Cross-site validation of lung cancer diagnosis by electronic nose with deep learning: a multicenter prospective study.

BACKGROUND: Although electronic nose (eNose) has been intensively investigated for diagnosing lung cancer, cross-site validation remains a major obstacle to be overcome and no studies have yet been performed. METHODS: Patients with lung cancer, as well as healthy control and diseased control groups, were prospectively recruited from two referral centers between 2019 and 2022. Deep learning models for detecting lung cancer with eNose breathprint were developed using training cohort from one site and then tested on cohort from the other site. Semi-Supervised Domain-Generalized (Semi-DG) Augmentation (SDA) and Noise-Shift Augmentation (NSA) methods with or without fine-tuning was applied to improve performance. RESULTS: In this study, 231 participants were enrolled, comprising a training/validation cohort of 168 individuals (90 with lung cancer, 16 healthy controls, and 62 diseased controls) and a test cohort of 63 individuals (28 with lung cancer, 10 healthy controls, and 25 diseased controls). The model has satisfactory results in the validation cohort from the same hospital while directly applying the trained model to the test cohort yielded suboptimal results (AUC, 0.61, 95% CI: 0.47─0.76). The performance improved after applying data augmentation methods in the training cohort (SDA, AUC: 0.89 [0.81─0.97]; NSA, AUC:0.90 [0.89─1.00]). Additionally, after applying fine-tuning methods, the performance further improved (SDA plus fine-tuning, AUC:0.95 [0.89─1.00]; NSA plus fine-tuning, AUC:0.95 [0.90─1.00]). CONCLUSION: Our study revealed that deep learning models developed for eNose breathprint can achieve cross-site validation with data augmentation and fine-tuning. Accordingly, eNose breathprints emerge as a convenient, non-invasive, and potentially generalizable solution for lung cancer detection. CLINICAL TRIAL REGISTRATION: This study is not a clinical trial and was therefore not registered.

Gut microbiota composition can reflect immune responses of latent tuberculosis infection in patients with poorly controlled diabetes.

BACKGROUND: Diabetes mellitus (DM) is a major risk factor for tuberculosis (TB). Evidence has linked the DM-related dysbiosis of gut microbiota to modifiable host immunity to Mycobacterium tuberculosis infection. However, the crosslinks between gut microbiota composition and immunological effects on the development of latent TB infection (LTBI) in DM patients remain uncertain. METHODS: We prospectively obtained stool, blood samples, and medical records from 130 patients with poorly-controlled DM (pDM), defined as ever having an HbA1c > 9.0% within previous 1 year. Among them, 43 had LTBI, as determined by QuantiFERON-TB Gold in-Tube assay. The differences in the taxonomic diversity of gut microbiota between LTBI and non-LTBI groups were investigated using 16S ribosomal RNA sequencing, and a predictive algorithm was established using a random forest model. Serum cytokine levels were measured to determine their correlations with gut microbiota. RESULTS: Compared with non-LTBI group, the microbiota in LTBI group displayed a similar alpha-diversity but different beta-diversity, featuring decrease of Prevotella_9, Streptococcus, and Actinomyces and increase of Bacteroides, Alistipes, and Blautia at the genus level. The accuracy was 0.872 for the LTBI prediction model using the aforementioned 6 microbiome-based biomarkers. Compared with the non-LTBI group, the LTBI group had a significantly lower serum levels of IL-17F (p = 0.025) and TNF-α (p = 0.038), which were correlated with the abundance of the aforementioned 6 taxa. CONCLUSIONS: The study results suggest that gut microbiome composition maybe associated with host immunity relevant to TB status, and gut microbial signature might be helpful for the diagnosis of LTBI.

Deep Learning-based Diagnosis and Localization of Pneumothorax on Portable Supine Chest X-ray in Intensive and Emergency Medicine: A Retrospective Study.

