{"_asset":{"@search.score":1,"Name":"Diagnostic accuracy of lung ultrasound for pulmonary tuberculosis in out-patients from Papua New Guinea, a resource-limited setting","AlternateTitleCommaDelimited":"","Id":"9463","AssetLocationUri":null,"Abstract":"

OBJECTIVE

To evaluate the diagnostic accuracy of lung ultrasound (LUS) for pulmonary tuberculosis (PTB) in Papua New Guinea.

METHODS

Prospective observational study including out-patients with presumptive PTB (cough ≥2 weeks) attending a TB clinic in Port Moresby, Papua New Guinea. Participants underwent clinical assessment, GeneXpert Ultra-MTB/RIF, Chest X-ray (CXR), and LUS. Diagnostic accuracy of LUS findings was assessed using GeneXpert as reference standard. Accuracy of CXR and agreement with LUS were also evaluated. A LUS scoring system was developed using LASSO logistic regression.

RESULTS

Between May 2022 and May 2023, 512 participants were enrolled and 488 with GeneXpert were included in the analyses. Median age was 30 [IQR 23-44] and 55% were male. GeneXpert was positive in 149 (30.5%). Combined LUS findings sensitivity was 92.6% (95%CI 90.1–95.1) and specificity 46.8% (95% CI: 42.1–51.5). CXR sensitivity was 91.3% (95% CI: 86.6–93.9) and specificity 65.1% (95% CI: 60.7–69.6). Agreement between LUS and CXR was moderate (kappa=0.48). The LUS score achieved 82.8% sensitivity and 84.2% specificity.

CONCLUSIONS

demonstrates high sensitivity for detecting PTB, equivalent to CXR, but low specificity. LUS may have diagnostic value, particularly where radiography is unavailable. The scoring system that showed higher specificity should be validated in future studies.

","PublicationOrPresentationDate":"2026-02-16T10:00:00Z","LastModifiedDate":"2026-01-27T20:45:07.199Z","IsDeleted":false,"DigitalObjectIdentifier":"10.1016/j.ijid.2025.108239","ContentTypeName":"Research","ParentContentTypeName":"Journal Article","JournalName":"International Journal of Infectious Diseases","ContentType":"Journal Article > Research","AvailabilityTypeId":1,"AvailabilityType":"Full Text","JournalIssue":"","JournalVolume":"","JournalPages":"108239","JournalAbbreviation":"Int J Infect Dis","Conference":null,"ConferenceStartDate":null,"ConferenceEndDate":null,"content":"\nInternational Journal of Infectious Diseases 163 (2026) 108239 \n\nContents lists available at ScienceDirect \n\nInternational Journal of Infectious Diseases \n\njournal homepage: www.elsevier.com/locate/ijid \n\nDiagnostic accuracy of lung ultrasound for pulmonary tuberculosis in \n\noutpatients from Papua New Guinea, a resource-limited setting \n\nMaria Lightowler 1 , ∗, Juno Min \n\n2 , Anna Nape \n\n3 , Thelma Galowa \n\n3 , Carolyn Hemo \n\n3 , \nErin Stratta \n\n4 , Matthew Fentress 5 , Stefan Weber 6 , Mischa Huson \n\n7 , Laura Sannino \n\n8 , \nSarala Nicholas 1 , Helena Huerga \n\n1 \n\n1 Epicentre, Paris, France \n2 Médecins Sans Frontières, Amsterdam, The Netherlands \n3 Médecins Sans Frontières, Port Moresby, Papua New Guinea \n4 Médecins Sans Frontières, New York, NY, USA \n5 Contra Costa/University of California, San Francisco, CA, USA \n6 Department for Infectious Disease and Tropical Medicine, University Hospital Heidelberg, Heidelberg, Germany \n7 Erasmus Medical Center, Rotterdam, The Netherlands \n8 Médecins Sans Frontières, Paris, France \n\na r t i c l e i n f o \n\nArticle history: \n\nReceived 1 October 2025 \n\nRevised 14 November 2025 \n\nAccepted 18 November 2025 \n\nKeywords: \n\nPulmonary tuberculosis \n\nLung ultrasound \n\nDiagnostic accuracy \n\nPapua New Guinea \n\nResource-limited settings \n\na b s t r a c t \n\nObjectives: to evaluate the diagnostic accuracy of lung ultrasound (LUS) for pulmonary tuberculosis (PTB) \n\nin Papua New Guinea (PNG). \n\nMethods: Prospective observational study including outpatients with presumptive PTB (cough ≥2 weeks) \n\nattending a TB clinic in Port Moresby, PNG. Participants underwent clinical assessment, GeneXpert Ultra- \n\nMTB/RIF, chest X-ray (CXR), and LUS. Diagnostic accuracy of LUS findings was assessed using GeneXpert \n\nas reference standard. Accuracy of CXR and agreement with LUS were also evaluated. A LUS scoring sys- \n\ntem was developed using LASSO logistic regression. \n\nResults: Between May 2022 and May 2023, 512 participants were enrolled, and 488 with GeneXpert were \n\nincluded in the analyses. Median age was 30 [inter-quartile range 23-44], and 55% were male. GeneXpert \n\nwas positive in 149 (30.5%). Combined LUS findings sensitivity was 92.6% (95% confidence intervals [CI] \n\n90.1-95.1) and specificity 46.8% (95% CI: 42.1-51.5). CXR sensitivity was 91.3% (95% CI: 86.6-93.9) and \n\nspecificity 65.1% (95% CI: 60.7-69.6). Agreement between LUS and CXR was moderate (kappa = 0.48). The \n\nLUS score achieved 82.8% sensitivity and 84.2% specificity. \n\nConclusion: LUS demonstrates high sensitivity for detecting PTB, equivalent to CXR, but low specificity. \n\nLUS may have diagnostic value, particularly where radiography is unavailable. The scoring system that \n\nshowed higher specificity should be validated in future studies. \n\n© 2025 The Authors. Published by Elsevier Ltd on behalf of International Society for Infectious Diseases. \n\nThis is an open access article under the CC BY-NC-ND license \n\n( http://creativecommons.org/licenses/by-nc-nd/4.0/ ) \n\n1\n\nt\n\na\n\ne\n\nd\n\nfi\n\na\n\nF\n\ni\n\np\n\nl \n\na\n\nu\n\nd\n\nt\n\nh\n\n1\n\nl\n\n. Introduction \n\nTuberculosis (TB) remains one of the most devastating infec- \n\nious diseases globally, ranking as the leading cause of death from \n\n single pathogen. In 2023, the World Health Organization (WHO) \n\nstimated 10.8 million new TB cases and 1.25 million TB-related \n\neaths worldwide. Notably, only 8.2 million of these cases were of- \n\ncially diagnosed, underscoring a substantial gap in case detection \n\nnd access to diagnostic services [ 1 ]. \n∗ Corresponding author: Maria Lightowler, 34 Avenue Jean Jaurès, 70519 Paris, \n\nrance. \n\nE-mail address: maria.lightowler@brussels.msf.org (M. Lightowler) . \n\no\n\nt\n\nd\n\ni\n\nttps://doi.org/10.1016/j.ijid.2025.108239 \n\n201-9712/© 2025 The Authors. Published by Elsevier Ltd on behalf of International Socie\n\nicense ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ) \nPapua New Guinea (PNG) bears a substantial high TB burden, \n\nncluding multidrug and rifampicin-resistant TB. In 2022, PNG re- \n\norted an estimated TB incidence of 424 cases per 10 0,0 0 0 popu- \n\nation, one of the highest in the Western Pacific region [ 1 ]. Oper-\n\ntional challenges in achieving bacteriological confirmation, partic- \n\nlarly in rural and remote regions, contribute to delayed or missed \n\niagnoses [ 2 ]. \n\nEffective TB diagnosis is essential to closing the current detec- \n\nion gap. Diagnostic innovations in recent years include the devel- \n\npment of WHO-endorsed rapid molecular assays that can simul- \n\naneously detect Mycobacterium tuberculosis and resistance to key \n\nrugs [ 3 ]. Chest X-ray (CXR) is recommended by WHO as screen- \n\nng and diagnostic tool and is particularly useful when bacteri- \nty for Infectious Diseases. This is an open access article under the CC BY-NC-ND \n\nhttps://doi.org/10.1016/j.ijid.2025.108239\nhttp://www.ScienceDirect.com\nhttp://www.elsevier.com/locate/ijid\nhttp://crossmark.crossref.org/dialog/?doi=10.1016/j.ijid.2025.108239&domain=pdf\nhttp://creativecommons.org/licenses/by-nc-nd/4.0/\nmailto:maria.lightowler@brussels.msf.org\nhttps://doi.org/10.1016/j.ijid.2025.108239\nhttp://creativecommons.org/licenses/by-nc-nd/4.0/\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\no\n\nr\n\ns\n\nh\n\nc\n\n[\n\nC\n\ni\n\nl\n\n[\n\nt\n\ni\n\nd \n\nc\n\nT \n\nr\n\ns \n\nh\n\ni\n\na\n\n1\n\nd\n\ng\n\no\n\ne\n\nu\n\nfi\n\ns\n\na\n\na\n\na\n\nr\n\nb\n\na\n\ni\n\nt\n\nt\n\nt\n\nm\n\nt\n\n2\n\n2\n\ni\n\nT\n\nb\n\nc\n\ns\n\n2\n\no\n\nc\n\nr\n\nC\n\na\n\nt\n\nc\n\n<\n\np\n\ne\n\nd\n\n2\n\nm\n\nc\n\ns\n\nC\n\ne\n\n2\n\n2\n\nt\n\nc\n\nh\n\nh\n\nb\n\nd\n\nc\n\ni\n\n1\n\nu\n\nw\n\n(\n\ni\n\nt\n\ne\n\na\n\nc\n\nw\n\na\n\ns\n\nf\n\na\n\nd\n\nt\n\n2\n\nw\n\ns\n\nm\n\no\n\nt\n\nm\n\nt\n\nv\n\nd\n\nT\n\n2\n\nb\n\np\n\no\n\ni\n\nc\n\nt\n\nlogical confirmation is lacking and in children [ 4 ]. CXR is also \n\necommended for TB screening and has a higher sensitivity than \n\nymptom screening alone. Despite its utility, CXR deployment in \n\nigh-burden, resource-limited settings is often constrained by high \n\nosts, limited infrastructure, and shortage of trained radiologists \n\n 5 ]. While artificial intelligence (AI)-driven software for automated \n\nXR interpretation has been recommended by WHO for TB screen- \n\nng [ 6 ], its widespread implementation remains limited by cost and \n\nogistical barriers in low-resource settings [ 7 ]. \n\nUltrasound (US) offers several advantages, including lower cost \n\n 8 ], safety profile and radiation-free [ 9 ], and minimal infrastruc- \n\nure [ 10 ]. Advances in portability, image resolution, and affordabil- \n\nty have facilitated the use as a point-of-care modality performed \n\nirectly by clinicians (i.e., point-of-care US or POCUS) [ 11 ]. US is in-\n\nreasingly being used to support the detection of extrapulmonary \n\nB (EPTB) in people living with HIV [ 12 ]. It aids in identifying pleu-\n\nal and pericardial effusions, abdominal lymphadenopathy, and le- \n\nions in the liver, spleen, and kidneys suggestive of TB [ 13–15 ] and\n\nas also shown utility in detecting mediastinal lymphadenopathy \n\nn paediatric TB [ 16 ]. Some studies have reported lung US (LUS) \n\nbnormalities in patients diagnosed with pulmonary TB (PTB) [ 17–\n\n9 ] and have proposed LUS as a potential triage tool for PTB [ 20 ]. \n\nDespite these promising findings, the evidence regarding the \n\niagnostic accuracy of LUS for PTB remains limited and hetero- \n\neneous. A 2021 systematic review of six studies (five in adults, \n\nne in children) reported variable sensitivity using different ref- \n\nrence standards and high risk of bias. In adults, subpleural nod- \n\nles (SUNs) and lung consolidation emerged as the most sensitive \n\nndings, although specificity data was scarce [ 21 ]. In a paediatric \n\ntudy from a high TB/HIV burden setting, LUS sensitivity was 64% \n\nnd specificity 42.7% using CXR as the reference standard [ 22 ], and \n\n proof-of-concept study in adults reported LUS sensitivity of 91% \n\nnd specificity of 46% against bacteriological confirmation [ 23 ]. \n\nThis study aims to expand the limited evidence regarding the \n\nole of LUS in TB diagnosis and its potential use as an accessi- \n\nle, point-of-care imaging modality, particularly when CXR is un- \n\nvailable. We evaluated the diagnostic accuracy of LUS for PTB us- \n\nng bacteriological TB confirmation as reference standard in outpa- \n\nients with presumed PTB in PNG. We also assessed the diagnos- \n\nic accuracy of CXR using the same reference standard, as well as \n\nhe agreement between LUS and CXR in identifying TB. Further- \n\nore, we developed a novel diagnostic scoring system to enhance \n\nhe performance of LUS for PTB diagnosis. \n\n. Methods \n\n.1. Study setting and design \n\nThis prospective observational diagnostic study was conducted \n\nn the TB clinic at the Gerehu Hospital in Port Moresby, PNG. \n\nhe TB clinic is an outpatient Ministry of Health facility that has \n\neen supported by Médecins Sans Frontières (MSF) since 2016. The \n\nlinic provides diagnostic and treatment services for both drug- \n\nusceptible TB and drug-resistant TB. \n\n.2. Study population \n\nThe study population consisted of individuals aged 10 years and \n\nlder who presented to the Gerehu TB clinic for evaluation and TB \n\nare. Inclusion criteria were: (1) presumptive PTB (defined as self- \n\neported cough lasting 2 weeks or more) or clinician referral for \n\nXR. Exclusion criteria were: (1) patients requiring inpatient man- \n\ngement; and (2) individuals already receiving TB treatment at the \n\nime of screening. All adult participants provided written informed \n\nonsent (or written informed parental consent and child assent if \n\n 18 years). \n2\n\nParticipants were mostly enrolled consecutively among eligible \n\natients. However, on days when the number of eligible patients \n\nxceeded US capacity, a maximum of four participants were ran- \n\nomly selected to maintain imaging quality. \n\n.3. Study procedures \n\nAll consenting participants underwent an initial clinical assess- \n\nent, including medical history, evaluation of TB exposure, physi- \n\nal examination, and sputum collection for Xpert MTB/RIF Ultra as- \n\nay (Cepheid, Sunnyvale, CA, USA). All participants underwent both \n\nXR at an accredited radiology facility and LUS conducted on-site \n\nither on the same or following day by a study-trained clinician. \n\n.3.1. Ultrasound \n\n.3.1.1. Scanning technique. US was performed by two newly \n\nrained clinicians who completed a 10-hour introductory POCUS \n\nourse led by a certified instructor, followed by a 2-week, 50- \n\nour hands-on training. Competency was demonstrated through 25 \n\nigh-quality scans covering the full protocol, reviewed with feed- \n\nack by the instructor, followed by 25 additional scans indepen- \n\nently reviewed remotely. After meeting these requirements, both \n\nlinicians jointly performed all study US, each contributing approx- \n\nmately 50% of scans. \n\nA portable M-Turbo US machine (Sonosite, Fujifilm) with a 6.0- \n\n3.0 MHz linear transducer (“Small Parts” preset, 4 cm depth) was \n\nsed for initial scans. For deeper structures or study participants \n\nith a larger body habitus, a 2.0-5.0 mHz curvilinear transducer \n\nabdominal preset) was used. Scans were conducted with patients \n\nn supine or seated positions (pleural effusion measurements, pos- \n\nerior views). \n\nSixteen lung zones were systematically examined ( Figure 1 ). In \n\nach zone, longitudinal and transverse scans were performed, with \n\nt least one 10-second longitudinal video clip recorded. Pathologi- \n\nal findings (e.g., consolidation, cavity, effusion) were documented \n\nith images and measured, and mapped to specific zones to en- \n\nble separate tracking of overlapping pathologies. \n\nAdditionally, a subxiphoid view was obtained in the supine po- \n\nition to assess and measure for pericardial effusion, using a low- \n\nrequency probe angled cranially. Alternatively, a parasternal long- \n\nxis view was used. For consistency, we refer to the entire proce- \n\nure (including pleural and pericardial scans) as LUS throughout \n\nhis manuscript. \n\n.3.1.2. LUS interpretation. Pseudonymized LUS images and clips \n\nere remotely reviewed by POCUS experts (experienced re- \n\nearchers in TB-specific US interpretation) and findings docu- \n\nented in standardized electronic case report forms (CRFs). Based \n\nn the combination of these findings, three nonexclusive diagnos- \n\nic categories were predefined per-protocol for TB detection, one \n\nain and two alternative definitions ( Table 1 ; Figure 2 ). \n\nDefinitions were adapted from prior studies [ 18 , 19 ] and refined \n\nhrough research expert discussions to reach consensus. \n\nConsolidation was subcategorized as ≤1 or > 1 cm. To reduce \n\nariability in interpretation between SUN and