Deep learning offers considerable promise for medical diagnostics. We aimed to evaluate the diagnostic accuracy of deep learning algorithms versus health-care professionals in classifying diseases ...using medical imaging.
In this systematic review and meta-analysis, we searched Ovid-MEDLINE, Embase, Science Citation Index, and Conference Proceedings Citation Index for studies published from Jan 1, 2012, to June 6, 2019. Studies comparing the diagnostic performance of deep learning models and health-care professionals based on medical imaging, for any disease, were included. We excluded studies that used medical waveform data graphics material or investigated the accuracy of image segmentation rather than disease classification. We extracted binary diagnostic accuracy data and constructed contingency tables to derive the outcomes of interest: sensitivity and specificity. Studies undertaking an out-of-sample external validation were included in a meta-analysis, using a unified hierarchical model. This study is registered with PROSPERO, CRD42018091176.
Our search identified 31 587 studies, of which 82 (describing 147 patient cohorts) were included. 69 studies provided enough data to construct contingency tables, enabling calculation of test accuracy, with sensitivity ranging from 9·7% to 100·0% (mean 79·1%, SD 0·2) and specificity ranging from 38·9% to 100·0% (mean 88·3%, SD 0·1). An out-of-sample external validation was done in 25 studies, of which 14 made the comparison between deep learning models and health-care professionals in the same sample. Comparison of the performance between health-care professionals in these 14 studies, when restricting the analysis to the contingency table for each study reporting the highest accuracy, found a pooled sensitivity of 87·0% (95% CI 83·0-90·2) for deep learning models and 86·4% (79·9-91·0) for health-care professionals, and a pooled specificity of 92·5% (95% CI 85·1-96·4) for deep learning models and 90·5% (80·6-95·7) for health-care professionals.
Our review found the diagnostic performance of deep learning models to be equivalent to that of health-care professionals. However, a major finding of the review is that few studies presented externally validated results or compared the performance of deep learning models and health-care professionals using the same sample. Additionally, poor reporting is prevalent in deep learning studies, which limits reliable interpretation of the reported diagnostic accuracy. New reporting standards that address specific challenges of deep learning could improve future studies, enabling greater confidence in the results of future evaluations of this promising technology.
None.
Rapid rotation is a fundamental characteristic of classical Be stars and a crucial property allowing for the formation of their circumstellar disks. Past evolution in a mass and angular momentum ...transferring binary system offers a plausible solution to how Be stars attained their fast rotation. Although the subdwarf remnants of mass donors in such systems should exist in abundance, only a few have been confirmed due to tight observational constraints. An indirect method of detecting otherwise hidden companions is offered by their effect on the outer parts of Be star disks, which are expected to be disrupted or truncated. In the context of the infrared and radio continuum excess radiation originating in the disk, the disk truncation can be revealed by a turndown in the spectral energy distribution due to reduced radio flux levels. In this work, we search for signs of spectral turndown in a sample of 57 classical Be stars with radio data, which include new data for 23 stars and the longest-wavelength detections so far (λ 10 cm) for two stars. We confidently detect the turndown for all 26 stars with sufficient data coverage (20 of which are not known to have close binary companions). For the remaining 31 stars, the data are inconclusive as to whether the turndown is present or not. The analysis suggests that many if not all Be stars have close companions influencing their outer disks. If confirmed to be subdwarf companions, the mass transfer spin-up scenario might explain the existence of the vast majority of classical Be stars.
Deep learning has the potential to transform health care; however, substantial expertise is required to train such models. We sought to evaluate the utility of automated deep learning software to ...develop medical image diagnostic classifiers by health-care professionals with no coding-and no deep learning-expertise.
We used five publicly available open-source datasets: retinal fundus images (MESSIDOR); optical coherence tomography (OCT) images (Guangzhou Medical University and Shiley Eye Institute, version 3); images of skin lesions (Human Against Machine HAM 10000), and both paediatric and adult chest x-ray (CXR) images (Guangzhou Medical University and Shiley Eye Institute, version 3 and the National Institute of Health NIH dataset, respectively) to separately feed into a neural architecture search framework, hosted through Google Cloud AutoML, that automatically developed a deep learning architecture to classify common diseases. Sensitivity (recall), specificity, and positive predictive value (precision) were used to evaluate the diagnostic properties of the models. The discriminative performance was assessed using the area under the precision recall curve (AUPRC). In the case of the deep learning model developed on a subset of the HAM10000 dataset, we did external validation using the Edinburgh Dermofit Library dataset.
