Canada

&lt;p&gt;**Introduction:** MRI’s low sensitivity, caused by the use of nuclei with low-gyromagnetic ratios or low magnetic field strength, can presently be improved with expensive high-field MRI-hardware and/or expensive enriched-isotopes. We propose a new method that does not require any extra signal-averaging or hardware to improve the quality of MRI images. We will use a significant k-space under-sampling acquisition method where only a certain percentage of the k-space points will be acquired per image, corresponding to the acceleration-factor (AF); it follows that one can acquire ten under-sampled images in the same time as one fully-sampled image. Averaging each possible combination of images of the under-sampled set, a density decay curve can then be fitted and reconstructed using the Stretched-Exponential-Model (SEM) combined with Compressed Sensing (CS).&lt;sup&gt;1&lt;/sup&gt;&lt;/p&gt;
&lt;p&gt;**Method:** &lt;sup&gt;1&lt;/sup&gt;H MR was performed on a resolution-phantom at the low-field (0.074T) MRI scanner using a home-built RF coil. Nine 2D fully-sampled k-spaces were acquired. Combinations of 2, 3, and 4 averages were carried out for each possible permutation, resulting in 14 k-spaces total (2 combinations for 4 averages, etc.); these were retroactively under-sampled for three AF’s (7, 10, 14). 3 Cartesian sampling schemes (FGRE, x-Centric&lt;sup&gt;2&lt;/sup&gt;, &amp; 8-sector FE Sectoral&lt;sup&gt;3&lt;/sup&gt;) were used. The SNR attenuation is assumed to represent a decrease of the resonant isotope density in phantom after diluting it with the non-resonant isotope. The resulting signal decay (density) curve was fitted using the Abascal method.&lt;sup&gt;1&lt;/sup&gt;&lt;/p&gt;
&lt;p&gt;**Results:** The SNR of the 9 k-space averaged image and the original image was 16 and 5, respectively. The SNR of the three sampling schemes is 15 for FGRE, 19 for x-Centric, and 17 for FE Sectoral.&lt;/p&gt;
&lt;p&gt;**Conclusion:** The improved SNR of the generated images for all sampling schemes demonstrate that the SEM equation can be adapted for fitting the SNR decay dependence of the MR signal. Since this technique does not require extra hardware, the proposed method could be implemented in current MRI-systems and yield improved images. Due to the CS-based reconstruction, the higher AF leads to more visible artefacting; this could be reduced by a Deep Learning-based correction after the fact.&lt;sup&gt;4&lt;/sup&gt;&lt;/p&gt;
&lt;p&gt;**References:**
**1** Abascal et al. IEEE Trans Med Imaging (2018); **2** Ouriadov et al. MRM (2017); **3** Khrapitchev et al. JMR (2006); **4** Duan et al. MRM (2019)&lt;/p&gt;

2021 CAP Virtual Congress

The Use of a Novel Sampling/Reconstruction Method for Non-Proton and Low Field MRI

hyperpolarization

low field mri

neural network

**WELCOME TO THE 2021 CAP VIRTUAL CONGRESS UNDERLINE PLATFORM**

The CAP staff and our organizing and program committees are pleased to welcome you to our 2021 Congress filled with plenary talks, technical sessions/talks, student presentations and competition sessions, workshops, and several symposia offered by the CAP divisions and its partners.  

You will also have opportunities to network with other delegates through the Gathertown platform accessed by selecting the lounge or posters button on the left.

Registered delegates should use the schedule button to call up the details of the session/event they wish to attend.  Each sessions includes a link to access the online platform where the event is taking place, either within Underline or through other Congress-related Zoom meetings.  

Invited speaker talks are identified with an (I), and student competitors include a (G*) or (U*) to identify graduate or undergraduate presenters.

Please take the time to check out the information about our sponsors and exhibitors, request further information, or leave a note of thanks for their support.

General information about the Congress can be found on the CAP website Congress pages at https://www.cap.ca/congress/2021.

Looking for a specific talk?   A full searchable list is available on our Indico Congress site at https://indico.cern.ch/event/985448/contributions/

We look forward to sharing this exciting week with you!


