arXiv subjects

Mark Walker

Publications and source records attributed to Mark Walker.

At least 19 recordsLinked to original sources

arXiv2025

Improved cystic hygroma detection from prenatal imaging using ultrasound-specific self-supervised representation learning

Cystic hygroma is a high-risk prenatal ultrasound finding that portends high rates of chromosomal abnormalities, structural malformations, and adverse pregnancy outcomes. Automated detection can increase reproducibility and support scalable early screening programs, but supervised deep learning methods are limited by small labelled datasets. This study assesses whether ultrasound-specific self-supervised pretraining can facilitate accurate, robust deep learning detection of cystic hygroma in first-trimester ultrasound images. We fine-tuned the Ultrasound Self-Supervised Foundation Model with Masked Autoencoding (USF-MAE), pretrained on over 370,000 unlabelled ultrasound images, for binary classification of normal controls and cystic hygroma cases used in this study. Performance was evaluated on the same curated ultrasound dataset, preprocessing pipeline, and 4-fold cross-validation protocol as for the DenseNet-169 baseline, using accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (ROC-AUC). Model interpretability was analyzed qualitatively using Score-CAM visualizations. USF-MAE outperformed the DenseNet-169 baseline on all evaluation metrics. The proposed model yielded a mean accuracy of 0.96, sensitivity of 0.94, specificity of 0.98, and ROC-AUC of 0.98 compared to 0.93, 0.92, 0.94, and 0.94 for the DenseNet-169 baseline, respectively. Qualitative Score-CAM visualizations of model predictions demonstrated clinical relevance by highlighting expected regions in the fetal neck for both positive and negative cases. Paired statistical analysis using a Wilcoxon signed-rank test confirmed that performance improvements achieved by USF-MAE were statistically significant (p = 0.0057).

cs.CV↗

arXiv2025

Deep Learning Analysis of Prenatal Ultrasound for Identification of Ventriculomegaly

The proposed study aimed to develop a deep learning model capable of detecting ventriculomegaly on prenatal ultrasound images. Ventriculomegaly is a prenatal condition characterized by dilated cerebral ventricles of the fetal brain and is important to diagnose early, as it can be associated with an increased risk for fetal aneuploidies and/or underlying genetic syndromes. An Ultrasound Self-Supervised Foundation Model with Masked Autoencoding (USF-MAE), recently developed by our group, was fine-tuned for a binary classification task to distinguish fetal brain ultrasound images as either normal or showing ventriculomegaly. The USF-MAE incorporates a Vision Transformer encoder pretrained on more than 370,000 ultrasound images from the OpenUS-46 corpus. For this study, the pretrained encoder was adapted and fine-tuned on a curated dataset of fetal brain ultrasound images to optimize its performance for ventriculomegaly detection. Model evaluation was conducted using 5-fold cross-validation and an independent test cohort, and performance was quantified using accuracy, precision, recall, specificity, F1-score, and area under the receiver operating characteristic curve (AUC). The proposed USF-MAE model reached an F1-score of 91.76% on the 5-fold cross-validation and 91.78% on the independent test set, with much higher scores than those obtained by the baseline models by 19.37% and 16.15% compared to VGG-19, 2.31% and 2.56% compared to ResNet-50, and 5.03% and 11.93% compared to ViT-B/16, respectively. The model also showed a high mean test precision of 94.47% and an accuracy of 97.24%. The Eigen-CAM (Eigen Class Activation Map) heatmaps showed that the model was focusing on the ventricle area for the diagnosis of ventriculomegaly, which has explainability and clinical plausibility.

