Defining an inter-reader reproducible radiomic feature set associated with treatment-related changes in PET/CT images of TB-infected marmosets

Radiomic analysis of preclinical tuberculosis imaging offers quantitative imaging features that may serve as biomarkers of lesion evolution and treatment response; however, feature robustness is highly sensitive to lesion size, segmentation variability, and acquisition parameters. This study aimed to define methodological considerations affecting radiomic feature reliability in marmosets infected with Mycobacterium tuberculosis and to identify a reproducible subset of features associated with longitudinal treatment-associated changes. Longitudinal PET/CT scans from infected marmosets, either untreated or treated with the standard of care regimen (isoniazid, rifampicin, pyrazinamide, and ethambutol), were analysed. Lesions were independently segmented by multiple readers, and radiomic features were evaluated for inter-reader reproducibility using Lin’s concordance correlation coefficient and screened for scanner-associated differences using the Mann-Whitney U-test. Linear mixed-effects models were then applied to characterise within-group temporal trajectories in treated and untreated lesions. Lesions < 16 mm³ yielded no reproducible shape features and substantially fewer reproducible second-order features than larger lesions. Sequential filtering for inter-reader reproducibility and detectable scanner-associated differences excluded 58.9% of the initial feature set. Linear mixed-effects models identified 176 features exhibiting significant time-dependent trajectories after multiple-testing correction, further reduced to 110 after excluding remaining scanner-level effects. Hierarchical clustering reduced this set to 26 representative features capturing distinct patterns of treatment-associated change. Variance decomposition indicated that lesion-level variability exceeded subject-level variability for the evaluated longitudinal measurements. The resulting feature set was further evaluated in three additional cohorts of animals treated with alternative regimens or agents. This study defines an inter-reader reproducible subset of predominantly PET-derived features that captures treatment-associated changes, supporting standardized radiomic pipelines for longitudinal assessment of therapeutic response in preclinical tuberculosis models.