The presence of confounding effects (or biases) is one of the most critical challenges in using deep learning to advance discovery in medical imaging studies. Confounders affect the relationship ...between input data (e.g., brain MRIs) and output variables (e.g., diagnosis). Improper modeling of those relationships often results in spurious and biased associations. Traditional machine learning and statistical models minimize the impact of confounders by, for example, matching data sets, stratifying data, or residualizing imaging measurements. Alternative strategies are needed for state-of-the-art deep learning models that use end-to-end training to automatically extract informative features from large set of images. In this article, we introduce an end-to-end approach for deriving features invariant to confounding factors while accounting for intrinsic correlations between the confounder(s) and prediction outcome. The method does so by exploiting concepts from traditional statistical methods and recent fair machine learning schemes. We evaluate the method on predicting the diagnosis of HIV solely from Magnetic Resonance Images (MRIs), identifying morphological sex differences in adolescence from those of the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA), and determining the bone age from X-ray images of children. The results show that our method can accurately predict while reducing biases associated with confounders. The code is available at https://github.com/qingyuzhao/br-net .
ABSTRACT
A very small fraction of (runaway) massive stars have masses exceeding $60\!-\!70\, \rm M_{\odot }$ and are predicted to evolve as luminous blue variable and Wolf–Rayet stars before ending ...their lives as core-collapse supernovae. Our 2D axisymmetric hydrodynamical simulations explore how a fast wind ($2000\, \rm km\, \rm s^{-1}$) and high mass-loss rate ($10^{-5}\, \rm M_{\odot }\, \rm yr^{-1}$) can impact the morphology of the circumstellar medium. It is shaped as 100 pc-scale wind nebula that can be pierced by the driving star when it supersonically moves with velocity $20\!-\!40\, \rm km\, \rm s^{-1}$ through the interstellar medium (ISM) in the Galactic plane. The motion of such runaway stars displaces the position of the supernova explosion out of their bow shock nebula, imposing asymmetries to the eventual shock wave expansion and engendering Cygnus-loop-like supernova remnants. We conclude that the size (up to more than $200\, \rm pc$) of the filamentary wind cavity in which the chemically enriched supernova ejecta expand, mixing efficiently the wind and ISM materials by at least $10{{\ \rm per\ cent}}$ in number density, can be used as a tracer of the runaway nature of the very massive progenitors of such $0.1\, \rm Myr$ old remnants. Our results motivate further observational campaigns devoted to the bow shock of the very massive stars BD+43°3654 and to the close surroundings of the synchrotron-emitting Wolf–Rayet shell G2.4+1.4.
Since the detection of nonthermal radio emission from the bow shock of the massive runaway star BD +43°3654, simple models have predicted high-energy emission, at X-rays and gamma-rays, from these ...Galactic sources. Observational searches for this emission so far give no conclusive evidence but a few candidates at gamma-rays. In this work we aim at developing a more sophisticated model for the nonthermal emission from massive runaway star bow shocks. The main goal is to establish whether these systems are efficient nonthermal emitters, even if they are not strong enough yet to be detected. For modeling the collision between the stellar wind and the interstellar medium we use 2D hydrodynamic simulations. We then adopt the flow profile of the wind and the ambient medium obtained with the simulation as the plasma state for solving the transport of energetic particles injected in the system, as well as the nonthermal emission they produce. For this purpose we solve a 3D (two spatial + energy) advection-diffusion equation in the test-particle approximation. We find that a massive runaway star with a powerful wind converts 0.16%-0.4% of the power injected in electrons into nonthermal emission, mostly produced by inverse Compton scattering of dust-emitted photons by relativistic electrons, and second by synchrotron radiation. This represents a fraction of ∼10−5 to 10−4 of the wind kinetic power. Given the better sensibility of current instruments at radio wavelengths, these systems are more prone to be detected at radio through the synchrotron emission they produce rather than at gamma energies.
This work revisits the electrostatic instability for blazar-induced pair beams propagating through the intergalactic medium (IGM) using linear analysis and PIC simulations. We study the impact of the ...realistic distribution function of pairs resulting from the interaction of high-energy gamma-rays with the extragalactic background light. We present analytical and numerical calculations of the linear growth rate of the instability for the arbitrary orientation of wave vectors. Our results explicitly demonstrate that the finite angular spread of the beam dramatically affects the growth rate of the waves, leading to the fastest growth for wave vectors quasi-parallel to the beam direction and a growth rate at oblique directions that is only a factor of 2-4 smaller compared to the maximum. To study the nonlinear beam relaxation, we performed PIC simulations that take into account a realistic wide-energy distribution of beam particles. The parameters of the simulated beam-plasma system provide an adequate physical picture that can be extrapolated to realistic blazar-induced pairs. In our simulations, the beam looses only 1% of its energy, and we analytically estimate that the beam would lose its total energy over about 100 simulation times. An analytical scaling is then used to extrapolate the parameters of realistic blazar-induced pair beams. We find that they can dissipate their energy slightly faster by the electrostatic instability than through inverse-Compton scattering. The uncertainties arising from, e.g., details of the primary gamma-ray spectrum are too large to make firm statements for individual blazars, and an analysis based on their specific properties is required.