eess.IV↗

arXiv2025

USF-MAE: Ultrasound Self-Supervised Foundation Model with Masked Autoencoding

Ultrasound imaging is one of the most widely used diagnostic modalities, offering real-time, radiation-free assessment across diverse clinical domains. However, interpretation of ultrasound images remains challenging due to high noise levels, operator dependence, and limited field of view, resulting in substantial inter-observer variability. Current Deep Learning approaches are hindered by the scarcity of large labeled datasets and the domain gap between general and sonographic images, which limits the transferability of models pretrained on non-medical data. To address these challenges, we introduce the Ultrasound Self-Supervised Foundation Model with Masked Autoencoding (USF-MAE), the first large-scale self-supervised MAE framework pretrained exclusively on ultrasound data. The model was pre-trained on 370,000 2D and 3D ultrasound images curated from 46 open-source datasets, collectively termed OpenUS-46, spanning over twenty anatomical regions. This curated dataset has been made publicly available to facilitate further research and reproducibility. Using a Vision Transformer encoder-decoder architecture, USF-MAE reconstructs masked image patches, enabling it to learn rich, modality-specific representations directly from unlabeled data. The pretrained encoder was fine-tuned on three public downstream classification benchmarks: BUS-BRA (breast cancer), MMOTU-2D (ovarian tumors), and GIST514-DB (gastrointestinal stromal tumors). Across all tasks, USF-MAE consistently outperformed conventional CNN and ViT baselines, achieving F1-scores of 81.6%, 79.6%, and 82.4%, respectively. Despite not using labels during pretraining, USF-MAE approached the performance of the supervised foundation model UltraSam on breast cancer classification and surpassed it on the other tasks, demonstrating strong cross-anatomical generalization.

eess.IV↗

arXiv2024

Free-floating "planets'' in the macrolensed quasar Q2237+0305

It has been claimed that the variability of field quasars resembles gravitational lensing by a large cosmological population of free-floating planets with mass of about 10 Earths. But Galactic photometric monitoring experiments, on the other hand, exclude a large population of such planetary-mass gravitational lenses. These apparently contradictory pieces of evidence can be reconciled if the objects under consideration have a mean column-density that lies between the critical column-densities for gravitational lensing in these two contexts. Dark matter in that form is known to be weakly collisional, so that a core develops in galaxy halo density profiles, and a preferred model has already been established. Here we consider what such a model implies for Q2237+0305, which is the best-studied example of a quasar that is strongly lensed by an intervening galaxy. We construct microlensing magnification maps appropriate to the four macro-images of the quasar -- all of which are seen through the bulge of the galaxy. Each of these maps exhibits a caustic network arising from the stars, plus many small, isolated caustics arising from the free-floating "planets" in the lens galaxy. The "planets" have little influence on the magnification histograms but a large effect on the statistics of the magnification gradients. We compare our predictions to the published OGLE photometry of Q2237+0305 and find that these data are consistent with the presence of the hypothetical "planets". However, the evidence is relatively weak because the OGLE dataset is not well suited to testing our predictions and requires low-pass filtering for this application. New data from a large, space-based telescope are desirable to address this issue.

astro-ph.CO↗

arXiv2023

Radio Variable and Transient Sources on Minute Timescales in the ASKAP Pilot Surveys

We present results from a radio survey for variable and transient sources on 15-min timescales, using the Australian SKA Pathfinder (ASKAP) pilot surveys. The pilot surveys consist of 505 h of observations conducted at around 1 GHz observing frequency, with a total sky coverage of 1476 deg$^2$. Each observation was tracked for approximately 8-10h, with a typical rms sensitivity of $\sim$30 $\mu$jy/beam and an angular resolution of $\sim$12 arcsec. The variability search was conducted within each 8-10h observation on a 15-min timescale. We detected 38 variable and transient sources. Seven of them are known pulsars, including an eclipsing millisecond pulsar, PSR J2039$-$5617. Another eight sources are stars, only one of which has been previously identified as a radio star. For the remaining 23 objects, 22 are associated with active galactic nuclei or galaxies (including the five intra-hour variables that have been reported previously), and their variations are caused by discrete, local plasma screens. The remaining source has no multi-wavelength counterparts and is therefore yet to be identified. This is the first large-scale radio survey for variables and transient sources on minute timescales at a sub-mJy sensitivity level. We expect to discover $\sim$1 highly variable source per day using the same technique on the full ASKAP surveys.

astro-ph.HE↗

arXiv2022

The Effect of Gravitational Lensing on Fast Transient Event Rates

Fast cosmological transients such as fast radio bursts (FRBs) and gamma-ray bursts (GRBs) represent a class of sources more compact than any other cosmological object. As such they are sensitive to significant magnification via gravitational lensing from a class of lenses which are not well-constrained by observations today. Low-mass primordial black holes are one such candidate which may constitute a significant fraction of the Universe's dark matter. Current observations only constrain their density in the nearby Universe, giving fast transients from cosmological distances the potential to form complementary constraints. Motivated by this, we calculate the effect that gravitational lensing from a cosmological distribution of compact objects would have on the observed rates of FRBs and GRBs. For static lensing geometries, we rule out the prospect that all FRBs are gravitationally lensed for a range of lens masses and show that lens masses greater than $10^{-5}M_\odot$ can be constrained with 8000 un-localised high fluence FRBs at 1.4GHz, as might be detected by the next generation of FRB-finding telescopes.