ICPR 2020 is the premier international conference on Pattern Recognition, gathering leading researchers from around the world to present their latest achievements in Pattern Analysis and Recognition, applied Artificial Intelligence and Machine Learning. The topic of the year is " Putting Artificial intelligence to work on Patterns". 
For the first time, ICPR has employed a two-round review process similar to journal submissions, with new papers allowed to be submitted in either the first or second round and papers submitted in the first round and not accepted allowed to be revised and re-submitted for second round review. 

In this edition of the **ICPR 2020** Conference, there are 1407 Oral/Poster presentations organized in five Tracks: 
- Artificial Intelligence, Machine Learning for Pattern Analysis; 
- Biometrics, Human Analysis and Behavior Understanding; Computer Vision, Robotics and Intelligent Systems; 
- Document and Media Analysis; and 
- Image and Signal Processing. 

The **ICPR 2020** Main Conference program also includes 7 Keynotes from prominent world-renowned scientists, 8 Tutorials, 28 Demos, 8 Challenges, 2 Panels, as well as an Industry/Sponsor Exhibition. In addition, 41 Workshops on a range of important current and emerging research topics complete the ICPR 2020 program. 


ICPR 2020 MAIN CONFERENCE

**Introduction:** MRI’s low sensitivity, caused by the use of nuclei with low-gyromagnetic ratios or low magnetic field strength, can presently be improved with expensive high-field MRI-hardware and/or expensive enriched-isotopes. We propose a new method that does not require any extra signal-averaging or hardware to improve the quality of MRI images. We will use a significant k-space under-sampling acquisition method where only a certain percentage of the k-space points will be acquired per image, corresponding to the acceleration-factor (AF); it follows that one can acquire ten under-sampled images in the same time as one fully-sampled image. Averaging each possible combination of images of the under-sampled set, a density decay curve can then be fitted and reconstructed using the Stretched-Exponential-Model (SEM) combined with Compressed Sensing (CS).1
**Method:** 1H MR was performed on a resolution-phantom at the low-field (0.074T) MRI scanner using a home-built RF coil. Nine 2D fully-sampled k-spaces were acquired. Combinations of 2, 3, and 4 averages were carried out for each possible permutation, resulting in 14 k-spaces total (2 combinations for 4 averages, etc.); these were retroactively under-sampled for three AF’s (7, 10, 14). 3 Cartesian sampling schemes (FGRE, x-Centric2, & 8-sector FE Sectoral3) were used. The SNR attenuation is assumed to represent a decrease of the resonant isotope density in phantom after diluting it with the non-resonant isotope. The resulting signal decay (density) curve was fitted using the Abascal method.1
**Results:** The SNR of the 9 k-space averaged image and the original image was 16 and 5, respectively. The SNR of the three sampling schemes is 15 for FGRE, 19 for x-Centric, and 17 for FE Sectoral.
**Conclusion:** The improved SNR of the generated images for all sampling schemes demonstrate that the SEM equation can be adapted for fitting the SNR decay dependence of the MR signal. Since this technique does not require extra hardware, the proposed method could be implemented in current MRI-systems and yield improved images. Due to the CS-based reconstruction, the higher AF leads to more visible artefacting; this could be reduced by a Deep Learning-based correction after the fact.4
**References:**
**1** Abascal et al. IEEE Trans Med Imaging (2018); **2** Ouriadov et al. MRM (2017); **3** Khrapitchev et al. JMR (2006); **4** Duan et al. MRM (2019)

technical paper

**Introduction:** A non-invasive imaging method: inhaled hyperpolarized (HP) 129Xe magnetic resonance imaging (MRI) is currently used to measure lung structure and function.***1*** Simultaneous ventilation/perfusion (V/P) lung measurements of functional gas exchange within the lungs can be obtained using this MRI approach because of the high solubility of xenon in lung tissue as compared to other imaging gases. This measurement is possible due to the distinct and large range of chemical shifts (~200ppm) of 129Xe when residing within barrier and RBC (e.g., barrier and RBC phase xenon) compared to the gas phase. Therefore, 129Xe is a unique probe for exploring xenon within and beyond the lung, such as lung parenchyma (barrier), RBC, and even other organs such as the **brain**, heart and kidney.
[15O]water positron emission tomography (PET) is the gold standard imaging method for determining cerebral perfusion.***2,3*** In this study, simultaneous 129Xe-based MRI and [15O]water PET images were collected and compared. 