PURPOSE: To develop two deep learning-based systems for diagnosing and localizing pneumothorax on portable supine chest X-rays (SCXRs). METHODS: For this retrospective study, images meeting the following inclusion criteria were included: (1) patient age ≥ 20 years; (2) portable SCXR; (3) imaging obtained in the emergency department or intensive care unit. Included images were temporally split into training (1571 images, between January 2015 and December 2019) and testing (1071 images, between January 2020 to December 2020) datasets. All images were annotated using pixel-level labels. Object detection and image segmentation were adopted to develop separate systems. For the detection-based system, EfficientNet-B2, DneseNet-121, and Inception-v3 were the architecture for the classification model; Deformable DETR, TOOD, and VFNet were the architecture for the localization model. Both classification and localization models of the segmentation-based system shared the UNet architecture. RESULTS: In diagnosing pneumothorax, performance was excellent for both detection-based (Area under receiver operating characteristics curve [AUC]: 0.940, 95% confidence interval [CI]: 0.907-0.967) and segmentation-based (AUC: 0.979, 95% CI: 0.963-0.991) systems. For images with both predicted and ground-truth pneumothorax, lesion localization was highly accurate (detection-based Dice coefficient: 0.758, 95% CI: 0.707-0.806; segmentation-based Dice coefficient: 0.681, 95% CI: 0.642-0.721). The performance of the two deep learning-based systems declined as pneumothorax size diminished. Nonetheless, both systems were similar or better than human readers in diagnosis or localization performance across all sizes of pneumothorax. CONCLUSIONS: Both deep learning-based systems excelled when tested in a temporally different dataset with differing patient or image characteristics, showing favourable potential for external generalizability.

Whole-Blood 3-Gene Signature as a Decision Aid for Rifapentine-based Tuberculosis Preventive Therapy.

BACKGROUND: Systemic drug reaction (SDR) is a major safety concern with weekly rifapentine plus isoniazid for 12 doses (3HP) for latent tuberculosis infection (LTBI). Identifying SDR predictors and at-risk participants before treatment can improve cost-effectiveness of the LTBI program. METHODS: We prospectively recruited 187 cases receiving 3HP (44 SDRs and 143 non-SDRs). A pilot cohort (8 SDRs and 12 non-SDRs) was selected for generating whole-blood transcriptomic data. By incorporating the hierarchical system biology model and therapy-biomarker pathway approach, candidate genes were selected and evaluated using reverse-transcription quantitative polymerase chain reaction (RT-qPCR). Then, interpretable machine learning models presenting as SHapley Additive exPlanations (SHAP) values were applied for SDR risk prediction. Finally, an independent cohort was used to evaluate the performance of these predictive models. RESULTS: Based on the whole-blood transcriptomic profile of the pilot cohort and the RT-qPCR results of 2 SDR and 3 non-SDR samples in the training cohort, 6 genes were selected. According to SHAP values for model construction and validation, a 3-gene model for SDR risk prediction achieved a sensitivity and specificity of 0.972 and 0.947, respectively, under a universal cutoff value for the joint of the training (28 SDRs and 104 non-SDRs) and testing (8 SDRs and 27 non-SDRs) cohorts. It also worked well across different subgroups. CONCLUSIONS: The prediction model for 3HP-related SDRs serves as a guide for establishing a safe and personalized regimen to foster the implementation of an LTBI program. Additionally, it provides a potential translational value for future studies on drug-related hypersensitivity.

Impact of Age on Outcome of Rifapentine-Based Weekly Therapy for Latent Tuberculosis Infection.