small consolidations, \n\netailed classification criteria were standardized (Supplementary \n\nable 1). \n\n.3.2. Chest X-ray \n\nPseudonymized DICOM CXR images were remotely interpreted \n\ny an expert radiologist (15 years of experience in TB CXR inter- \n\nretation) for the presence of cavities, miliary pattern, reticulon- \n\ndular infiltrates, lobar or segmental consolidation, nodular lesions, \n\nntrathoracic lymphadenopathy, pleural effusion, and an enlarged \n\nardiac silhouette with a “water-bottle” configuration (Supplemen- \n\nary Table 2). Finding location was specified (apical, upper, or lower \n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\nFigure 1. Lung zones and apical sweep zones (16 zones): anterior (mid-clavicular), lateral (mid-axillary), and posterior (mid-scapular) zones on both sides, each divided into \n\nupper (3, 1, 5, 7, 11 and 9) and lower zones (4, 2, 6, 8, 12 and 10), plus bilateral apical sweeps (1a, 5a, 9a, 10a). Scanning began in the supine position (zones 1-8), followed \n\nby seated posterior views (zones 9-16). X: Subxiphoid view. \n\nTable 1 \n\nDefinitions of individual ultrasound findings and predefined combination of ultrasound and chest X-ray findings for detecting TB. \n\nDefinition \n\nIndividual ultrasound findings \n\nSubpleural nodule (SUN) • Well-defined, round or oval nodule immediately beneath the pleura \n• Vertical diameter < 1 cm (measured from the pleural line) \n• Hypoechoic interior with no internal echoes \n• Clear hyperechoic posterior wall \n• Posterior acoustic enhancement \n\nConsolidation • Hypoechoic area within lung parenchyma \n• Ill-defined margins with irregular, ragged posterior border \n\n◦ Speckled internal appearance, possibly with hyperechoic punctate or linear foci \n\n(air bronchograms) \n• Posterior acoustic enhancement inconsistent \n\nCavity • Consolidation containing a central hypoechoic, avascular area (no flow on color Doppler) \n\nMiliary pattern • Diffuse subpleural granular appearance with multiple B-lines originating from subpleural \n\ngranules in at least one lung zone \n\nPleural effusion • Fluid in the pleural space (between the parietal and visceral pleura) \n• Measured > 0.5 cm in depth (the diaphragm to lung) in seated position if feasible \n• Simple effusion: entirely anechoic and free-flowing \n• Complex effusion: contains echogenic debris (e.g., fibrin strands) or is loculated and not \n\nentirely anechoic \n\nPericardial effusion • Fluid in the pericardial space, measured > 0.3 cm between the right ventricle and parietal \n\npericardium \n\n◦ Simple effusion: is entirely anechoic \n\n◦ Complex effusion contains echogenic material or is loculated, not free-flowing \n\nPredefined combination of ultrasound findings for detecting TB \n\nHighly suggestive of TB (main definition) • Presence of any of the following findings on US—SUN in ≥2 zones (with at least one in the \n\napical or upper zone), consolidation > 1 cm in any region, miliary pattern, cavity, pleural \n\neffusion or pericardial effusion > 1 cm \n\nHighly suggestive (alternative definition 1) • SUN in ≥2 zones (with at least one in the apical zone), consolidation > 1 cm in apical region, \n\nmiliary pattern, cavity or pericardial effusion > 1 cm \n\nHighly suggestive (alternative definition 2) SUN in ≥2 zones (with at least one in the apical or upper zone), consolidation > 1 cm in apical \n\nregion; miliary pattern, cavity or pericardial effusion > 1 cm \n\nPredefined combination of chest X-ray findings for detecting TB \n\nHighly suggestive of TB (main definition) • Presence of any of the following findings on CXR—reticulonodular infiltrates in any lung zone, \n\nlobar or segmental consolidation, nodular lesion, intrathoracic lymphadenopathy, pleural \n\neffusion, or an enlarged cardiac silhouette with a “water-bottle” configuration \n\nHighly suggestive (alternative definition 1) • Presence of reticulonodular infiltrates in any lung zone, cavities, or a miliary pattern \n\nHighly suggestive (alternative definition 2) • Presence of reticulonodular infiltrates localized to the apical or upper lung zones, cavities, or \n\na miliary pattern \n\n3\n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\nFigure 2. Typical thoracic ultrasound findings in tuberculosis. (a) Subpleural nodule: Subpleural round hypoechoic nodule measuring less than 1 cm and demonstrating \n\nposterior acoustic enhancement. (b) Consolidation: Hypoechoic area within lung parenchyma. (c) Cavity: Hypoechoic area with ill-defined margins and an avascular center \n\n(no flow on color Doppler). (d) Miliary pattern: Subpleural granular appearance with diffuse B-lines that appear to emanate from the subpleural granules. (e) Pleural effusion: \n\nanechoic fluid collection in the pleural space. (f) Pericardial effusion: anechoic fluid surrounding the heart. This image is used for illustration purposes only and should not be \n\nreproduced by others. \n\nl\n\ni\n\nd\n\nt\n\nt\n\nw\n\nm\n\nP\n\nd\n\ni\n\n2\n\np\n\ns\n\np\n\n2\n\nu\n\n(\n\nt\n\nd\n\no\n\nfi\n\ni\n\nn\n\na\n\na\n\nb\n\nung zones; right or left). Based on the combination of these find- \n\nngs, three nonexclusive diagnostic categories were predefined for \n\netecting TB. \n\nTo reduce interpretation bias, all image reviewers were blinded \n\no clinical and laboratory data, including GeneXpert results and \n\nreatment decisions. The experts interpreting the CXR and LUS \n\nere additionally blinded to the results of the other imaging \n\nodality. \n\nTo enhance consistency across LUS expert interpretations, the \n\nOCUS experts initially reviewed a common set of 20 LUS scans, \n\niscussed the findings, and aligned on study definitions prior to \n\nnitiating formal image review. \n\n.4. Data collection and management \n\nClinical, laboratory, and treatment data were recorded on \n\naper-based CRF and secondarily on REDCAP [ 24 ] by local study \n4\n\ntaff, overseen by the principal investigator. Remote imaging inter- \n\nretation was also documented in standardized REDCap CRFs. \n\n.5. Study outcomes and data analyses \n\nThe primary outcome was the diagnostic accuracy of individ- \n\nal and a predefined combination of LUS findings for detecting TB \n\nmain definition), assessed against the reference standard. Addi- \n\nionally, diagnostic accuracy was estimated using two alternative \n\nefinitions for predefined combined LUS findings highly suggestive \n\nf TB. LUS was considered not highly suggestive of TB when the \n\nndings did not fulfil the definition. Diagnostic accuracy metrics—\n\nncluding sensitivity, specificity, positive predictive value (PPV), and \n\negative predictive value (NPV)—were calculated. The diagnostic \n\nccuracy of individual and predefined combined CXR findings was \n\nlso evaluated using the same metrics. Additionally, the agreement \n\netween LUS and CXR in identifying cases classified as “Highly sug- \n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\nFigure 3. Study flowchart. \n\ng\n\n(\n\ne\n\nt\n\nn\n\nm\n\na\n\nm\n\nc\n\nt\n\ns\n\nt\n\nf\n\nm\n\nc\n\nd\n\ng\n\nL\n\na\n\nf\n\nt\n\nt\n\ni\n\nf\n\na\n\ni\n\nm\n\nM\n\n(\n\nL\n\nm\n\nt\n\n2\n\ns\n\nv\n\nw\n\ni\n\nC\n\n3\n\n3\n\ns\n\nt\n\nc\n\nm\n\np\n\nT\n\np\n\na\n\n(\n\np\n\na\n\nw\n\na\n\nt\n\na\n\np\n\n3\n\nt\n\nestive of TB” (main definition) was assessed using Cohen’s kappa \n\n κ) coefficient. \n\nThe Xpert MTB/RIF Ultra assay result was used as the refer- \n\nnce standard for TB diagnosis. Participants were classified as ei- \n\nher having confirmed TB or non-TB based on the assay positive or \n\negative results, respectively. Those without bacteriological confir- \n\nation