**Methods:** A 60mL plastic syringe was used in which 30mL of the hyperpolarized 129Xe gas was dissolved in [15O]-water solution (30mL) After dissolving, all leftover xenon gas was removed from the syringe. A turn-key, spin-exchange polarizer system (Polarean 9820 129Xe polarizer) was used for obtaining Hyperpolarized 129Xe gas. 129Xe dissolved phase images were acquired in a 3T PET/MRI (Siemens Biograph mMR) scanner. [15O]water PET data were acquired simultaneously with 129Xe MRI using the integrated PET system in the 3T PET/MRI. 

**Results:** Two consecutive 2D axial 129Xe MRI images and two (2D and 3D) [15O]water PET images were acquired simultaneously. 129Xe/PET images indicate that the diameter of the phantom from both PET and MRI images are similar. Both 129Xe images demonstrate a sufficient SNR level (80 and 10 respectively) suggesting that 3D 129Xe imaging is possible. 

**Conclusions:** The results of this proof-of-concept study clearly indicate the feasibility of the simultaneous hyperpolarized 129Xe MRI and [15O]water PET measurements. This demonstration enables the next step, namely, in-vivo double tracer brain perfusion imaging which we plan to perform using a small animal model.

***References:***
***1***.Kaushik, S. S. et al. MRM (2016); ***2***. Fan, A., et. al. JCBFM (2016); ***3.*** Ssali. T., et. al. JNM (2018).

Simultaneous Hyperpolarized 129Xe MRI and [15O]water PET Multi-Modal Imaging: A Proof of Concept Study

We are developing a time-reversal breaking test in radiative $\beta$ decay, using just the momenta of three outgoing particles. This type of time reversal is independent of nuclear spin, so explores time reversal-breaking physics unrelated to electric dipole moments (though there are model-dependent constraints at 1-loop order from null measurements of the neutron EDM). The scalar triple product of three momenta $\vec{p_1}\cdot \vec{p_2} \times \vec{p_3}$ provides a unique time-reversal odd observable, but trivially vanishes in ordinary $\beta$ decay when the three momenta sum to zero. So we need the fourth outgoing particle in radiative $\beta$ decay, considering the correlation between $\beta$, $\nu$, and $\gamma$. We add $\gamma$-ray detectors (GAGG scintillator with SiPM readout) to TRIUMF's magneto-optical trap for beta decay (TRINAT), which includes a uniform electrostatic field for efficient recoil ion detection. Explicit models produce this observable with an antisymmetric Chern-Simons term from QCD-like new interactions, interfering with the standard model vector electroweak interaction within the nucleon [S. Gardner and D. He, Phys. Rev. D 87 116012 (2013)], and among the predicted features are a quite different gamma-ray spectrum than normal bremsstrahlung. We will show initial data from the decay of 92Rb, a case without vector interactions not yet testing the explicit models.

Time-reversal test in radiative beta decay: progress

Much of what we know of the early universe comes from observations of the cosmic microwave background (CMB): a 13 billion-year-old field of microwave radiation that permeates the entire universe. Recent technological advances have made real the possibility of combining CMB measurements with other large data sets to extract hitherto inaccessible cosmological information. One such example is the novel technique of kinetic Sunyaev Zel’dovich (kSZ) tomography in which one combines a CMB temperature map with the positions and distances of galaxies across the sky, using statistical correlations between high-fidelity maps to reconstruct the velocity of those galaxies at the largest angular scales and over an appreciable fraction of the volume of the universe. At these scales and distances the motion of large-scale structure owes its statistical properties to the conditions of the early universe; measuring this velocity map would therefore provide an independent probe into the physics of that era.

The primary challenge in extracting a velocity map using kSZ tomography is characterizing the dominant sources of uncertainty introduced by non-idealities such as redshift measurement error, incomplete sky coverage, galactic and extragalactic contaminants, and confusion with other physical effects in the CMB. To this end we have designed a pipeline to perform the reconstruction using next-generation CMB and galaxy survey data that incorporates these contaminants into its design. We account for redshift errors and demonstrate that masking and other cut sky effects do not influence the reconstruction fidelity. We show that galactic processes do not contribute to the reconstruction noise, and estimate the impact of extragalactic sources that mimic the kSZ signal. We demonstrate that reconstruction is possible with data from next generation surveys and forecast how well the pipeline will perform with realistic contaminants. We estimate a strong signal-to-noise of the velocity map on large scales, making it a new data product on the cutting edge of cosmological research, as the field turns its attention to upcoming high-resolution datasets.

