Search Results
Evolutionary design of explainable algorithms for biomedical image segmentation
An unresolved issue in contemporary biomedicine is the overwhelming number and diversity of complex images that require annotation, analysis and interpretation. Recent advances in Deep Learning have revolutionized the field of computer vision, creating algorithms that compete with human experts in image segmentation tasks. However, these frameworks require large human-annotated datasets for training and the resulting “black box” models are difficult to interpret. In this study, we introduce Kartezio, a modular Cartesian Genetic Programming-based computational strategy that generates fully transparent and easily interpretable image processing pipelines by iteratively assembling and parameterizing computer vision functions. The pipelines thus generated exhibit comparable precision to state-of-the-art Deep Learning approaches on instance segmentation tasks, while requiring drastically smaller training datasets. This Few-Shot Learning method confers tremendous flexibility, speed, and functionality to this approach. We then deploy Kartezio to solve a series of semantic and instance segmentation problems, and demonstrate its utility across diverse images ranging from multiplexed tissue histopathology images to high resolution microscopy images. While the flexibility, robustness and practical utility of Kartezio make this fully explicable evolutionary designer a potential game-changer in the field of biomedical image processing, Kartezio remains complementary and potentially auxiliary to mainstream Deep Learning approaches.
Wireless magnetic-based closed-loop control of self-propelled microjets
In this study, we demonstrate closed-loop motion control of self-propelled microjets under the influence of external magnetic fields. We control the orientation of the microjets using external magnetic torque, whereas the linear motion towards a reference position is accomplished by the thrust and pulling magnetic forces generated by the ejecting oxygen bubbles and field gradients, respectively. The magnetic dipole moment of the microjets is characterized using the U-turn technique, and its average is calculated to be 1.3x10-10 A.m2 at magnetic field and linear velocity of 2 mT and 100 μm/s, respectively. The characterized magnetic dipole moment is used in the realization of the magnetic force-current map of the microjets. This map in turn is used for the design of a closed-loop control system that does not depend on the exact dynamical model of the microjets and the accurate knowledge of the parameters of the magnetic system. The motion control characteristics in the transient- and steady-states depend on the concentration of the surrounding fluid (hydrogen peroxide solution) and the strength of the applied magnetic field. Our control system allows us to position microjets at an average velocity of 115 μm/s, and within an average region-of-convergence of 365 μm.
Parameterization for subgrid-scale motion of ice-shelf calving fronts
In order to explore the response of the Greenland ice sheet (GIS) to climate change on long (centennial to multi-millennial) time scales, a regional energy-moisture balance model has been developed. This model simulates seasonal variations of temperature and precipitation over Greenland and explicitly accounts for elevation and albedo feedbacks. From these fields, the annual mean surface temperature and surface mass balance can be determined and used to force an ice sheet model. The melt component of the surface mass balance is computed here using both a positive degree day approach and a more physically-based alternative that includes insolation and albedo explicitly. As a validation of the climate model, we first simulated temperature and precipitation over Greenland for the prescribed, present-day topography. Our simulated climatology compares well to observations and does not differ significantly from that of a simple parameterization used in many previous simulations. Furthermore, the calculated surface mass balance using both melt schemes falls within the range of recent regional climate model results. For a prescribed, ice-free state, the differences in simulated climatology between the regional energy-moisture balance model and the simple parameterization become significant, with our model showing much stronger summer warming. When coupled to a three-dimensional ice sheet model and initialized with present-day conditions, the two melt schemes both allow realistic simulations of the present-day GIS.
Self-assembly of Co/Pt stripes with current-induced domain wall motion towards 3D racetrack devices
Modification of the magnetic properties under the induced strain and curvature is a promising avenue to build three-dimensional magnetic devices, based on the domain wall motion. So far, most of the studies with 3D magnetic structures were performed in the helixes and nanowires, mainly with stationary domain walls. In this study, we demonstrate the impact of 3D geometry, strain and curvature on the current-induced domain wall motion and spin-orbital torque efficiency in the heterostructure, realized via a self-assembly rolling technique on a polymeric platform. We introduce a complete 3D memory unit with write, read and store functionality, all based on the field-free domain wall motion. Additionally, we conducted a comparative analysis between 2D and 3D structures, particularly addressing the influence of heat during the electric current pulse sequences. Finally, we demonstrated a remarkable increase of 30% in spin-torque efficiency in 3D configuration.
