arXiv subjects

Rui Santos

Publications and source records attributed to Rui Santos.

At least 19 recordsLinked to original sources

arXiv2026

A Comprehensive Analysis of the R2HDM Vacuum Evolution and the Induced GW and Collider Phenomenology

Extended Higgs sectors beyond the Standard Model (BSM) allow to dynamically generate the observed baryon asymmetry of the Universe through electroweak baryogenesis and thereby solve one of the most prominent open problems of the SM. The strong first-order phase transitions (PTs), required to preserve the generated asymmetry in the electroweak vacuum, source gravitational waves (GW) that can be tested at future experiments like LISA. Gravitational waves hence provide the exciting possibility to probe BSM physics through cosmological processes. In order to be able to eventually pin down the specific underlying physics, a good understanding of the evolution of the vacuum of the model under investigation is indispensable as well as of the uncertainties that are involved in the derivation of the GW spectrum. We use our code BSMPTv3 that allows to reliably derive the finite temperature vacuum structure of extended Higgs sectors with multiple vacuum directions and calculates the GW spectrum of the found (multiple) strong first-order PTs, and we apply it to the real, i.e. CP-conserving, 2-Higgs-Doublet Model. Taking into account all relevant theoretical and experimental constraints, we perform a thorough analysis of its vacuum evolution, the related collider and GW phenomenology and complement it by an uncertainty discussion.

hep-ph↗

arXiv2026

Full Next-To Leading-Order Electroweak and QCD Corrections to the Relic Density in the CxSM

In this work, we present the calculation of the next-to-leading order (NLO) QCD and electroweak corrections to the thermally averaged cross section of Dark Matter (DM) annihilation in the framework of the complex singlet extended Standard Model (CxSM) with one DM candidate. We derive the corresponding NLO QCD and electroweak corrected relic density and discuss the impact of the corrections as well as the remaining theoretical uncertainty due to missing higher-order corrections. This is estimated through a variation of the renormalization schemes for the singlet vacuum expectation value $v_S$ and the Higgs mixing angle $\alpha$. For the bulk of the scanned parameter points, the QCD and electroweak corrections are found to be of typical size with the remaining theoretical uncertainty ranging at the percent level. The larger corrections that are found, can be attributed to a large counterterm for $v_S$ emerging in the case of large mass gaps in the decay channel used for the renormalization. The corrections are of phenomenological impact. There are parameter points that are allowed at leading order (LO), but are excluded after including the NLO corrections to the relic density, and vice versa. Our corrections have been implemented in the new code RelExt@NLO based on the LO code RelExt, which has been made publicly available. This work marks a first step towards the calculation of the QCD and electroweak corrected relic densities in models with DM candidates stabilized by a discrete $\mathbb{Z}_2$ symmetry.

hep-ph↗

arXiv2026

$t \bar{t}$ production as a window to invisible new physics

We present a phenomenological study where we probe the sensitivity to invisible dark matter (DM) mediators produced in association with a $t\bar{t}$ pair at the Large Hadron Collider (LHC). Building on previous work focused on scalar mediators, we extend the analysis to include spin-1 mediators, $Y_1$, with both vector and axial-vector couplings to top quarks. The mediator mass is fixed to 5 GeV. Signal samples of $pp \rightarrow t\bar{t}Y_i$ ($i = 0, 1$) are generated using a MadGraph5_aMC@NLO simplified DM model. Only dileptonic final states of the $t\bar{t}$ system are considered, and the reconstruction is performed through a kinematic fit without explicitly reconstructing the invisible mediator. All relevant Standard Model backgrounds are included. We consider several exclusion scenarios to assess the sensitivity to the presence of a spin-1 mediator, as well as the ability to distinguish a pure vector or axial-vector mediator from alternative hypotheses with different spin and CP properties. We find that the analysis is sensitive to light spin-1 mediators and that CP-sensitive angular observables provide discrimination power between vector, axial-vector, scalar and pseudoscalar scenarios. These results highlight the potential of $t\bar{t}$ final states not only to search for invisible particles, but also to characterize their spin and parity properties in case of discovery.

