Cardiovascular
Analysis of rhythm information, heart-rate variability and haemodynamic changes.
Clinical AI Development
Our development scope covers physiological forecasting, multimodal modelling and the presentation of predictive information within clinical workflows.
Each project begins with a defined clinical question, intended patient population and proposed use of the software.
Physiological Parameter Forecasting
We develop methods to predict the direction and magnitude of changes in physiological parameters.
These predicted trajectories can then support assessment of deterioration risk, giving healthcare professionals more context than a risk category alone.
Multimodal Modelling
We investigate relationships across physiological signals, treatment information and clinical history.
- Patient-monitor measurements and waveforms
- Ventilator parameters and respiratory waveforms
- Infusion data and treatment context
- Connected wearables and biosensors
- Electronic health records and laboratory results
Inputs depend on clinical relevance, data availability and validation requirements.
Clinical Research Areas
Respiratory
Analysis of oxygenation, respiratory patterns and patient–ventilator interactions.
Cross-System Deterioration
Exploration of interacting physiological changes, including potential future research into sepsis-related deterioration.
From Development to Evidence
Retrospective testing
Assess prediction accuracy, calibration and failure patterns.
Silent prospective evaluation
Test live performance without directing care.
Usability evaluation
Assess interpretation, workload and potential use errors.
External validation
Assess performance across independent clinical settings.
Clinical use that influences care will require evidence and approvals appropriate to the intended purpose.
Responsible Development
We consider data quality, privacy, cybersecurity and human oversight throughout development.
Evaluation will examine clinically relevant measures such as sensitivity, positive predictive value, warning lead time and alert burden, alongside overall model performance.