Research Areas from the Laboratory of AI for Digital Health | Engineering Science Department | University of Oxford

These AI models assist GPs by identifying subtle gastrointestinal cancer patterns from longitudinal data. This enables real-time, personalised risk assessments at the point of care, streamlining timely specialist referrals and guiding appropriate investigations to expedite diagnosis for high-risk patients while minimising unnecessary testing.

Moving beyond static scoring, these AI models provide dynamic, real-time risk predictions across the entire cardiovascular care trajectory—from primary care to the ICU. By tracking evolving physiological patterns, the models identify stage-specific danger signals, enabling earlier intervention and transparent, data-driven decision-making for conditions like heart failure and myocardial infarction.

We develop models for early diagnosis and risk prediction that alert clinicians to patient deterioration up to 24 hours before critical events occur. We also provide personalised treatment recommendations by using reinforcement learning to optimise drug dosages and intervention timing based on individual patient trajectories. Finally, we leverage transferable knowledge to support the management of rare diseases in data-scarce intensive care environments.

We develop AI models for cardiovascular screening in HIV populations using low-cost wearables in resource-limited settings. By analysing pulse oximeter waveforms, these tools provide accessible diagnostics that outperform traditional risk scores. This enables life-saving cardiac care without requiring expensive imaging or specialised clinical equipment.

We develop models for kidney transplantation to support critical decisions on whether to accept or decline specific organ offers. By providing explainable survival predictions and quantifying uncertainty, these tools optimise global organ allocation and improve patient outcomes, ensuring AI-driven recommendations are both transparent and clinically trustworthy.