Derivation of gravity wave intrinsic parameters and vertical wavelength using a single scanning OH(3-1) airglow spectrometer
For the first time, we present an approach to derive zonal, meridional, and vertical wavelengths as well as periods of gravity waves based on only one OH* spectrometer, addressing one vibrational-rotational transition. Knowledge of these parameters is a precondition for the calculation of further information, such as the wave group velocity vector. OH(3-1) spectrometer measurements allow the analysis of gravity wave ground-based periods but spatial information cannot necessarily be deduced. We use a scanning spectrometer and harmonic analysis to derive horizontal wavelengths at the mesopause altitude above Oberpfaffenhofen (48.09∘ N, 11.28∘ E), Germany for 22 nights in 2015. Based on the approximation of the dispersion relation for gravity waves of low and medium frequencies and additional horizontal wind information, we calculate vertical wavelengths. The mesopause wind measurements nearest to Oberpfaffenhofen are conducted at Collm (51.30∘ N, 13.02∘ E), Germany, ca. 380 km northeast of Oberpfaffenhofen, by a meteor radar. In order to compare our results, vertical temperature profiles of TIMED-SABER (thermosphere ionosphere mesosphere energetics dynamics, sounding of the atmosphere using broadband emission radiometry) overpasses are analysed with respect to the dominating vertical wavelength.
Direct estimation of the global distribution of vertical velocity within cirrus clouds
Cirrus clouds determine the radiative balance of the upper troposphere and the transport of water vapor across the tropopause. The representation of vertical wind velocity, W, in atmospheric models constitutes the largest source of uncertainty in the calculation of the cirrus formation rate. Using global atmospheric simulations with a spatial resolution of 7 km we obtain for the first time a direct estimate of the distribution of W at the scale relevant for cirrus formation, validated against long-term observations at two different ground sites. The standard deviation in W, σ w, varies widely over the globe with the highest values resulting from orographic uplift and convection, and the lowest occurring in the Arctic. Globally about 90% of the simulated σ w values are below 0.1 m s-1 and about one in 104 cloud formation events occur in environments with σ w > 0.8 m s-1. Combining our estimate with reanalysis products and an advanced cloud formation scheme results in lower homogeneous ice nucleation frequency than previously reported, and a decreasing average ice crystal concentration with decreasing temperature. These features are in agreement with observations and suggest that the correct parameterization of σ w is critical to simulate realistic cirrus properties.
Discovery of ZrCoBi based half Heuslers with high thermoelectric conversion efficiency
Thermoelectric materials are capable of converting waste heat into electricity. The dimensionless figure-of-merit (ZT), as the critical measure for the material's thermoelectric performance, plays a decisive role in the energy conversion efficiency. Half-Heusler materials, as one of the most promising candidates for thermoelectric power generation, have relatively low ZTs compared to other material systems. Here we report the discovery of p-type ZrCoBi-based half-Heuslers with a record-high ZT of ∼1.42 at 973 K and a high thermoelectric conversion efficiency of ∼9% at the temperature difference of ∼500 K. Such an outstanding thermoelectric performance originates from its unique band structure offering a high band degeneracy (N v) of 10 in conjunction with a low thermal conductivity benefiting from the low mean sound velocity (v m ∼2800 m s-1). Our work demonstrates that ZrCoBi-based half-Heuslers are promising candidates for high-temperature thermoelectric power generation.
Continuous electroosmotic sorting of particles in grooved microchannels
We propose a novel microfluidic fractionation concept suitable for neutrally buoyant micron-sized particles. This approach takes advantage of the ability of grooved channel walls oriented at an angle to the direction of an external electric field to generate a transverse electroosmotic flow. Using computer simulations, we first demonstrate that the velocity of this secondary transverse flow depends on the distance from the wall, so neutrally buoyant particles, depending on their size and initial location, will experience different lateral displacements. We then optimize the geometry and orientation of the surface texture of the channel walls to maximize the efficiency of particle fractionation. Our method is illustrated in a full scale computer experiment where we mimic the typical microchannel with a bottom grooved wall and a source of polydisperse particles that are carried along the channel by the forward electroosmotic flow. Our simulations show that the particle dispersion can be efficiently separated by size even in a channel that is only a few texture periods long. These results can guide the design of novel microfluidic devices for efficient sorting of microparticles.
Impact of energy dissipation on interface shapes and on rates for dewetting from liquid substrates
We revisit the fundamental problem of liquid-liquid dewetting and perform a detailed comparison of theoretical predictions based on thin-film models with experimental measurements obtained by atomic force microscopy. Specifically, we consider the dewetting of a liquid polystyrene layer from a liquid polymethyl methacrylate layer, where the thicknesses and the viscosities of both layers are similar. Using experimentally determined system parameters like viscosity and surface tension, an excellent agreement of experimentally and theoretically obtained rim profile shapes are obtained including the liquid-liquid interface and even dewetting rates. Our new energetic approach additionally allows to assess the physical importance of different contributions to the energy-dissipation mechanism, for which we analyze the local flow fields and the local dissipation rates. Using this approach, we explain why dewetting rates for liquid-liquid systems follow no universal power law, despite the fact that experimental velocities are almost constant. This is in contrast to dewetting scenarios on solid substrates and in contrast to previous results for liquid-liquid substrates using heuristic approaches.