Samuel Ruiperez Campillo


Last Name

Ruiperez Campillo

First Name

Samuel

ORCID

0000-0002-5425-4175

Organisational unit

09670 - Vogt, Julia / Vogt, Julia

Filters

2023 - 202641
Reset filters

Search Results

Publications1 - 10 of 41
  • Structure is Supervision: Multiview Masked Autoencoders for Radiology
    Item type: Working Paper
    Laguna Cillero, Sonia; Agostini, Andrea; Ryser, Alain; et al. (2025)
    arXiv
    Building robust medical machine learning systems requires pretraining strategies that exploit the intrinsic structure present in clinical data. We introduce Multiview Masked Autoencoder (MVMAE), a self-supervised framework that leverages the natural multi-view organization of radiology studies to learn view-invariant and disease-relevant representations. MVMAE combines masked image reconstruction with cross-view alignment, transforming clinical redundancy across projections into a powerful self-supervisory signal. We further extend this approach with MVMAE-V2T, which incorporates radiology reports as an auxiliary text-based learning signal to enhance semantic grounding while preserving fully vision-based inference. Evaluated on a downstream disease classification task on three large-scale public datasets, MIMIC-CXR, CheXpert, and PadChest, MVMAE consistently outperforms supervised and vision-language baselines. Furthermore, MVMAE-V2T provides additional gains, particularly in low-label regimes where structured textual supervision is most beneficial. Together, these results establish the importance of structural and textual supervision as complementary paths toward scalable, clinically grounded medical foundation models.
  • RadVLM model
    Item type: Other Publication
    Deperrois, Nicolas; Matsuo, Hidetoshi; Ruiperez-Campillo, Samuel; et al. (2025)
    We present RadVLM, a compact (7B) multitask conversational foundation model designed for CXR interpretation. Its development relies on the curation of a large-scale instruction dataset comprising over 1 million image-instruction pairs containing both single-turn tasks - such as report generation, abnormality classification, and visual grounding - and multi-turn, multi-task conversational interactions. Our experiments show that RadVLM, fine-tuned on this instruction dataset, achieves state-of-the-art performance in conversational capabilities and visual grounding while remaining competitive in other radiology tasks (report generation, classification). Ablation studies further highlight the benefit of joint training across multiple tasks, particularly for scenarios with limited annotated data. Together, these findings highlight the potential of the RadVLM model as a clinically relevant AI assistant, providing structured CXR interpretation and conversational capabilities to support more effective and accessible diagnostic workflows.
  • Biophysics-inspired deep learning for improved denoising in ventricular signals in ischemic cardiomyopathy
    Item type: Other Conference Item
    Ruiperez Campillo, Samuel; Rau, Moritz; Ganesan, Prasanth; et al. (2025)
    European Heart Journal
    Background Ventricular arrhythmias in ischemic cardiomyopathy are a major cause of sudden cardiac death, yet their mapping is compromised by noise from pacing artifacts, catheter movement, or electrical interferences. Although traditional filtering techniques mitigate some artifacts, they often fail to preserve fine-grained features. We propose a conditional denoising diffusion probabilistic model (cDDPM), a biophysics-inspired deep learning model that enhances intracardiac signal fidelity while maintaining the integrity of critical diagnostic features. Objective This study aims to develop and evaluate our cDDPM for denoising intracardiac EGMs. We developed a proof-of-concept approach using monophasic action potentials (MAP), in order to assess preservation of signal morphology. By adopting a diffusion-based framework, we hypothesize that our model will address the limitations of classical filtering and other deep learning methods, providing improved signal fidelity. Methods We utilized 5706 MAP recordings from 42 ischemic cardiomyopathy patients, acquired during ventricular stimulation via a 7F catheter. The dataset was augmented with simulated and real EP noise extracted from patient recordings to emulate clinically relevant contamination. We implemented a cDDPM, where noisy EGMs were progressively denoised in an iterative reverse diffusion process. Performance was assessed using Pearson's correlation coefficient (PCC) to measure waveform