Search

Developing an interpretable system for generating reports in chest X-ray (CXR) analysis is becoming increasingly crucial in Computer-aided Diagnosis (CAD) systems, enabling radiologists to comprehend the decisions made by these systems. Despite the growth of diverse datasets and methods focusing on report generation, there remains a notable gap in how closely these models' generated reports align with the interpretations of real radiologists. In this study, we tackle this challenge by initially introducing Fine-Grained CXR (FG-CXR) dataset, which provides fine-grained paired information between the captions generated by radiologists and the corresponding gaze attention heatmaps for each anatomy. Unlike existing datasets that include a raw sequence of gaze alongside a report, with significant misalignment between gaze location and report content, our FG-CXR dataset offers a more grained alignment between gaze attention and diagnosis transcript. Furthermore, our analysis reveals that simply applying black-box image captioning methods to generate reports cannot adequately explain which information in CXR is utilized and how long needs to attend to accurately generate reports. Consequently, we propose a novel explainable radiologist's attention generator network (Gen-XAI) that mimics the diagnosis process of radiologists, explicitly constraining its output to closely align with both radiologist's gaze attention and transcript. Finally, we perform extensive experiments to illustrate the effectiveness of our method. Our datasets and checkpoint is available at https://github.com/UARK-AICV/FG-CXR., Comment: ACCV 2024

Medical eye-tracking data is an important information source for understanding how radiologists visually interpret medical images. This information not only improves the accuracy of deep learning models for X-ray analysis but also their interpretability, enhancing transparency in decision-making. However, the current eye-tracking data is dispersed, unprocessed, and ambiguous, making it difficult to derive meaningful insights. Therefore, there is a need to create a new dataset with more focus and purposeful eyetracking data, improving its utility for diagnostic applications. In this work, we propose a refinement method inspired by the target-present visual search challenge: there is a specific finding and fixations are guided to locate it. After refining the existing eye-tracking datasets, we transform them into a curated visual search dataset, called GazeSearch, specifically for radiology findings, where each fixation sequence is purposefully aligned to the task of locating a particular finding. Subsequently, we introduce a scan path prediction baseline, called ChestSearch, specifically tailored to GazeSearch. Finally, we employ the newly introduced GazeSearch as a benchmark to evaluate the performance of current state-of-the-art methods, offering a comprehensive assessment for visual search in the medical imaging domain. Code is available at \url{https://github.com/UARK-AICV/GazeSearch}., Comment: Aceepted WACV 2025

Semi-supervised 3D object detection is a common strategy employed to circumvent the challenge of manually labeling large-scale autonomous driving perception datasets. Pseudo-labeling approaches to semi-supervised learning adopt a teacher-student framework in which machine-generated pseudo-labels on a large unlabeled dataset are used in combination with a small manually-labeled dataset for training. In this work, we address the problem of improving pseudo-label quality through leveraging long-term temporal information captured in driving scenes. More specifically, we leverage pre-trained motion-forecasting models to generate object trajectories on pseudo-labeled data to further enhance the student model training. Our approach improves pseudo-label quality in two distinct manners: first, we suppress false positive pseudo-labels through establishing consistency across multiple frames of motion forecasting outputs. Second, we compensate for false negative detections by directly inserting predicted object tracks into the pseudo-labeled scene. Experiments on the nuScenes dataset demonstrate the effectiveness of our approach, improving the performance of standard semi-supervised approaches in a variety of settings.

Predicting human gaze behavior within computer vision is integral for developing interactive systems that can anticipate user attention, address fundamental questions in cognitive science, and hold implications for fields like human-computer interaction (HCI) and augmented/virtual reality (AR/VR) systems. Despite methodologies introduced for modeling human eye gaze behavior, applying these models to medical imaging for scanpath prediction remains unexplored. Our proposed system aims to predict eye gaze sequences from radiology reports and CXR images, potentially streamlining data collection and enhancing AI systems using larger datasets. However, predicting human scanpaths on medical images presents unique challenges due to the diverse nature of abnormal regions. Our model predicts fixation coordinates and durations critical for medical scanpath prediction, outperforming existing models in the computer vision community. Utilizing a two-stage training process and large publicly available datasets, our approach generates static heatmaps and eye gaze videos aligned with radiology reports, facilitating comprehensive analysis. We validate our approach by comparing its performance with state-of-the-art methods and assessing its generalizability among different radiologists, introducing novel strategies to model radiologists' search patterns during CXR image diagnosis. Based on the radiologist's evaluation, MedGaze can generate human-like gaze sequences with a high focus on relevant regions over the CXR images. It sometimes also outperforms humans in terms of redundancy and randomness in the scanpaths., Comment: Submitted to the Journal