hep-ph↗

arXiv2026

A Deep Dive into Baryon Asymmetry -- the C2HDM

In this paper, we present our new implementation of the computation of the baryon asymmetry in the code BSMPT. It is based on the WKB ansatz generalizing the transport equations to an arbitrary number of moments. Two different truncation schemes are implemented, and the profile of the vacuum expectation value (VEV) is derived from the equations of motion in addition to the modeling with the kink profile. We validate our implementation with a simple benchmark model and perform a detailed analysis within the CP-violating 2-Higgs-Doublet Model (C2HDM). Barring the collision term, however, our implementation can readily be applied to any extended Higgs sector with an arbitrary number of VEV directions. We study in detail the dependencies of the baryon asymmetry on the number of moment equations, the applied truncation scheme, the wall velocity, the wall velocity times wall width, the VEV profile, the strength of the phase transition, and the amount of CP violation in the model and present a detailed uncertainty analysis. We investigate the interplay of the generated baryon asymmetry and the gravitational waves signal at LISA. Our results guide the way for future improvements in the computation of the baryon asymmetry and give directions for model building. The uncertainty analysis is the basis for any investigation aiming at deducing model parameters from cosmological processes.

hep-ph↗

arXiv2026

AIdentifyAGE Ontology for Decision Support in Forensic Dental Age Assessment

Age assessment is crucial in forensic and judicial decision-making, particularly in cases involving undocumented individuals and unaccompanied minors, where legal thresholds determine access to protection, healthcare, and judicial procedures. Dental age assessment is widely recognized as one of the most reliable biological approaches for adolescents and young adults, but current practices are challenged by methodological heterogeneity, fragmented data representation, and limited interoperability between clinical, forensic, and legal information systems. These limitations hinder transparency and reproducibility, amplified by the increasing adoption of AI- based methods. The AIdentifyAGE ontology is domain-specific and provides a standardized, semantically coherent framework, encompassing both manual and AI-assisted forensic dental age assessment workflows, and enabling traceable linkage between observations, methods, reference data, and reported outcomes. It models the complete medico-legal workflow, integrating judicial context, individual-level information, forensic examination data, dental developmental assessment methods, radiographic imaging, statistical reference studies, and AI-based estimation methods. It is being developed together with domain experts, and it builds on upper and established biomedical, dental, and machine learning ontologies, ensuring interoperability, extensibility, and compliance with FAIR principles. The AIdentifyAGE ontology is a fundamental step to enhance consistency, transparency, and explainability, establishing a robust foundation for ontology-driven decision support systems in medico-legal and judicial contexts.

cs.AI↗

arXiv2026

Reassessing CP Violation in the C2HDM with Machine Learning

We provide a study of the parameter space of the complex 2-Higgs Doublet Model (C2HDM), focusing on signs of large CP-violating couplings of the 125 GeV Higgs boson with the fermions. The study is performed utilizing Machine Learning (ML) techniques developed recently for parameter space exploration, including an Evolutionary Strategy Algorithm and Novelty Reward. We give particular attention to the electron electric dipole moment (eEDM). We confirm that the recently found kite diagrams are crucial for the outcome of the analysis. Moreover, their use also mitigates the dependence of the results on the scale and scheme choice of the masses in the loop diagrams. We furthermore point out that, already at the current level of experimental precision, the Barr-Zee diagrams with charm quark loops must be taken into account. The combined use of kite diagrams and ML techniques allows for the resurrection of large fermion CP-odd couplings for Type-II and Flipped C2HDM when the 125 GeV Higgs coincides with the second lightest neutral scalar. This arises due to cancellations, typically of the per-mil order, which, moreover, will still be possible for a foreseeable eEDM precision down to $10^{-33}$ e.cm. For these cases, the constraints on the CP-odd couplings arises from the precision LHC measurements.

hep-ph↗

arXiv2025

When Do Domain-Specific Foundation Models Justify Their Cost? A Systematic Evaluation Across Retinal Imaging Tasks

Large vision foundation models have been widely adopted for retinal disease classification without systematic evidence justifying their parameter requirements. In the present work we address two critical questions: First, are large domain-specific foundation models essential, or do compact general-purpose architectures suffice? Second, does specialized retinal pretraining justify its computational cost? To answer this, we benchmark initialization strategies across four retinal imaging classification tasks spanning Optical Coherence Tomography (OCT) and Color Fundus Photography (CFP) modalities: 8-class OCT classification, 3-class diabetic macular edema (DME), 5-class diabetic retinopathy (DR), and 3-class glaucoma (GL) detection. We evaluate 12-13 model configurations per task, including vision transformers (22.8M-86.6M parameters), Swin Transformers (27.6M-28.3M), ConvNeXt (28.6M), and the domain-specific RETFound models (303M), under identical training conditions. Our results challenge prevailing assumptions: First, we demonstrate that pretraining provides universal benefits (5.18-18.41% improvement), scaling with task difficulty. Second, compact architectures (27-29M) dominate Pareto frontiers; SwinV2-tiny achieves top-1 performance on three datasets. Third, RETFound (303M) justifies its computational cost only for challenging DR grading (accuracy of 71.15%), while ImageNet pretraining proves to be sufficient with all other tasks (DME accuracy: 99.24%, OCT accuracy: 97.96%). CFP tasks show larger pretraining accuracy gains (9.13-18.41%) than OCT (5.18%). Thus, the evidence suggests that compact general-purpose models deliver near-optimal performance for most retinal classification tasks; specialized foundation models warranted only for fine-grained discrimination under extreme class imbalance.