BACKGROUND: Weekly rifapentine and isoniazid (3HP) is gaining popularity for latent tuberculosis infection treatment because of its short course and high completion rate. Prior to widespread use, comprehensive 3HP treatment assessment covering an all-age population is essential. METHODS: Participants receiving ≥1 3HP dose from September 2014 to December 2019 were stratified into elderly (≥65 years), middle-aged (>35 & <65 years), and younger (≤35 years) age groups. This study investigated the impact of age on treatment outcome, particularly systemic drug reactions (SDRs) and 3HP discontinuation. RESULTS: Overall, 134 of 579 (23.1%) participants were elderly. The completion rate was 83.1% overall and was highest and lowest in the younger group (94.5%) and elderly (73.9%) group, respectively. However, the 3HP discontinuation rate was not significantly different among the 3 groups in multivariate logistic regression analysis. In total, 362 (62.5%) participants experienced 1 or more adverse drug reactions (ADRs), of which 38 (10.5%) and 98 (27.1%) required temporary and permanent treatment interruption, respectively. The SDR risk was 11.2% in overall and 17.1% in the middle-aged group, 3.04-fold higher than that in the elderly group (P = .025). This finding was consistently observed in different clinical settings. Hypertensive events accompanied with flu-like symptoms occurred in 11.2% of elderly participants, and accounted for 50% of grade ≥3 ADRs. CONCLUSIONS: With proper medical support and programmatic follow-up, the 3HP completion rate is >70% even in elderly participants. In middle-aged and elderly individuals, 3HP should be employed with caution because of risk of SDRs and hypertensive events, respectively. Summary: Under programmatic medical support, widespread use of weekly rifapentine and isoniazid (3HP) for latent tuberculosis treatment is possible for its high completion rate. 3HP should be employed with caution for risk of systemic drug reactions and hypertensive events in middle-aged and elderly individuals, respectively.

Completion Rate and Safety of Programmatic Screening and Treatment for Latent Tuberculosis Infection in Elderly Patients With Poorly Controlled Diabetic Mellitus: A Prospective Multicenter Study.

BACKGROUND: Poor control of diabetes mellitus (DM) increases active tuberculosis (TB) risk. Understanding risk factors for latent TB infection (LTBI) in this population and intervention completion rates is crucial for policy making. METHODS: Under a collaborative multidisciplinary team consisting of public health professionals, endocrinologists, and pulmonologists, patients aged >45 years with poorly controlled DM (pDM), defined as having a glycated hemoglobin level of ≥9% within the preceding year, were enrolled by endocrinologists from 2 hospitals; these patients underwent LTBI screening by using QuantiFERON (QFT). Once-weekly isoniazid and rifapentine for 12 weeks (3HP) or daily isoniazid for 9 months (9H) was administered by pulmonologists. QFT-positivity predictors were evaluated using logistic regression. Completion rates and safety were also investigated. RESULTS: Among 980 patients with pDM (age: 64.2 ±â€…9.7 years), 261 (26.6%) were QFT-positive. Age, DM duration, chronic kidney disease stage ≥3, and dipeptidyl peptidase-4 inhibitor use, not using metformin, were associated with QFT-positivity. Preventive therapy (3HP: 138; 9H: 62) was administered in 200 (76.6%) QFT-positive patients. The completion rates of 3HP and 9H were 84.1% and 79.0%, respectively (P = .494). Nine (6.5%) and zero patients in the 3HP and 9H groups, respectively, developed systemic drug reactions (P = .059); 78.3% and 45.2% had ≥1 adverse drug reactions (P < .001); and post-treatment QFT conversion rates were 32% and 20%, respectively (P = .228). CONCLUSIONS: LTBI prevalence exceeds 25% in elderly patients with pDM. Under care from a collaborative multidisciplinary team, the completion rate of preventive therapy, regardless of regimen could approach, or even exceed 80% in this population.

Predictors of Successful Weaning from Noninvasive Ventilation in Patients with Acute Exacerbation of Chronic Obstructive Pulmonary Disease: A Single-Center Retrospective Cohort Study.