Human-AI collaboration to identify and correct perceptual errors in chest radiographs has not been previously explored. This study aimed to develop a collaborative AI system, CoRaX, which integrates eye gaze data and radiology reports to enhance diagnostic accuracy in chest radiology by pinpointing perceptual errors and refining the decision-making process. Using public datasets REFLACX and EGD-CXR, the study retrospectively developed CoRaX, employing a large multimodal model to analyze image embeddings, eye gaze data, and radiology reports. The system's effectiveness was evaluated based on its referral-making process, the quality of referrals, and performance in collaborative diagnostic settings. CoRaX was tested on a simulated error dataset of 271 samples with 28% (93 of 332) missed abnormalities. The system corrected 21% (71 of 332) of these errors, leaving 7% (22 of 312) unresolved. The Referral-Usefulness score, indicating the accuracy of predicted regions for all true referrals, was 0.63 (95% CI 0.59, 0.68). The Total-Usefulness score, reflecting the diagnostic accuracy of CoRaX's interactions with radiologists, showed that 84% (237 of 280) of these interactions had a score above 0.40. In conclusion, CoRaX efficiently collaborates with radiologists to address perceptual errors across various abnormalities, with potential applications in the education and training of novice radiologists., Comment: Under Review in Journal

Binary mass transfer can occur at high rates due to rapid expansion of the donor's envelope. In the case where mass transfer is unstable, the binary can rapidly shrink its orbit and lead to a merger. In this work we consider the appearance of the system preceding merger, specifically for the case of a low-mass ($\approx 2.5$-$3~M_\odot$) helium star with a neutron star (NS) companion. Modeling the mass transfer history as well as the wind launched by super-Eddington accretion onto the NS, we find that such systems can power slowly rising transients with timescales as long as years, and luminosities of $\sim 10^{40}$-$10^{41}$ erg s$^{-1}$ from optical to UV. The final explosion following the merger (or core-collapse of the helium star in some cases) leads to an interaction-powered transient with properties resembling Type Ibn supernovae (SNe), possibly with a bright early peak powered by shock cooling emission for merger-powered explosions. We apply our model to the Type Ibn SN 2023fyq, that displayed a long-term precursor activity from years before the terminal explosion., Comment: Accepted to the Open Journal of Astrophysics. A typographic error in equation 34 has been corrected, and correspondingly Figure 5 has been slightly updated (other results are unaffected). Light curve source code available here: https://github.com/DTsuna/merger_precursor.git

The triangular-lattice system $\alpha$-RuI$_3$ is isostructural to the widely-studied $\alpha$-RuCl$_3$ compound which was identified as a potential Kitaev system but exhibits, instead of spin liquid behaviour, a magnetically ordered zig-zag ground state which sets in below 14~K. Here we show experimentally that, in contrast, the spins in $\alpha$-RuI$_3$ remain dynamic down to at least 50~mK. We study the spin dynamics using muon-spin relaxation methods and determine the presence of low-frequency fluctuations which are characteristic of a two-dimensional system., Comment: 5 pages, 5 figures

We present the first limit on $g_{A\gamma}$ coupling constant using the Bragg-Primakoff conversion based on an exposure of 1107.5 kg days of data from the CDEX-1B experiment at the China Jinping Underground Laboratory. The data are consistent with the null signal hypothesis, and no excess signals are observed. Limits of the coupling $g_{A\gamma}<2.08\times10^{-9}$ GeV$^{-1}$ (95\% C.L.) are derived for axions with mass up to 100 eV/$c^2$. Within the hadronic model of KSVZ, our results exclude axion mass $>5.3~\rm{eV}/c^2$ at 95\% C.L., Comment: 7 pages, 5 figures

We present an approach to designing 3D Iterated Function Systems (IFS) within the Unity Editor and rendered to VR in real-time. Objects are modeled as a hierarchical tree of primitive shapes and operators, editable using a graphical user interface allowing artists to develop psychedelic scenes with little to no coding knowledge, and is easily extensible for more advanced users to add their own primitive shapes and operators., Comment: 4 pages, 5 figures. Submitted to IEEE VIS 2024

