Divergent Evolution of Tuberculosis Lesions During Treatment: A Longitudinal CT-Based Analysis of Progression and Regression Patterns

TL;DR

This study uses CT scans to track how tuberculosis lesions change during treatment, finding that certain patterns may indicate poor treatment outcomes.

Contribution

The study introduces lesion-level trajectory analysis using serial CT scans to identify treatment failure indicators in tuberculosis.

Findings

Five lesion volume patterns were identified, including 'Mix-D-I' associated with treatment failure.

66.7% of treatment success patients had multiple lesion patterns, compared to all treatment failure patients.

Mix-D-I lesions were more common in treatment failure cases and statistically linked to poor outcomes.

Abstract

Objectives: Lesion-level dynamics may reveal pulmonary tuberculosis (PTB) heterogeneity and help identify factors associated with treatment outcomes. Methods: A total of 288 serial Computed Tomography (CT) scans from 125 PTB patients were obtained from the National Institute of Allergy and Infectious Diseases (NIAID) TB Portals database (2008–2023). Lesions were segmented and annotated to obtain volume and imaging features, and a conservative longitudinal volume quantification method was used to characterize dynamic volume patterns. The proportion of lesions with different patterns was analyzed at the patient level to assess trajectory diversity. Firth’s penalized logistic regression was used to identify factors associated with treatment outcomes. Results: Among 435 lesions in 125 patients, five patterns emerged: Stable, Decrease, Increase, Mix-I-D (increase then decrease), and Mix-D-I (decrease then increase). Multiple patterns coexisted in 66.7% of treatment success patients and all treatment failure patients. Mix-D-I lesions were identified more frequently in treatment failure patients (25.0% vs. 1.4%, p = 0.027), and in multivariable analysis, the presence of Mix-D-I lesions was statistically associated with treatment failure (p = 0.024). Conclusions: PTB lesions showed high trajectory heterogeneity. The presence of Mix-D-I lesions may point to an unfavorable treatment course, suggesting lesion dynamics could serve as a potential indicator for poor outcomes. By quantifying lesion-level trajectories on serial CT scans, we extend PET/CT-based evidence and support the value of routine monitoring in clinical management of tuberculosis.

Figures