-
Latent Diffusion Models with Image-Derived Annotations for Enhanced AI-Assisted Cancer Diagnosis in Histopathology
Authors:
Pedro Osorio,
Guillermo Jimenez-Perez,
Javier Montalt-Tordera,
Jens Hooge,
Guillem Duran-Ballester,
Shivam Singh,
Moritz Radbruch,
Ute Bach,
Sabrina Schroeder,
Krystyna Siudak,
Julia Vienenkoetter,
Bettina Lawrenz,
Sadegh Mohammadi
Abstract:
Artificial Intelligence (AI) based image analysis has an immense potential to support diagnostic histopathology, including cancer diagnostics. However, develo** supervised AI methods requires large-scale annotated datasets. A potentially powerful solution is to augment training data with synthetic data. Latent diffusion models, which can generate high-quality, diverse synthetic images, are promi…
▽ More
Artificial Intelligence (AI) based image analysis has an immense potential to support diagnostic histopathology, including cancer diagnostics. However, develo** supervised AI methods requires large-scale annotated datasets. A potentially powerful solution is to augment training data with synthetic data. Latent diffusion models, which can generate high-quality, diverse synthetic images, are promising. However, the most common implementations rely on detailed textual descriptions, which are not generally available in this domain. This work proposes a method that constructs structured textual prompts from automatically extracted image features. We experiment with the PCam dataset, composed of tissue patches only loosely annotated as healthy or cancerous. We show that including image-derived features in the prompt, as opposed to only healthy and cancerous labels, improves the Fréchet Inception Distance (FID) from 178.8 to 90.2. We also show that pathologists find it challenging to detect synthetic images, with a median sensitivity/specificity of 0.55/0.55. Finally, we show that synthetic data effectively trains AI models.
△ Less
Submitted 15 December, 2023;
originally announced December 2023.
-
A Decade of SCUBA-2: A Comprehensive Guide to Calibrating 450 $μ$m and 850 $μ$m Continuum Data at the JCMT
Authors:
Steve Mairs,
Jessica T. Dempsey,
Graham S. Bell,
Harriet Parsons,
Malcolm J. Currie,
Per Friberg,
Xue-Jian Jiang,
Alexandra J. Tetarenko,
Dan Bintley,
Jamie Cookson,
Shaoliang Li,
Mark G. Rawlings,
Jan Wouterloot,
David Berry,
Sarah Graves,
Izumi Mizuno,
Alexis Ann Acohido,
Alyssa Clark,
Jeff Cox,
Miriam Fuchs,
James Hoge,
Johnathon Kemp,
E'lisa Lee,
Callie Matulonis,
William Montgomerie
, et al. (2 additional authors not shown)
Abstract:
The Submillimetre Common User Bolometer Array 2 (SCUBA-2) is the James Clerk Maxwell Telescope's continuum imager, operating simultaneously at 450 and 850~$μ$m. SCUBA-2 was commissioned in 2009--2011 and since that time, regular observations of point-like standard sources have been performed whenever the instrument is in use. Expanding the calibrator observation sample by an order of magnitude com…
▽ More
The Submillimetre Common User Bolometer Array 2 (SCUBA-2) is the James Clerk Maxwell Telescope's continuum imager, operating simultaneously at 450 and 850~$μ$m. SCUBA-2 was commissioned in 2009--2011 and since that time, regular observations of point-like standard sources have been performed whenever the instrument is in use. Expanding the calibrator observation sample by an order of magnitude compared to previous work, in this paper we derive updated opacity relations at each wavelength for a new atmospheric-extinction correction, analyze the Flux-Conversion Factors (FCFs) used to convert instrumental units to physical flux units as a function of date and observation time, present information on the beam profiles for each wavelength, and update secondary-calibrator source fluxes. Between 07:00 and 17:00 UTC, the portion of the night that is most stable to temperature gradients that cause dish deformation, the total-flux uncertainty and the peak-flux uncertainty measured at 450~$μ$m are found to be 14\% and 17\%, respectively. Measured at 850~$μ$m, the total-flux and peak-flux uncertainties are 6\%, and 7\%, respectively. The analysis presented in this work is applicable to all SCUBA-2 projects observed since 2011.