but treated based on clinical assessment were categorized \n\ns “unconfirmed TB” and were excluded from the diagnostic pri- \n\nary accuracy analyses. \n\nAdditionally, two secondary analyses were conducted: one in- \n\nluding patients with unconfirmed TB and another excluding par- \n\nicipant under 15 years of age. \n\nBaseline demographic and clinical characteristics were pre- \n\nented for the entire cohort and stratified by diagnostic categories. \n\nMean and standard deviation were used for normally dis- \n\nributed numerical data, and median and inter-quartile range (IQR) \n\nor non-normally distributed data. Categorical variables were sum- \n\narized as frequencies and proportions. \n\nTo explore combinations of LUS findings beyond predefined \n\nombinations with potentially increased diagnostic accuracy and \n\nevelop a clinically applicable diagnostic score, we used logistic re- \n\nression with least absolute shrinkage and selection operator (LR- \n\nASSO). The dataset was randomly divided into a training set (70%) \n\nnd a test set (30%). Model generalizability was optimized using k- \n\nold cross-validation on the training set to select the most suitable \n\nuning parameter ( λ). The model was then retrained on the full \n\nraining set with this λ, and the resulting coefficients were scaled \n\nnto a simplified point-based score. Risk scores were generated \n\nor the test set, and an optimal cut-off was determined via ROC \n\nnalysis to dichotomize PTB status. Diagnostic metrics (sensitiv- \n\nty, specificity, accuracy) were derived from the resulting confusion \n\natrix. \n\nAll estimates are reported with 95% confidence intervals (CI). \n\nost statistical analyses were performed using Stata version 15.1 \n\nStataCorp LLC, College Station, TX, USA). Analyses involving LR- \n\nASSO modelling, data splitting, cross-validation, and diagnostic \n\netric calculations were conducted in R (version 4.4.2; R Founda- \n\nion for Statistical Computing, Vienna, Austria). \n5\n\n.6. Ethical considerations \n\nThe study was approved by the Medical Research Advi- \n\nory Committee of PNG (No. 19.01) and the MSF Ethical Re- \n\niew Board (ID: 1976). The study was conducted in accordance \n\nith the Declaration of Helsinki [ 25 ] and the International Eth- \n\ncal Guidelines for Health-Related Research Involving Humans by \n\nIOMS [ 26 ]. \n\n. Results \n\n.1. Participants’ characteristics \n\nBetween May 16, 2022, and May 8, 2023, 1503 patients were \n\ncreened at the Gerehu TB clinic. Of these, 512 were enrolled in \n\nhe study, and 488 were tested with GeneXpert MTB/RIF and in- \n\nluded in the analyses ( Figure 3 ). Participants were predominantly \n\nale (270, 55.3%), and median age was 30 years [IQR: 23-44]. HIV \n\nrevalence was 2.0%. Approximately 50% reported contact with a \n\nB case, primarily within the household, and 21% had a history of \n\nrevious TB treatment ( Table 2 ). \n\nThe 488 participants tested with GeneXpert MTB/RIF ultra- \n\nssay were classified as follows: 149 (30.5%) confirmed TB, 55 \n\n11.3%) unconfirmed TB, and 284 (58.2%) as non-TB. In total, 433 \n\narticipants with confirmed TB and non-TB were included in the \n\nccuracy analyses. Among the confirmed TB patients, 129 (86.6%) \n\nere classified as only PTB, and 20 (13.4%) had both pulmonary \n\nnd extrapulmonary involvement. Rifampicin resistance was iden- \n\nified in two participants. \n\nAmong the 55 individuals excluded from the primary accuracy \n\nnalyses, a higher proportion were male and slightly older com- \n\nared to the included population ( Table 2 ). \n\n.2. LUS findings \n\nThe average duration of the LUS examination performed by the \n\nrained clinician was 39 minutes (range: 20-69 minutes). Image \n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\nTable 2 \n\nDemographic and clinical characteristics at baseline, stratified by TB diagnosis ( N = 488). \n\nNon-TB (284) Confirmed TB (149) Unconfirmed TB (55) Total (488) \n\nn (%) n (%) n (%) n (%) \n\nGender \n\nFemale 141 (49.6) 57 (38.3) 20 (36.4) 218 (44.7) \n\nMale 143 (50.4) 92 (61.7) 35 (63.6) 270 (55.3) \n\nAge, median [IQR] 30.5 [23-44] 26 [22-36] 34 [26-43] 30 [23-41] \n\nAge group categories (years) \n\n10 to ≤11 8 (2.8) 1 (0.7) 0 (0.0) 9 (1.8) \n\n12 to ≤14 8 (2.8) 3 (2.0) 2 (3.6) 13 (2.7) \n\n15 to ≤20 33 (11.6) 24 (16.1) 7 (12.7) 64 (13.1) \n\n21 to ≤30 93 (32.7) 66 (44.3) 14 (25.5) 173 (35.5) \n\n31 to ≤40 61 (21.5) 27 (18.1) 15 (27.3) 103 (21.1) \n\n> 40 81 (28.5) 28 (18.8) 17 (30.9) 126 (25.8) \n\nKnown TB contact 141 (49.8) 80 (53.7) 27 (49.1) 248 (50.9) \n\nHousehold TB contact 128 (92.1) 71 (88.8) 26 (96.3) 225 (91.5) \n\nPast Tuberculosis treatment 71 (25) 20 (13.4) 11 (20) 102 (20.9) \n\nPrevious treatment outcome ( n = 102) \n\nCured or completed 42 (59.1) 11 (57.9) 4 (36.4) 57 (56.5) \n\nInterrupted or failure 4 (5.6) 1 (5.3) 1 (9.1) 6 (5.9) \n\nUnknown 25 (35.3) 8 (36.8) 6 (54.5) 38 (37.6) \n\nCough 289 (100) 149 (100) 55 (100) 496 (100) \n\nProduction of sputum 253 (89.1) 131 (87.9) 50 (90.9) 434 (88.9) \n\nWeight loss 242 (85.2) 145 (97.3) 49 (89.1) 436 (89.3) \n\nWeakness 240 (84.5) 129 (86.6) 51 (92.7) 420 (86.1) \n\nFever 194 (68.3) 123 (82.6) 46 (83.6) 363 (74.4) \n\nDifficulty to breath 195 (68.7) 115 (77.2) 50 (90.9) 360 (73.8) \n\nNight sweats 193 (68) 121 (81.2) 42 (76.4) 356 (73) \n\nChest pain 194 (68.3) 114 (76.5) 47 (85.5) 355 (72.7) \n\nLoss of appetite 139 (49.1) 92 (61.7) 40 (72.7) 271 (55.6) \n\nOther symptoms 58 (20.9) 37 (25.2) 16 (29.6) 111 (23.2) \n\nHemoptysis 63 (22.2) 34 (22.8) 9 (16.4) 106 (21.7) \n\nLymph nodes 57 (20.1) 47 (31.5) 13 (23.6) 117 (24) \n\nCervical/supraclavicular 46 (80.7) 45 (95.7) 11 (84.6) 102 (87.2) \n\nOther 11 (19.3) 2 (4.2) 2 (15.4) 15 (12.8) \n\nAscites 4 (1.4) 2 (1.3) 6 (10.9) 12 (2.5) \n\nOther signs 11 (4.0) 7 (4.7) 0 (0.0) 18 (3.7) \n\nDiabetes 2 (0.7) 2 (1.3) 1 (1.8) 5 (1) \n\nHIV status \n\nPositive 5 (1.8) 1 (0.7) 2 (3.6) 8 (1.6) \n\nNegative 6 (2.1) 9 (6.1) 2 (3.6) 17 (3.5) \n\nUnknown 271 (96.1) 138 (93.2) 51 (92.7) 460 (94.8) \n\nPlace of residency \n\nNorthwest province resident 192 (67.6) 89 (59.7) 36 (65.5) 317 (65.0) \n\nNorthwest province suburb \n\nGerehu 100 (52.1) 51 (57.3) 19 (52.8) 170 (53.6) \n\nOther suburbs 92 (47.9) 38 (42.7) 17 (47.2) 147 (46.4) \n\nq\n\ni\n\nc\n\n>\n\np\n\no\n\nt \n\nn\n\nm\n\ni\n\na\n\ns\n\ns\n\ns\n\ni\n\nq \n\nt\n\ns\n\n3\n\nr\n\na\n\ni\n\n(\n\nT\n\nu\n\na\n\nh\n\nfi\n\n3\n\nd\n\ns\n\no\n\nuality was rated by the POCUS experts as “good” in 225 partic- \n\npants (46.5%), “adequate” in 246 (50.8%), and “poor” in 13 (2.7%). \n\nThe most common LUS findings among participants were small \n\nonsolidations < 1 cm (67%), SUNs (48%), and larger consolidations \n\n 1 cm (46%). Less frequent findings were miliary pattern (11%), \n\nericardial effusion (9%), and cavities (3%). All LUS findings were \n\nbserved more frequently in participants with TB compared to \n\nhose without TB ( Table 3 ). Full details of LUS findings by diag-\n\nostic group are provided in the Supplementary Material (Supple- \n\nentary Table 3). \n\nDiagnostic performance varied among the different LUS find- \n\nngs. SUNs showed moderate accuracy, with a sensitivity of 64% \n\nnd specificity of 60%. Small consolidations demonstrated higher \n\nensitivity (79%) but lower specificity (39%), whereas larger con- \n\nolidations showed slightly lower sensitivity (76%) and moderate \n\npecificity (69%) ( Table 3 ). \n\nThe diagnostic accuracy of SUN varied depending on the spec- \n\nfications of the SUN characteristics. Restricting SUN finding to a \n\nuantity of more than 5 or a size of more than 5 mm increased\n\nhe specificity