The role of medical imaging in translational research into early pulmonary tuberculosis

Abstract

Early detection of tuberculosis (TB) is central to global efforts for TB care and prevention. Conventional symptom-based screening and sputum microbiology fail to identify a substantial proportion of cases, particularly those that are asymptomatic or have low bacillary burden. Non-invasive medical imaging offers a critical solution for visualizing pulmonary pathology before individuals with TB develop characteristic symptoms. Chest X-ray (CXR) has long been central to community screening, and recent advances in computer-aided detection (CAD) systems endorsed by the World Health Organization have improved scalability and reduced reliance on expert interpretation. However, limitations in sensitivity for the earliest stages of disease and reduced specificity in individuals with prior lung pathology persist, underscoring the need for further research to improve performance for early TB detection.

High-resolution modalities such as computed tomography (CT), Magnetic Resonance Imaging (MRI) and functional imaging approaches such as positron emission tomography (PET) and Single-Photon Emission Computed Tomography (SPECT) provide unparalleled insights into lesion dynamics, disease activity, and treatment response. They can be utilised to benchmark development of novel diagnostic tests for bacteriologically unconfirmed TB and define imaging correlates of early disease progression and resolution. Furthermore, emerging innovations in pathogen-specific radiotracers may enable localisation of viable bacilli in vivo. As a result, these new high-resolution imaging technologies offer transformative potential to address key knowledge gaps in the natural history, pathogenesis, dissemination, and therapeutic approaches to early pulmonary TB.

Together, clinical imaging provides a framework for the development and validation of clinically relevant biomarkers and quantitative readouts that capture early, asymptomatic TB pathology and disease activity prior to conventional microbiologic confirmation. Future work should focus on integrating advanced imaging with microbiology, host-response biomarkers, and artificial intelligence to define actionable imaging phenotypes that inform early diagnosis, risk stratification, and treatment decision-making across diverse settings.

Introduction

National prevalence surveys conducted across Asia and Africa over the past two decades consistently demonstrate that approximately half of individuals with undiagnosed, bacteriologically confirmed TB do not report symptoms.[1] These cases are typically detected through chest X-ray (CXR) screening, which reveals radiological evidence of pulmonary TB in otherwise asymptomatic individuals. Furthermore, globally, more than one-third of pulmonary TB diagnoses are bacteriologically unconfirmed.[2] Individuals with bacteriologically unconfirmed TB are often diagnosed based on CXR abnormalities when sputum tests are negative and no alternative cause is identified. The extent to which prevalent TB may be missed by symptom screening, and frequency with which clinical TB remains paucibacillary even after symptoms develop, underscore the critical role of imaging in detecting early stages of tuberculous pathology.

The range of tools available for imaging early TB is broad and growing. The utility of CXR in community-based TB screening is well established, but it has limitations, including variable sensitivity and specificity, and reliance on expert interpretation. Recent advances in artificial intelligence (AI) and computer-aided diagnostic (CAD) software are addressing these challenges by providing standardized and accurate interpretations in real time, thus improving scalability.[3] Beyond CXR, cross-sectional imaging modalities such as computed tomography (CT) offer superior resolution, enabling detailed characterization of pulmonary lesions.[4] Nuclear medicine techniques, including positron emission tomography (PET) with radio-labelled tracers, provide functional insights into disease activity and host-pathogen interactions.[5] Collectively, these imaging approaches not only support early diagnosis but also offer a unique window into TB pathogenesis, lesion dynamics, and treatment response—making them invaluable for translational research.

When using radiographic imaging to visualize TB pathology, no single feature is pathognomonic but characteristic patterns exist. In immunocompetent adolescents and adults, pulmonary disease typically begins in the upper lobes following failure of granulomatous containment.[6–8] Early changes include cellular exudates within terminal bronchioles, progressing to lipid-rich pneumonia as bronchi become obstructed.[9] On CT, these changes appear as centrilobular nodules connected by branching linear opacities, the classic “tree-in-bud” pattern, which can coalesce into lobular consolidation (Figure 1).[9, 10] On CXR, early disease manifests as poorly defined nodules and patchy upper-zone infiltrates.[8] As disease advances, pro-inflammatory cytokines like TNF-α, IL-1β, and IL-6 contribute to a process of necrosis and extracellular matrix degradation through hydrolytic enzymes and matrix metalloproteinases (MMPs) leading to cavitation, visible radiographically as radiolucent cavities surrounded by dense consolidation.[9, 10] Over the span of weeks to years, lesions may expand more slowly, or can regress spontaneously or with treatment, leaving fibrotic scars, volume loss, and calcification.[4, 11]

Figure 1.