PURPOSE: Noninvasive ventilation (NIV) is often required for patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD), and it can significantly reduce the need for endotracheal intubation. Currently, there is no standard method for predicting successful weaning from NIV. Therefore, we aimed to evaluate whether a weaning index can predict NIV outcomes of patients with AECOPD. METHODS: This study was conducted at a single academic public hospital in northern Taiwan from February 2019 to January 2021. Patients with AECOPD admitted to the hospital with respiratory failure who were treated with NIV were included in the study. Univariate and multivariate logistic regression analyses were used to identify independent predictors of successful weaning from NIV. Receiver operating characteristic curve methodology was used to assess the predictive capacity. RESULTS: A total of 85 patients were enrolled, 65.9% of whom were successfully weaned from NIV. The patients had a mean age of 75.8 years and were mostly men (89.4%). The rapid shallow breathing index (RSBI) (P < 0.001), maximum inspiratory pressure (P = 0.014), and maximum expiratory pressure (P = 0.004) of the successful group were significant while preparing to wean. The area under the receiver operating characteristic curve for the RSBI was 0.804, which was considered excellent discrimination. CONCLUSION: The RSBI predicted successful weaning from NIV in patients with AECOPD with hypercapnic respiratory failure. This index may be useful for selecting patients with AECOPD that are suitable for NIV weaning.

Nontuberculous mycobacterial lung disease epidemiology in Taiwan: A systematic review.

Nontuberculous mycobacteria (NTM) are critical emerging global infectious pathogens. Though NTM can be mere colonizers when isolated from human specimens, NTM are also responsible for diverse human infections. NTM-lung disease (NTM-LD) is the most common human disease entity. The present review aims to provide general insight into NTM-LD epidemiology in Taiwan. In reviewing NTM epidemiology in Taiwan, we discovered three distinguishing features. First, NTM disease incidence has increased in Taiwan over the past decade. Second, the distribution of NTM varies geographically in Taiwan. Mycobacterium avium-intracellulare complex (MAC) is the dominant species in northern Taiwan, whereas Mycobacterium abscessus complex and MAC may be equally dominant in southern Taiwan. Third, researchers in Taiwan have published valuable research investigating NTM among special patient populations, including patients in intensive care units, with ventilator dependency, with pulmonary tuberculosis, and who are infected with specific NTM species. The largest obstacle to clarifying NTM epidemiology in Taiwan may be the lack of routine NTM species identification in laboratories. Increased awareness of NTM diseases and acknowledgment that NTM species identification is crucial and guides clinical management are essential steps for facilitating the identification of NTM species in laboratories.

Clinical impact of using fluoroquinolone with low antimycobacterial activity on treatment delay in tuberculosis: Hospital-based and population-based cohort study.

BACKGROUND/PURPOSE: Little remains known regarding whether newer FQ with less anti-mycobacterial activity (gemifloxacin) would reduce treatment delay. METHODS: We identified one hospital-based cohort (HBC) and one population-based cohort (PBC) including patients receiving amoxicillin/clavulanate acid (Beta-lactam), gemifloxacin (Gemi), and fluoroquinolones other than gemifloxacin (Non-Gemi FQ) prior to TB treatment. RESULTS: A total of 201 patients in the HBC and 3544 patients in the PBC were recruited. After 1:1 propensity score matching, TB treatment delay was statistically insignificant between Beta-lactam, Gemi group, and Non-Gemi FQ group in HBC (Beta-lactam vs Gemi: 22.3 ± 21.4 d vs 28.6 ± 27.9 d, p = 0.292; Beta-lactam vs Non-Gemi FQ: 33.3 ± 26.5 d vs 50.3 ± 47.3 d, p = 0.135) and PBC (Beta-lactam vs Gemi: 26.4 ± 29.1 vs 25.0 ± 28.1, p = 0.638; Beta-lactam vs Non-Gemi FQ: 29.4 ± 36.0 d vs 32.7 ± 35.0 d, p = 0.124, Non-Gemi FQ vs Gemi: 28.4 ± 33.0 d vs 25.0 ± 28.1 d, p = 0.29). CONCLUSION: While limited by relatively low case number, our study showed that use of gemifloxacin neither results in nor reduces delay in TB treatment. The issue of FQ use on TB treatment delay was also not observed in our study. Early survey and maintaining high clinical alertness remains the key to reducing TB treatment delay.

Impact of metformin use among tuberculosis close contacts with diabetes mellitus in a nationwide cohort study.