preservation, root mean square error (RMSE) for time-domain fidelity, and peak signal-to-noise ratio (PSNR) for overall denoising effectiveness. Our method was benchmarked against widely clinically used filters and a state-of-the-art deep learning-based denoising model (VAE). Results The cDDPM significantly outperformed both baseline approaches in all key metrics. On the test set, cDDPM achieved a PCC of 0.984±0.004, compared to 0.970±0.009 for the VAE and 0.899±0.017 for classic filtering (p<0.001). RMSE marked a 2.5-fold improvement over VAEs and a nearly 9-fold reduction compared to classic filters. PSNR increased to 30.07±0.85 dB, highlighting the superior noise suppression capacity of the proposed model while maintaining critical EGM features essential for clinical decision-making. Conclusion Our biophysics-inspired deep learning framework provides a robust approach for denoising intracardiac EGMs, preserving fine-grained electrophysiological features crucial for arrhythmia diagnosis. By iteratively refining signal reconstructions, this method overcomes limitations of both rule-based filtering and latent-space-based generative approaches. These findings suggest that diffusion models could significantly enhance the ability to interpret noisy EGMs. Future work should focus on optimizing real-time clinical application and validating performance across diverse populations.
  • Mathematical models and computational approaches in CAR-T therapeutics
    Item type: Review Article
    Putignano, Guido; Ruiperez Campillo, Samuel; Yuan, Zhou; et al. (2025)
    Frontiers in Immunology
    BackgroundThe field of synthetic biology aims to engineer living organisms for specific therapeutic applications, with CAR-T cell therapy emerging as a groundbreaking approach in cancer treatment due to its potential for flexibility, specificity, predictability, and controllability. CAR-T cell therapies involve the genetic modification of T cells to target tumor-specific antigens. However, challenges persist because the limited spatio-temporal resolution in current models hinders the therapy’s safety, cost-effectiveness, and overall potential, particularly for solid tumorsMain bodyThis manuscript explores how mathematical models and computational techniques can enhance CAR-T therapy design and predict therapeutic outcomes, focusing on critical factors such as antigen receptor functionality, treatment efficacy, and potential adverse effects. We examine CAR-T cell dynamics and the impact of antigen binding, addressing strategies to overcome antigen escape, cytokine release syndrome, and relapse.ConclusionWe propose a comprehensive framework for using these models to advance CAR-T cell therapy, bridging the gap between existing therapeutic methods and the full potential of CAR-T engineering and its clinical application.
  • A Denoising VAE for Intracardiac Times Series in Ischemic Cardiomyopathy
    Item type: Conference Paper
    Ruiperez Campillo, Samuel; Ryser, Alain; Sutter, Thomas M.; et al. (2024)
    ICLR 2024 Workshop on Learning from Time Series For Health
    In the field of cardiac electrophysiology (EP), effectively reducing noise in intra-cardiac signals is crucial for the accurate diagnosis and treatment of arrhythmias and cardiomyopathies. However, traditional noise reduction techniques fall short in addressing the diverse noise patterns from various sources, often non-linear and non-stationary, present in these signals. This work introduces a Variational Autoencoder (VAE) model, aimed at improving the quality of intra-ventricular monophasic action potential (MAP) signal recordings. By constructing representations of clean signals from a dataset of 5706 time series from 42 patients diagnosed with ischemic cardiomyopathy, our approach demonstrates superior denoising performance when compared to conventional filtering methods commonly employed in clinical settings. We assess the effectiveness of our VAE model using various metrics, indicating its superior capability to denoise signals across different noise types, including time-varying non-linear noise frequently found in clinical settings. These results reveal that VAEs can eliminate diverse sources of noise in single beats, outperforming state-of-the-art denoising techniques and potentially improving treatment efficacy in cardiac EP.
  • RadVLM: A Multitask Conversational Vision-Language Model for Radiology
    Item type: Working Paper
    Deperrois, Nicolas; Matsuo, Hidetoshi; Ruiperez Campillo, Samuel; et al. (2025)
    arXiv