Imaging of early stage of TB – Panels A, B and C show 3D volume rendering of fused FDG-PET/CT images of the lung highlighting metabolically active infiltrates (in red) in the upper lobe consistent with TB in asymptomatic HIV infected adults living in high TB burden settings in South Africa undergoing community screening for TB. The participant in panel C was culture positive at time of scan and participants in panels A and B culture negative (cohort described in detail in ref 4). Panel D shows baseline CXR and CT scan of an asymptomatic individual undergoing screening who was initially Xpert trace but then Xpert negative on repeat Xpert and sputum culture. Initial CXR and CT showed right apical 2–5mm nodules and fibrosis (green arrows). CAD (qXRv4) score 0.70. Progressed to Xpert positive (very low) and culture positive at 3-month follow up, with progression of right apical lesion on repeat CT (red arrow).

Given its utility for both diagnostic classification and pathophysiologic description, imaging thus serves a dual purpose, identifying early, often asymptomatic TB and providing mechanistic insights into disease evolution. These capabilities position imaging as a cornerstone for translational research in early TB, enabling development of diagnostics, evaluation of novel interventions, and deeper understanding of TB natural history. In this article we lay out advances in the role of medical imaging for translational research into early TB, describe recent insights and highlight the key knowledge gaps.

Conventional Imaging Modalities

CXR and CAD advances and limitations

Chest X-Ray has had a central role in TB screening for nearly a century.[12] Recent advances in machine learning techniques have led to the development of CAD systems that automate the interpretation of digital CXRs and are now endorsed by the WHO for use in screening.[3, 13] CAD systems use convolutional neural networks trained on large datasets of CXR’s to detect abnormalities associated with TB.[3] They typically assign a score from 0–100 or 0–1 (with a higher values indicating a greater likelihood of TB), from which a threshold score is used to identify those for further confirmatory bacteriological testing.[3] CAD represents a major advancement in TB screening in high burden settings. The lack of reliance on radiologists reduces inter-reader variability and enables screening at scale – a particular advantage in resource limited settings –and has led to widespread uptake of CAD in efforts to screen for TB in high-burden countries.[14]

However, there are some limitations to the ability of CXR and CAD to identify pulmonary TB disease at an early stage. Lesions must evolve to a minimum size and radiographic density to be detectable.[15] The visibility of small lesions can also be affected by location in relation to overlying soft tissue and bony structures. These factors make CXR relatively insensitive for early and mild forms of disease. Because most data on CXR and CAD performance for early TB derives from screening approaches that screen concurrently with symptoms and CXR before microbiological testing, microbiologically confirmed TB in asymptomatic individuals with false-negative CXRs is typically not captured and not considered in accuracy estimates. This verification bias results in preferential inclusion of individuals with radiographically apparent disease and systematically tends to exclude those with early or paucibacillary TB. Using data from such screening activities to evaluate CXR and CAD thus overestimates their sensitivity, particularly for asymptomatic TB.[16] Few studies have evaluated the sensitivity of CXR for early TB without these biases. In one such study, an observational cohort study of adult household contacts of pulmonary TB patients in South Africa, an investigator interpreted CXRs as abnormal for only 62% (31/50) of contacts with microbiologically confirmed TB overall, and only 56% (23/41) among the 82% who were asymptomatic.[17] These findings underscore the limited sensitivity of CXR for detecting microbiologically confirmed TB at an early stage, even when interpreted by experienced readers.