kSZ Tomography with Foregrounds and Systematics

Vertically stacked heterostructures of two-dimensional materials provide a platform for realizing novel electronic states due to proximity effects. In particular, moiré patterns in two-dimensional material heterostructures have been shown to create flat bands that favor the occurrence of correlated electronic states. In this work, we use scanning tunneling microscopy and spectroscopy to study moiré patterns in mechanically assembled heterostructures of graphene and transition metal dichalcogenides. By comparing to a theoretical model, in the case of graphene - ReS2 heterostructures we find the presence of stripped moiré patterns, reflecting the different crystal symmetry of the two lattices.

The experimental work was supported by: National Sciences and Engineering Research Council (NSERC) No. RGPIN-2016-06717, No. 521420 and NSERC Scholarship program. Theory work was supported by Singapore Ministry of Education AcRF Tier 2 (No. MOE2017-T2-2-140).

Moiré patterns in graphene and transition metal dichalcogenide heterostructures

**THESIS:** The ontology of spacetime should not be modelled as a single universally-shared 4D spacetime manifold.

**THE CONSENSUS SPACETIME ONTOLOGY:** The accepted ontology of classical spacetime is of a 4D differentiable manifold of events. Although different observers may assign different coordinate labels to events, it is conventionally asserted that the underlying reality is that of a single spacetime manifold. 
This single classical manifold ontology prohibits nonlocality, because there is no possibility for any form of physical interaction, except those occurring along paths in the manifold. Consequently, Bell nonlocality (EPR) remains mysterious [1].

**DIAGNOSIS:** The existence of an invariant distance measure (the spacetime interval), and smooth coordinate manifolds has been taken to imply the existence of a universally-shared smooth spacetime manifold. However, this move involves an improper conflation of two different modes of description.
The causal structure possesses invariant properties, but lacks the smooth neighborhood properties required for a manifold, whereas exactly the converse is true of the coordinate manifolds. Manifolds are smooth, but not invariant. We cannot borrow invariance from the causal structure and glue it to coordinate manifolds to create an invariant manifold.

**GR:** The tensor calculus of GR is "coordinate-free" in the sense that it gives consistent results, irrespective of coordinate system. However, the existence of coordinate mappings between spacetime manifolds does not imply the physical existence of a universal shared manifold. By analogy, the existence of an atlas of the human brain (allowing anatomical mapping between individual brains) does not imply the physical existence of a shared `universal brain’. A mathematical mapping does not imply a physical process.

**CONCLUSION:** To move beyond classical GR, it is necessary to abandon the image of spacetime as a single deformable 4D entity, and to replace it with a structure acknowledging the reality of the multiple coordinate manifolds. See [2] for such a `many-spaces’ proposal, including an account of quantum nonlocality.

**REFERENCES**

[1] Nicolas Brunner et al. “Bell nonlocality”. In: Rev. Mod. Phys. 86 (2 Apr. 2014), p. 419.

[2] J. C. Sharp. One Universe, Many Spaces: A Non-Local, Relativistic Quantum Spacetime
https://doi.org/10.20944/preprints201805.0003.v1

The Collapse of the Manifold

Introduction: A great challenge in quantitative dynamic positron emission tomography (PET) imaging is to determine the exact volumes of interest (VOI) with which one wants to work. They have a tremendous impact on the time-activity curves that are used to extract the pharmacokinetic coefficients. Since PET images are functional and not anatomical, using a bijective relationship with a computed tomography (CT) co-image is neither the sole nor the best possibility. In recent years, many publications have come with ingenious methods to work directly with the PET images, ranging from machine learning algorithms to manual toil to define regions. These techniques have different uses, mainly in the hope of enabling easier and more efficient tumor delimitation. In the case of dynamic images, the temporal aspect of the imaging procedure changes the methods that can be implemented. Furthermore, the need to delimitate specific and precise functional sites render the whole operation computationally and physically challenging, especially in the absence of a common and well-established methodology. 
Methodology: In this project, a novel approach using a gradient-based segmentation was used on pre-clinical dynamic PET images on rats. Fourteen different animals were used under similar pre-clinical conditions. The developed segmentation technique uses properties of the image itself, relying on already known properties of the radio-drug used in order to segment automatically the kidney of the animal. The work was conducted using the mini-PET scanner at the Montreal Neurological Institute, according to the ethical guideline from the University of Montréal and the Canadian Tri-Council.
Results: From the preliminary data, the proposed method has a relevant rate of success on clinical images in delimitating the volumes of interest. The respective time-activity curves follow the general pattern of the manual delimitations done by experts, yet with non-negligible differences. So far, the proposed technique offers good result on 8 of the 14 rats, as compared to 12 rats when using a manual segmentation. The greatest strength of the algorithm is its ability to reproduce the same results notwithstanding the operator. The technique can also quantify movement in the organ of interest and work in spite of a great amount of Gaussian noise. 
Keywords: Nuclear Medicine, pre-clinical dynamic PET imaging, pharmacokinetic

