Tuberculosis (TB) is an infectious disease, usually affecting the lungs, and remains the world’s leading infectious killer. Current TB treatment follows rigid protocols designed decades ago, with little scope for personalisation.

Our approach

This project will develop a more tailored approach by grouping patients according to bacterial burden and complicating conditions such as diabetes or HIV.

Using lung scans from a South African clinical trial, the project will create AI algorithms to automatically detect and map TB infection and assess its severity. Mathematical models will then simulate disease progression for each patient type, incorporating differences in immune response.

Aims and objectives

Working with clinical, biological, and computational experts, the project will refine these models using the latest laboratory data and validate them against trial results.

Once robust, the models will be used with doctors to design patient-specific treatment protocols that can later be tested in clinical trials, aiming to improve outcomes, reduce side effects, cut costs, and lower the risk of antibiotic resistance.

Methods

We’re combining mathematical analysis, AI, and computational modelling: developing neural network algorithms to analyse lung scans and automatically detect and map TB infection, building mechanistic systems of differential equations to describe within-host disease dynamics, and constructing hybrid individual-based models to capture spatial interactions between bacteria and the immune response.

These components are integrated with pharmacokinetic–pharmacodynamic frameworks to simulate treatment effects, with analytical methods and large-scale simulations used to explore disease progression and design personalised therapy strategies.

Project team

• Principal Investigator: Dr Ruth Bowness, Reader, Department of Mathematical Sciences
• Postdoctoral Researcher: Dr Roselyn Kaondera-Shava, Research Associate, Department of Mathematical Sciences
• Contributor: Dr Chris Rowlatt – IMI MIRA