PET-CT scans and AI-read X rays show potential to predict tuberculosis diagnosis over 5 years

On World TB Day, researchers from CIDRI Africa have published landmark findings showing that advanced imaging technologies can detect evidence of asymptomatic tuberculosis (TB) in the lungs years before symptoms appear or routine tests are positive. The study, published in The Lancet Respiratory Medicine, is the largest of its kind to follow high risk individuals with sensitive imaging over five years.

Global estimates suggest that up to one quarter of the world’s population have been infected by the bacteria Mycobacterium tuberculosis. However, this figure is based largely on inference from tests showing an immune response to the bacterium, not direct evidence of infection. Most people with a positive immune response never develop disease. Predicting who will progress to TB remains one of the most urgent gaps in TB prevention. 

“Identifying those most likely to develop TB is crucial if we want to prevent transmission and intervene earlier” said first author Professor Hanif Esmail.

In the study by Esmail et al., researchers followed 250 HIV-negative, asymptomatic people in Khayelitsha, Cape Town, South Africa who were household contacts of drug-resistant TB cases. Each participant received an 18 fluorodeoxyglucose PET CT scan as well as an AI read digital chest X ray. Participants were monitored for up to five years.

During the follow up period, 18 individuals were diagnosed and treated for TB. Six were identified early through enhanced screening at the start of the study, five of whom would have been missed by routine rapid molecular testing, while the remaining 12 were diagnosed with TB after a median of 3 years. Many of these participants were still asymptomatic at the time bacteria was detected in their sputum, highlighting that transmission may occur before routine systems detect disease.

PET CT is the most sensitive imaging tool for research use and revealed a wide spectrum of lung abnormalities. However, those whose PET CT scans displayed a specific set of abnormalities associated with TB disease processes at the start of the study were more than 28 times more likely to be diagnosed with TB during follow-up compared with individuals whose scans appeared normal. Thus, whilst 205 of the 250 showed an immune response to the bacterium, it was the 29 with lung abnormalities on PET CT associated with TB that were at the highest risk of TB diagnosis.

“These findings position PET CT as a powerful research tool for understanding how TB progresses in the body. While the method is too costly and complex for large-scale public health use, its precision offers valuable insights for clinical research studies to develop improved diagnostics and therapeutics.”, said co-senior author Associate Professor Anna Coussens.

More immediately impactful, however, was the performance of AI interpreted chest X rays. Although less sensitive than PET CT, the AI readings showed good alignment with PET CT predictions, suggesting significant promise for mass screening.

“AI-read chest X rays could play a vital role in strengthening TB control strategies through mass-screening efforts.”, said co-senior author Professor Robert J Wilkinson.

The study highlights a critical advancement: the ability to identify people likely to have TB long before symptoms appear. With earlier detection comes the possibility of timely intervention—reducing transmission, preventing lung damage, and improving global TB prevention efforts.

Read the paper in The Lancet Respiratory Medicine "PET/CT benchmarked detection and 5-year progression of asymptomatic tuberculosis: a longitudinal, prospective cohort study"