On the rationale for a standardized pre-clinical segmentation technique in PET pharmacokinetic

Two dimensional (2D) materials are tempting for nano-electromechanical systems (NEMS), as they are atomically thin, have low mass and high flexibility. 2D materials have tunable optical and electronic properties, and can display other exotic properties, such as superconductivity. Over the last decade, scientists have begun stacking 2D materials into heterostructures that have desirable properties. One heterostructure is twisted bilayer graphene (TBG), which is comprised of two monolayers of graphene stacked at a rotated angle. If the angle of rotation is 1.1 degree, this material can display superconductivity at cryogenic temperatures.

The goal of this work is to model 2D drums, and to prepare sample fabrication techniques for NEMS devices. We developed analytic and numerical approximations for the motion of a 2D drum resonator. We use thin plate solutions due to a non-zero bending rigidity, and assumed low strain . We contrasted these solutions to linear and nonlinear membranes, where tension forces dominate. We consider nonlinearities inherent in the deflection driven by electrostatic forces and where strain dominates over rigidity. We also describe how finite element models can be constructed in Comsol, where the vanishing thinness makes conventional models impractical. These agree with the analytic plate approximations. 

We use flake exfoliation, pick up and placement methods to fabricate heterostructures. We produce graphene flakes via mechanical exfoliation from low residue tape. We detail techniques for optimized cleanliness and monolayer production. The flake pick up is verified via optical and atomic force microscopy (AFM). Our eventual aim is to study a 2D material when it forms the top plate of a capacitor in an LC circuit – allowing a light mass resonator for quantum nanomechanics.

Towards 2D materials as Quantum Nano-Electromechanical Systems

Molybdenum possesses seven stable isotopes and the relative amounts of these isotopes are found to vary in nature. This is because physical and chemical processes can redistribute the isotopes in a system due to the differences atomic masses. Specific processes can leave an “isotopic fingerprint” that may be recorded in the isotopic composition of the element in a given sample. The interpretation of these data can enable one to elucidate source(s) and processes that may have affected the element. An important example of a potential Mo source is petroleum coke (PC). This is a by-product from the extraction of crude oil from the Oil Sands in northern Alberta and although it is employed as a fuel source, it is not used as quickly as it is produced, allowing it to accumulate on site [1]. There is evidence PC dust spread by wind contributes to an increase in polycyclic aromatic hydrocarbons (PAH) and polycyclic aromatic compounds (PAC) accumulating in lichen samples in forests in the Athabasca Oil Sands Region [2]. This could mean that trace metals are deposited in forests or bodies of water. Natural sources of the surrounding oil sands and bitumen seeps mean that the Athabasca River contains relatively high concentrations of metals compared to glacial counterparts. It is vital to distinguish between the natural and anthropogenic inputs so that appropriate procedures can be put in place to minimize human impacts on the region. Concentration measurements of Mo in aqueous environments provide ambiguous results as they do not necessarily distinguish between natural and industrial sources. Isotope abundance data can provide additional information on the source and history of the material. This project measured the isotopic composition of Mo leached from PC. Isotope determination was done using the double spike method measured with a Neptune multi-collector ICP-MS. The δ97/95 values for three oven dried PC leachate samples were determined to be +0.13 ‰, +0.34 ‰, and +0.81 ‰ with a 2σ uncertainty of 0.06 ‰. These samples had concentrations of 3.16, 9.47, and 0.55 µg/L respectively (uncertainty of 5 %). The isotopic composition of samples from this region gives a better understanding of the sources and sinks of Mo and can be combined with snow, lichens, and water to identify potential environmental concerns caused by PC distributed by wind.
1. J. M. Robertson et al. Aqueous- and solid-phase molybdenum geochemistry of oil sands fluid petroleum coke deposits, Alberta, Canada. Chemosphere 217, (2019).
2. M. S. Landis et al. Source apportionment of an epiphytic lichen biomonitor to elucidate the sources and spatial distribution of polycyclic aromatic hydrocarbons in the Athabasca Oil Sands Region, Alberta, Canada. Science of the Total Environment 654, (2019).