to 92% and 95% respectively, while decreasing the \n\nensitivity to 20% and 21%, respectively ( Table 3 ). \n6\n\n.3. CXR findings \n\nCXR image quality, as assessed by the expert radiologist, was \n\nated as “good” for 153 CXR (31.4%), “adequate” for 286 (58.7%), \n\nnd “poor” for 48 (9.9%). \n\nThe most common CXR findings among confirmed TB partic- \n\npants were reticulonodular infiltrates (70%) and cavities (23%) \n\n Table 4 ). Cavities were often multiple and > 1 cm (Supplementary \n\nable 4). \n\nCXR findings showed variable diagnostic accuracy for TB. Retic- \n\nlonodular infiltrates and cavities had the highest sensitivities, 70% \n\nnd 54%, respectively, and lymphadenopathy and miliary pattern \n\nad the highest specificities, 97% and 100%, respectively. Other \n\nndings demonstrated lower sensitivity despite high specificity. \n\n.4. Diagnostic accuracy of combined imaging findings for TB \n\netection \n\nUsing a predefined combination of imaging findings for “Highly \n\nuggestive of TB” (main definition), LUS demonstrated a sensitivity \n\nf 92.6% (95% CI: 90.1-95.1) and specificity of 46.8% (95% CI: 42.1- \n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\nTable 3 \n\nDistribution of lung ultrasound findings among non-TB ( n = 284) and confirmed TB ( n = 149) participants and diagnostic accuracy measures (sensitivity, specificity, PPV, \n\nNPV) ( N = 433). \n\nUltrasound finding Non-TB \n\n( N = 284) \n\nConfirmed TB \n\n( N = 149) \n\nTotal \n\n( N = 433) \n\nSensitivity Specificity PPV NPV \n\nn % n % n % (95% CI) (95% CI) (95% CI) (95% CI) \n\nSubpleural nodules (SUN) 113 39.8 95 63.8 208 48 63.8 (59.2-68.3) 60.2 (55.6-64.8) 45.7 (41.0-50.4) 76.0 (72.0-80.0) \n\nConsolidation ≤1 cm 174 61.3 117 78.5 291 67.2 78.5 (74.6-82.4) 38.7 (34.1-43.3) 40.2 (35.6-44.8) 77.5 (73.5-81.4) \n\nConsolidation > 1 cm 89 31.3 112 75.2 201 46.4 75.7 (71.6-79.7) 68.7 (64.3-73.0) 55.7 (51.0-60.4) 84.4 (81.0-87.8) \n\nCavity 4 1.4 9 6.0 13 3.0 6.0 (3.8-8.3) 98.6 (97.5-99.7) 69.2 (64.9-73.6) 66.7 (62.2-71.1) \n\nMiliary pattern 17 6.0 29 19.5 46 10.6 19.5 (15.7-23.2) 94.0 (91.8-96.3) 63.0 (58.5-67.6) 69.0 (64.6-73.4) \n\nPleural effusion 39 13.7 60 40.3 99 22.9 40.5 (35.9-45.1) 86.3 (83.0-89.5) 60.6 (56.0-65.2) 73.6 (69.4-77.7) \n\nPericardial effusion 18 6.3 19 12.8 37 8.5 12.7 (9.6-16.0) 93.7 (91.4-96.0) 51.4 (46.6-56.1) 67.2 (62.8-71.6) \n\nSUN quantity > 5a 24 8.5 29 19.5 53 12.2 19.5 (15.7-23.2) 91.6 (89.0-94.2) 54.7 (50.0-59.4) 68.4 (64.0-72.8) \n\nSUN size > 5 mm 13 4.6 31 20.8 44 10.2 20.8 (17.0-24.6) 95.4 (93.5-97.4) 70.5 (66.2-74.8) 69.7 (65.3-74.0) \n\nSUN in apical 34 12 49 32.9 83 19.2 32.9 (28.5-37.3) 88.0 (85.0-91.1) 59.0 (54.4-63.7) 71.4 (67.2-75.7) \n\nSUN apical or upper 99 34.9 83 55.7 182 42 55.7 (51.0-60.4) 65.1 (60.7-69.6) 45.6 (40.9-50.3) 73.7 (69.6-77.9) \n\nSUN at least in 2 zones 77 27.1 60 40.3 137 31.6 40.3 (35.7-44.9) 72.9 (68.7-77.1) 43.8 (39.1-48.5) 69.9 (65.6-74.3) \n\na “Quantity > 5” refers to the total number of SUNs observed across all lung zones. \n\nTable 4 \n\nDistribution of chest X-ray findings among non-TB ( n = 284) and confirmed TB ( n = 149) participants and diagnostic accuracy measures (sensitivity, specificity, PPV, NPV) \n\n( N = 433). \n\nNon-TB \n\n( N = 284) \n\nConfirmed TB \n\n( N = 149) \n\nTotal \n\n( N = 433) \n\nSensitivity Specificity PPV NPV \n\nCXR findings n (%) n (%) n (%) (95% CI) (95% CI) (95% CI) (95% CI) \n\nReticulonodular infiltrates 60 21.1 104 69.8 164 37.9 69.8 (65.5-74.1) 78.9 (75.0-82.7) 63.4 (58.9-68.0) 83.3 (79.8-86.8) \n\nLobar or segmental \n\nconsolidation \n\n9 3.2 13 8.7 22 5.1 8.7 (6.1-11.4) 96.8 (95.1-98.5) 59.1 (54.5-63.7) 66.9 (54.5-63.7) \n\nCavity 17 6 81 54.4 98 22.6 54.4 (49.7 59.1) 94 (91.8-96.3) 82.7 (79.1-86.2) 79.7 (75.9-83.5) \n\nNodular lesion 25 8.8 15 10.1 40 9.2 10.1 (7.2-12.9) 91.2 (88.5-93.8) 37.5 (32.9-42.1) 65.9 (61.4-70.4) \n\nMiliary pattern 0 0 8 5.4 8 1.8 5.3 (3.3-7.5) 100 (100.0-100.0) 100 (100.0-100.0) 66.8 (62.4-71.3) \n\nPleural effusion 22 7.7 41 27.5 63 14.5 27.5 (23.3-31.7) 92.3 (89.7-94.8) 65.1 (60.6-69.6) 70.8 (66.5-75.1) \n\nLymphadenopathy 8 2.8 22 14.8 30 6.9 14.8 (11.4-18.1) 97.2 (95.6-98.7) 73.3 (69.2-77.5) 68.5 (69.2-77.5) \n\nEnlarged cardiac silhouette 22 7.7 5 3.4 27 6.2 3.4 (1.7-5.1) 92.3 (89.7-94.8) 18.5 14.9-22.2) 64.5 (60.0-69.0) \n\nApical pleural thickening 21 7.4 18 12.1 39 9 12.1 (9.0-15.2) 92.6 (90.1-95.1) 46.2 (41.5-50.9) 66.8 (62.3-71.2) \n\nFibrosis/scarring 24 8.5 10 6.7 34 7.9 6.7 (4.3-9.1) 91.6 (89.3-94.2) 29.4 (25.1-33.7) 65.2 (60.7-69.7) \n\nTable 5 \n\nDiagnostic accuracy of lung ultrasound (LUS) and chest X-ray (CXR) for tuberculosis diagnosis: Sensitivity, specificity, positive predictive value (PPV), and negative predictive \n\nvalue (NPV) using main and alternative definitions. \n\nSensitivity Specificity PPV NPV \n\n(95% CI) (95% CI) (95% CI) (95% CI) \n\nMain definition \n\nLUS a 92.6 (90.1-95.1) 46.8 (42.1-51.5) 47.8 (43.1-52.5) 92.3 (89.9-94.9) \n\nCXR b 91.3 (88.6-93.9) 65.1 (60.7-69.6) 57.9 (53.2-62.5) 93.4 (91.1-95.8) \n\nAlternative definitions \n\nLUS definition 1 c 69.1 (64.8-73.5) 63.7 (59.2-68.3) 50.0 (45.3-54.7) 79.7 (76.0-83.5) \n\nLUS definition 2 d 65.1 (60.6-69.5) 74.3 (70.2-78.4) 57.1 (52.4-61.7) 80.2 (76.5-84.0) \n\nCXR definition 1 e 74.5 (70.4-78.6) 77.5 (73.5-81.4) 63.4 (58.9-68.0) 85.3 (81.9-88.6) \n\nCXR definition 2 f 71.1 (66.9-75.4) 88.0 (85.0-91.1) 75.7 (71.7-79.8) 85.3 (82.05-88.7) \n\na Highly suggestive of TB (LUS main definition): Presence of SUN in ≥2 zones (with at least one in the apical or upper), consolidation > 1 cm, a miliary pattern, cavities, \n\npleural effusion or pericardial effusion > 1 cm. Not suggestive of TB: all others. \nb Highly suggestive of TB (CXR main definition): Presence of reticulonodular infiltrates in any lung region, lobar or segmental consolidation, nodular lesions, lymphadenopa- \n\nthy, pleural effusion, or an enlarged cardiac silhouette with a “water-bottle” shape. Not suggestive of TB: all others. \nc Highly suggestive of TB: Presence of SUN in ≥2 zones (with at least one in the apical region), consolidation > 1 cm, a miliary pattern or cavities or pericardial effusion \n\n> 1 cm. Not suggestive of TB: all others. \nd Highly suggestive of TB: Presence of SUN in ≥2 zones (with at least one in the apical or upper), consolidation > 1 cm; a miliary pattern or cavities or pericardial effusion \n\n> 1 cm. Not suggestive of TB: all others. \ne Highly suggestive of TB: Presence of reticulonodular infiltrates in any lung zone, cavities, or a miliary pattern. Not suggestive of TB: all others. \nf Highly suggestive of TB: Presence of reticulonodular infiltrates localized to the apical or upper lung zones, cavities, or a miliary pattern. Not suggestive of TB: all others. \n\n5\n\n(\n\n9\n\ns\n\na\n\nr\n\ny\n\nT\n\nr\n\n3\n\np\n\nm\n\n1.5). The PPV was 47.8% (95% CI 43.1-52.5), and the NPV was 92.3% \n\n95% CI 89.9-94.9). CXR showed a sensitivity of 91.3% (95% CI: 86.6- \n\n3.9) and specificity of 65.1% (95% CI: 60.7-69.6) ( Table 5 ). \n\nUsing alternative definitions for “highly suggestive of TB”, LUS \n\nensitivity decreased, 69% and 65% using alternative definitions 1 \n\nnd 2, respectively, while the specificity increased, 64% and 74%, \n\nespectively ( Table 5 ). \n\nSecondary analysis showed that including unconfirmed TB \n\nielded a sensitivity of 91.7% (95% CI: 89.2-94.1; Supplementary \n7\n\nable 5), while excluding participants < 15 years produced similar \n\nesults than the primary analyses (Supplementary Table 6). \n\n.5. Agreement between CXR and LUS \n\nThe agreement between CXR and LUS in categorizing partici- \n\nants as “Highly