Mathematically modelling tuberculosis: using lung scans to map infection, and a hybrid individual-based model to simulate infection and treatment

Tuberculosis (TB) is an infectious disease that usually affects the lungs. It can develop when bacteria spread through droplets in the air. In the past TB or "consumption" was a major cause of death worldwide. After the discovery of antibiotics and general improvement in living conditions, prevalence of the disease fell. However, since the 1980s cases have been rising again. TB is now the biggest infectious disease killer (above HIV/AIDS and now COVID-19).

This project seeks to take a step forward in personalising TB treatment. Currently with treatment for TB disease, doctors must follow rigid treatment protocols that only allow for variations in patients' weights. These treatment regimens were defined years ago when very little was understood about this disease. We now know more about TB bacteria and how the infection dynamics can change depending on particular patients' immune responses. For example, people who have diabetes and/or HIV tend to have more complex and severe TB disease. We also know that the severity of infection, i.e. the amount of lung tissue affected, plays a part in how successful treatment will be.

This project seeks to group TB patients according to their bacterial burden, i.e. how much infection is present, and the presence of any other conditions (such as diabetes or HIV) that could make their TB disease more complex, in order to find optimal ways of treating them.
I will use a collection of lung scans taken from a clinical trial in South Africa to develop Artificial Intelligence (AI) algorithms to automatically identify TB infection in patients. This algorithm will be able to identify where in the lungs the infection appears and how severe it is. This will mean that in future TB doctors could take an individual TB patient's lung scan and feed it into the AI algorithm to automatically map that patient's TB infection onto a computer. Once on the computer, I will use mathematical modelling to simulate what would happen in that patient's lungs (also taking into account their particular immune response, by factoring in whether they are diabetic or HIV-positive). I have already developed mathematical models that are capable of simulating a typical immune response during TB infection and will work with relevant biologists to integrate the differences seen in infection dynamics when patients are also diabetic/HIV-positive.
Building mathematical models of this type is complex and there are many unknowns, this is why I will work closely with my biological collaborators to ensure that the latest laboratory data is used to quantify the processes involved. I will also work with mathematical/computational colleagues to use relevant techniques to help with model development, and to test how accurate the models are. I will also use additional data from the South African clinical trial to test model predictions.

Once I am confident that the AI algorithms and models are robust, I will work with doctors to try to find more patient-specific treatment protocols. This will mean in future that some patients won't need as much treatment (hence cutting costs and reducing side-effects for these patients), and some will need variations in the antibiotic combinations/doses that are currently prescribed. Ultimately this will help to increase treatment success, prevent future TB relapses, and reduce the chance of antibiotic resistance emerging.

Tuberculosis (TB) is the leading infectious killer (now above COVID-19). Currently treatment of the disease follows rigid protocols. This project will take a significant step towards personalising treatment, by taking disease burden and co-morbidities/co-infections into account before deciding treatment.
I will employ Artificial Intelligence techniques on PET/CT scans of patients' lungs to train a neural network to automatically identify the location and extent of TB disease. This will then inform the spatial domain, a computational representation of lungs, for a mathematical model. I have previously developed 2 mathematical frameworks: a hybrid individual-based model (IBM) concentrating on a small section of lung tissue that mimics TB granuloma formation. This model simulates individual bacteria and immune cells, as well as cytokines, oxygen and antibiotic dynamics. A second framework is a networked metapopulation model that simulates the entire lung, including environmental heterogeneity. In this fellowship I will combine these frameworks, embedding the IBM into the network model, where the IBMs will act as nodes. Information from individual PET/CT scans will inform node location, ensuring granularity where it is needed, i.e. more detail where infection is concentrated.
I will also develop the IBM framework to allow for immunological changes typical in subgroups of TB patients. I will concentrate on TB/HIV and TB/diabetes as these are the leading causes of complex TB disease. This will allow me to simulate different treatment options (via integrated Pharmacokinetic/Pharmacodynamic models) for particular subsets of patients. Specific questions: which patients could have a shorter course of treatment? Which patients would benefit from higher doses of or non-standard antibiotics? This work will help to find improved treatments: those that are cost-effective, have fewer sides effects, and result in fewer relapses and reduced antibiotic resistance emergence.