    The widespread use of chest X-rays (CXRs), coupled with a shortage of radiologists, has driven growing interest in automated CXR analysis and AI-assisted reporting. While existing vision-language models (VLMs) show promise in specific tasks such as report generation or abnormality detection, they often lack support for interactive diagnostic capabilities. In this work we present RadVLM, a compact, multitask conversational foundation model designed for CXR interpretation. To this end, we curate a large-scale instruction dataset comprising over 1 million image-instruction pairs containing both single-turn tasks -- such as report generation, abnormality classification, and visual grounding -- and multi-turn, multi-task conversational interactions. After fine-tuning RadVLM on this instruction dataset, we evaluate it across different tasks along with re-implemented baseline VLMs. Our results show that RadVLM achieves state-of-the-art performance in conversational capabilities and visual grounding while remaining competitive in other radiology tasks. Ablation studies further highlight the benefit of joint training across multiple tasks, particularly for scenarios with limited annotated data. Together, these findings highlight the potential of RadVLM as a clinically relevant AI assistant, providing structured CXR interpretation and conversational capabilities to support more effective and accessible diagnostic workflows.
  • Omnipolar Technology in Intracardiac signals: Are We Fulfilling the Assumptions?
    Item type: Conference Paper
    Ramírez, Elisa; Tonko, Johanna; Alós, Raúl; et al. (2025)
    2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)
    The HD-Grid catheter is widely in clinical practice for intracavitary electrical mapping. However, its interelectrode spacing does not always ensure compliance with the assumptions of waveform homogeneity, amplitude consistency, and planar wavefront propagation required by the traveling wave theory underlying omnipolar reconstruction. In this study we aim to quantify the extend to which these assumptions hold by introducing the Amplitude Variability, Morphology, and Non-Planarity parameters. Additionally, we propose a solution to this limitation through unipolar signal interpolation to increase virtual spatial resolution and ensure accurate omnipolar signal reconstruction and biomarkers extraction. Out method employs a linear combination of unipolar HD-Grid signals, weighted using spline interpolation the inverse of the distance. Results indicate that compliance with the theoretical assumptions is influenced by interelectrode distance, with optimal adherence achieved at electrode spacings of 0.5 mm, ensuring remains within a 5% tolerance across all parameters. The proposed method improves adherence to the theoretical assumptions, enabling more reliable omnipolar signal reconstruction, thereby enhancing the characterization of intracardiac propagation patterns towards a more accurate localization of ablation targets.Clinical Relevance— Non-adherence to the theoretical assumptions in omnipolar technology can lead to inaccurate characterization of cardiac propagation patterns. The proposed method enhances compliance with these assumptions, yielding a more accurate representation of the omnipolar signal.
  • Optimizing patient selection for primary prevention implantable cardioverter-defibrillator implantation: utilizing multimodal machine learning to assess risk of implantable cardioverter-defibrillator non-benefit
    Item type: Journal Article
    Kolk, Maarten Z.H.; Ruiperez Campillo, Samuel; Deb, Brototo; et al. (2023)
    EP Europace
    Aims Left ventricular ejection fraction (LVEF) is suboptimal as a sole marker for predicting sudden cardiac death (SCD). Machine learning (ML) provides new opportunities for personalized predictions using complex, multimodal data. This study aimed to determine if risk stratification for implantable cardioverter-defibrillator (ICD) implantation can be improved by ML models that combine clinical variables with 12-lead electrocardiograms (ECG) time-series features. Methods and results A multicentre study of 1010 patients (64.9 ± 10.8 years, 26.8% female) with ischaemic, dilated, or non-ischaemic cardiomyopathy, and LVEF ≤ 35% implanted with an ICD between 2007 and 2021 for primary prevention of SCD in two academic hospitals was performed. For each patient, a raw 12-lead, 10-s ECG was obtained within 90 days before ICD implantation, and clinical details were collected. Supervised ML models were trained and validated on a