suggestive of TB” using the main definition was \n\noderate with a Cohen’s kappa of 0.48 (95% CI 0.40-0.56). This \n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\ni\n\ns\n\n3\n\nd\n\ni\n\nz\n\np\n\ni\n\nt\n\nfi\n\n(\n\n(\n\n4\n\na\n\nu\n\nL\n\na\n\nm\n\n4\n\nc\n\ns\n\nt\n\nr\n\nT\n\ni\n\na\n\nc\n\nb\n\nb\n\nm\n\nt\n\nd\n\nr\n\nc\n\nfi\n\n>\n\nfi\n\n(\n\np\n\ns\n\nr\n\nL\n\nw\n\ni\n\na\n\nb\n\nc\n\nz\n\nr\n\na\n\ng\n\nfi\n\ni\n\nd\n\ne\n\ni\n\nd\n\ns\n\na\n\nr\n\nt\n\ni\n\nm\n\nb\n\nq\n\nH\n\ne\n\na\n\np\n\np\n\n4\n\ne\n\ns\n\na\n\nb\n\ni\n\nt\n\np\n\ne\n\nb\n\ns\n\nb\n\na\n\nb\n\na\n\ns\n\nv\n\nt\n\nc\n\nt\n\nU\n\np \n\nA\n\nt\n\np\n\nd\n\na\n\nm\n\nd\n\nh\n\nt\n\nv\n\nt\n\nr\n\nd\n\nn\n\ns\n\np\n\ni\n\n5\n\na\n\nt\n\nndicates that the two imaging modalities often identify different \n\nubsets of patients with TB (Supplementary File, Table 7). \n\n.6. LUS-based diagnostic score \n\nThe final LR-LASSO model selected several LUS findings as pre- \n\nictors of PTB, which were incorporated into a point-based scor- \n\nng system ranging from 0 to 6 points: miliary pattern in ≥2 lung \n\nones (2 points), consolidation > 1 cm in apical or upper zones (2 \n\noints), pleural effusion (1 point), and SUN or consolidation ≤1 cm \n\nn the apical zone (1 point) (Supplementary File, Table 8). Using a \n\nhreshold of 1.5, individuals with scores of 2 or higher were classi- \n\ned as PTB positive. The score demonstrated a sensitivity of 82.8% \n\n95% CI: 64.2-94.2) and a specificity of 84.2% (95% CI: 72.1-92.5) \n\nSupplementary File, Table 9 and Figure 1 ). \n\n. Discussion \n\nThis study provides evidence regarding the potential of LUS as \n\n diagnostic tool for PTB in outpatients aged over 10 years, partic- \n\nlarly in settings without access to CXR. Our findings suggest that \n\nUS has high sensitivity and low specificity for PTB. The LUS has \n\n comparable sensitivity to that of CXR but lower specificity. Both \n\nodalities have similar performance in ruling out PTB. \n\n.1. Study findings \n\nOur study findings align with previous studies. A proof-of- \n\noncept study in South Africa reported 85% sensitivity and 52% \n\npecificity for LUS in detecting thoracic abnormalities (consolida- \n\nion, subpleural pathology, cavity, any B-patterns, pleural line ir- \n\negularities, or pleural effusion) using bacteriologically confirmed \n\nB as the reference standard [23] . Montuori et al. found a sensitiv- \n\nty of 86% and specificity of 63% using a combination of SUNs and \n\npical consolidation, though 14% of TB cases lacked bacteriological \n\nonfirmation [19] . The slightly higher specificity reported in studies \n\ny Montuori et al. [19] and Fentress et al. [23] may be influenced \n\ny differences in how LUS findings were defined and applied. They \n\nay have applied stricter criteria for lesion size, location, or pat- \n\nern recognition, which could reduce false positives. Additionally, \n\nifferences in real-time image acquisition vs interpretation from \n\necorded clips, as well as variations in study populations, may also \n\nontribute to these discrepancies. \n\nImportantly, the most frequent LUS findings present in con- \n\nrmed TB cases were consolidations ≤1 cm (78.5%), consolidations \n\n 1 cm (75.2%), and SUNs (63.8%). Findings with the highest speci- \n\ncity were SUNs > 5 mm (95.4%), SUNs located in the apical zones \n\n88%), and pericardial effusion (93.7%). These results align with \n\nrevious studies, such as Montuori et al. [19] , which reported a \n\npecificity of 66.7% for SUN, and Fentress et al. [23] . Cavities were \n\narely detected in our study, corroborating known limitations of \n\nUS in identifying cavitations [18] . \n\nMost of our lung imaging was performed using a linear probe \n\nith 4 cm starting depth, likely improving detection of subtle find- \n\nngs, such as small SUNs, compared to a curvilinear probe on an \n\nbdominal preset. This approach may have increased sensitivity \n\nut reduced specificity. \n\nWe developed an LR-LASSO-derived 6-point scoring model in- \n\norporating key LUS findings—miliary pattern in at least 2 lung \n\nones, consolidation > 1 cm in the apical or upper lung zone, pleu- \n\nal effusion, and either SUN or small consolidations ( ≤1 cm) in the \n\npical lung zone. At a threshold of 2 points, the model showed \n\nood diagnostic performance of the score (sensitivity 82.8%, speci- \n\ncity 84.2%). The scoring model simplifies interpretation by group- \n\nng together SUN and small consolidations, which are intrinsically \n\nifficult to differentiate. It also streamlines image acquisition by \n8\n\nxcluding pericardial effusion. These advantages may further facil- \n\ntate the implementation use of LUS in resource-limited settings. \n\nTraining nonspecialist clinicians to perform LUS using a stan- \n\nardized acquisition protocol was successful, with Health Exten- \n\nion Officers consistently obtaining high-quality US images. This \n\nligns with evidence from the Focused Assessment with Sonog- \n\naphy for HIV-associated TB (FASH) protocol, which demonstrated \n\nhat novice clinicians can be trained not only to acquire but also to \n\nnterpret sonographic findings for EPTB [27] . Unlike other POCUS \n\nodules that are typically performed and interpreted immediately \n\ny performing clinicians, our study focused solely on image ac- \n\nuisition, with expert reviewers interpreting LUS exams remotely. \n\nowever, the time required for LUS image acquisition followed by \n\nxternal interpretation may limit practical implementation of this \n\npproach in busy clinical settings. Nevertheless, this time is ex- \n\nected to be reduced following the training and implementation \n\nhase. \n\n.2. Strengths and limitations \n\nThe strengths of this study include the use of a robust refer- \n\nnce standard (bacteriologically confirmed TB), a large outpatient \n\nample that addresses limitations of previous studies [21] , and \n\n standardized image acquisition protocol Importantly, we report \n\noth sensitivity and specificity for individual and combined imag- \n\nng findings, and diagnostic accuracy estimates remained consis- \n\nent when including patients with unconfirmed TB, reinforcing our \n\nrimary results. LUS and CXR images were interpreted by blinded \n\nxperts, minimizing the risk of bias. Furthermore, LUS acquisition \n\ny trained clinicians on-site supports the applicability of our re- \n\nults to real-world, low-resource settings. \n\nOur study has also limitations. It was conducted in a high TB \n\nurden setting, where lung sequelae from previous TB may have \n\nffected image interpretation. Therefore, generalizability to low- \n\nurden TB settings may be limited. Inter-reader interpretation vari- \n\nbility may have occurred, as four POCUS experts reviewed sub- \n\nets of US examinations. Ideally, each exam would have been re- \n\niewed by multiple readers, but this was not possible due to \n\nime constraints and expert reviewer availability. Additionally, sub- \n\nentimeter findings such as SUNs are difficult to identify and dis- \n\ninguish on recorded clips without hands-on manipulation of the \n\nS transducer and real-time evaluation of the images. Operator de- \n\nendency remains a notable limitation of US in general [ 28 , 29 ].