The volume of a brain lesion (e.g. infarct or tumor) is a powerful indicator of patient prognosis and can be used to guide the therapeutic strategy. Lesional volume estimation is usually performed by segmentation with deep convolutional neural networks (CNN), currently the state-of-the-art approach. However, to date, few work has been done to equip volume segmentation tools with adequate quantitative predictive intervals, which can hinder their usefulness and acceptation in clinical practice. In this work, we propose TriadNet, a segmentation approach relying on a multi-head CNN architecture, which provides both the lesion volumes and the associated predictive intervals simultaneously, in less than a second. We demonstrate its superiority over other solutions on BraTS 2021, a large-scale MRI glioblastoma image database.

Volumetry is one of the principal downstream applications of 3D medical image segmentation, for example, to detect abnormal tissue growth or for surgery planning. Conformal Prediction is a promising framework for uncertainty quantification, providing calibrated predictive intervals associated with automatic volume measurements. However, this methodology is based on the hypothesis that calibration and test samples are exchangeable, an assumption that is in practice often violated in medical image applications. A weighted formulation of Conformal Prediction can be framed to mitigate this issue, but its empirical investigation in the medical domain is still lacking. A potential reason is that it relies on the estimation of the density ratio between the calibration and test distributions, which is likely to be intractable in scenarios involving high-dimensional data. To circumvent this, we propose an efficient approach for density ratio estimation relying on the compressed latent representations generated by the segmentation model. Our experiments demonstrate the efficiency of our approach to reduce the coverage error in the presence of covariate shifts, in both synthetic and real-world settings. Our implementation is available at this https URL

Planets are thought to form from dust and gas in protoplanetary disks, and debris disks are the remnants of planet formation. Aged a few Myr up to a few Gyr, debris disks have lost their primordial gas, and their dust is produced by steady-state collisions between larger, rocky bodies. Tens of debris disks, with sizes of tens, sometimes hundreds of au, have been resolved with high spatial resolution, high contrast imagers at optical/near-IR or (sub)-millimeter interferometers. They commonly show cavities, ring-like structures, and gaps, which are often regarded as indirect signatures of the presence of planets that gravitationally interact with unseen planetesimals. However, no planet responsible for these features has been detected yet, probably because of the limited sensitivity (typically 2-10 MJ) of high contrast imaging instruments prior to JWST. We have used the unprecedented sensitivity of JWST/MIRI in the thermal IR to search for such planets in the disk of the ~ 6.4 Myr old star TWA 7. With its pole-on orientation, this three-ring debris disk is indeed ideally suited for such a detection. We unambiguously detected a source 1.5 arsec from the star, that is best interpreted as a cold, sub-Jupiter mass planet. Its estimated mass (~ 0.3 MJ) and position (~ 52 au, de-projected) can thoroughly account for the main disk structures.

Deep Learning models are easily disturbed by variations in the input images that were not seen during training, resulting in unpredictable behaviours. Such Out-of-Distribution (OOD) images represent a significant challenge in the context of medical image analysis, where the range of possible abnormalities is extremely wide, including artifacts, unseen pathologies, or different imaging protocols. In this work, we evaluate various uncertainty frameworks to detect OOD inputs in the context of Multiple Sclerosis lesions segmentation. By implementing a comprehensive evaluation scheme including 14 sources of OOD of various nature and strength, we show that methods relying on the predictive uncertainty of binary segmentation models often fails in detecting outlying inputs. On the contrary, learning to segment anatomical labels alongside lesions highly improves the ability to detect OOD inputs.