Diagnostic properties and discriminative performance from internal validations were high in the binary classification tasks (sensitivity 73·3-97·0%; specificity 67-100%; AUPRC 0·87-1·00). In the multiple classification tasks, the diagnostic properties ranged from 38% to 100% for sensitivity and from 67% to 100% for specificity. The discriminative performance in terms of AUPRC ranged from 0·57 to 1·00 in the five automated deep learning models. In an external validation using the Edinburgh Dermofit Library dataset, the automated deep learning model showed an AUPRC of 0·47, with a sensitivity of 49% and a positive predictive value of 52%.
All models, except the automated deep learning model trained on the multilabel classification task of the NIH CXR14 dataset, showed comparable discriminative performance and diagnostic properties to state-of-the-art performing deep learning algorithms. The performance in the external validation study was low. The quality of the open-access datasets (including insufficient information about patient flow and demographics) and the absence of measurement for precision, such as confidence intervals, constituted the major limitations of this study. The availability of automated deep learning platforms provide an opportunity for the medical community to enhance their understanding in model development and evaluation. Although the derivation of classification models without requiring a deep understanding of the mathematical, statistical, and programming principles is attractive, comparable performance to expertly designed models is limited to more elementary classification tasks. Furthermore, care should be placed in adhering to ethical principles when using these automated models to avoid discrimination and causing harm. Future studies should compare several application programming interfaces on thoroughly curated datasets.
National Institute for Health Research and Moorfields Eye Charity.
Abstract
Be stars are main-sequence massive stars with emission features in their spectrum, which originates in circumstellar gaseous discs. Even though the viscous decretion disc model can ...satisfactorily explain most observations, two important physical ingredients, namely the magnitude of the viscosity (α) and the disc mass injection rate, remain poorly constrained. The light curves of Be stars that undergo events of disc formation and dissipation offer an opportunity to constrain these quantities. A pipeline was developed to model these events that use a grid of synthetic light curves, computed from coupled hydrodynamic and radiative transfer calculations. A sample of 54 Be stars from the OGLE survey of the Small Magellanic Cloud (SMC) was selected for this study. Because of the way our sample was selected (bright stars with clear disc events), it likely represents the densest discs in the SMC. Like their siblings in the Galaxy, the mass of the disc in the SMC increases with the stellar mass. The typical mass and angular momentum loss rates associated with the disc events are of the order of ∼10−10 M⊙ yr−1 and ∼5 × 1036 g cm2 s−2, respectively. The values of α found in this work are typically of a few tenths, consistent with recent results in the literature and with the ones found in dwarf novae, but larger than current theory predicts. Considering the sample as a whole, the viscosity parameter is roughly two times larger at build-up (〈αbu〉 = 0.63) than at dissipation (〈αd〉 = 0.26). Further work is necessary to verify whether this trend is real or a result of some of the model assumptions.
