Starting date: October 2026
Application deadline date: March 20th, 2026
Final decision date: June 4th, 2026
Contact: hugues.patural@univ-st-etienne.fr, olivier.alata@univ-st-etienne.fr
Position Overview
This PhD project corresponds to a fully funded 3-year doctoral position within the framework of the Doctoral School EDSIS. The thesis will be conducted at Université Jean Monnet, Saint-Étienne, France, with an expected start in October 2026.
Scientific Context and Rationale
Premature birth and perinatal brain injury remain major risk factors for long-term neurodevelopmental impairments, including cognitive delay, motor dysfunction, and behavioral disorders. In current clinical practice, neurodevelopmental prognosis primarily relies on standardized psychometric scales such as the Bayley Scales of Infant and Toddler Development, which are administered several months or years after birth. While informative, these tools do not enable early risk stratification during the neonatal period, thereby delaying targeted interventions.
Recent advances in artificial intelligence (AI) have opened new perspectives for early prediction of neurodevelopmental outcomes using physiological signals. Electroencephalography-based approaches have shown promising results; however, EEG interpretation remains complex, time-consuming, and partly subjective. In contrast, continuous and non-invasive assessment of the autonomic nervous system through heart rate variability (HRV) derived from electrocardiographic recordings provides a robust and clinically accessible proxy of cortical and subcortical regulation.
Numerous studies have demonstrated strong associations between HRV indices and brain maturation, as well as between altered autonomic regulation and adverse neurodevelopmental outcomes. Leveraging recent advancements in deep learning and explainable AI, this project aims to develop predictive and interpretable models that link neonatal HRV patterns to neurodevelopmental trajectories assessed at two years of age.
PhD thesis subject
The primary objective of this PhD project is to design and validate explainable deep learning models capable of predicting neurodevelopmental alterations at two years of age from neonatal HRV data.
The specific objectives are:
- To characterize sympathetic and parasympathetic maturation profiles in premature infants using linear, spectral, and non-linear HRV indices.
- To develop regression and deep learning models predicting Bayley-III cognitive and developmental scores from neonatal HRV features.
- To integrate state-of-the-art explainability techniques to identify the most informative HRV biomarkers and enhance clinical interpretability.
- To assess the robustness and clinical relevance of the proposed models in infants with normal versus impaired neurodevelopmental outcomes.
Methodology
- Study Population and data collection
The study will include approximately 100 premature infants (gestational age < 37 weeks) recruited at term-equivalent age. Two subgroups will be defined according to cerebral integrity assessed by cranial ultrasound or MRI:
- Infants with routine brain imaging.
- Infants presenting cerebral lesions at high risk of sequelae (e.g., severe intraventricular hemorrhage, cystic periventricular leukomalacia, congenital brain malformations, hypoxic–ischemic encephalopathy).
Clinical perinatal data and longitudinal follow-up information up to two years of age will be extracted from standardized hospital databases. Neurodevelopmental assessment will be performed at two years of corrected age using the Bayley Scales of Infant and Toddler Development (Bayley-III) by certified neuropsychologists. Continuous neonatal ECG recordings will be collected over 12–24 hour windows in the neonatal intensive care unit. These recordings will be obtained using bedside monitors connected to a centralized and secure data storage infrastructure.
- HRV Analysis
HRV features will be extracted using validated software and will include:
- Time-domain indices (e.g., SDNN, SDANN, RMSSD, pNN20, SD1, SD2).
- Frequency-domain indices (e.g., total power, VLF, LF, HF).
- Non-linear indices (e.g., detrended fluctuation analysis, entropy-based measures, fractal and chaos-related indices).
- Artificial Intelligence and Explainability
Feature normalization and statistical preprocessing will be applied before the modeling process. Initial regression models will be developed to identify predictive HRV markers. These models will subsequently be extended to deep learning architectures, including ensemble approaches.
Explainable AI techniques will be integrated to quantify the contribution of individual HRV features to model predictions, thereby ensuring transparency and facilitating clinical interpretation. Model performance will be evaluated using appropriate cross-validation strategies and clinically relevant outcome metrics.
Expected Outcomes and Impact
This project is expected to deliver:
- Novel predictive models enabling ultra-early identification of infants at risk of neurodevelopmental impairment.
- Clinically interpretable HRV biomarkers of autonomic and neurodevelopmental maturation.
- Scientific publications in high-impact peer-reviewed journals and presentations at international conferences.
In the longer term, the proposed approach could be integrated into routine neonatal monitoring systems as a decision-support tool, contributing to personalized follow-up strategies and early intervention programs.
Work Environment and Supervision
The PhD will be conducted within the SAINBIOSE-DVH laboratory (INSERM UMR 1059), in close collaboration with the Hubert Curien Laboratory (CNRS UMR 5516). The project will benefit from a multidisciplinary environment combining expertise in neonatology, physiology, signal processing, artificial intelligence, and explainable machine learning.
The doctoral candidate will be supervised by:
- Prof. Hugues Patural (SAINBIOSE-DVH), primary supervisor.
- Dr. Sébastien Celle (SAINBIOSE-DVH), co-supervisor.
- Prof. Olivier Alata (Hubert Curien Laboratory), co-supervisor.
Candidate Profile
The ideal candidate holds a Master’s degree in computer science, biomedical engineering, applied mathematics, or a related field. Strong skills in data analysis, signal processing, machine learning, and programming (Python, MATLAB, C/C++) are expected. Prior experience or strong interest in deep learning and medical applications will be considered a significant asset.
Career Perspectives
The PhD will provide advanced training at the interface of digital health, physiology, and artificial intelligence, opening career opportunities in academic research, clinical research, or industrial R&D in biomedical data science and AI-driven healthcare.
Salary Net : 1650€ per month for 3 years.
Additional paid teaching activities can be envisaged on demand.
Application process : The application should include the following documents:
– Letter of intent
– Grades and ranking during Master 1 and Master 2
– Scientific CV – List of publications (if applicable)
– Names of Referees (at least 2)