ABSTRACT
A signification fraction of Galactic massive stars (${\ge}8\, \rm M_{\odot }$) are ejected from their parent cluster and supersonically sail away through the interstellar medium (ISM). The ...winds of these fast-moving stars blow asymmetric bubbles thus creating a circumstellar environment in which stars eventually die with a supernova explosion. The morphology of the resulting remnant is largely governed by the circumstellar medium of the defunct progenitor star. In this paper, we present 2D magneto-hydrodynamical simulations investigating the effect of the ISM magnetic field on the shape of the supernova remnants of a $35\, \mathrm{M}_{\odot }$ star evolving through a Wolf–Rayet phase and running with velocity 20 and $40\, \rm km\, \rm s^{-1}$, respectively. A $7\, \mu \rm G$ ambient magnetic field is sufficient to modify the properties of the expanding supernova shock front and in particular to prevent the formation of filamentary structures. Prior to the supernova explosion, the compressed magnetic field in the circumstellar medium stabilizes the wind/ISM contact discontinuity in the tail of the wind bubble. A consequence is a reduced mixing efficiency of ejecta and wind materials in the inner region of the remnant, where the supernova shock wave propagates. Radiative transfer calculations for synchrotron emission reveal that the non-thermal radio emission has characteristic features reflecting the asymmetry of exiled core-collapse supernova remnants from Wolf–Rayet progenitors. Our models are qualitatively consistent with the radio appearance of several remnants of high-mass progenitors, namely the bilateral G296.5+10.0 and the shell-type remnants CTB109 and Kes 17, respectively.
Context.
Supernova remnants (SNRs) are known to accelerate particles to relativistic energies, on account of their nonthermal emission. The observational progress from radio to gamma-ray observations ...reveals more and more morphological features that need to be accounted for when modeling the emission from those objects.
Aims.
We use our time-dependent acceleration code RATPaC to study the formation of extended gamma-ray halos around supernova remnants and the morphological implications that arise when the high-energetic particles start to escape from the remnant.
Methods.
We performed spherically symmetric 1D simulations in which we simultaneously solved the transport equations for cosmic rays, magnetic turbulence, and the hydrodynamical flow of the thermal plasma in a volume large enough to keep all cosmic rays in the simulation. The transport equations for cosmic rays and magnetic turbulence were coupled via the cosmic-ray gradient and the spatial diffusion coefficient of the cosmic rays, while the cosmic-ray feedback onto the shock structure can be ignored. Our simulations span 25 000 yr, thus covering the free-expansion and the Sedov-Taylor phase of the remnant’s evolution.
Results.
We find a strong difference in the morphology of the gamma-ray emission from supernova remnants at later stages dependent on the emission process. At early times, both the inverse-Compton and the Pion-decay morphology are shell-like. However, as soon as the maximum-energy of the freshly accelerated particles starts to fall, the inverse-Compton morphology starts to become center-filled, whereas the Pion-decay morphology keeps its shell-like structure. Escaping high-energy electrons start to form an emission halo around the remnant at this time. There are good prospects for detecting this spectrally hard emission with the future Cerenkov Telescope Array, as there are for detecting variations in the gamma-ray spectral index across the interior of the remnant. Further, we find a constantly decreasing nonthermal X-ray flux that makes a detection of X-ray unlikely after the first few thousand years of the remnants’ evolution. The radio flux is increasing throughout the SNR’s lifetime and changes from a shell-like to a more center-filled morphology later on.
Context.
Supernova remnants are known to accelerate cosmic rays on account of their nonthermal emission of radio waves, X-rays, and gamma rays. Although there are many models for the acceleration of ...cosmic rays in supernova remnants, the escape of cosmic rays from these sources has not yet been adequately studied.
Aims.
We aim to use our time-dependent acceleration code RATPaC to study the acceleration of cosmic rays and their escape in post-adiabatic supernova remnants and calculate the subsequent gamma-ray emission from inverse-Compton scattering and Pion decay.
Methods.