eess.IV↗

arXiv2025

Generalist versus Specialist Vision Foundation Models for Ocular Disease and Oculomics

Medical foundation models, pre-trained with large-scale clinical data, demonstrate strong performance in diverse clinically relevant applications. RETFound, trained on nearly one million retinal images, exemplifies this approach in applications with retinal images. However, the emergence of increasingly powerful and multifold larger generalist foundation models such as DINOv2 and DINOv3 raises the question of whether domain-specific pre-training remains essential, and if so, what gap persists. To investigate this, we systematically evaluated the adaptability of DINOv2 and DINOv3 in retinal image applications, compared to two specialist RETFound models, RETFound-MAE and RETFound-DINOv2. We assessed performance on ocular disease detection and systemic disease prediction using two adaptation strategies: fine-tuning and linear probing. Data efficiency and adaptation efficiency were further analysed to characterise trade-offs between predictive performance and computational cost. Our results show that although scaling generalist models yields strong adaptability across diverse tasks, RETFound-DINOv2 consistently outperforms these generalist foundation models in ocular-disease detection and oculomics tasks, demonstrating stronger generalisability and data efficiency. These findings suggest that specialist retinal foundation models remain the most effective choice for clinical applications, while the narrowing gap with generalist foundation models suggests that continued data and model scaling can deliver domain-relevant gains and position them as strong foundations for future medical foundation models.

eess.IV↗

arXiv2025

Towards a Unified Framework for Pseudo-Nambu-Goldstone Dark Matter and Electroweak Baryogenesis

We propose the complex singlet-extended 2-Higgs-Doublet Model (cS2HDM), a spin-0 Dark Matter (DM) model with a Higgs sector consisting of two Higgs doublets and a complex singlet, as a benchmark for LHC DM searches. The model predicts a pseudo-Nambu-Goldstone DM candidate whose interactions with nuclei are naturally suppressed, while allowing for all sources of CP-violation under the assumption of flavour alignment in the Yukawa sector, which enables CP-violating interactions of the Higgs bosons even in the alignment limit. This feature makes the model attractive for studies of electroweak baryogenesis while accommodating a Higgs-portal DM candidate with standard thermal freeze-out. We confront the model with a comprehensive set of theoretical and experimental constraints, including Higgs-boson signal strength measurements, searches for additional Higgs bosons, DM relic abundance and direct detection, as well as electroweak precision observables and the electron EDM, with emphasis on the impact of the new CP-violating sources. For DM direct detection, we perform a one-loop computation of DM-nucleon scattering including CP-violating effects. We provide a public software package to facilitate future phenomenological studies of the cS2HDM.

hep-ph↗

arXiv2025

FusionFM: Fusing Eye-specific Foundational Models for Optimized Ophthalmic Diagnosis

Foundation models (FMs) have shown great promise in medical image analysis by improving generalization across diverse downstream tasks. In ophthalmology, several FMs have recently emerged, but there is still no clear answer to fundamental questions: Which FM performs the best? Are they equally good across different tasks? What if we combine all FMs together? To our knowledge, this is the first study to systematically evaluate both single and fused ophthalmic FMs. To address these questions, we propose FusionFM, a comprehensive evaluation suite, along with two fusion approaches to integrate different ophthalmic FMs. Our framework covers both ophthalmic disease detection (glaucoma, diabetic retinopathy, and age-related macular degeneration) and systemic disease prediction (diabetes and hypertension) based on retinal imaging. We benchmarked four state-of-the-art FMs (RETFound, VisionFM, RetiZero, and DINORET) using standardized datasets from multiple countries and evaluated their performance using AUC and F1 metrics. Our results show that DINORET and RetiZero achieve superior performance in both ophthalmic and systemic disease tasks, with RetiZero exhibiting stronger generalization on external datasets. Regarding fusion strategies, the Gating-based approach provides modest improvements in predicting glaucoma, AMD, and hypertension. Despite these advances, predicting systemic diseases, especially hypertension in external cohort remains challenging. These findings provide an evidence-based evaluation of ophthalmic FMs, highlight the benefits of model fusion, and point to strategies for enhancing their clinical applicability.