Conversely, CXR can also diagnose some early TB that sputum microbiological testing cannot, when CXR abnormalities suggestive of TB may precede microbiological positivity. This may occur because people with no or limited symptoms produce lower quality sputum specimens, or because early TB lesions have limited connectivity to airways and respiratory secretions. For instance, in a household contact cohort study in South Africa, more than 30% of individuals with abnormal CXR’s and CAD scores above threshold and initial negative sputum Xpert on spontaneous sampling were subsequently diagnosed with TB either by enhanced sputum investigation by sputum induction or during five years follow up.[18]

One option for improving the sensitivity of screening with CXR CAD is to reduce the threshold score that triggers TB testing. However, the trade-off with specificity is substantial, and number of confirmatory tests required for high sensitivity may be unaffordable in high-burden settings. For example, a cross-sectional study in Uganda performed non-symptom-gated screening using CAD qXR version 3.2 in community and health facility settings, offering follow-up Xpert MTB/RIF Ultra testing to all individuals with CAD scores above a low threshold of 0.1. Assuming an 0.1% prevalence of Xpert-positive TB at scores <0.1, the AUC for CAD in detecting TB in all those screened was 0.92 (95% CI 0.90–0.94). Had investigators used a more typical CAD score threshold of 0.5, they would have needed three times fewer confirmatory tests (for 5.4% rather than 16.9% of the population) and specificity would increase from 83.6% to 94.8%, but sensitivity would have dropped from 89.9% to 79.1% demonstrating how threshold setting may exclude a meaningful proportion of early disease.[19]

In addition to limitations in sensitivity, the appearance of early TB lesions on CXR are often also non-specific, challenging readers and CAD systems to differentiate active or progressive TB lesions from old or inactive TB lesions, or non-TB pathology in whom residual changes such as fibrosis, calcification and volume loss tend to lower CAD specificity.[18] Similarly, abnormalities caused by other conditions for example silicosis may produce a high CAD abnormality score that is not due to TB.[20] Children with pulmonary TB and individuals with advanced HIV are other groups for whom performance of CAD remains suboptimal as presentation of disease particularly in relation to localisation in the lung differs reducing AUC and sensitivity of standard thresholds.[21]

New and updated software versions have improved the accuracy of CAD for TB detection still, several research gaps remain. Key priorities include improving algorithms to better detect early TB, integrating demographic and clinical data to refine risk scores, and optimising CAD threshold selection. In addition, given that CAD systems are capable of identifying lung abnormalities unrelated TB, implementation of broader lung-pathology detection algorithms into TB screening practices requires evaluation for impact on clinical decision making and onward management.[20]

Point-of-care Ultrasound (POCUS) advances and limitations.

Point-of-care ultrasound (POCUS) for tuberculosis (TB) offers practical and conceptual advantages over chest radiography (CXR) for triage and early disease detection in resource-limited settings, including portability, lack of ionizing radiation, bedside deployability, and sensitivity to peripheral and pleural-based pathology that may be under-recognized or occult on CXR.[22] Recent work in developing streamlined POCUS acquisition protocols for TB have enabled reproducible image capture suitable for both human interpretation and downstream algorithm development.[23] These approaches align with established extrapulmonary frameworks such as the Focused Assessment with Sonography for HIV-associated Tuberculosis (FASH), which has demonstrated utility for detecting pleural, pericardial, lymph node, and solid-organ TB in HIV-endemic settings.[24] Whether POCUS can contribute meaningfully to early TB detection remains unknown and will require studies specifically designed to assess its performance in identifying subtle lung abnormalities preceding clinical or microbiological disease. Key priorities include multi-site external validation across diverse settings, where challenges analogous to CXR-CAD are expected, including limited detection of small and central lesions, reduced specificity in individuals with prior TB, and limited generalizability in children and people living with HIV or other immunocompromising conditions.[25]

Role for high resolution imaging in translational research

High resolution thoracic imaging has become a critical adjunct in translational research efforts toward tuberculosis (TB) elimination. It enables detection of early pathology, quantification of lesion burden, and longitudinal assessment of treatment response that complement, and in some contexts exceed, the resolution of conventional sputum-based and radiographic tools.

Computed tomography (CT) and PET/CT.

