algorithms used in sensorless AO setups, nor the complex and/or expensive use of fluorescent beads or a pulsed laser to generate a guide star.
The benefit of the extended scene SHWFS has been demonstrated for scattering samples, delivering improved accuracy at large depths in low signal-to-background ratio conditions [2]. The sensor was implemented in an AO two-photon fluorescence microscopy (AO-TPFM) setup, with a dual-colour labelling strategy: the wavefront was measured using red anatomical labelling, thus preserving the photon budget of the green labelling of interest. The figure shows the ability to correct aberrations up to 350 microns deep inside a fixed brain tissue, resulting in a large gain in intensity and resolution. Since the scattering length in fixed brain tissue is two-times smaller than in intact tissue, the technique promises deep in vivo imaging at depths reaching 700 microns, with contrast typically increased by
a factor of five for small structures.
The sensor was also implemented in an AO light-sheet fluorescence microscope (AO-LSFM) for in vivo imaging of densely labelled samples [3]. Combined with an approach based on axially swept LSFM, the novel AO strategy enabled coverage of a large field of view (typically 400 microns), whilst maintaining a high 3D resolution (typically 0.4 x 0.4 x 1 micron) and increased signal deep inside a zebrafish larva. The setup delivered enhanced 3D in vivo images down to 300 μm deep in various areas, such as the heart and brain, demonstrating its performance capabilities.
AO-TPFM for closed-loop correction in fixed brain slices. 2D maximal intensity projection of a stack of images in cortical brain slices of GAD-GFP mice without (left) and with (right) AO correction. Cell depths are colour-coded. Inserts: original Zstacks (S. Imperato, ESPCI and L. Bourdieu, IBENS)
Alexandra Fragola (ISMO, Université Paris Saclay) and Fabrice Harms (Imagine Optic)
[1] A. Hubert et al., Opt. Lett. 44: 2514-2517 (2019)
[2] S. Imperato et al., Opt. Express 30: 15250-15265 (2022)
[3] A. Hubert et al., http://biorxiv.org/lookup/doi/10.1101/2023.01.06.522997 (2023)