astro-ph.CO↗

arXiv2022

Cosmological nanolensing by dense gas clouds

We study the influence of a cosmological population of dense gas clouds on distant sources, with emphasis on quasar optical variability. In addition to gravitational lensing such clouds affect flux measurements via refraction in the neutral gas and via dust extinction, leading to a variety of possible light curves even in the low optical depth limit. We classify and illustrate the types of light curves that can arise. For sources as large as quasars we show that gravitational lensing and extinction are the dominant effects, with gas refraction playing only a minor role. We find that clouds with mass ~10^{-4.5+/-0.5} M_\odot can reproduce the observed distribution of quasar variation amplitudes, but only if such clouds make up a large fraction of the closure density. In that case there may also be substantial extinction of distant optical sources, which can in principle be constrained by data on "standard candles" such as type Ia supernovae. Unfortunately that extinction is essentially grey, even when the material opacity is strongly wavelength dependent, making it difficult to distinguish from the influence of the background geometry. We propose a novel statistical test of the origin of quasar variability, based on the angular structure of the variation timescale for a large number of quasars distributed all over the sky. If quasar variability is primarily due to nanolensing that angular structure is expected to include a quadrupole term of amplitude $\sim5\%$, which ought to be measurable with future data from the Gaia mission.

astro-ph.CO↗

arXiv2021

Down-drifting, bandwidth and beaming of fast radio bursts

'Downwards drifting' structures observed in fast radio bursts (FRBs) could naturally arise from a screen occurring after a small initial dispersion measure region. The screen imprints temporally sharp but broadband structure on the pulse that has already been dispersed, and the `structure-maximising' bulk of the dispersion measure is acquired further along the path to observer. If so, scaling of the drift rate of repeating FRBs with frequency suggests that the emission is beamed -- out of our line of sight -- with the circum-burst plasma deflecting the beam towards us. This in turn explains the observed limited, variable bandwidths of bursts despite broadband nature of the underlying emission. We summarise the geometric constraints on this simple model delivering much of the observed FRB spectro-temporal phenomenology, present generic predictions and discuss possible nature of modulation.

astro-ph.HE↗

arXiv2021

ASKAP observations of multiple rapid scintillators reveal a degrees-long plasma filament

We present the results from an Australian Square Kilometre Array Pathfinder search for radio variables on timescales of hours. We conducted an untargeted search over a 30 deg$^2$ field, with multiple 10-hour observations separated by days to months, at a central frequency of 945 MHz. We discovered six rapid scintillators from 15-minute model-subtracted images with sensitivity of $\sim 200\,\mu$Jy/beam; two of them are extreme intra-hour variables with modulation indices up to $\sim 40\%$ and timescales as short as tens of minutes. Five of the variables are in a linear arrangement on the sky with angular width $\sim 1$ arcmin and length $\sim 2$ degrees, revealing the existence of a huge plasma filament in front of them. We derived kinematic models of this plasma from the annual modulation of the scintillation rate of our sources, and we estimated its likely physical properties: a distance of $\sim 4$ pc and length of $\sim 0.1$ pc. The characteristics we observe for the scattering screen are incompatible with published suggestions for the origin of intra-hour variability leading us to propose a new picture in which the underlying phenomenon is a cold tidal stream. This is the first time that multiple scintillators have been detected behind the same plasma screen, giving direct insight into the geometry of the scattering medium responsible for enhanced scintillation.

astro-ph.GA↗

arXiv2019

Cosmic snow clouds: self-gravitating gas spheres manifesting hydrogen condensation

