Searcharxiv⌕ Search

arXiv subjects

Hui Meng

Publications and source records attributed to Hui Meng.

9 recordsLinked to original sources
arXiv2026

FUGC: Benchmarking Semi-Supervised Learning Methods for Cervical Segmentation

Accurate segmentation of cervical structures in transvaginal ultrasound (TVS) is critical for assessing the risk of spontaneous preterm birth (PTB), yet the scarcity of labeled data limits the performance of supervised learning approaches. This paper introduces the Fetal Ultrasound Grand Challenge (FUGC), the first benchmark for semi-supervised learning in cervical segmentation, hosted at ISBI 2025. FUGC provides a dataset of 890 TVS images, including 500 training images, 90 validation images, and 300 test images. Methods were evaluated using the Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), and runtime (RT), with a weighted combination of 0.4/0.4/0.2. The challenge attracted 10 teams with 82 participants submitting innovative solutions. The best-performing methods for each individual metric achieved 90.26\% mDSC, 38.88 mHD, and 32.85 ms RT, respectively. FUGC establishes a standardized benchmark for cervical segmentation, demonstrates the efficacy of semi-supervised methods with limited labeled data, and provides a foundation for AI-assisted clinical PTB risk assessment.

eess.IV↗
arXiv2025

Deep Disentangled Representation Network for Treatment Effect Estimation

Estimating individual-level treatment effect from observational data is a fundamental problem in causal inference and has attracted increasing attention in the fields of education, healthcare, and public policy.In this work, we concentrate on the study of disentangled representation methods that have shown promising outcomes by decomposing observed covariates into instrumental, confounding, and adjustment factors. However, most of the previous work has primarily revolved around generative models or hard decomposition methods for covariates, which often struggle to guarantee the attainment of precisely disentangled factors. In order to effectively model different causal relationships, we propose a novel treatment effect estimation algorithm that incorporates a mixture of experts with multi-head attention and a linear orthogonal regularizer to softly decompose the pre-treatment variables, and simultaneously eliminates selection bias via importance sampling re-weighting techniques. We conduct extensive experiments on both public semi-synthetic and real-world production datasets. The experimental results clearly demonstrate that our algorithm outperforms the state-of-the-art methods focused on individual treatment effects.

cs.LG↗
arXiv2024

GuidedNet: Semi-Supervised Multi-Organ Segmentation via Labeled Data Guide Unlabeled Data

Semi-supervised multi-organ medical image segmentation aids physicians in improving disease diagnosis and treatment planning and reduces the time and effort required for organ annotation.Existing state-of-the-art methods train the labeled data with ground truths and train the unlabeled data with pseudo-labels. However, the two training flows are separate, which does not reflect the interrelationship between labeled and unlabeled data.To address this issue, we propose a semi-supervised multi-organ segmentation method called GuidedNet, which leverages the knowledge from labeled data to guide the training of unlabeled data. The primary goals of this study are to improve the quality of pseudo-labels for unlabeled data and to enhance the network's learning capability for both small and complex organs.A key concept is that voxel features from labeled and unlabeled data that are close to each other in the feature space are more likely to belong to the same class.On this basis, a 3D Consistent Gaussian Mixture Model (3D-CGMM) is designed to leverage the feature distributions from labeled data to rectify the generated pseudo-labels.Furthermore, we introduce a Knowledge Transfer Cross Pseudo Supervision (KT-CPS) strategy, which leverages the prior knowledge obtained from the labeled data to guide the training of the unlabeled data, thereby improving the segmentation accuracy for both small and complex organs. Extensive experiments on two public datasets, FLARE22 and AMOS, demonstrated that GuidedNet is capable of achieving state-of-the-art performance. The source code with our proposed model are available at https://github.com/kimjisoo12/GuidedNet.

cs.CV↗
arXiv2021

Cerebral Aneurysm Flow Diverter Modeled as a Thin Inhomogeneous Porous Medium in Hemodynamic Simulations

Rapid and accurate simulation of cerebral aneurysm flow modifications by flow diverters (FDs) can help improving patient-specific intervention and predicting treatment outcome. However, with explicit FD devices being placed in patient-specific aneurysm model, the computational domain must be resolved around the thin stent wires, leading to high computational cost in computational fluid dynamics (CFD). Classic homogeneous porous medium (PM) methods cannot accurately predict the post-stenting aneurysmal flow field due to the inhomogeneous FD wire distributions on anatomic arteries. We propose a novel approach that models the FD flow modification as a thin inhomogeneous porous medium (iPM). It improves over classic PM approaches in that, first, FD is treated as a screen, which is more accurate than the classic Darcy-Forchheimer relation based on 3D PM. second, the pressure drop is calculated using local FD geometric parameters across an inhomogeneous PM, which is more realistic. To test its accuracy and speed, we applied the iPM technique to simulate the post stenting flow field in three patient-specific aneurysms and compared the results against CFD simulations with explicit FD devices. The iPM CFD ran 500% faster than the explicit CFD while achieving 94%-99% accuracy. Thus iPM is a promising clinical bedside modeling tool to assist endovascular interventions with FD and stents.

