NeoCurve
Predicting Oxyhaemoglobin Dissociation Curve Shift in Preterm Infants.
NeoCurve bridges physiology, clinical neonatology, and artificial intelligence to improve the understanding and bedside assessment of bronchopulmonary dysplasia (BPD) in preterm infants. The project, conducted as part of my MSc in Medical Informatics at the Universitäts-Kinderspital beider Basel (UKBB), aimed to develop a machine learning framework capable of predicting the oxyhaemoglobin dissociation curve (ODC) shift — a physiologically grounded marker of pulmonary gas-exchange efficiency.
Traditional methods for estimating this curve require multiple paired measurements of inspired oxygen pressure (PiO₂) and arterial oxygen saturation (SpO₂), which limits their feasibility in daily clinical use. To address this, NeoCurve introduced a data-efficient, explainable ML model that predicts the ODC shift from as little as one or two non-invasive SpO₂–PiO₂ data points.
The project integrated feature engineering, physiological modeling, and rigorous external validation using multi-centre neonatal datasets. The final linear regression-based model achieved strong performance (MAE = 0.63 kPa on test set; 0.49 kPa on validation set) and robust generalisation across datasets, enabling a simplified yet physiologically interpretable representation of pulmonary impairment.
In addition to the core model, the project delivered two functional software tools:
- a Streamlit-based annotation platform for clinician-guided labelling and data curation, and
- a React/Flask web application deploying the trained model as a lightweight decision-support tool for bedside use.
By combining clinical realism, physiological interpretability, and reproducible AI design, PhysioAI demonstrates how machine learning can translate complex physiological principles into actionable bedside insights, providing a foundation for future non-invasive monitoring and early risk stratification in neonatal care.
The prototype of this webapplication is available here.
This work, a collaboration with the University Children’s Hospital Basel (UKBB) is currently in preparation for publication.
