arXiv subjects

Marc D. Ryser

Publications and source records attributed to Marc D. Ryser.

6 recordsLinked to original sources

arXiv2026

Unsupervised Domain Adaptation for Calcification Classification in Mammography Across Multi-Site Datasets

Deep learning-based computer-aided diagnosis (CAD) systems have shown strong performance in breast cancer diagnosis, particularly for classification tasks in mammography. However, domain shifts across multi-site datasets remain a challenge, especially when models are applied to unseen domains. In this work, we proposed a calcification classification framework to improve malignant versus benign breast disease classification across multi-site mammography datasets. The framework consisted of two components: (1) an unsupervised domain adaptation module based on style transfer models (AdaIN and CycleGAN) to generate vendor-specific and technique-specific training samples without additional annotations, and (2) a supervised classification module using Swin Transformer V2 as the backbone. We evaluated the proposed method on three datasets: cross-validation on OPTIMAM (National Health Service, United Kingdom; n=2994), followed by external validation on EMBED (Emory University; n=125), and Duke Calcification Dataset v1 (n=788). These datasets cover multiple vendors and include both full-field digital mammography and synthetic 2D images derived from digital breast tomosynthesis. The proposed framework improved cross-site performance for both EMBED (AUC 0.68 to 0.72) and the Duke Calcification Dataset (AUC 0.68 to 0.73). These findings indicate that domain adaptation can reduce domain shifts and improve the generalization for calcification classification across multi-site datasets.

cs.CV↗

arXiv2025

MammoTracker: Mask-Guided Lesion Tracking in Temporal Mammograms

Accurate lesion tracking in temporal mammograms is essential for monitoring breast cancer progression and facilitating early diagnosis. However, automated lesion correspondence across exams remains a challenges in computer-aided diagnosis (CAD) systems, limiting their effectiveness. We propose MammoTracker, a mask-guided lesion tracking framework that automates lesion localization across consecutively exams. Our approach follows a coarse-to-fine strategy incorporating three key modules: global search, local search, and score refinement. To support large-scale training and evaluation, we introduce a new dataset with curated prior-exam annotations for 730 mass and calcification cases from the public EMBED mammogram dataset, yielding over 20000 lesion pairs, making it the largest known resource for temporal lesion tracking in mammograms. Experimental results demonstrate that MammoTracker achieves 0.455 average overlap and 0.509 accuracy, surpassing baseline models by 8%, highlighting its potential to enhance CAD-based lesion progression analysis. Our dataset will be available at https://gitlab.oit.duke.edu/railabs/LoGroup/mammotracker.

cs.CV↗

arXiv2016

Bone Remodeling as a Spatial Evolutionary Game

Bone remodeling is a complex process involving cell-cell interactions, biochemical signaling and mechanical stimuli. Early models of the biological aspects of remodeling were non-spatial and focused on the local dynamics at a fixed location in the bone. Several spatial extensions of these models have been proposed, but they generally suffer from two limitations: first, they are not amenable to analysis and are computationally expensive, and second, they neglect the role played by bone-embedded osteocytes. To address these issues, we developed a novel model of spatial remodeling based on the principles of evolutionary game theory. The analytically tractable framework describes the spatial interactions between zones of bone resorption, bone formation and quiescent bone, and explicitly accounts for regulation of remodeling by bone-embedded, mechanotransducing osteocytes. Using tools from the theory of interacting particle systems we systematically classified the different dynamic regimes of the spatial model and identified regions of parameter space that allow for coexistence of resorption, formation and quiescence, as observed in physiological remodeling. In coexistence scenarios, three-dimensional simulations revealed the emergence of sponge-like bone clusters. Comparison between spatial and non-spatial dynamics revealed substantial differences and suggested a stabilizing role of space. Our findings emphasize the importance of accounting for spatial structure and bone-embedded osteocytes when modeling the process of bone remodeling. Thanks to the lattice-based framework, the proposed model can easily be coupled to a mechanical model of bone loading.

q-bio.PE↗

arXiv2012

Triviality of the 2D stochastic Allen-Cahn equation

We consider the stochastic Allen-Cahn equation driven by mollified space-time white noise. We show that, as the mollifier is removed, the solutions converge weakly to 0, independently of the initial condition. If the intensity of the noise simultaneously converges to 0 at a sufficiently fast rate, then the solutions converge to those of the deterministic equation. At the critical rate, the limiting solution is still deterministic, but it exhibits an additional damping term.

math.PR↗

arXiv2011

On the well-posedness of the stochastic Allen-Cahn equation in two dimensions

White noise-driven nonlinear stochastic partial differential equations (SPDEs) of parabolic type are frequently used to model physical and biological systems in space dimensions d = 1,2,3. Whereas existence and uniqueness of weak solutions to these equations are well established in one dimension, the situation is different for d \geq 2. Despite their popularity in the applied sciences, higher dimensional versions of these SPDE models are generally assumed to be ill-posed by the mathematics community. We study this discrepancy on the specific example of the two dimensional Allen-Cahn equation driven by additive white noise. Since it is unclear how to define the notion of a weak solution to this equation, we regularize the noise and introduce a family of approximations. Based on heuristic arguments and numerical experiments, we conjecture that these approximations exhibit divergent behavior in the continuum limit. The results strongly suggest that a series of published numerical studies are problematic: shrinking the mesh size in these simulations does not lead to the recovery of a physically meaningful limit.

math.NA↗

arXiv2007

Thermal quantum electrodynamics of nonrelativistic charged fluids

The theory relevant to the study of matter in equilibrium with the radiation field is thermal quantum electrodynamics (TQED). We present a formulation of the theory, suitable for non relativistic fluids, based on a joint functional integral representation of matter and field variables. In this formalism cluster expansion techniques of classical statistical mechanics become operative. They provide an alternative to the usual Feynman diagrammatics in many-body problems which is not perturbative with respect to the coupling constant. As an application we show that the effective Coulomb interaction between quantum charges is partially screened by thermalized photons at large distances. More precisely one observes an exact cancellation of the dipolar electric part of the interaction, so that the asymptotic particle density correlation is now determined by relativistic effects. It has still the $r^{-6}$ decay typical for quantum charges, but with an amplitude strongly reduced by a relativistic factor.

quant-ph↗