Supervisory team
LSHTM
Nagasaki University
- Nobuyuki Nishikiori ([email protected])
- Nobuo Saito ([email protected])
Project
Background
Tuberculosis (TB) remains a global health threat. Many developed countries, including Japan, historically used mass X-ray campaigns to help reduce TB incidence. However, until recently, such campaigns were not recommended for current high burden countries. With the advent of ultra-portable digital chest X-ray systems supported by Computer Aided Detection (CAD) software to automate reading, there is renewed interest and investment in using this technology as a tool for TB screening.
Currently recommended CAD software systems use artificial intelligence (AI), specifically deep neural networks to interpret chest X-ray (CXR) images, detect abnormalities consistent with TB, and predict the likelihood of TB. Most of these AI algorithms provide a TB probability score ranging 0 to 100, with higher score being indicative of higher likelihood of having TB. However, concerns remain over the ability of these models to correctly process these images across diverse patient demographics. For example, the sensitivity and specificity of AI score across different algorithms tested varied by age groups, gender, as well as within and between countries. As TB prevalence varies strongly by e.g. gender, and previous TB history, this variation poses a challenge for wider roll-out of CXR+CAD screening.
To improve performance of CAD algorithms, locally relevant (i.e. from high TB burden countries) and well-described training data is crucial.
Proposed project
This project will examine the hypothesis that combining the CAD score with contextualized risk factors including socio-demographic characteristics (age, sex, ethnicity etc.) and available clinical history (duration of cough/fever, DM status, HIV status, history of TB) will improve performance, rather than the CAD software alone.
Broad objective of the PhD
To develop a model to predict the likelihood TB by utilizing CAD scores from digital chest X-rays, socio-demographic characteristics, and clinical history of individuals being screened for TB.
Specific objectives of the PhD:
- To quantify risk levels associated with socio-demographic characteristics and clinical history in high TB burden countries through review of existing literature
- To develop TB likelihood prediction model utilizing CAD scores from digital chest X-rays, socio-demographic characteristics, and clinical history of individuals from a high burden setting (Bangladesh)
- To compare performance of the newly developed model with the performance of TB likelihood predictions from only CAD scores at various thresholds
- To model the epidemiological impact and financial implications of implementing the combined algorithm in high burden settings.
The role of LSHTM and NU in this collaborative project
The four supervisors from the two institutions bring in distinct sets of expertise and experience which will be instrumental in training a capable student to develop skills required for this project. The supervisor at LSHTM (Dr Katherine Horton) is a world leading TB modeller with specialisation in community screening and gender differences. She will work as a primary supervisor, supported by Prof Houben, who has extensive experience at PhD supervision, and co-leads the world leading LSHTM TB Modelling Group. Both will ensure necessary training of the student required to develop skills in data analysis and mathematical modelling.
At Nagasaki University, Prof Nishikiori brings extensive experience in TB epidemiology and policy, gained through more than a decade of work with the World Health Organization. He led the design and evaluation of systematic TB screening initiatives and developed ScreenTB, a WHO decision-support tool used globally to optimise screening strategies based on epidemiological and cost considerations. He also led global work on TB patient cost surveys, linking economic burden assessment with programmatic decision-making. His expertise at the intersection of TB epidemiology, health-systems analysis, and programme implementation will guide the student in developing models that are both analytically rigorous and operationally relevant.
Dr Nobuo Saito, Associate Professor at the Institute of Tropical Medicine, Nagasaki University, complements this expertise with strong clinical and field-based perspectives on TB control in resource-limited settings. With a background in infectious diseases and extensive experience leading international research collaborations in Asia and Africa, he brings valuable insights into the practical implementation of TB interventions. Dr Saito will help ensure that modelled findings are grounded in real-world clinical and programmatic realities, strengthening the translational impact of the project.
The student will be primarily based at LSHTM, as a larger part of the work will require close supervision by the LSHTM supervisors, while maintaining active collaboration and guidance from Nagasaki University throughout the project period.
Particular prior educational requirements for a student undertaking this project
- A student having a medical degree along with demonstrable experience with quantitative analyses is important.
- Experience in TB would be desirable.
Skills we expect a student to develop/acquire whilst pursuing this project
The student will be expected to acquire necessary skills required for infectious disease modelling as well as modelling financial implications.