We performed spherically symmetric 1D simulations in which we simultaneously solved the transport equations for cosmic rays, magnetic turbulence, and the hydrodynamical flow of the thermal plasma in a volume large enough to keep all cosmic rays in the simulation. The transport equations for cosmic rays and magnetic turbulence were coupled via the cosmic-ray gradient and the spatial diffusion coefficient of the cosmic rays, while the cosmic-ray feedback onto the shock structure can be ignored. Our simulations span 100 000 years, thus covering the free-expansion, the Sedov–Taylor, and the beginning of the post-adiabatic phase of the remnant’s evolution.
Results.
At later stages of the evolution, cosmic rays over a wide range of energy can reside outside of the remnant, creating spectra that are softer than predicted by standard diffusive shock acceleration, and feature breaks in the 10 − 100 GeV-range. The total spectrum of cosmic rays released into the interstellar medium has a spectral index of
s
≈ 2.4 above roughly 10 GeV which is close to that required by Galactic propagation models. We further find the gamma-ray luminosity to peak around an age of 4000 years for inverse-Compton-dominated high-energy emission. Remnants expanding in low-density media generally emit more inverse-Compton radiation, matching the fact that the brightest known supernova remnants – RCW86, Vela Jr., HESS J1731−347 and RX J1713.7−3946 – are all expanding in low density environments.
ABSTRACT
Wolf–Rayet stars are advanced evolutionary stages of massive stars. Despite their large mass-loss rates and high wind velocities, none of them displays a bow shock, although a fraction of ...them are classified as runaway. Our 2.5-D numerical simulations of circumstellar matter around a $60\mbox{-}\rm M_{\odot }$ runaway star show that the fast Wolf–Rayet stellar wind is released into a wind-blown cavity filled with various shocks and discontinuities generated throughout the preceding evolutionary phases. The resulting fast-wind–slow-wind interaction leads to the formation of spherical shells of swept-up dusty material similar to those observed in the near-infrared at $24\, \rm \mu \rm m$ with Spitzer, which appear to be comoving with the runaway massive stars, regardless of their proper motion and/or the properties of the local ambient medium. We interpret bright infrared rings around runaway Wolf–Rayet stars in the Galactic plane as an indication of their very high initial masses and complex evolutionary history. Stellar-wind bow shocks become faint as stars run in diluted media, therefore our results explain the absence of bow shocks detected around Galactic Wolf–Rayet stars, such as the high-latitude, very fast-moving objects WR71, WR124 and WR148. Our results show that the absence of a bow shock is consistent with the runaway nature of some Wolf–Rayet stars. This questions the in situ star formation scenario of high-latitude Wolf–Rayet stars in favour of dynamical ejection from birth sites in the Galactic plane.
Response inhibition is disturbed in several disorders sharing impulse control deficits as a core symptom. Since response inhibition is a cognitively and neurally multifaceted function which has been ...shown to rely on differing neural subprocesses and neurotransmitter systems, further differentiation to define neurophysiological endophenotypes is essential. Response inhibition may involve at least three separable cognitive subcomponents, i.e. interference inhibition, action withholding, and action cancelation. Here, we introduce a novel paradigm – the Hybrid Response Inhibition task – to disentangle interference inhibition, action withholding and action cancelation and their neural subprocesses within one task setting during functional magnetic resonance imaging (fMRI). To validate the novel task, results were compared to a battery of separate, standard response inhibition tasks independently capturing these subcomponents and subprocesses. Across all subcomponents, mutual activation was present in the right inferior frontal cortex (rIFC), pre-supplementary motor area (pre-SMA) and parietal regions. Interference inhibition revealed stronger activation in pre-motor and parietal regions. Action cancelation resulted in stronger activation in fronto-striatal regions. Our results show that all subcomponents share a common neural network and thus all constitute different subprocesses of response inhibition. Subprocesses, however, differ to the degree of regional involvement: interference inhibition relies more pronouncedly on a fronto-parietal–pre-motor network suggesting its close relation to response selection processes. Action cancelation, in turn, is more strongly associated with the fronto-striatal pathway implicating it as a late subcomponent of response inhibition. The new paradigm reliably captures three putatively subsequent subprocesses of response inhibition and might be a promising tool to differentially assess disturbed neural networks in disorders showing impulse control deficits.
► A novel task disentangling three subcomponents of response inhibition is introduced. ► All subcomponents share a common fronto-parietal inhibition network. ► Mutual activation in the rIFC and the pre-SMA is present in all subcomponents. ► Interference inhibition is based on a fronto-parietal–pre-motor network. ► Action cancelation relies on the (indirect) prefrontal–striatal pathway.