\n\nlthough image acquisition was relatively standardized, interpre- \n\nation is prone to inter-reader variability and subjectivity. This is \n\narticularly relevant when distinguishing SUN from small consoli- \n\nations, which often share overlapping features. Despite efforts to \n\npply consistent distinguishing criteria, a degree of subjectivity re- \n\nains. Similarly, differentiating cavities from consolidation remains \n\nifficult; however, given the rarity of cavities in our dataset, this \n\nad minimal practical impact on our results. Emerging AI solu- \n\nions have potential to increase objectivity and reduce inter-reader \n\nariability. Deep-learning approaches are showing promise in au- \n\nomating US image interpretation and improving diagnostic accu- \n\nacy [ 30 ]. Specific analysis in age subgroups, particularly those un- \n\ner 15 years—who more frequently present with EPTB and lymph \n\node involvement [ 31 ]—was not possible due to the small sample \n\nize in this group. While the model-derived scoring system showed \n\nromising results, prospective validation will be required to assess \n\nts diagnostic value. \n\n. Conclusions \n\nIn summary, LUS demonstrates a high sensitivity for PTB, equiv- \n\nlent to CXR, although the specificity is low. LUS may have poten- \n\nial for PTB diagnosis as an imaging modality, particularly when \n\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\nC\n\ni\n\nP\n\np\n\ni\n\np\n\nd\n\nE\n\nC\n\nS\n\nF\n\nO\n\nD\n\nc\n\ni\n\nA\n\ns\n\nf\n\nh\n\n(\n\nt\n\np\n\nd\n\ni\n\nA\n\ns\n\nN\n\nt\n\nt\n\nW\n\nt\n\nc\n\nt\n\na\n\na\n\nv\n\nS\n\nf\n\nR\n\n \n\n \n\n[ \n\n \n\n \n\n \n\n[\n\n[ \n\n[ \n\n[ \n\n[ \n\n[\n\n[\n\n[\n\n[\n\nXR is unavailable. We did not identify a combination of LUS find- \n\nngs providing an optimal balance of sensitivity and specificity for \n\nTB diagnosis. However, our diagnostic scoring system offers a \n\nragmatic alternative, with notably higher specificity. Future stud- \n\nes are needed to externally validate the proposed score and to ex- \n\nlore whether integrating the FASH protocol for extra-PTB, enhance \n\niagnostic yield. \n\nthical approval \n\nThe study was approved by the Medical Research Advisory \n\nommittee of Papua New Guinea (No. 19.01) and the Medecins \n\nans Frontieres Ethical Review Board (ID: 1976). \n\nunding \n\nThis work was funded by Médecins Sans Frontières—\n\nperational Center Paris. \n\neclaration of competing interest \n\nThe authors declare that they have no known competing finan- \n\nial interests or personal relationships that could have appeared to \n\nnfluence the work reported in this article. \n\ncknowledgments \n\nWe thank all study participants, and the Gerehu study team, \n\npecially Anna Nape for the study coordination and Carolyn Hemo \n\nor performing the ultrasounds, study Community Health worker \n\nealth extension officers, laboratory technicians, data supervisors \n\nIbrahim Seyni Am) a as well as the entire Médecins Sans Fron- \n\nières team in Port Moresby for supporting the preparation and im- \n\nlementation phase of the study, including MSF field project coor- \n\ninator, mission epidemiologist, medical coordinator and cell med- \n\ncal responsible. \n\nuthor contributions \n\nM. Lightowler, L. Sannino and H. Huerga conceived and de- \n\nigned the study. M. Lightowler performed the data analysis. A. \n\nappe, C. Hemo and T. Galowa contributed to study implementa- \n\nion and data collection. J. Min assisted with data collection and \n\nhe design of radiology and ultrasound procedures. M. Fentress, S. \n\neber and M. Hudson contributed to data collection and interpre- \n\nation. E. Stratta assisted with study preparation and training and \n\nontributed to the design and interpretation of the ultrasound pro- \n\nocol and related data collection. S. Nichola performed additional \n\nnalyses on scoring. M. Lightowler drafted the initial manuscript, \n\nnd H. Huerga provided major revisions. All authors critically re- \n\niewed the manuscript and approved the final version. \n\nupplementary materials \n\nSupplementary material associated with this article can be \n\nound, in the online version, at doi:10.1016/j.ijid.2025.108239 . \n\neferences \n\n[1] World Health Organization Global tuberculosis report 2024 . Geneva: World \n\nHealth Organization; 2024 . Available from: https://iris.who.int/bitstream/ \n\nhandle/10665/379339/9789240101531-eng.pdf?sequence=1 . \n[2] Bita S, Kelebi T, Holmes A, Vaccher S, Majumdar SS, Greig J. TB burden and\n\ndiagnostic challenges at Sandaun Provincial Hospital in West Sepik Province \nof PNG, 2016–2021. Public Health Action 2024; 14 (3):91–6. doi: 10.5588/pha.24. \n\n0016 . \n9\n\n[3] World Health Organization WHO consolidated guidelines on tuberculosis. Module \n3: diagnosis – rapid diagnostics for tuberculosis detection . 3rd ed. Geneva: World \n\nHealth Organization; 2024 . \n[4] World Health Organization Chest radiography in tuberculosis detection . Geneva: \n\nWorld Health Organization; 2016 . \n[5] Pomykala K, Desai I, Jardon M, Naik P, Pool K–L. Imaging of tuberculo- \n\nsis in resource–limited settings. Curr Radiol Rep 2019; 7 :23. doi: 10.1007/ \ns40134-019-0335-7 . \n\n[6] Use of computer-aided detection software for tuberculosis screening: WHO policy \n\nstatement , Geneva: World Health Organization; 2025. (License: CC BYNCSA 3.0 \nIGO). Cataloguing-in-Publication (CIP) data available at: . \n\n[7] Qin ZZ, Naheyan T, Ruhwald M, Denkinger CM, Gelaw S, Nash M, et al. A\nnew resource on artificial intelligence powered computer automated detection \n\nsoftware products for tuberculosis programmes and implementers. Tuberculosis \n(Edinb) 2021; 127 :102049 PMID: 33440315. doi: 10.1016/j.tube.2020.102049 . \n\n[8] Parker L, Nazarian LN, Carrino JA, Morrison WB, Grimaldi G, Frangos AJ, et al. \n\nMusculoskeletal imaging: Medicare use, costs, and potential for cost substitu- \ntion. J Am Coll Radiol 2008; 5 (3):182–8 . \n\n[9] Smith-Bindman R, Aubin C, Bailitz J, Bengiamin RN, Camargo CA, Corbo J, et al. \nUltrasonography versus computed tomography for suspected nephrolithiasis. N \n\nEngl J Med 2014; 371 (12):1100–10 . \n[10] Lichtenstein D, Goldstein I, Mourgeon E, Cluzel P, Grenier P, Rouby JJ. \n\nComparative diagnostic performances of auscultation, chest radiography, and \n\nlung ultrasonography in acute respiratory distress syndrome. Anesthesiology \n20 04; 10 0 :9–15. doi: 10.1097/0 0 0 0 0542-20 04010 0 0-0 0 0 06 . \n\n[11] Sorensen B, Hunskaar S. Point-of-care ultrasound in primary care: a sys- \ntematic review of generalist performed point-of-care ultrasound in unse- \n\nlected populations. Ultrasound J 2019; 11 (1):31 PMID: 31749019. doi: 10.1186/ \ns13089- 019- 0145- 4 . \n\n12] Bobbio F, Di Gennaro F, Marotta C, Kok J, Akec G, Norbis L, et al. Focused\n\nultrasound to diagnose HIV-associated tuberculosis (FASH) in the extremely \nresource-limited setting of South Sudan: a cross-sectional study. BMJ Open \n\n2019; 9 :e027179. doi: 10.1136/bmjopen-2018-027179 . \n[13] Goblirsch S, Bahlas S, Ahmed M, Brunetti E, Wallrauch C, Heller T. Ultrasound \n\nfindings in cases of extrapulmonary TB in patients with HIV infection in Jed- \ndah, Saudi Arabia. Asian Pac J Trop Dis 2014; 4 (1):14–17 . \n\n[14] Griesel R, Cohen K, Mendelson M, Maartens G. Abdominal ultrasound for the \n\ndiagnosis of tuberculosis among human immunodeficiency virus-positive in- \npatients with World Health Organization danger signs. Open Forum Infect Dis \n\n2019; 6 (4):ofz094. doi: 10.1093/ofid/ofz094 . \n[15] Kahn D, Pool KL, Phiri L, Chibwana F, Schwab K, Longwe BA, et al. Diagnos-\n\ntic utility and impact on clinical decision making of focused assessment with \nsonography for HIV-associated tuberculosis in Malawi: a prospective cohort \n\nstudy. Glob Health Sci Pract 2020; 8 (1):28–37. doi: 10.9745/GHSP- D- 19- 00251 . \n\n[16] Pool KL, Heuvelings CC, Bélard S, Grobusch MP, Zar HJ, Bulas D, et al. Technical\naspects of mediastinal ultrasound for pediatric pulmonary tuberculosis. Pediatr \n\nRadiol 2017; 47 (13):1839–48 . \n[17] Babasa R. III, Aralar K.C., Bithao J. Diagnostic accuracy of point–of–care \n\nlung ultrasound among patients suspected of having pulmonary tuberculo- \nsis. WINFOCUS Case Study [Internet]. 