physics.flu-dyn↗
arXiv2021

Hydrodynamic interactions and extreme particle clustering in turbulence

From new detailed experimental data, we found that the Radial Distribution Function (RDF) of inertial particles in turbulence grows explosively with $r^{-6}$ scaling as the collision radius is approached. We corrected a theory by Yavuz et al. (Phys. Rev. Lett. 120, 244504 (2018)) based on hydrodynamic interactions between pairs of weakly inertial particles, and demonstrate that even this corrected theory cannot explain the observed RDF behavior. We explore several alternative mechanisms for the discrepancy that were not included in the theory and show that none of them are likely the explanation, suggesting new, yet to be identified physical mechanisms are at play.

physics.flu-dyn↗
arXiv2021

Particle Radial Distribution Function and Relative Velocity Measurement in Turbulence at Small Particle-Pair Separations

The collision rate of particles suspended in turbulent flow is critical to particle agglomeration and droplet coalescence. The collision kernel can be evaluated by the radial distribution function (RDF) and radial relative velocity (RV) between particles at small separations $r$. Previously, the smallest $r$ was limited to roughly the Kolmogorov length $\eta$ due to particle position uncertainty and image overlap. We report a new approach to measure RDF and RV near contact ($r/a\: \approx$ 2.07, $a$ particle radius) overcoming these limitations. Three-dimensional particle tracking velocimetry using four-pulse Shake-the-Box algorithm recorded short particle tracks with the interpolated midpoints registered as particle positions to avoid image overlap. This strategy further allows removal of mismatched tracks using their characteristic false RV. We measured RDF and RV in a one-meter-diameter isotropic turbulence chamber with Taylor Reynolds number $Re_\lambda=324$ with particles of 12-16 $\mu$m radius and Stokes number $\approx$ 0.7. While at large $r$ the measured RV agrees with the literature, when $r<20\eta$ the first moment of negative RV is 10 times higher than direct numerical simulations of non-interacting particles. Likewise, when $r>\eta$, RDF scales as $r^{-0.39}$ reflecting RDF scaling for polydisperse particles in the literature , but when $r\lessapprox\eta$ RDF scales as $r^{-6}$, yielding 1000 times higher near-contact RDF than simulations. Such extreme clustering and relative velocity enhancement can be attributed to particle-particle interactions. Uncertainty analysis substantiates the observed trends. This first-ever simultaneous RDF and RV measurement at small separations provides a clear glimpse into the clustering and relative velocities of particles in turbulence near-contact.

physics.flu-dyn↗
arXiv2020

A Nonlinear Mechanics-based Virtual Coiling Method For Intracranial Aneurysm

Enodvascular coils treat intracranial aneurysms (IAs) by causing them to occlude by thrombosis. Ideally, coiled IAs eventually occlude in the long-term. However, 20.8% are found incompletely occluded at treatment follow-up. Computer simulations of coiling and its effect on aneurysmal flow could help clinicians predict treatment outcomes a priori, but it requires accurate modeling of coils and their deployment procedure. In addition to being accurate, coiling simulations must be efficient to be used as a bedside tool. To date, several virtual coiling techniques have been developed, but they lack either accuracy or efficiency. For example, finite-element-based virtual coiling methods model the mechanics of coiling and are highly accurate, at the expense of high computational cost (and thus low efficiency). Geometric-rule-based coiling techniques ignore the mechanics and therefore are computationally efficient, but may produce unrealistic coil deployments. In order to develop a virtual coiling method that combines accuracy and efficiency, we propose a novel virtual coiling algorithm that models coil deployment with nonlinear mechanics and nonlinear contact. Our approach is potentially more accurate than existing "simple" techniques because we model coil mechanics. It is also potentially faster than finite-element techniques because it models the most time-consuming part of these algorithms-namely contact resolution-with a novel formulation that resolves contact faster with exponential functions. Moreover, we model the coil's pre-shape as well as coil packaging into the catheter, both of which are important to model but are lacking from most existing techniques.

physics.app-ph↗
arXiv2017

Particle-Pair Relative Velocity Measurement in High-Reynolds-Number Homogeneous and Isotropic Turbulence Using 4-Frame Particle Tracking Velocimetry

The radial relative velocity between particles suspended in turbulent flow plays a critical role in droplet collision and growth. We present a simple and accurate approach to RV measurement in isotropic turbulence - planar 4-frame particle tracking velocimetry - using routine PIV hardware. This study demonstrates the feasibility of accurately measuring RV using routine hardware and verifies, for the first time, the path-history and inertial filtering effects on particle-pair RV at large particle separations experimentally.

physics.flu-dyn↗