BACKGROUND: The protective effect of metformin against active tuberculosis (TB) among TB close contacts is unknown. METHODS: TB close contacts with diabetes mellitus (DM) and normal renal function were selected from the National Health Insurance Research Database of Taiwan. Metformin users were patients who received ≥90 cumulative defined daily doses within 1 year before the index date. For each metformin user, a propensity-score matched metformin nonuser and an age- and sex-matched healthy TB close contact were selected. The outcome was incident TB, identified using previously validated diagnostic criteria. Independent predictors were investigated using stratified Cox regression analysis. Interaction analysis was also performed. RESULTS: A total of 5846 TB close contacts who were metformin users, metformin non-users, and healthy contacts were analysed. The incidence of active TB was 755 (600-938), 1117 (927-1335), and 526 (393-689) cases per 100,000 person-years in each group, respectively. Multivariate analysis revealed that healthy contacts had the lowest risk of developing active TB (adjusted hazard ratio [aHR]: 0.42 [0.30-0.60]) and metformin use partially reversed the risk associated with DM (aHR: 0.73 [0.54-0.98]). Subpopulation analysis revealed a significant interaction between insulin use and metformin use. CONCLUSIONS: Metformin use is associated with a lower risk of developing active TB among TB close contacts with DM, especially for insulin users. It may be an alternative choice for primary prevention against active TB if no contraindications exist. However, prospective studies are needed to confirm the findings.

Acute biliary events during anti-tuberculosis treatment: hospital case series and a nationwide cohort study.

BACKGROUND: Tuberculosis (TB) remains one of the major infectious diseases worldwide. Adverse reactions are common during TB treatment. Few reports, however, are available on treatment-related acute biliary events (ABEs), such as cholelithiasis, biliary obstruction, acute cholecystitis, and cholangitis. METHODS: We first report four pulmonary TB patients who developed ABEs during anti-TB treatment. Abdominal sonography revealed multiple gall stones with dilated intrahepatic ducts in three patients and cholecystitis in one patient. To investigate the incidence of and risk factors for ABEs during anti-TB treatment, we subsequently conducted a nationwide cohort study using the National Health Insurance Research Database of Taiwan. RESULTS: A total of 159,566 pulmonary TB patients were identified from the database between 1996 and 2010, and among them, 195 (0.12%) developed ABEs within 180 days after beginning anti-TB treatment. Logistic regression analysis revealed that the risk factors associated with ABEs are older age (relative risk [RR]: 1.32 [1.21-1.44] per 10-year increment) and diabetes mellitus (RR: 1.59 [1.19-2.13]). CONCLUSIONS: Although infrequently encountered, ABEs should be considered among patients with TB who experience abdominal discomfort with hyperbilirubinemia, especially patients who have older age or diabetes.

Update on tuberculosis biomarkers: From correlates of risk, to correlates of active disease and of cure from disease.

Tuberculosis (TB) remains a devastating disease, yet despite its enormous toll on global health, tools to control TB are insufficient and often outdated. TB Biomarkers (TB-BM) would constitute extremely useful tools to measure infection status and predict outcome of infection, vaccination or therapy. There are several types of TB-BM: Correlate of Infection; Correlate of TB Disease; Correlate of Increased Risk of Developing Active TB Disease; Correlate of the Curative Response to Therapy; and Correlate of Protection (CoP). Most TB-BM currently studied are host-derived BM, and consist of transcriptomic, proteomic, metabolomic, cellular markers or marker combinations ('signatures'). In particular, vaccine-inducible CoP are expected to be transformative in developing new TB vaccines as they will de-risk vaccine research and development (R&D) as well as human testing at an early stage. In addition, CoP could also help minimizing the need for preclinical studies in experimental animals. Of key importance is that TB-BM are tested and validated in different well-characterized human TB cohorts, preferably with complementary profiles and geographically diverse populations: genetic and environmental factors such as (viral) coinfections, exposure to non-tuberculous mycobacteria, nutritional status, metabolic status, age (infants vs children vs adolescents vs adults) and other factors impact host immune set points and host responses across different populations. In this study, we review the most recent advances in research into TB-BM for the diagnosis of active TB, risk of TB development and treatment-induced TB cure.