Context. The structure of the inner parts of Be star disks (≲ 20 stellar radii) is well explained by the viscous decretion disk (VDD) model, which is able to reproduce the observable properties of ...most of the objects studied so far. The outer parts, on the other hand, are not observationally well-explored, as they are observable only at radio wavelengths. A steepening of the spectral slope somewhere between infrared and radio wavelengths was reported for several Be stars that were previously detected in the radio, but a convincing physical explanation for this trend has not yet been provided. Aims. We test the VDD model predictions for the extended parts of a sample of six Be disks that have been observed in the radio to address the question of whether the observed turndown in the spectral energy distribution (SED) can be explained in the framework of the VDD model, including recent theoretical development for truncated Be disks in binary systems. Methods. We combine new multi-wavelength radio observations from the Karl. G. Jansky Very Large Array (JVLA) and Atacama Pathfinder Experiment (APEX) with previously published radio data and archival SED measurements at ultraviolet, visual, and infrared wavelengths. The density structure of the disks, including their outer parts, is constrained by radiative transfer modeling of the observed spectrum using VDD model predictions. In the VDD model we include the presumed effects of possible tidal influence from faint binary companions. Results. For 5 out of 6 studied stars, the observed SED shows strong signs of SED turndown between far-IR and radio wavelengths. A VDD model that extends to large distances closely reproduces the observed SEDs up to far IR wavelengths, but fails to reproduce the radio SED. Using a truncated VDD model improves the fit, leading to a successful explanation of the SED turndown observed for the stars in our sample. The slope of the observed SEDs in the radio is however not well reproduced by disks that are simply cut off at a certain distance. Rather, some matter seems to extend beyond the truncation radius, where it still contributes to the observed SEDs, making the spectral slope in the radio shallower. This finding is in agreement with our current understanding of binary truncation from hydrodynamical simulations, in which the disk does extend past the truncation radius. Therefore, the most probable cause for the SED turndown is the presence of binary companions that remain undetected for most of our sources.
Multitechnique study of the B[e] supergiant RMC 82 Seriacopi, D B; Carciofi, A C; de Amorim, T H ...
Monthly notices of the Royal Astronomical Society,
01/2024, Letnik:
528, Številka:
1
Journal Article
Recenzirano
Odprti dostop
ABSTRACT
Be supergiants (sgBe) are rare objects whose evolutionary stage remains uncertain. Observationally, they display strong Balmer emission lines, infrared excess, and intrinsic polarization, ...indicating a non-spherical circumstellar envelope. We present a study of the sgBe RMC 82, using new spectropolarimetric data complemented by photometry from the ultraviolet (UV) to the mid-infrared. Our two-component model comprises a slow, dense equatorial wind wherein dust grains form and a fast polar wind. We applied the hdust radiative transfer code and Bayesian statistics to infer the parameters from a grid of 3240 pre-computed models. The model accurately reproduces the spectral energy distribution and polarized spectrum, but struggles to match the H α emission. Our results suggest a large mass-loss rate of $6.6 \times 10^{-6}\, \mathrm{{\rm M}_{\odot }\, yr^{-1}\, sr^{-1}}$. The dense wind is confined within an opening angle of 11°. The hottest dust grains are located at 277 R* with a temperature of 870 K. The dust grains are porous, with a density of 0.051 $\rm {g\, cm^{-3}}$. The central star was found to be significantly hotter than previous estimates (Teff = $27\, 000$ K). By comparing models with different components, we find that gas reprocesses a significant amount of UV radiation, shielding the dust. However, the dust also scatters UV photons back to the inner disc, increasing its temperature and H α emission. We conclude that self-consistent models, that account for the gas–dust interplay in the envelope, are essential for studying sgBe and similar objects.
ABSTRACT
Light scattered off particles can become linearly polarized. Stars surrounded by oblique, co-rotating envelopes are therefore expected to manifest periodic linear polarimetric variations. ...The electron scattering magnetospheres of magnetic massive stars are expected to be suitable candidates to observe this effect. In this paper, we present the first semi-analytical model capable of synthesizing the continuum polarimetric signatures of magnetic O-type stars in an optically thin, single electron scattering limit. The purpose of this investigation is to improve our general understanding of magnetic hot stars by characterizing their polarimetric behaviour. Our linear polarization model is constructed by combining the analytical expressions for the polarimetric variations of an obliquely rotating envelope with the analytic dynamical magnetosphere model to represent a physical model for the envelope density structure. We compute grids of model Stokes Q and U curves and show that their shapes are unique to the choice of inclination and obliquity angles. We apply our model to HD 191612, a prototypical Of?p-type star, having both polarimetric and photometric observations. We find that the polarimetric modulations are best reproduced with $i=19^{+12}_{-3}$○, $\beta =71^{+3}_{-9}$○, and $\log \dot{M}_{B=0}=-6.11^{+0.12}_{-0.06}$ M⊙ yr−1. These results agree with previous investigations of this star. By combining both polarimetric and photometric synthesis tools, we simultaneously model the observations thus adding further refinement of the wind and magnetic properties of HD 191612.