We present the first results of the search for sub-MeV fermionic dark matter absorbed by electron targets of Germanium using the 205.4~kg$\cdot$day data collected by the CDEX-10 experiment, with the analysis threshold of 160~eVee. No significant dark matter (DM) signals over the background are observed. Results are presented as limits on the cross section of DM--electron interaction. We present new constraints of cross section in the DM range of 0.1--10 keV/$c^2$ for vector and axial-vector interaction. The upper limit on the cross section is set to be $\rm 5.5\times10^{-46}~cm^2$ for vector interaction, and $\rm 1.8\times10^{-46}~cm^2$ for axial-vector interaction at DM mass of 5 keV/$c^2$., Comment: 6 pages, 4 figures

We report new constraints on light dark matter (DM) boosted by blazars using the 205.4 kg day data from the CDEX-10 experiment located at the China Jinping Underground Laboratory. Two representative blazars, TXS 0506+56 and BL Lacertae are studied. The results derived from TXS 0506+56 exclude DM-nucleon elastic scattering cross sections from $4.6\times 10^{-33}\ \rm cm^2$ to $1\times10^{-26}\ \rm cm^2$ for DM masses between 10 keV and 1 GeV, and the results derived from BL Lacertae exclude DM-nucleon elastic scattering cross sections from $2.4\times 10^{-34}\ \rm cm^2$ to $1\times10^{-26}\ \rm cm^2$ for the same range of DM masses. The constraints correspond to the best sensitivities among solid-state detector experiments in the sub-MeV mass range., Comment: 7 pages, 4 figures

Dark matter (DM) is a major constituent of the Universe. However, no definite evidence of DM particles (denoted as ``$\chi$") has been found in DM direct detection (DD) experiments to date. There is a novel concept of detecting $\chi$ from evaporating primordial black holes (PBHs). We search for $\chi$ emitted from PBHs by investigating their interaction with target electrons. The examined PBH masses range from 1$\times$10$^{15}$ to 7$\times$10$^{16}$ g under the current limits of PBH abundance $f_{PBH}$. Using 205.4 kg$\cdot$day data obtained from the CDEX-10 experiment conducted in the China Jinping Underground Laboratory, we exclude the $\chi$--electron ($\chi$--$e$) elastic-scattering cross section $\sigma_{\chi e} \sim 5\times10^{-29}$ cm$^2$ for $\chi$ with a mass $m_{\chi}\lesssim$ 0.1 keV from our results. With the higher radiation background but lower energy threshold (160 eV), CDEX-10 fill a part of the gap in the previous work. If ($m_{\chi}$, $\sigma_{\chi e}$) can be determined in the future, DD experiments are expected to impose strong constraints on $f_{PBH}$ for large $M_{PBH}$s., Comment: 9 pages, 6 figures, 3 tables. Version updated to match SCPMA version

Gauge fields are ubiquitous in modern quantum physics. In superfluids, quantized vortices can be induced by gauge fields. Here we demonstrate the first experimental observation of vortex nucleations in spinor Bose-Einstein Condensates under radially-localized synthetic magnetic fields. The associated gauge potentials $\vec{A}$ are azimuthal and created by light-induced spin-orbital-angular-momentum coupling, generating circulating azimuthal velocity fields $\propto \vec{p}-\vec{A}$ even when the canonical momentum $\vec{p}= 0$. A sufficiently large azimuthal velocity peaked near the condensate center results in a dynamically unstable localized excitation that initiates vortex nucleations. This excitation appears as a spontaneously-formed vortex-antivortex pair near the cloud center. Following the initially developed instability, the dynamics is governed by the asymmetry and dissipation, where the atomic orbital angular momentum evolves and can reach the value of the ground state. Our system exhibits dynamical and Landau instabilities and agrees reasonably with time-dependent Gross-Pitaevskii simulations.