High-resolution CT is a cornerstone for assessing structural pathology associated with pulmonary TB. CT enables three-dimensional, high-contrast visualization of lung parenchyma and airways, allowing detection of pathology at an earlier stage and with greater anatomic specificity (Figure 1). CT enable identification of classic hallmarks of TB such as centrilobular nodules, tree-in-bud patterns, cavitation, consolidation, and residual fibrosis. Additionally, thin-slice, multi-phase protocols allow detailed delineation of pulmonary architecture for both baseline assessments of disease burden and serial assessments of structural resolution or progression.[10] As such, CT provides an essential anatomic reference standard for translational TB studies focused on early disease detection and characteristization.

Hybrid positron emission tomography/computed tomography (PET/CT) builds on anatomical CT by integrating metabolic imaging (Figure 1). Using radiotracers such as ¹⁸F-Fluorodeoxyglucosefluorodeoxyglucose (FDG), PET/CT detects regions of increased glucose metabolism associated with host inflammatory responses to Mtb, including lesions that may lack overt structural correlates on CT alone. PET/CT thus provides a window into spatially-resolved inflammatory activity and bacterial burden, making it valuable for treatment monitoring, lesion-level heterogeneity analysis, relapse risk stratification, and investigation of asymptomatic, subclinical, or incipient TB. This capability is particularly relevant for early stage of TB, where microbiologic confirmation may be absent and thus cannot be used as reference standard. FDG-PET/CT has been shown to identify metabolically active lesions in asymptomatic individuals with recent exposure or latent infection who subsequently progress to active disease.[4] In high-risk cohorts, such as household contacts of infectious TB cases, individuals with HIV, or populations undergoing active case finding reported sensitivity of FDG-PET/CT for culture-confirmed pulmonary TB frequently exceeds 80%, although estimates vary by study design, reference standard, and disease spectrum. Specificity is more variable, reflecting non-specific FDG uptake in other infectious, inflammatory, or malignant processes, a limitation shared with many host-response–based diagnostics.[26] This capability to identify metabolically active lesions in individuals without bacteriologically confirmed disease supports its use as a benchmark modality for evaluating emerging biomarkers, including blood transcriptional signatures and circulating mycobacterial DNA.[27]

Nevertheless, PET/CT as has some limitations. High cost, constrained access, and radiation exposure constrain general use. Non-specific tracer uptake in other infections, inflammation, or malignancy diminishes specificity, raising the risk of false positives. However, to some extent this can be mitigated by conducting studies in settings where the pre-test probability is high and confounding conditions is low. Radiation exposure from both CT and PET raises concerns for repeated imaging, healthy participants, and paediatric populations, necessitating rigorous ethical review and transparent risk communication in research protocols. In this context, PET/CT functions less as a population-level diagnostic tool and more as a high-resolution reference standard for early diagnostics development and pathogenesis research.

Single-Photon Emission Computed Tomography

Single-Photon Emission Computed Tomography (SPECT) provides an adjunct modality for three-dimensional functional imaging of TB. SPECT uses gamma-emitting tracers such as ^99mTc compounds and ⁶⁷Ga-citrate that may be taken up by physiological processes. Clinical and preclinical studies have demonstrated accumulation of ²⁰¹Tl, ^99mTc-sestamibi (MIBI), and ^99mTc-tetrofosmin in active TB lesions, resulting in variable ability to distinguish active from inactive pulmonary TB and, in some settings, tuberculous from malignancy.[28] ⁶⁷Ga-citrate scintigraphy, now often combined with SPECT or SPECT/CT, has historically been used to localize active pulmonary and extrapulmonary TB, assess disease extent, and monitor treatment response, but is limited by imperfect specificity, prolonged imaging protocols, and relatively high radiation dose.[28] Hybrid SPECT/CT improves lesion localization and characterization over planar imaging by correlating tracer uptake with CT-defined cavities, nodules, and miliary patterns in systemic granulomatous diseases, including TB. Additional SPECT agents, including radiolabeled anti-tuberculous drugs and immune cell–targeted tracers, have been explored primarily in preclinical or early translational studies to assess disease distribution or host immune trafficking.[29] Compared with FDG PET/CT, SPECT generally has lower spatial resolution and quantitative accuracy, but its benefits of lower cost, broad ^99mTc availability, make it attractive for research and selected clinical questions in high-burden, resource-limited settings. However, non-specific uptake of most tracers and radiation exposure have kept its role largely adjunctive and exploratory rather than frontline for TB diagnosis.