Aims. The X-shooter archive of several thousand telluric standard star spectra was skimmed for Be and Be shell stars to derive the stellar fundamental parameters and statistical properties, in ...particular for the less investigated late-type Be stars and the extension of the Be phenomenon into early A stars. Methods. An adapted version of the BCD method is used, using the Balmer discontinuity parameters to determine effective temperature and surface gravity. This method is optimally suited for late B stars. The projected rotational velocity was obtained by profile fitting to the Mg ii lines of the targets, and the spectra were inspected visually for the presence of peculiar features such as the infrared Ca ii triplet or the presence of a double Balmer discontinuity. The Balmer line equivalent widths were measured, but they are only useful for determining the pure emission contribution in a subsample of Be stars owing to uncertainties in determining the photospheric contribution. Results. A total of 78, mostly late-type, Be stars, were identified in the X-shooter telluric standard star archive, out of which 48 had not been reported before. We confirm the general trend that late-type Be stars have more tenuous disks and are less variable than early-type Be stars. The relatively large number (48) of relatively bright (V> 8.5) additional Be stars casts some doubt on the statistics of late-type Be stars; they are more common than currently thought. The Be/B star fraction may not strongly depend on spectral subtype.
Aims.
We present a detailed visible and near-infrared spectro-interferometric analysis of the Be-shell star
o
Aquarii from quasi-contemporaneous CHARA/VEGA and VLTI/AMBER observations.
Methods.
We ...analyzed spectro-interferometric data in the H
α
(VEGA) and Br
γ
(AMBER) lines using models of increasing complexity: simple geometric models, kinematic models, and radiative transfer models computed with the 3D non-LTE code HDUST.
Results.
We measured the stellar radius of
o
Aquarii in the visible with a precision of 8%: 4.0 ± 0.3
R
⊙
. We constrained the circumstellar disk geometry and kinematics using a kinematic model and a MCMC fitting procedure. The emitting disk sizes in the H
α
and Br
γ
lines were found to be similar, at ~10–12 stellar diameters, which is uncommon since most results for Be stars indicate a larger extension in H
α
than in Br
γ
. We found that the inclination angle
i
derived from H
α
is significantly lower (~15°) than the one derived from Br
γ
:
i
~ 61.2° and 75.9°, respectively. While the two lines originate from a similar region of the disk, the disk kinematics were found to be near to the Keplerian rotation (i.e.,
β
= −0.5) in Br
γ
(
β
~ −0.43), but not in H
α
(
β
~ −0.30). After analyzing all our data using a grid of HDUST models (BeAtlas), we found a common physical description for the circumstellar disk in both lines: a base disk surface density Σ
0
= 0.12 g cm
−2
and a radial density law exponent
m
= 3.0. The same kind of discrepancy, as with the kinematic model, is found in the determination of
i
using the BeAtlas grid. The stellar rotational rate was found to be very close (~96%) to the critical value. Despite being derived purely from the fit to interferometric data, our best-fit HDUST model provides a very reasonable match to non-interferometric observables of
o
Aquarii: the observed spectral energy distribution, H
α
and Br
γ
line profiles, and polarimetric quantities. Finally, our analysis of multi-epoch H
α
profiles and imaging polarimetry indicates that the disk structure has been (globally) stable for at least 20 yr.
Conclusions.
Looking at the visible continuum and Br
γ
emission line only,
o
Aquarii fits in the global scheme of Be stars and their circumstellar disk: a (nearly) Keplerian rotating disk well described by the viscous decretion disk (VDD) model. However, the data in the H
α
line shows a substantially different picture that cannot fully be understood using the current generation of physical models of Be star disks. The Be star
o
Aquarii presents a stable disk (close to the steady-state), but, as in previous analyses, the measured
m
is lower than the standard value in the VDD model for the steady-state regime (
m
= 3.5). This suggests that some assumptions of this model should be reconsidered. Also, such long-term disk stability could be understood in terms of the high rotational rate that we measured for this star, the rate being a main source for the mass injection in the disk. Our results on the stellar rotation and disk stability are consistent with results in the literature showing that late-type Be stars are more likely to be fast rotators and have stable disks.