Segmentation of pneumonia lesions from CT scans of COVID-19 patients is important for accurate diagnosis and follow-up. Deep learning has a potential to automate this task but requires a large set of high-quality annotations that are difficult to collect. Learning from noisy training labels that are easier to obtain has a potential to alleviate this problem. To this end, we propose a novel noise-robust framework to learn from noisy labels for the segmentation task. We first introduce a noise-robust Dice loss that is a generalization of Dice loss for segmentation and Mean Absolute Error (MAE) loss for robustness against noise, then propose a novel COVID-19 Pneumonia Lesion segmentation network (COPLE-Net) to better deal with the lesions with various scales and appearances. The noiserobust Dice loss and COPLE-Net are combined with an adaptive self-ensembling framework for training, where an Exponential Moving Average (EMA) of a student model is used as a teacher model that is adaptively updated by suppressing the contribution of the student to EMA when the student has a large training loss. The student
We present a science forecast for the eBOSS survey. Focusing on discrete tracers, we forecast the expected accuracy of the baryonic acoustic oscillation (BAO), the redshift-space distortion (RSD) measurements, the f NL parameter quantifying the primordial non-Gaussianity, the dark energy and modified gravity parameters. We also use the line-of-sight clustering in the Ly-α forest to constrain the total neutrino mass. We find that eBOSS LRGs, ELGs and Clustering Quasars (CQs) can achieve a precision of 1%, 2.2% and 1.6%, respectively, for spherically averaged BAO distance measurements. Using the same samples, the constraint on f σ 8 is expected to be 2.5%, 3.3% and 2.8% respectively. For primordial non-Gaussianity, eBOSS alone can reach an accuracy of σ(f NL ) ∼ 10−15. eBOSS can at most improve the dark energy Figure of Merit (FoM) by a factor of 3 for the Chevallier-Polarski-Linder (CPL) parametrisation, and can well constrain three eigenmodes for the general equation-of-state parameter. eBOSS can also significantly improve constraints on modified gravity parameters by providing the RSD information, which is highly complementary to constraints obtained from weak lensing measurements. A principle component analysis (PCA) shows that eBOSS can measure the eigenmodes of the effective Newton's constant to 2% precision; this is a factor of 10 improvement over that achievable without eBOSS. Finally, we derive the eBOSS constraint (combined with Planck, DES and BOSS) on the total neutrino mass, σ(Σm ν ) = 0.03eV (68% CL), which in principle makes it possible to distinguish between the two scenarios of neutrino mass hierarchies.
We present the [Oii] (λλ3729, 3726) luminosity function measured in the redshift range 0.1 < z < 1.65 with unprecedented depth and accuracy. Our measurements are based on medium resolution flux-calibrated spectra of emission line galaxies with the FORS2 instrument at VLT and with the SDSS-III/BOSS spectrograph. The FORS2 spectra and the corresponding catalog containing redshifts and line fluxes are released along with this paper. In this work we use a novel method to combine the aforementioned surveys with GAMA, zCOSMOS and VVDS, which have different target selection, producing a consistent weighting scheme to derive the [Oii] luminosity function.The measured luminosity function is in good agreement with previous independent estimates. The comparison with two state-of-theart semi-analytical models is good, which is encouraging for the production of mock catalogs of [Oii] flux limited surveys. We observe the bright end evolution over 8.5 Gyr: we measure the decrease of log L * from 42.4 erg/s at redshift 1.44 to 41.2 at redshift 0.165 and we find that the faint end slope flattens when redshift decreases. This measurement confirms the feasibility of the target selection of future baryonic acoustic oscillation surveys aiming at observing [Oii] flux limited samples.
Highlights d SARS-CoV-2 genome sequencing and phylogenetic analyses identify 35 recurrent mutations d Association with 117 clinical phenotypes reveals potentially important mutations d D500-532 in Nsp1 coding region correlates with lower viral load and serum IFN-b d Viral isolates with D500-532 mutation induce lower IFN-I response in the infected cells
A sample of 109 Type Ia supernovae (SNe Ia) with recession velocity P30,000 km s À1 is compiled from published SN Ia light curves to explore the expansion rate of the local universe. Based on the color parameter ÁC 12 and the decline rate Ám 15 , we found that the average absorption-to-reddening ratios for SN Ia host galaxies are R UBVI ¼ 4:37 AE 0:25, 3:33 AE 0:11, 2:30 AE 0:11, and 1:18 AE 0:11, which are systematically lower than the standard values in the Galaxy. We investigated the correlations of the intrinsic luminosity with light-curve decline rate, color index, and SN environmental parameters. In particular, we found that SNe Ia in E/S0 galaxies close to the central region are brighter than those in the outer region, which may suggest a possible metallicity effect on SN luminosity. The dependence of SN luminosity on galactic environment disappears after corrections for the extinction and ÁC 12 . The Hubble diagrams constructed using 73 Hubble flow SNe Ia yield a 1 scatter of P0.12 mag in BVI bands and $0.16 mag in U band. The luminosity difference between normal SNe Ia and peculiar objects (including SN 1991bg-like and SN 1991T-like events) has now been reduced to within 0.15 mag via ÁC 12 correction. We use the same precepts to correct the nearby SNe Ia with Cepheid distances and found that the fully corrected absolute magnitudes of SNe Ia are M B ¼ À19:33 AE 0:06 and M V ¼ À19:27 AE 0:05. We deduced a value for the Hubble constant of H 0 ¼ 72 AE 6 (total) km s À1 Mpc À1 .
We present an updated U BV RI photometric catalog containing 970 objects in the field of M31, selected from the Revised Bologna Catalog (RBC v.4.0), including 965, 967, 965, 953, and 827 sources in the individual U BV RI bands, respectively, of which 205, 123, 14, 126, and 109 objects do not have previously published photometry. Photometry is performed using archival images from the Local Group Galaxies Survey, which covers 2.2 deg 2 along the major axis of M31. Detailed comparisons show that our photometry is fully consistent with previous measurements in all filters. We focus on 445 confirmed 'globular-like' clusters and candidates, comprising typical globular and young massive clusters. The ages and masses of these objects are derived by comparison of their observed spectralenergy distributions with simple stellar population synthesis. Approximately half of the clusters are younger than 2 Gyr, suggesting that there has been significant recent active star formation in M31, which is consistent with previous results. We note that clusters in the halo (r projected > 30 kpc) are composed of two different components, older clusters with ages > 10 Gyr and younger clusters with ages around 1 Gyr. The spatial distributions show that the young clusters (< 2 Gyr) are spatially coincident with the galaxy's disk, including the '10 kpc ring,' the 'outer ring,' and the halo of M31, while the old clusters (> 2 Gyr) are spatially correlated with the bulge and halo. We also estimate the masses of the 445 confirmed clusters and candidates in M31 and find that our estimates agree well with previously published values. We find that none of the young disk clusters can survive the inevitable encounters with giant molecular clouds in the galaxy's disk and that they will eventually disrupt on timescales of a few Gyr. Specifically, young disk clusters with a mass of 10 4 M ⊙ are expected to dissolve within 3.0 Gyr and will, thus, not evolve to become globular clusters.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.