Magnetic resonance imaging (MRI)

Magnetic resonance imaging (MRI) is another advanced imaging modality that, although most commonly used for extrapulmonary and CNS TB, has a growing role in imaging pulmonary TB pathology with advances in motion correction and sequence optimization. MRI has several key operational advantages over CT alone or in combination with PET - including lack of ionizing radiation, making it particularly appealing for use in children, pregnant women, and serial imaging.[30] Emerging research in TB studies offers promising accuracy for detecting nodules, ground-glass opacities, tree-in-bud morphology, necrotic lesions, and cavitation, with some studies reporting substantial agreement with CT.[30, 31] MRI can also be combined with PET (PET/MRI)), which has the benefit of reducing overall ionising radiation dose while allowing combined anatomical and metabolic imaging in asymptomatic TB, however availability is limited.[32] Overall, despite its advantages, pulmonary MRI remains best suited to research and selected clinical scenarios, as low proton density of lung parenchyma, susceptibility to motion and air–tissue artifacts, and longer acquisition times currently limit routine clinical use.

Pathogen-specific and other novel radiotracers.

Traditional PET imaging in TB has relied heavily on FDG, which identifies increased glucose metabolism predominantly associated with host inflammatory responses rather than pathogen-specific activity. While FDG PET/CT has provided invaluable insights into lesion heterogeneity and treatment associated changes, it lacks specificity for Mtb itself, with uptake driven largely by activated host immune cell glycolysis, a feature shared across infectious, inflammatory, and malignant processes. This limited specificity constrains the ability of FDG to distinguish viable bacillary activity from sterile inflammation, motivating translational efforts to develop pathogen-specific radiotracers that exploit metabolic pathways unique to bacteria.[5, 33]

Para-aminobenzoic acid (PABA) is one promising scaffold for pathogen-specific PET tracers because it is incorporated into the bacterial folate biosynthesis pathway, which is absent in mammalian cells, and was identified in systematic screens as a substrate that is selectively taken up and metabolized by viable bacteria, including mycobacteria.[34] Radiolabelled PABA analogs (e.g., ¹¹C-PABA and ¹⁸F–labeled derivatives) have been developed to support PET imaging of bacterial infection in preclinical models, such as distinguishing sites of infection from sterile inflammation and monitoring treatment response.[35] A combined preclinical and early human study of non-tuberculous mycobacterial pulmonary infection (e.g., M. abscessus) demonstrate that ¹¹C-PABA PET can visualize infected lung lesions and track antibiotic response, purely.[36] Although PABA radiotracer data in M. tuberculosis–specific human disease remain to be established, the underlying biochemical basis, suggests that appropriately labelled PABA analogs (e.g., ¹⁸F-PABA) could offer pathogen-specific PET imaging in TB, potentially enabling detection of viable bacilli even in subclinical or low-burden settings.

Another potential tracer for specifically imaging M. tuberculosis leverages trehalose metabolism, a pathway central to mycobacterial cell wall biosynthesis and largely absent in mammalian tissues. Trehalose is processed by the mycobacterial Antigen 85 (Ag85) complex and related enzymes involved in constructing trehalose-containing glycolipids essential for mycobacterial viability. Early biochemical studies showed that M. tuberculosis incorporates trehalose analogs into its cell envelope when engineered as fluorescent probes, demonstrating an avenue for selective bacterial labelling.[37] Building on this principle, 2-[¹⁸F]fluoro-2-deoxytrehalose (¹⁸F-FDT) has been developed as a mechanism-based PET tracer that exploits mycobacterial trehalose processing pathways. Preclinical studies - including in murine and non-human primate TB models- demonstrate that ¹⁸F-FDT can specifically image TB-associated lesions and dynamically report changes with treatment, providing a more pathogen-specific signal than generic metabolic tracers like FDG.[38] Because trehalose transport and incorporation pathways are largely absent in mammalian tissues, trehalose-based tracers hold promise for distinguishing viable pathogen burden vs sterile inflammatory responses; addressing a key limitation of FDG-based imaging paradigms. Pathogen-specific radiotracers could undoubtably provide novel insights into disease in vivo. However, early stages of TB will be paucibacillary in nature resulting in potentially more limited uptake of tracers which may prove challenging for visualisation during the initial phases of development of these tools

