Imaging lung lesions predicts TB treatment outcomes in marmosets

Researchers used PET/CT imaging to track lung lesions in marmosets infected with tuberculosis and treated with a range of drug regimens. Animals were assigned to 22 treatment arms for two months, and radiographic changes were quantified alongside bacterial counts per lesion. Unsupervised clustering combined imaging and bacterial data to create treatment response profiles at the lesion level. These profiles matched clinical outcomes, distinguished lesions that improved from those that failed, and indicated that imaging measures gave more insight than bacterial burden alone.

What the study examined

This work examined whether radiographic measurements of lung lesions could clarify how well tuberculosis drug regimens work. Infected marmosets were treated for two months across 22 different drug arms, including single drugs and combinations. PET/CT imaging was used to measure changes in lesions over the treatment period, and lesion-level bacterial burden was assessed at the end of the study.

Key findings

Quantitative imaging data and terminal bacterial counts were combined using unsupervised clustering to produce multivariate treatment response profiles. These profiles aligned with known clinical outcomes and offered lesion-level insight into when treatments succeeded or failed. The approach separated cavitary lesions that responded from those that did not or that worsened after the first month of therapy.

The study successfully predicted the observed inferiority of one shorter regimen compared with a longer standard regimen for cavitary lung disease. Overall, a blend of quantitative imaging measures provided more informative signals about treatment outcome than bacterial burden measured alone.

Why it matters

• Improved interpretation: Combining imaging changes with bacterial data created profiles that better reflected how individual lesions reacted to treatment.
• Lesion-level view: The method revealed that different lesions in the same subject can have divergent responses, highlighting complexity beyond aggregate bacterial counts.
• Clinical alignment: The profiles produced in animals matched clinical patterns observed in human treatment outcomes, offering a way to bridge preclinical and clinical data.