development cohort (n = 550) from Hospital A to predict ICD non-arrhythmic mortality at three-year follow-up (i.e. mortality without prior appropriate ICD-therapy). Model performance was evaluated on an external patient cohort from Hospital B (n = 460). At three-year follow-up, 16.0% of patients had died, with 72.8% meeting criteria for non-arrhythmic mortality. Extreme gradient boosting models identified patients with non-arrhythmic mortality with an area under the receiver operating characteristic curve (AUROC) of 0.90 [95% confidence intervals (CI) 0.80–1.00] during internal validation. In the external cohort, the AUROC was 0.79 (95% CI 0.75–0.84). Conclusions ML models combining ECG time-series features and clinical variables were able to predict non-arrhythmic mortality within three years after device implantation in a primary prevention population, with robust performance in an independent cohort.
  • Reducing diverse sources of noise in ventricular electrical signals using variational autoencoders
    Item type: Journal Article
    Ruiperez Campillo, Samuel; Ryser, Alain; Sutter, Thomas M.; et al. (2026)
    Expert Systems with Applications
    Reducing electrophysiological (EP) signal noise is essential for diagnosis, mapping, and ablation procedures in patients with arrhythmias or conditions such as cardiomyopathies. However, traditional approaches have been suboptimal due to the varied sources of noise. We hypothesized that variational autoencoders (VAEs) can learn key components of ’clean’ electrophysiological signals by creating robust internal representations, thereby enabling automatic denoising of diverse noise in clinical recordings. We set out to apply a β-VAE model to a dataset of 5706 intra-ventricular monophasic action potential (MAP) signals, selected because their morphology is verifiable and measurable against a reference, from 42 patients with ischemic cardiomyopathy at risk for sudden death. We designed a noise library, and implemented baselines based on state-of-the-art clinical filtering techniques. The proposed β-VAE model was assessed for various noise types, including challenging non-stationary real EP noise. Comprehensive evaluation using general metrics and clinical action potential duration labels by domain experts revealed that our β-VAE outperformed current state-of-the-art filters in denoising efficacy, with key physiological information encoded in the reconstruction. We performed a sensitivity analysis that confirmed the robustness of the β-VAE model to increasing noise levels. These results demonstrate the ability of our model to denoise various sources, including those of time-varying nature. The application to well-studied MAPs verifies that clinically meaningful features were reconstructed in the EP context. This work enhances traditional signal processing approaches to ensure ’clean’ electrical signals, and may have promising applications for diagnosis, tracking therapy and prognostication in patients with EP disorders in real-world clinical environments.
  • EEG Tensorization Enhances CNN-Based Outcome Classification in Comatose Patients Following a Cardiac Arrest
    Item type: Conference Paper
    Ors-Quixal, R. Teodoro; Ruiperez Campillo, Samuel; Castells-Ramón, Francisco; et al. (2024)
    2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)
    Standard diagnostic methods for evaluating the severity of brain injuries resulting from cardiac arrest, such as the Glasgow Coma Scale, exhibit subjective biases that lead to potentially fatal misclassifications, where life-support systems are prematurely withdrawn from patients who might otherwise recover. This study utilizes an open dataset from the International Cardiac Arrest Research Consortium to develop and evaluate a 3D convolutional neural network (CNN) model for classifying outcomes in comatose patients after cardiac arrest. The electroencephalographic (EEG) signals from the dataset are preprocessed by resampling, filtering, and standardizing signal length (10 seconds) and channel count. The model’s architecture comprises 3D convolutional neural networks with subsequent layers for vectorization, compression, and further automatic feature extraction. Evaluation metrics focus on the area under the receiver operating characteristic curve, confusion matrix, accuracy, and F1 score. Results show that the 3D-CNN model outperforms existing 2D-CNN models in classifying outcomes for comatose patients, exhibiting a higher area under the receiver operating characteristic curve.
Publications1 - 10 of 41