Molybdenum stable isotope measurements of petroleum coke in the Athabasca Oil Sands Region

The potential breakdown of the notion of a metric at high energy scales could imply the existence of a fundamental minimal length scale below which distances cannot be resolved. One approach to realizing this minimum length scale is construct a quantum field theory with a bandlimit on the field. We report on an investigation of the effects of imposing a field bandlimit on a curved and compact spacetime. To achieve this operationally, we couple two Gaussian-smeared UDW detectors to a scalar field on a S^2 x R spherical spacetime through Dirac-delta switching. Delta switching allows for a non-perturbative analysis that includes higher order effects due to the bandlimit. The bandlimit is implemented through a cut-off of the allowable angular momentum modes of the field. We find that the detectors are less sensitive to the bandlimit on the spherical spacetime, and observe features similar to flat spacetime. These include the response of the detectors depending on their geometry and that smaller detectors couple in a stronger manner to the field. 
We also explore two squeezed detector setups in both flat and spherical spacetimes, and find notable difference between the two cases. Due to the compact nature of spherical space, the lack of dissipation of any perturbation to the field results in locally excited signals of the field traveling from pole to pole in the spacetime. Quite strikingly, squeezing increases the response of the detectors in flat space but decreases the response on the spherical spacetime. Moreover, we find that squeezing on a sphere introduces extra anisotropies that could be exploited to amplify or weaken the response of the second detector.

Bandlimited UDW detector dynamics on 2+1 flat and spherical spacetimes

Liquid-phase exfoliation (LPE) is a low-cost and scalable technique for producing a wide range of van der Waals nanomaterials that can be incorporated into existing laboratory sample and industrial material production. Liquid-phase exfoliated nanomaterials have the potential to produce devices quickly and at low-cost, with colloidal dispersions easily adaptable to existing production methods. Other methods of nanomaterial production, such as chemical vapour deposition (CVD) and mechanical exfoliation can be costly, time-consuming and require complicated equipment to produce comparatively small area devices. Such methods are excellent for laboratory-scale samples to examine physical properties and produce proof of concept devices. However, LPE can and will bridge the gap towards real-world applications that require faster, easier and more cost-effective production methods. In this work, we investigate the saturable absorption and Kerr nonlinearity of graphene fabricated by LPE and CVD. 

Thin films of LPE graphene were produced on BK7 glass substrates, and compared to CVD graphene transferred onto an identical substrate. Through careful consideration of the concentration of graphene dispersion and deposition methods, very thin films of graphene can be prepared. Atomic force microscopy (AFM) measurements showed effective bi-layer graphene thickness for the LPE samples. Z-scan measurements performed with 180 fs pulses at a wavelength of 1030 nm reveal that both LPE and CVD graphene display strong saturable absorption characteristics, with a nonlinear absorption coefficient (*β*) approaching -104 cm/GW and a Kerr nonlinearity (*n2*) of -1 cm2/GW. Such strong saturable absorption is ideal for devices such as mode-lockers for ultrafast pulsed lasers. The magnitude of the nonlinear absorption coefficient of the LPE graphene increases with pulse duration, *τ*, up to 105 cm/GW at around *τ* = 10 ps. These results pave the way for the use of LPE graphene in nonlinear optical applications such as frequency generation and mode locking.

The Use of a Novel Sampling/Reconstruction Method for Non-Proton and Low Field MRI

Downloads

Next from 2021 CAP Virtual Congress

Simultaneous Hyperpolarized 129Xe MRI and [15O]water PET Multi-Modal Imaging: A Proof of Concept Study

Similar lecture

A Comparison of Neural Network Approaches for Melanoma Classification

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES