Postdoctoral Research Position(H/F)

Description du poste

Sepsis Prediction by Intelligent Continuous Evaluation (SPICE)

Artificial Intelligence for Continuous Physiological Monitoring in Critical Care

1. Context and Subject

Sepsis remains one of the leading causes of mortality worldwide and a major burden for intensive care units. Despite advances in monitoring, many episodes of hemodynamic deterioration or inappropriate fluid administration remain difficult to predict in real time. Continuous high-frequency physiological signals, including electrocardiography, invasive and non-invasive arterial pressure, plethysmography, and respiratory traces, contain rich yet underexploited information on circulatory responsiveness and impending instability.

The SPICE project, developed within the IHU Prometheus initiative, aims to build a multimodal, intelligent surveillance framework capable of identifying early physiopathological transitions in septic patients. A particular focus is placed on predicting fluid responsiveness (e.g., passive leg raising response), detecting cardiovascular mal-adaptation, and anticipating transitions between rescue, stabilization, and weaning phases.

Drawing inspiration from recent work on PLR detection through perfusion index analysis (Beurton et al., Crit Care 2019), deep temporal models for sepsis onset (Hyland et al., Nat Med 2020), and multimodal fusion for patient monitoring (Nguyen et al., IEEE TBME 2022), the project proposes to push beyond classical feature engineering and leverage modern generative and representation-learning approaches to uncover latent physiological signatures that are robust, explainable, and clinically actionable.

2. Position of the Problem

Early identification of hemodynamic deterioration relies today on threshold-based alarms and subjective interpretation of continuous data streams. These systems often fail to detect subtle precursors of circulatory collapse, suffer from alarm fatigue, and rarely adapt to individual patient trajectories.

Fluid responsiveness evaluation, although essential for avoiding fluid overload and guiding resuscitation, often requires invasive cardiac output monitoring. Non-invasive alternatives based on plethysmographic variability or perfusion index remain promising but suffer from noise, inter-patient variability, and limited generalizability across clinical contexts.

Meanwhile, AI-based methods have demonstrated impressive results in retrospective ICU datasets; however, most rely heavily on low-frequency vital signs, electronic health record data, or narrow temporal windows. High-frequency physiological waveforms, available at 100-500 Hz, remain largely underutilized despite their potential to reveal the microdynamics of perfusion and autonomic regulation.

The challenge is therefore to (i) structure and curate large-scale repositories of raw signals, (ii) extract reliable latent representations capturing patient-specific physiological states, (iii) detect clinically meaningful transitions with quantified uncertainty, and (iv) provide explainable outputs that align with the needs of bedside physicians. The postdoctoral researcher will address these challenges by combining signal processing, deep learning, Bayesian modeling, explainability tools, and ontologically guided clinical reasoning.

Profil

Profile and skills sought

PhD in AI, computer science, biomedical engineering, applied mathematics, or related fields.

Strong expertise in machine learning, deep learning, or signal processing.

Programming skills in Python, PyTorch/TensorFlow, and familiarity with HPC environments.

Experience with physiological data, ICU systems, or clinical AI is an asset.

Intitulé du poste proposé : Agent de laboratoire H/F

Poste basé à EVRY COURCOURONNES au sein du laboratoire IBISC

Secteur d’activité : Biotechnologie

Nature du contrat : CDD de 12 mois

Disponibilité : MArs 2026

Niveau d’études : Bac +5 et plus

Rémunération envisagée : Selon votre profil