2019 [cited 2025 Dec 12]; Available \n\nfrom: https://www.winfocus.org/case-study/diagnostic-accuracy-of-point-of- \ncare- lung- ultrasound- among- patients- suspected- of- having- pulmonary- \n\ntuberculosis/ . \n\n[18] Agostinis P, Copetti R, Lapini L, Badona Monteiro G, N’Deque A, Bari- \ntussio A. Chest ultrasound findings in pulmonary tuberculosis. Trop Doct \n\n2017; 47 (4):320–8 PMID: 28541140. doi: 10.1177/0049475517709633 . \n[19] Montuori M, Casella F, Casazza G, Franzetti F, Pini P, Invernizzi C, et al. Lung\n\nultrasonography in pulmonary tuberculosis: a pilot study on diagnostic accu- \nracy in a high-risk population. Eur J Intern Med 2019; 66 :29–34 . \n\n20] Nathavitharana RR, Garcia-Basteiro AL, Ruhwald M, Cobelens F, Theron G. \n\nReimagining the status quo: how close are we to rapid sputum-free tubercu- \nlosis diagnostics for all? EBioMedicine 2022; 78 :103939 . \n\n21] Bigio J, Kohli M, Klinton JS, MacLean E, Gore G, Small PM, et al. Diagnostic\naccuracy of point-of-care ultrasound for pulmonary tuberculosis: a systematic \n\nreview. PLoS One 2021; 16 (5):e0251236 PMID: 33961639. doi: 10.1371/journal. \npone.0251236 . \n\n22] Erem G, Otike C, Okuja M, Ameda F, Nalyweyiso DI, Mubuuke AG, et al. Diag-\n\nnostic accuracy of chest ultrasound scan in the diagnosis of childhood tuber- \nculosis. PLoS One 2023; 18 (9):e0287621 . \n\n23] Fentress M, Henwood PC, Maharaj P, Mitha M, Khan D, Caligiuri P, et al. High\nsensitivity of ultrasound for the diagnosis of tuberculosis in adults in South \n\nAfrica: a proof-of-concept study. PLOS Glob Public Health 2022; 2 :e0 0 0 080 0 . \n24] Harris PA, Taylor R, Minor BL, Elliott V, Fernandez M, O’Neal L, et al. The RED-\n\nCap consortium: building an international community of software platform \n\npartners. J Biomed Inform 2019; 95 :103208. doi: 10.1016/j.jbi.2019.103208 . \n25] World Medical Association. Declaration of Helsinki: ethical principles for med- \n\nical research involving human participants. JAMA 2024; 333 (1):71–4 . \n26] Council for International Organizations of Medical Sciences (CIOMS) World \n\nHealth Organization. International ethical guidelines for health-related research in- \nvolving humans . 4th ed. Geneva: CIOMS; 2016 . \n\n27] Heller T, Wallrauch C, Lessells RJ, Goblirsch S, Brunetti E. Short course for \n\nfocused assessment with sonography for HIV associated tuberculosis: pre- \nliminary results in a rural South African hospital. Am J Trop Med Hyg \n\n2010; 82 (3):512–15 . \n28] Suttels V, Du Toit JD, Fiogbé AA, Wachinou AP, Guendehou B, Alovokpin- \n\nhou F, et al. Point-of-care ultrasound for tuberculosis management in Sub- \n\nhttps://doi.org/10.1016/j.ijid.2025.108239\nhttps://iris.who.int/bitstream/handle/10665/379339/9789240101531-eng.pdf?sequence=1\nhttps://doi.org/10.5588/pha.24.0016\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0003\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0004\nhttps://doi.org/10.1007/s40134-019-0335-7\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0006\nhttps://doi.org/10.1016/j.tube.2020.102049\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0008\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0009\nhttps://doi.org/10.1097/00000542-200401000-00006\nhttps://doi.org/10.1186/s13089-019-0145-4\nhttps://doi.org/10.1136/bmjopen-2018-027179\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0013\nhttps://doi.org/10.1093/ofid/ofz094\nhttps://doi.org/10.9745/GHSP-D-19-00251\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0016\nhttps://www.winfocus.org/case-study/diagnostic-accuracy-of-point-of-care-lung-ultrasound-among-patients-suspected-of-having-pulmonary-tuberculosis/\nhttps://doi.org/10.1177/0049475517709633\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0019\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0020\nhttps://doi.org/10.1371/journal.pone.0251236\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0022\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0023\nhttps://doi.org/10.1016/j.jbi.2019.103208\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0025\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0026\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0027\n\n\nM. Lightowler, J. Min, A. Nape et al. International Journal of Infectious Diseases 163 (2026) 108239\n\n[\n\n[\n\n[ \nSaharan Africa—a balanced SWOT analysis. Int J Infect Dis 2022; 123 :46–51 \nPMID: 35811083. doi: 10.1016/j.ijid.2022.07.009 . \n\n29] Conlon TW, Yousef N, Mayordomo-Colunga J, Tissot C, Fraga MV, Bhom- \nbal S, et al. Establishing a risk assessment framework for point-of-care ul- \n\ntrasound. Eur J Pediatr 2022; 181 (4):1449–57 PMID: 34846557. doi: 10.1007/ \ns00431- 021- 04324- 4 . \n10\n30] Shen YT, Chen L, Yue WW, Xu HX. Artificial intelligence in ultrasound. Eur J \nRadiol 2021; 139 :109717 . \n\n31] Dubois MM, Brooks MB, Malik AA, Siddiqui S, Ahmed JF, Jaswal M, et al. Age-\nspecific clinical presentation and risk factors for extrapulmonary tuberculosis \n\ndisease in children. Pediatr Infect Dis J 2022; 41 (8):620–5 PMID: 35849791; \nPMCID: PMC9281512. doi: 10.1097/INF.0 0 0 0 0 0 0 0 0 0 0 03584 . \n\nhttps://doi.org/10.1016/j.ijid.2022.07.009\nhttps://doi.org/10.1007/s00431-021-04324-4\nhttp://refhub.elsevier.com/S1201-9712(25)00461-8/sbref0030\nhttps://doi.org/10.1097/INF.0000000000003584\n\n\tDiagnostic accuracy of lung ultrasound for pulmonary tuberculosis in outpatients from Papua New Guinea, a resource-limited setting\n\t1 Introduction\n\t2 Methods\n\t2.1 Study setting and design\n\t2.2 Study population\n\t2.3 Study procedures\n\t2.3.1 Ultrasound\n\t2.3.2 Chest X-ray\n\n\t2.4 Data collection and management\n\t2.5 Study outcomes and data analyses\n\t2.6 Ethical considerations\n\n\t3 Results\n\t3.1 Participants’ characteristics\n\t3.2 LUS findings\n\t3.3 CXR findings\n\t3.4 Diagnostic accuracy of combined imaging findings for TB detection\n\t3.5 Agreement between CXR and LUS\n\t3.6 LUS-based diagnostic score\n\n\t4 Discussion\n\t4.1 Study findings\n\t4.2 Strengths and limitations\n\n\t5 Conclusions\n\tEthical approval\n\tFunding\n\tDeclaration of competing interest\n\tAcknowledgments\n\tAuthor contributions\n\tSupplementary materials\n\tReferences\n\n\n","UrlSlug":"diagnostic-accuracy-lung-ultrasound-pulmonary-tuberculosis-out-patients-papua-new-guinea-resource-limited-setting","collections":[],"countries":[{"Name":"Papua New Guinea","Id":171}],"asset_languages":[{"AlternateTitle":"","language":[{"Id":1,"Name":"English"}],"contentTranslation":[],"contentLocation":[]}],"creators":[{"Name":"Huerga H","Id":17224,"ac":[{"Order":11}]},{"Name":"Nicholas S","Id":18073,"ac":[{"Order":10}]},{"Name":"Sannino L","Id":23886,"ac":[{"Order":9}]},{"Name":"Lightowler M","Id":24725,"ac":[{"Order":0}]},{"Name":"Fentress M","Id":25652,"ac":[{"Order":6}]},{"Name":"Stratta E","Id":34696,"ac":[{"Order":5}]},{"Name":"Min J","Id":36237,"ac":[{"Order":1}]},{"Name":"Nape A","Id":37588,"ac":[{"Order":2}]},{"Name":"Galowa T","Id":37589,"ac":[{"Order":3}]},{"Name":"Hemo C","Id":37590,"ac":[{"Order":4}]},{"Name":"Weber S","Id":37591,"ac":[{"Order":7}]},{"Name":"Huson M","Id":37592,"ac":[{"Order":8}]}],"keywords":[{"Name":"tuberculosis","Id":3},{"Name":"diagnostics","Id":7},{"Name":"ultrasound","Id":26}]}}

Science Portal

Diagnostic accuracy of lung ultrasound for pulmonary tuberculosis in out-patients from Papua New Guinea, a resource-limited setting