Diabetes Mellitus and Latent Tuberculosis Infection: A Systematic Review and Metaanalysis.

Background: Despite the well-documented association between diabetes and active tuberculosis, evidence of the association between diabetes and latent tuberculosis infection (LTBI) remains limited and inconsistent. Methods: We included observational studies that applied either the tuberculin skin test or the interferon gamma release assay for diagnosis of LTBI and that provided adjusted effect estimate for the association between diabetes and LTBI. We searched PubMed and EMBASE through 31 January 2016. The risk of bias of included studies was assessed using a quality assessment tool modified from the Newcastle-Ottawa scale. Results: Thirteen studies (1 cohort study and 12 cross-sectional studies) were included, involving 38263 participants. The cohort study revealed an increased but nonsignificant risk of LTBI among diabetics (risk ratio, 4.40; 95% confidence interval [CI], 0.50-38.55). For the cross-sectional studies, the pooled odds ratio from the random-effects model was 1.18 (95% CI, 1.06-1.30), with a small statistical heterogeneity across studies (I2, 3.5%). The risk of bias assessment revealed several methodological issues, but the overall direction of biases would reduce the positive causal association between diabetes and LTBI. Conclusions: Diabetes was associated with a small but statistically significant risk for LTBI. Findings from this review could be used to inform future cost-effectiveness analysis on the impact of LTBI screening programs among diabetics.

Mycobacterium abscessus Complex Infections in Humans.

Mycobacterium abscessus complex comprises a group of rapidly growing, multidrug-resistant, nontuberculous mycobacteria that are responsible for a wide spectrum of skin and soft tissue diseases, central nervous system infections, bacteremia, and ocular and other infections. M. abscessus complex is differentiated into 3 subspecies: M. abscessus subsp. abscessus, M. abscessus subsp. massiliense, and M. abscessus subsp. bolletii. The 2 major subspecies, M. abscessus subsp. abscessus and M. abscessus subsp. massiliense, have different erm(41) gene patterns. This gene provides intrinsic resistance to macrolides, so the different patterns lead to different treatment outcomes. M. abscessus complex outbreaks associated with cosmetic procedures and nosocomial transmissions are not uncommon. Clarithromycin, amikacin, and cefoxitin are the current antimicrobial drugs of choice for treatment. However, new treatment regimens are urgently needed, as are rapid and inexpensive identification methods and measures to contain nosocomial transmission and outbreaks.

Quantifying the contribution of smear-negative, culture-positive pulmonary tuberculosis to nosocomial transmission.

BACKGROUND: Despite current guidelines for tuberculosis (TB) control in health care settings, which focused on smear-positive cases, prevention of nosocomial TB transmission continues to be a challenge. Here, we report the results of the first hospital-wide prospective study applying interferon-gamma release assay to investigate the role of smear-negative, culture-positive index cases in nosocomial TB transmission. METHODS: We prospectively identified cases of culture-confirmed smear-negative pulmonary TB receiving aerosol-generating procedures (AGPs) and cases of culture-confirmed smear-positive pulmonary TB admitted at a medical center. Nosocomial transmission was evaluated by screening their close contacts for latent TB infection (LTBI) using an interferon-gamma release assay. RESULTS: A total of 93 smear-negative index receiving AGP and 122 smear-positive index were enrolled. Among them, 13 (14.0%) and 43 (35.2%) index cases, respectively, had secondary cases of LTBI (P < .001). Sputum smear negativity (adjusted odds ratio: 0.20 [0.08-0.48]) and AGP (sputum suction; adjusted odds ratio: 3.48 [1.34-9.05]) are independent factors of transmission. A similar proportion in the close contacts of the 2 index groups had LTBI (17 [15.3%] and 63 [16.0%], respectively), and the former index group contributed to 21.3% of the nosocomial transmission. CONCLUSIONS: Smear-negative, culture-positive index cases receiving AGPs could be as infectious as smear-positive index cases. Hospital TB control policy should also focus on the former group.