cs.CV↗

arXiv2025

BSM: Extended Scalar Sectors

In particle physics the world is described by a function, the Lagrangian. Each of its sectors characterizes the interactions between the particles of the Standard Model (SM). The addition of hypothetical new particles is done by including new terms in the Lagrangian. The scalar or Higgs sector of the SM is built with only one scalar complex field and it is extended by including new spin zero fields. This can help to solve questions that cannot be answered by the SM alone, like introducing dark matter candidates or new sources of CP-violation required to explain the matter-antimatter asymmetry of the universe. The corresponding theories have to be probed experimentally. For the high energy region, the standard tools are collider experiments such as the Large Hadron Collider, or other possible future facilities. Dark matter experiments scrutinize the connection between the visible and the dark world.

hep-ph↗

arXiv2025

V-Associated Production & Vector Boson Fusion as an LHC Signature of CP Violation

We investigate the role of vector boson fusion (VBF) and associated production with an electroweak boson (V-AP) of beyond-the-Standard-Model Higgses at the LHC in probing CP violation in extended Higgs sectors. Resonant production of a new Higgs boson through V-AP/VBF subsequently decaying into a Z boson and a 125 GeV Higgs boson h would be a robust sign of CP violation, since h has been measured to be (predominantly) a CP-even state. After identifying this and other sets of signatures which rely on V-AP/VBF to jointly uncover CP violation, we analyze the prospects to measure CP violation in this way for the complex two-Higgs-doublet-model at the High-Luminosity LHC.

hep-ph↗

arXiv2025

Interplay between Electroweak Symmetry Breaking and Higgs Portal Dark Matter

Models of Dark Matter must contend with the fact that the presence of electroweak symmetry breaking along the thermal evolution of the Universe modifies the masses, interactions and, thus, the thermally averaged cross sections. We study in detail the impact of taking (not taking) the presence of the electroweak symmetry breaking into account in the calculations of the Dark Matter relic density in Higgs portal models, providing a model-independent measure of such differences. By focusing on a particular model, we show that ignoring this effect can lead to the inclusion (exclusion) of wrong (viable) regions of parameter space.

hep-ph↗

arXiv2025

Dark Matter in Multi-Singlet Extensions of the Standard Model

We study the simplest extensions of the Standard Model (SM) that provide Dark Matter (DM) candidates, built with the addition of real singlets and new $\mathcal{Z}_2$ symmetries. In this type of models the interactions between SM particles are not altered except for the new interactions stemming from the portal couplings that link the SM Higgs with the DM candidates. In the extension with just one singlet, DM masses below about 3.5 TeV are already excluded by the combination of relic density and direct detection (DD) constraints, except in the resonant case where the DM mass is close to half the Higgs mass, making them undetectable at the LHC. Adding just one more real singlet with an independent $\mathcal{Z}_2$ symmetry opens up a new mass window for one of the DM candidates and decreases the lower bound on the mass of the other. Adding more singlets with independent $\mathcal{Z}_2$ symmetries will not change this picture dramatically. If instead we add new singlets all odd under the same $\mathcal{Z}_2$ symmetry, the allowed mass region for the DM candidate (i.e., the lightest dark sector scalar) will span the entire mass range from half the Higgs mass to the TeV scale. In principle, such light particles could be probed at the LHC in mono-$X$ searches. Although they are still out of reach with the current LHC DM searches, there are good chances to probe the models in some final states at the High-Luminosity (HL-LHC) stage of the LHC.

hep-ph↗

arXiv2025

RelExt: A New Dark Matter Tool for the Exploration of Dark Matter Models

We present the C++ program RelExt for Standard Model (SM) extensions that feature a Dark Matter (DM) candidate. The tool allows to efficiently scan the parameter spaces of these models to find parameter combinations that lead to relic density values which are compatible with the measured value within the uncertainty specified by the user. The code computes the relic density for freeze-out (co-)annihilation processes. The user can choose between several pre-installed models or any arbitrary other model featuring a discrete $\mathbb{Z}_2$ symmetry, by solely providing the corresponding FeynRules model files. The code automatically generates the required (co-)annihilation amplitudes and thermally averaged cross sections, including the total widths in the $s$-channel mediators, and solves the Boltzmann equation to determine the relic density. It can easily be linked to other tools like e.g.~ScannerS to check for the relevant theoretical and experimental constraints, or to BSMPT to investigate the phase history of the model and possibly related gravitational waves signals.

hep-ph↗

arXiv2025

Is an Ultra Large Natural Image-Based Foundation Model Superior to a Retina-Specific Model for Detecting Ocular and Systemic Diseases?