△ Less
Submitted 28 July, 2021;
originally announced July 2021.
-
Phosphine Gas in the Cloud Decks of Venus
Authors:
Jane S. Greaves,
Anita M. S. Richards,
William Bains,
Paul B. Rimmer,
Hideo Sagawa,
David L. Clements,
Sara Seager,
Janusz J. Petkowski,
Clara Sousa-Silva,
Sukrit Ranjan,
Emily Drabek-Maunder,
Helen J. Fraser,
Annabel Cartwright,
Ingo Mueller-Wodarg,
Zhuchang Zhan,
Per Friberg,
Iain Coulson,
E'lisa Lee,
Jim Hoge
Abstract:
Measurements of trace-gases in planetary atmospheres help us explore chemical conditions different to those on Earth. Our nearest neighbor, Venus, has cloud decks that are temperate but hyper-acidic. We report the apparent presence of phosphine (PH3) gas in Venusian atmosphere, where any phosphorus should be in oxidized forms. Single-line millimeter-waveband spectral detections (quality up to ~15…
▽ More
Measurements of trace-gases in planetary atmospheres help us explore chemical conditions different to those on Earth. Our nearest neighbor, Venus, has cloud decks that are temperate but hyper-acidic. We report the apparent presence of phosphine (PH3) gas in Venusian atmosphere, where any phosphorus should be in oxidized forms. Single-line millimeter-waveband spectral detections (quality up to ~15 sigma) from the JCMT and ALMA telescopes have no other plausible identification. Atmospheric PH3 at ~20 parts-per-billion abundance is inferred. The presence of phosphine is unexplained after exhaustive study of steady-state chemistry and photochemical pathways, with no currently-known abiotic production routes in Venusian atmosphere, clouds, surface and subsurface, or from lightning, volcanic or meteoritic delivery. Phosphine could originate from unknown photochemistry or geochemistry, or, by analogy with biological production of phosphine on Earth, from the presence of life. Other PH3 spectral features should be sought, while in-situ cloud and surface sampling could examine sources of this gas.
△ Less
Submitted 14 September, 2020;
originally announced September 2020.
-
Rotation State of Comet 103P/Hartley 2 from Radio Spectroscopy at 1 mm
Authors:
Michal Drahus,
David Jewitt,
Aurelie Guilbert-Lepoutre,
Waclaw Waniak,
James Hoge,
Dariusz Lis,
Hiroshige Yoshida,
Ruisheng Peng,
Albrecht Sievers
Abstract:
The nuclei of active comets emit molecules anisotropically from discrete vents. As the nucleus rotates, we expect to observe periodic variability in the molecular emission line profiles, which can be studied through mm/submm spectroscopy. Using this technique we investigated the HCN atmosphere of comet 103P/Hartley 2, the target of NASA's EPOXI mission, which had an exceptionally favorable apparit…
▽ More
The nuclei of active comets emit molecules anisotropically from discrete vents. As the nucleus rotates, we expect to observe periodic variability in the molecular emission line profiles, which can be studied through mm/submm spectroscopy. Using this technique we investigated the HCN atmosphere of comet 103P/Hartley 2, the target of NASA's EPOXI mission, which had an exceptionally favorable apparition in late 2010. We detected short-term evolution of the spectral line profile, which was stimulated by the nucleus rotation, and which provides evidence for rapid deceleration and excitation of the rotation state. The measured rate of change in the rotation period is +1.00 \pm 0.15 min per day and the period itself is 18.32 \pm 0.03 hr, both applicable at the epoch of the EPOXI encounter. Surprisingly, the spin-down efficiency is lower by two orders of magnitude than the measurement in comet 9P/Tempel 1 and the best theoretical prediction. This secures rotational stability of the comet's nucleus during the next few returns, although we anticipate a catastrophic disruption from spin-up as its ultimate fate.
△ Less
Submitted 31 March, 2011;
originally announced April 2011.