Sputum bacterial microbiota signature as a surrogate for predicting disease progression of nontuberculous mycobacterial lung disease.

OBJECTIVES: Predicting progression of nontuberculous mycobacterial lung disease (NTM-LD) remains challenging. This study evaluated whether sputum bacterial microbiome diversity can be the biomarker and provide novel insights into related phenotypes and treatment timing. METHODS: We analyzed 126 sputum microbiomes of 126 patients with newly diagnosed NTM-LD due to Mycobacterium avium complex, M. abscessus complex, and M. kansasii between May 2020 and December 2021. Patients were followed for 2 years to determine their disease progression status. We identified consistently representative genera that differentiated the progressor and nonprogressor by using six methodologies. These genera were used to construct a prediction model using random forest with 5-fold cross validation. RESULTS: Disease progression occurred in 49 (38.6%) patients. Compared with nonprogressors, α-diversity was lower in the progressors. Significant compositional differences existed in the ß-diversity between groups (p=0.001). The prediction model for NTM-LD progression constructed using seven genera (Burkholderia, Pseudomonas, Sphingomonas, Candidatus Saccharibacteria, Phocaeicola, Pelomonas, and Phascolarctobacterium) with significantly differential abundance achieved an area under curve of 0.871. CONCLUSIONS: Identification of the composition of sputum bacterial microbiome facilitates prediction of the course of NTM-LD, and maybe used to develop precision treatment involving modulating the respiratory microbiome composition to ameliorate NTM-LD.

Deep Learning-based Diagnosis of Pulmonary Tuberculosis on Chest X-ray in the Emergency Department: A Retrospective Study.

Prompt and correct detection of pulmonary tuberculosis (PTB) is critical in preventing its spread. We aimed to develop a deep learning-based algorithm for detecting PTB on chest X-ray (CXRs) in the emergency department. This retrospective study included 3498 CXRs acquired from the National Taiwan University Hospital (NTUH). The images were chronologically split into a training dataset, NTUH-1519 (images acquired during the years 2015 to 2019; n = 2144), and a testing dataset, NTUH-20 (images acquired during the year 2020; n = 1354). Public databases, including the NIH ChestX-ray14 dataset (model training; 112,120 images), Montgomery County (model testing; 138 images), and Shenzhen (model testing; 662 images), were also used in model development. EfficientNetV2 was the basic architecture of the algorithm. Images from ChestX-ray14 were employed for pseudo-labelling to perform semi-supervised learning. The algorithm demonstrated excellent performance in detecting PTB (area under the receiver operating characteristic curve [AUC] 0.878, 95% confidence interval [CI] 0.854-0.900) in NTUH-20. The algorithm showed significantly better performance in posterior-anterior (PA) CXR (AUC 0.940, 95% CI 0.912-0.965, p-value < 0.001) compared with anterior-posterior (AUC 0.782, 95% CI 0.644-0.897) or portable anterior-posterior (AUC 0.869, 95% CI 0.814-0.918) CXR. The algorithm accurately detected cases of bacteriologically confirmed PTB (AUC 0.854, 95% CI 0.823-0.883). Finally, the algorithm tested favourably in Montgomery County (AUC 0.838, 95% CI 0.765-0.904) and Shenzhen (AUC 0.806, 95% CI 0.771-0.839). A deep learning-based algorithm could detect PTB on CXR with excellent performance, which may help shorten the interval between detection and airborne isolation for patients with PTB.

Matrix-assisted laser desorption ionization-time of flight mass spectrometry can accurately differentiate between Mycobacterium masilliense (M. abscessus subspecies bolletti) and M. abscessus (sensu stricto).

Among 36 Mycobacterium masilliense and 22 M. abscessus isolates identified by erm(41) PCR and sequencing analysis of rpoB and 23S rRNA genes, the rate of accurate differentiation between these two subspecies was 100% by cluster analysis of spectra generated by Bruker Biotyper matrix-assisted laser desorption ionization-time of flight mass spectrometry.