The advent of foundation models (FMs) is transforming medical domain. In ophthalmology, RETFound, a retina-specific FM pre-trained sequentially on 1.4 million natural images and 1.6 million retinal images, has demonstrated high adaptability across clinical applications. Conversely, DINOv2, a general-purpose vision FM pre-trained on 142 million natural images, has shown promise in non-medical domains. However, its applicability to clinical tasks remains underexplored. To address this, we conducted head-to-head evaluations by fine-tuning RETFound and three DINOv2 models (large, base, small) for ocular disease detection and systemic disease prediction tasks, across eight standardized open-source ocular datasets, as well as the Moorfields AlzEye and the UK Biobank datasets. DINOv2-large model outperformed RETFound in detecting diabetic retinopathy (AUROC=0.850-0.952 vs 0.823-0.944, across three datasets, all P<=0.007) and multi-class eye diseases (AUROC=0.892 vs. 0.846, P<0.001). In glaucoma, DINOv2-base model outperformed RETFound (AUROC=0.958 vs 0.940, P<0.001). Conversely, RETFound achieved superior performance over all DINOv2 models in predicting heart failure, myocardial infarction, and ischaemic stroke (AUROC=0.732-0.796 vs 0.663-0.771, all P<0.001). These trends persisted even with 10% of the fine-tuning data. These findings showcase the distinct scenarios where general-purpose and domain-specific FMs excel, highlighting the importance of aligning FM selection with task-specific requirements to optimise clinical performance.

eess.IV↗

arXiv2024

Tables with Critical Values for the Meta-Analysis of Genuine and Fake $\boldsymbol{p}$-Values

The classical theory for the meta-analysis of $p$-values is based on the assumption that if the overall null hypothesis is true, then all $p$-values used in a chosen combined test statistic are genuine, i.e., are observations from independent and identically distributed standard uniform random variables. However, the pressure felt by most researchers to publish, which is worsen by publication bias, can originate fake $p$-values to be reported, usually Beta(1,2) distributed. In general, the existence of fake $p$-values in a sample of $p$-values to be combined is unknown, and if, for some reason, there is information that they do exist, their number will most likely be unknown as well. Moreover, even if fake $p$-values are accounted for, the cumulative distribution function of classical combined test statistics does not have a closed-form expression that facilitates its practical usage. To overcome this problem, tables with estimated critical values are supplied for the commonly used combined tests for the meta-analysis of $p$-values when a few of them are fake ones, i.e., Beta(1,2) distributed.

stat.CO↗

arXiv2024

Block Expanded DINORET: Adapting Natural Domain Foundation Models for Retinal Imaging Without Catastrophic Forgetting

Integrating deep learning into medical imaging is poised to greatly advance diagnostic methods but it faces challenges with generalizability. Foundation models, based on self-supervised learning, address these issues and improve data efficiency. Natural domain foundation models show promise for medical imaging, but systematic research evaluating domain adaptation, especially using self-supervised learning and parameter-efficient fine-tuning, remains underexplored. Additionally, little research addresses the issue of catastrophic forgetting during fine-tuning of foundation models. We adapted the DINOv2 vision transformer for retinal imaging classification tasks using self-supervised learning and generated two novel foundation models termed DINORET and BE DINORET. Publicly available color fundus photographs were employed for model development and subsequent fine-tuning for diabetic retinopathy staging and glaucoma detection. We introduced block expansion as a novel domain adaptation strategy and assessed the models for catastrophic forgetting. Models were benchmarked to RETFound, a state-of-the-art foundation model in ophthalmology. DINORET and BE DINORET demonstrated competitive performance on retinal imaging tasks, with the block expanded model achieving the highest scores on most datasets. Block expansion successfully mitigated catastrophic forgetting. Our few-shot learning studies indicated that DINORET and BE DINORET outperform RETFound in terms of data-efficiency. This study highlights the potential of adapting natural domain vision models to retinal imaging using self-supervised learning and block expansion. BE DINORET offers robust performance without sacrificing previously acquired capabilities. Our findings suggest that these methods could enable healthcare institutions to develop tailored vision models for their patient populations, enhancing global healthcare inclusivity.

cs.CV↗