We present hydrostatic equilibrium models of spherical, self-gravitating clouds of helium and molecular hydrogen, focusing on the cold, high-density regime where solid- or liquid-hydrogen can form. The resulting structures have masses from 0.1 Msun down to several x 1.e-8 Msun, and span a broad range of radii: 1.e-4 < R(AU) < 1.e7. Our models are fully convective, but all have a two-zone character with the majority of the mass in a small, condensate-free core, surrounded by a colder envelope where phase equilibrium obtains. Convection in the envelope is unusual in that it is driven by a mean-molecular-weight inversion, rather than by an entropy gradient. In fact the entropy gradient is itself inverted, leading to the surprising result that envelope convection transports heat inwards. In turn that permits the outer layers to maintain steady state temperatures below the cosmic microwave background. Amongst our hydrostatic equilibria we identify thermal equilibria appropriate to the Galaxy, in which radiative cooling from H2 is balanced by cosmic-ray heating. These equilibria are all thermally unstable, albeit with very long thermal timescales in some cases. The specific luminosities of all our models are very low, and they therefore describe a type of baryonic dark matter. Consequently such clouds are thermally fragile: when placed in a harsh radiation field they will be unable to cool effectively and disruption will ensue as heat input drives a secular expansion. Disrupting clouds should leave trails of gas and H2 dust in their wake, which might make them easier to detect. Our models may be relevant to the cometary globules in the Helix Nebula, and the G2 cloud orbiting Sgr A*.

astro-ph.GA↗

arXiv2018

The character of the warm ionised medium

$\textit{Herschel}$ observations of far infrared N$^+$ emission lines have demonstrated that dense plasma, with $n_e\sim30\,{\rm cm^{-3}}$, is ubiquitous in the inner Galactic plane. By combining the information from $\textit{Herschel}$ with other tracers of ionised gas, we build a picture of this dense plasma. We adopt a collisional ionisation model, so the analysis is not tied to a specific energisation mechanism. We find that the dense plasma is concentrated in a disk that is $130\;{\rm pc}$ thick, and makes a significant contribution to radio pulsar dispersion measures in the inner Galactic plane. The strength of the far infrared N$^+$ emission requires high temperatures in the plasma, with $T \simeq 19{,}000\,{\rm K}$ indicated both by the ratio of N$^+$ to C$^+$, and by the ratio of N$^+$ to microwave bremsstrahlung in the inner Galactic plane. This parallels the situation at high Galactic latitudes, where strong optical emission is observed from N$^+$ (and S$^+$), relative to both H$\alpha$ and microwave bremsstrahlung, and suggests a common origin. If so, the same gas provides a natural explanation for the extreme radio-wave scattering phenomena that are sometimes observed in pulsars and quasars. We therefore propose a new picture of the warm ionised medium as seen in emission, in which the plasma is dense, hot, and localised in numerous structures of size $\sim 10^2\,{\rm AU}$ that are clustered around stars.

astro-ph.GA↗

arXiv2017

Extreme radio-wave scattering associated with hot stars

We use data on extreme radio scintillation to demonstrate that this phenomenon is associated with hot stars in the solar neighbourhood. The ionized gas responsible for the scattering is found at distances up to 1.75pc from the host star, and on average must comprise 1.E5 distinct structures per star. We detect azimuthal velocities of the plasma, relative to the host star, up to 9.7 km/s, consistent with warm gas expanding at the sound speed. The circumstellar plasma structures that we infer are similar in several respects to the cometary knots seen in the Helix, and in other planetary nebulae. There the ionized gas appears as a skin around tiny molecular clumps. Our analysis suggests that molecular clumps are ubiquitous circumstellar features, unrelated to the evolutionary state of the star. The total mass in such clumps is comparable to the stellar mass.

astro-ph.GA↗

arXiv2015

The infrared dielectric function of solid para-hydrogen

We report laboratory measurements of the absorption coefficient of solid para-H2, within the wavelength range from 1 to 16.7 micron, at high spectral resolution. In addition to the narrow rovibrational lines of H2 which are familiar from gas phase spectroscopy, the data manifest double transitions and broad phonon branches that are characteristic specifically of hydrogen in the solid phase. These transitions are of interest because they provide a spectral signature which is independent of the impurity content of the matrix. We have used our data, in combination with a model of the ultraviolet absorptions of the H2 molecule, to construct the dielectric function of solid para-H2 over a broad range of frequencies. Our results will be useful in determining the electromagnetic response of small particles of solid hydrogen. The dielectric function makes it clear that pure H2 dust would contribute to IR extinction predominantly by scattering starlight, rather than absorbing it, and the characteristic IR absorption spectrum of the hydrogen matrix itself will be difficult to observe.