The search of close (a<=5 au) giant planet(GP) companions with radial velocity(RV) around young stars and the estimate of their occurrence rates is important to constrain the migration timescales. Furthermore, this search will allow the giant planet occurrence rates to be computed at all separations via the combination with direct imaging techniques. The RV search around young stars is a challenge as they are generally faster rotators than older stars of similar spectral types and they exhibit signatures of spots or pulsation in their RV time series. Specific analyses are necessary to characterize, and possibly correct for, this activity. Our aim is to search for planets around young nearby stars and to estimate the GP occurrence rates for periods up to 1000 days. We used the SOPHIE spectrograph to observe 63 A-M young (<400 Myr) stars. We used our SAFIR software to compute the RVs and other spectroscopic observables. We then combined this survey with the HARPS YNS survey to compute the companion occurrence rates on a total of 120 young A-M stars. We report one new trend compatible with a planetary companion on HD109647. We also report HD105693 and HD112097 as binaries, and we confirm the binarity of HD2454, HD13531, HD17250A, HD28945, HD39587, HD131156, HD 142229, HD186704A, and HD 195943. We constrained for the first time the orbital parameters of HD195943B. We refute the HD13507 single brown dwarf (BD) companion solution and propose a double BD companion solution. Based on our sample of 120 young stars, we obtain a GP occurrence rate of 1_{-0.3}^{+2.2}% for periods lower than 1000 days, and we obtain an upper limit on BD occurrence rateof 0.9_{-0.9}^{+2}% in the same period range. We report a possible lack of close (1

The Gaia mission is expected to yield the detection of several thousands of exoplanets, perhaps at least doubling the number of known exoplanets. Although the harvest is expected to occur when the astrometric time series will be published with DR4 at the eve of 2026, the DR3 is already a precious database to search for exoplanet beyond 1 au. With this objective, we characterized multiple systems by exploiting two astrometric signatures derived from the DR3 astrometric solution of bright sources (G<16). We have the proper motion anomaly, or PMa, for sources also observed with Hipparcos, and the excess of residuals in the RUWE and the astrometric excess noise (AEN). Those astrometric signatures give an accurate measurement of the astrometric motion of a source seen with Gaia, even in the presence of calibration and measurement noises. We found that they can allow identifying stellar binaries and hint to companions with a mass in the planetary domain. We introduce a tool called GaiaPMEX, that is able, for a given source, to model its astrometric signatures, by a photocenter orbit due to a companion with certain mass and semi-major axis (sma). Comparing to their actual measurements from the DR3 and Hipparcos, GaiaPMEX calculates a confidence map of the possible companion's mass and sma. The constraints on mass are, as expected, degenerate, but when allowed, coupling the use of PMa and RUWE, may significantly narrow the space of solutions. Thanks to combining Gaia and Hipparcos, planets are expected to be most frequently found within 1-10 au from their star, at the scale of Earth-to-Saturn orbits. In this range, exoplanets with mass down to 0.1 MJup are more favorably detected around M-dwarfs closer than 10 pc. Some fraction, if not all, of companions identified with GaiaPMEX may be characterized in the future using the astrometric time series that will be published with the DR4.

In a previous paper, we introduced a new tool called GaiaPMEX. It characterizes the mass and semi-major axis relative to the central star (sma) of a possible companion around any source observed with Gaia. It uses the value of RUWE, or, with both Gaia and Hipparcos, the value of proper motion anomaly (PMa), alone or combined with the RUWE. Our goal is to exploit the large volume of sources in Gaia's DR3 and find new exoplanet candidates. We wish to create a new input catalog of planet-candidate hosting systems to the disposal of future follow-up projects. Beyond G=14, this catalog would prepare the arrival of powerful instruments on the ELTs, that could include RV follow-up of faint stars and direct imaging of planets around main sequence Gyr-old stars. We used the mass-sma degenerate set of solutions obtained by GaiaPMEX from any value of RUWE to select a sample of bright (G<16) Gaia sources whose companions could be planetary, with a mass <13.5 MJup. It led us to identify a sample of 9,698 planet candidate hosting sources, whose companion may have a mass <13.5 MJup in the range of 1-3-au sma. We identified 19 systems that are also reported in the Nasa exoplanet archive. We detected 8 substellar companions with a 1-3-au sma, initially discovered and characterised with RV and astrometry. Moreover, we found 6 transiting-planet systems and 2 wide-orbit systems for whom we predict the existence of supplementary companions. Focusing on the subsample of sources observed with Hipparcos, combining RUWE and PMa, we confirmed the identification of 4 new planetary candidate systems HD 187129, HD 81697, CD-42 883, and HD 105330. Given the degeneracy of mass-sma, many of the candidates in this 9,698 sources catalog might have a larger mass, in the brown-dwarf and stellar domain, if their sma departs from the 1-3-au range. The vetting of this large catalog will be the subject of future studies.