U-based binary alloys have been widely adopted in fast nuclear reactors, but their stability under extreme conditions of high-pressure is almost unknown, mounting up to latent risk in applications. Here, possible ordered phases in U-Zr system up to 200 GPa are comprehensively investigated by unbiased first-principles structure prediction. Stable U2Zr, metastable U3Zr and U4Zr phases are discovered for the first time, which exhibit strong stability under compression. They all are metallic, with 5f electrons of uranium dominating the electronic density of states near the Fermi level. Prominent ionic interactions between U and Zr atoms, as well as covalent interactions between adjacent uranium atoms, are found. The same strategy is applied to explore the stability of ordered phases in other U-based binary transition metal alloys, U-X (X= Sc, Ti, V, Cr, Y, Nb, Mo, Hf, Ta, W). Stable and metastable ordered phases similar to U-Zr alloy are unveiled, all with similar electronic structures. For these alloys, we find that the structure of U2X (X=Zr, Ti, Hf) hosts a unique hybrid phase transition similar to U2Nb, which is a superposition of a first-order transition and a second-order transition. The prediction of these novel phases not only refutes the stability of the long-believed ordered phase I4/mmm-U2Mo, but also rewrites the phase diagrams of U-X (X= Zr, Sc, Ti, V, Cr, Nb, Mo, Hf, Ta) alloys under high pressure. All of these findings promote our understanding of the high-pressure behavior of the broad category of U-based binary alloys with transition metals., Comment: 27 pages, 12 figures, with supplementary material

Ultrafast electric-field control of emergent electronic and magnetic states at oxide interfaces offers exciting prospects for the development of new generations of energy-efficient devices. Here, we demonstrate that the electronic structure and emergent ferromagnetic interfacial state in epitaxial LaNiO3/CaMnO3 superlattices can be effectively controlled using intense single-cycle THz electric-field pulses. We employ a combination of polarization-dependent X-ray absorption spectroscopy with magnetic circular dichroism and X-ray resonant magnetic reflectivity to measure a detailed magneto-optical profile and thickness of the ferromagnetic interfacial layer. Then, we use time-resolved and temperature-dependent magneto-optical Kerr effect, along with transient optical reflectivity and transmissivity measurements, to disentangle multiple correlated electronic and magnetic processes driven by ultrafast high-field (~1 MV/cm) THz pulses. These processes include an initial sub-picosecond electronic response, consistent with non-equilibrium Joule heating; a rapid (~270 fs) demagnetization of the ferromagnetic interfacial layer, driven by THz-field-induced nonequilibrium spin-polarized currents; and subsequent multi-picosecond dynamics, possibly indicative of a change in the magnetic state of the superlattice due to the transfer of spin angular momentum to the lattice. Our findings shed light on the intricate interplay of electronic and magnetic phenomena in this strongly correlated material system, suggesting a promising avenue for efficient control of two-dimensional ferromagnetic states at oxide interfaces using ultrafast electric-field pulses.

We predict novel topological phases with broken time-reversal symmetry supporting the coexistence of opposite chiral edge states, which are fundamentally different from the photonic spin-Hall, valley-Hall, and higher-order topological phases. We find a fine-grained categorization of Chern insulators, their band topologies characterized by identical Chern number are completely different. Furthermore, we prove that different topologies cause zeros in their Bloch wavefunction overlaps, which imprint the band gap closing and appear at the degenerate points of topological phase transition. The Bloch wavefunction overlaps predict the reflection and refraction at a topological time boundary, and the overlap zeros ensure the existence of vanishing revival amplitude at critical times even though different topologies before and after the time boundary have identical Chern number. Our findings create new opportunities for topological metamaterials, uncover the topological feature hidden in the time boundary effect as a probe of topology, and open a venue for the exploration of the rich physics originating from the long-range couplings., Comment: 8 pages, 5 figures

A fraction of core-collapse supernovae (SNe) with signs of interaction with a dense circumstellar matter are preceded by bright precursor emission. While the precursors are likely caused by a mass ejection before core-collapse, their mechanism to power energetic bursts, sometimes reaching $10^{48}$--$10^{49}\ {\rm erg}$ that are larger than the binding energies of red supergiant envelopes, is still under debate. Remarkably, such a huge energy-deposition should result in an almost complete envelope ejection and hence a strong sign of interaction, but the observed SNe with precursors show in fact typical properties among the interacting SNe. More generally, the observed luminosity of $10^{40-42}\,\rm erg\,s^{-1}$ is shown to be challenging for a single SN progenitor. To resolve these tensions, we propose a scenario where the progenitor is in a binary system with a compact object (CO), and an outburst from the star leads to a super-Eddington accretion onto the CO. We show that for sufficiently short separations, outbursts with moderate initial kinetic energies of $10^{46}$--$10^{47}$ erg can be energized by the accreting CO so that their radiative output can be consistent with the observed precursors. We discuss the implications of our model in relation to CO binaries detectable with Gaia and gravitational wave detectors., Comment: Accepted by ApJ; 18 pages, 5 Figures (+ appendix). Light curve source code available here: https://github.com/DTsuna/binary_precursor