astro-ph.GA↗

arXiv2012

VAST: An ASKAP Survey for Variables and Slow Transients

The Australian Square Kilometre Array Pathfinder (ASKAP) will give us an unprecedented opportunity to investigate the transient sky at radio wavelengths. In this paper we present VAST, an ASKAP survey for Variables and Slow Transients. VAST will exploit the wide-field survey capabilities of ASKAP to enable the discovery and investigation of variable and transient phenomena from the local to the cosmological, including flare stars, intermittent pulsars, X-ray binaries, magnetars, extreme scattering events, interstellar scintillation, radio supernovae and orphan afterglows of gamma ray bursts. In addition, it will allow us to probe unexplored regions of parameter space where new classes of transient sources may be detected. In this paper we review the known radio transient and variable populations and the current results from blind radio surveys. We outline a comprehensive program based on a multi-tiered survey strategy to characterise the radio transient sky through detection and monitoring of transient and variable sources on the ASKAP imaging timescales of five seconds and greater. We also present an analysis of the expected source populations that we will be able to detect with VAST.

astro-ph.IM↗

arXiv2011

Goonhilly Sparklers

Flux monitoring of compact radio quasars has revealed dramatic radio-wave lensing events which challenge our understanding of the interstellar medium. However, the data on these events remain very sparse. Here we consider how the Goonhilly radio astronomical facility can make an impact on this problem by dedicating one or more dishes to flux monitoring for a period of one year. Such an experiment would be able to identify \sim6 new events and study them in detail.

astro-ph.HE↗

arXiv2008

Extremely Anisotropic Scintillations

A small number of quasars exhibit interstellar scintillation on time-scales less than an hour; their scintillation patterns are all known to be anisotropic. Here we consider a totally anisotropic model in which the scintillation pattern is effectively one-dimensional. For the persistent rapid scintillators J1819+3845 and PKS1257-326 we show that this model offers a good description of the two-station time-delay measurements and the annual cycle in the scintillation time-scale. Generalising the model to finite anisotropy yields a better match to the data but the improvement is not significant and the two additional parameters which are required to describe this model are not justified by the existing data. The extreme anisotropy we infer for the scintillation patterns must be attributed to the scattering medium rather than a highly elongated source. For J1819+3845 the totally anisotropic model predicts that the particular radio flux variations seen between mid July and late August should repeat between late August and mid November, and then again between mid November and late December as the Earth twice changes its direction of motion across the scintillation pattern. If this effect can be observed then the minor-axis velocity component of the screen and the orientation of that axis can both be precisely determined. In reality the axis ratio is finite, albeit large, and spatial decorrelation of the flux pattern along the major axis may be observable via differences in the pairwise fluxes within this overlap region; in this case we can also constrain both the major-axis velocity component of the screen and the magnitude of the anisotropy.

astro-ph↗

arXiv2008

Interstellar Holography

The dynamic spectrum of a radio pulsar is an in-line digital hologram of the ionised interstellar medium. It has previously been demonstrated that such holograms permit image reconstruction, in the sense that one can determine an approximation to the complex electric field values as a function of Doppler-shift and delay, but to date the quality of the reconstructions has been poor. Here we report a substantial improvement in the method which we have achieved by simultaneous optimisation of the thousands of coefficients that describe the electric field. For our test spectrum of PSR B0834+06 we find that the model provides an accurate representation of the data over the full 63 dB dynamic range of the observations: residual differences between model and data are noise-like. The advent of interstellar holography enables detailed quantitative investigation of the interstellar radio-wave propagation paths for a given pulsar at each epoch of observation; we illustrate this using our test data which show the scattering material to be structured and highly anisotropic. The temporal response of the medium exhibits a scattering tail out to beyond 100 microsec and a pulse arrival time measurement at this frequency and this epoch of observation would be affected by a mean delay of 15 microsec due to multipath propagation in the interstellar medium.

astro-ph↗

arXiv2005

Electric field representation of pulsar intensity spectra

Pulsar dynamic spectra exhibit high visibility fringes arising from interference between scattered radio waves. These fringes may be random or highly ordered patterns, depending on the nature of the scattering or refraction. Here we consider the possibility of decomposing pulsar dynamic spectra -- which are intensity measurements -- into their constituent scattered waves, i.e. electric field components. We describe an iterative method of achieving this decomposition and show how the algorithm performs on data from the pulsar B0834+06. The match between model and observations is good, although not formally acceptable as a representation of the data. Scattered wave components derived in this way are immediately useful for qualitative insights into the scattering geometry. With some further development this approach can be put to a variety of uses, including: imaging the scattering and refracting structures in the interstellar medium; interstellar interferometric imaging of pulsars at very high angular resolution; and mitigating pulse arrival time fluctuations due to interstellar scattering.

astro-ph↗