Beyond FDG and pathogen-directed substrates, several host- or microenvironment-targeted radiotracers have been applied to TB. For example, TSPO-targeted tracers such as radioiodinated DPA-713 report macrophage-rich inflammation and have been used to quantify treatment-associated changes in murine TB.[39] ⁶⁸Ga-citrate PET/CT, an infection/inflammation tracer with transferrin/lactoferrin–related uptake, has shown accumulation in pulmonary and extrapulmonary TB lesions in pilot human studies.[40] SPECT tracers such as ⁹⁹mTc-ethambutol have also been evaluated for localizing TB lesions, although performance varies by setting and study design.[29]

Insights into early TB heterogeneity from high-resolution imaging.

High-resolution clinical imaging has transformed how we understand early disease biology in oncology: lesions can be detected and phenotyped before symptom onset, with imaging biomarkers capturing microenvironmental alternations – such as vascular remodelling, immune infiltration, and tissue heterogeneity - that precede clinical presentation. Quantitative imaging and radiomics are increasingly used to identify malignant lesions at an early stage and distinguish indolent from aggressive disease trajectories, and infer underlying biology, shifting the paradigm from symptom-triggered diagnosis to biology-triggered risk stratification.[29]

TB research is converging on a similar conceptual framework. Rather than a strict “latent vs active” dichotomy, imaging supports a continuum of disease states in which early pulmonary and nodal pathology can be present, metabolically active, and prognostically meaningful even when symptoms and sputum tests are negative (Figure 1). Longitudinal imaging studies indicate that some lesions may persist, wax and wane, or slowly progress over months to years before overt symptoms or high bacillary loads appear. In a five-year prospective household-contact study in South Africa, baseline PET/CT identified asymptomatic TB-pattern abnormalities in a substantial subset of household contacts, and those with lesions radiographically consistent with TB had markedly higher subsequent risk of culture-positive TB compared with contacts with normal lungs while symptom-based screening at baseline would have missed all of these cases.[11] This supports a key pathogenesis inference: early TB can be biologically active and prognostically meaningful without meeting conventional symptom or sputum thresholds.

Within this early disease spectrum, intrathoracic lymph nodes emerge as potentially informative compartments. PET/CT studies of recent contacts consistently demonstrate FDG-avid mediastinal and hilar lymph nodes, even in the absence of overt lung disease on chest radiograph, suggesting early immune activation and/or occult mycobacterial trafficking within lymphatic structures.[27] These findings emphasize that early TB pathogenesis is not confined to parenchymal lesions and that nodal biology may precede, or evolve independently of, airway-accessible disease detectable by sputum assays.

Key knowledge gaps

Trajectory

Bacteriologically negative TB is increasingly recognized as an important early TB state and a potential target for intervention. Its pre-infectious nature makes it a prime opportunity to avert future transmission, and its low prevalence relative to even earlier-stage targets such as Mtb infection (as measured by antigen reactivity) could make it a realistic target for large-scale treatment, if it could be detected accurately. Modelling results suggest that detecting and treating people with bacteriologically negative TB could allow active case-finding to achieve burden-reduction goals at least twice as quickly as with conventional CXR-based screening algorithms.[41] But intervening on this state will require first delineating features that predict progression and training CAD to detect them.

Among people with TB-suggestive CXR in historical cohorts, a systematic review estimated a ten-fold difference in risk of progression to bacteriological positivity depending whether the X-ray had active versus inactive disease features.[42] Thus, identifying radiographic features associated with progression is key, but these features are difficult to map onto the results of CAD algorithms. CAD rates the likelihood of TB along a single dimension intended to represent current disease (as identified by radiologist judgment and/or microbiological reference testing), and omission of progression information, from most CAD training datasets limits their accuracy for identifying bacteriologically negative TB or predicting its trajectory. Characterizing trajectory is also an issue for people previously treated for TB. They often have residual TB-related changes from prior disease that reduce the specificity of X-ray, but they are also recommended for screening based on their elevated risk for TB recurrence.

CAD-CXR algorithms for could be improved for early TB stages by training them against paired CT or PET/CT images (as radiological reference standard) from cohorts with longitudinal follow for microbiological progression. Furthermore development of algorithms to assess change in an individual’s imaging over time, could make CXR a more useful screening tool.

Treatment

Closely related to the question of trajectory is the need to better define treatment requirements for early TB that is diagnosed radiographically. Even with current screening practices, CXR-based screening interventions often result in clinical diagnoses of bacteriologically unconfirmed TB.[43] In current practice, these clinical diagnoses occur mainly in individuals highly TB-suggestive symptoms or particularly high-risk X-rays, and they typically receive a standard four-drug, six-month treatment regimen. As efforts to detect and treat early TB expand, and particularly if CAD improves in its ability to detect more subtle X-ray features with high risk of progression, it could be advantageous to treat such individuals with more limited treatment regimens. However, the appropriate regimen for a given X-ray-based diagnosis is not yet well defined. In historical trials of treatment for culture-negative TB detected by X-ray, regimens (many now obsolete) of a single drug or a very short duration were only partially effective at preventing progress to bacteriologically positive disease.[44] The RADIO-TB trial (NCT07163143) which is soon to start recruitment, is the first trial in 35 years in this population and will evaluate the 2, 3, 4, 5, and 6 months of the standard TB treatment regimen.[45] To reduce treatment requirements and thereby facilitate wider diagnosis and treatment of bacteriologically unconfirmed early TB, there is a need to better delineate the imaging features (such as indicators of low-burden or inactive disease or early treatment response) that may be associated with favourable response to shortened regimens. As microbiology cannot be used to evaluate treatment response in bacteriologically unconfirmed TB, CT and/or PET/CT may prove useful as surrogate marker of treatment response in phase 2 trials to assist with regimen development and evaluation of novel drugs in this area.

Infectiousness

Another knowledge gap is whether imaging features of early TB can help to characterize infectiousness. One of the strongest known correlates of infectiousness is sputum bacterial burden, but there is evidence that individuals with low or negative sputum bacterial burden may still generate aerosols with viable Mtb.[46] While these studies did not perform systematic radiologic characterization of sputum-negative aerosol-positive individuals with imaging, face-mask sampling with Mtb NAAT demonstrated that some individuals with detectable exhaled Mtb output had PET–CT–visible pulmonary lesions prior to sputum microbiological confirmation.[47] Among imaging features linked to infectiousness, lung cavitation is the best described, but it is typically linked to high sputum bacillary loads and its independent association with transmission is inconsistent[48]. It is possible that imaging could help to characterize the pathophysiologic changes that facilitate aerosolization and identify individuals who may be more likely to generate transmission at a given sputum bacterial burden – and, potentially, in those whose sputum markers are not yet positive.

Conclusion

Pulmonary TB begins as a localised process within the lung with host-pathogen interactions unfolding within the airways and surrounding tissue. Sampling of blood and respiratory secretions provide access to cells valuable for diagnosis and immunological insights, however, these approaches capture only a partial view of disease biology particularly during early stages of disease when lesions may be inaccessible and paucibacillary. Imaging offers a non-invasive, comprehensive, window into the site of disease from its inception Imaging. It anchors other diagnostic modalities by visualizing the precise site and extent of disease, while advanced techniques increasingly allow interrogation of metabolic activity and potentially pathogen-specific processes. Imaging can be repeated longitudinally, enabling detailed monitoring of lesion evolution, heterogeneity, and treatment response in ways that biological sampling alone cannot achieve.

Imaging is therefore a key tool in translational TB research, providing insights into early disease biology, progression, and therapeutic dynamics that are essential for accelerating the development of diagnostics, interventions, and strategies for TB elimination.