Review
Abstract
Background: Tuberculosis (TB) remains a significant health concern, contributing to the highest mortality among infectious diseases worldwide. However, none of the various TB diagnostic tools introduced is deemed sufficient on its own for the diagnostic pathway, so various artificial intelligence (AI)–based methods have been developed to address this issue.
Objective: We aimed to provide a comprehensive evaluation of AI-based algorithms for TB detection across various data modalities.
Methods: Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) 2020 guidelines, we conducted a systematic review to synthesize current knowledge on this topic. Our search across 3 major databases (Scopus, PubMed, Association for Computing Machinery [ACM] Digital Library) yielded 1146 records, of which we included 152 (13.3%) studies in our analysis. QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies version 2) was performed for the risk-of-bias assessment of all included studies.
Results: Radiographic biomarkers (n=129, 84.9%) and deep learning (DL; n=122, 80.3%) approaches were predominantly used, with convolutional neural networks (CNNs) using Visual Geometry Group (VGG)-16 (n=37, 24.3%), ResNet-50 (n=33, 21.7%), and DenseNet-121 (n=19, 12.5%) architectures being the most common DL approach. The majority of studies focused on model development (n=143, 94.1%) and used a single modality approach (n=141, 92.8%). AI methods demonstrated good performance in all studies: mean accuracy=91.93% (SD 8.10%, 95% CI 90.52%-93.33%; median 93.59%, IQR 88.33%-98.32%), mean area under the curve (AUC)=93.48% (SD 7.51%, 95% CI 91.90%-95.06%; median 95.28%, IQR 91%-99%), mean sensitivity=92.77% (SD 7.48%, 95% CI 91.38%-94.15%; median 94.05% IQR 89%-98.87%), and mean specificity=92.39% (SD 9.4%, 95% CI 90.30%-94.49%; median 95.38%, IQR 89.42%-99.19%). AI performance across different biomarker types showed mean accuracies of 92.45% (SD 7.83%), 89.03% (SD 8.49%), and 84.21% (SD 0%); mean AUCs of 94.47% (SD 7.32%), 88.45% (SD 8.33%), and 88.61% (SD 5.9%); mean sensitivities of 93.8% (SD 6.27%), 88.41% (SD 10.24%), and 93% (SD 0%); and mean specificities of 94.2% (SD 6.63%), 85.89% (SD 14.66%), and 95% (SD 0%) for radiographic, molecular/biochemical, and physiological types, respectively. AI performance across various reference standards showed mean accuracies of 91.44% (SD 7.3%), 93.16% (SD 6.44%), and 88.98% (SD 9.77%); mean AUCs of 90.95% (SD 7.58%), 94.89% (SD 5.18%), and 92.61% (SD 6.01%); mean sensitivities of 91.76% (SD 7.02%), 93.73% (SD 6.67%), and 91.34% (SD 7.71%); and mean specificities of 86.56% (SD 12.8%), 93.69% (SD 8.45%), and 92.7% (SD 6.54%) for bacteriological, human reader, and combined reference standards, respectively. The transfer learning (TL) approach showed increasing popularity (n=89, 58.6%). Notably, only 1 (0.7%) study conducted domain-shift analysis for TB detection.
Conclusions: Findings from this review underscore the considerable promise of AI-based methods in the realm of TB detection. Future research endeavors should prioritize conducting domain-shift analyses to better simulate real-world scenarios in TB detection.
Trial Registration: PROSPERO CRD42023453611; https://www.crd.york.ac.uk/PROSPERO/view/CRD42023453611
doi:10.2196/69068
Keywords
Introduction
Tuberculosis (TB) stood as the dominant infectious disease threat worldwide, affecting over 10.5 million individuals and claiming the lives of 1.3 million people in 2022 []. A pivotal strategy in halting this worldwide epidemic revolves around breaking the transmission chain by identifying and treating all individuals with infectious forms of TB []. Various diagnostic tools have been introduced, including chest radiography, tuberculin skin tests, interferon-gamma release assays, sputum smear microscopy, sputum mycobacterial culture, and an array of nucleic acid amplification tests []. Nonetheless, each of these diagnostic modalities faces significant implementation challenges, and none alone is deemed adequate for the diagnostic pathway. Consequently, researchers have endeavored to integrate artificial intelligence (AI)–based algorithms to augment the detection of TB.
AI encompasses a spectrum of techniques enabling computer programs to tackle intricate problems by emulating human cognitive processes []. Presently, 2 widely used terms denote AI techniques that necessitate minimal-to-no human intervention: machine learning (ML) and deep learning (DL). ML possesses the capability to discern meaningful patterns within datasets autonomously, without explicit programming []. Conversely, DL emerges as a subdomain within ML, leveraging deep and intricate neural networks to extract features or patterns from datasets for subsequent analysis [,].
In this study, we aimed to comprehensively evaluate the performance of AI-based algorithms, particularly ML and DL, for TB detection. This aligns closely with the first pillar of the World Health Organization (WHO) End TB Strategy, which includes systematic screening for TB in high-risk groups []. In fact, a novel recommendation was issued by WHO in 2021: the approval of AI tools to analyze chest X-rays (CXRs) for TB detection in place of human readers [].
Several notable review papers have emerged. Jimmy et al [] conducted a meticulous systematic review, delving into various DL methods for TB detection based on CXR. Similarly, Sharma et al [] contributed a narrative survey spotlighting DL-based convolutional neural networks (CNNs) tailored for TB diagnosis from CXR. Zeyu et al [] embarked on a narrative review aimed at dissecting diverse DL-based approaches for TB diagnosis. Oloko-Oba and Viriri [] scrutinized DL techniques for TB detection via chest radiographs. Santosh et al [] conducted a systematic review, albeit with a narrower time frame spanning 5 years (2016-2021), focusing on DL for TB screening using CXR. Carvalho et al [] contributed a systematic review on DL techniques used for classifying TB bacilli in microscopic Ziehl-Neelsen (ZN) sputum smear images. Da Silva Barros et al [] conducted a systematic review on ML models geared toward predicting TB treatment outcomes. Distinguished from these endeavors, our current review took a broader perspective, encompassing not only DL but also ML as integral components of AI-based methods for TB detection.
Siddiqui and Garg [] examined recent studies of intelligent techniques (ML and DL) for diagnosing pulmonary TB. Singh et al [] scrutinized the drawbacks of conventional TB diagnostics, while exploring the utility of ML and DL methods for TB diagnosis. Additionally, they highlighted several commercial computer-aided detection (CAD) tools as promising AI-driven instruments. Notably, both reviews were reported as narrative reviews, rather than using the systematic review methodology we adopted here.
Harris et al [] meticulously conducted a systematic review focusing on the diagnostic accuracy of AI-based tools in TB detection using CXR. Hansun et al [] recently published a systematic review examining the effectiveness of ML and DL methods for TB detection using CXR. Both studies aligned closely with our review but were limited to CXR. In contrast, our review encompassed a broader spectrum of diagnostic data, including radiographic, biochemical, physiological, and other clinical data. This broader scope enabled a more comprehensive evaluation of AI-based methods across diverse data modalities for TB detection.
Methods
Design, Registration, and Information Sources
In conducting this systematic review, we adhered to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines () []. Our review protocol, in accordance with the PRISMA-Protocol 2015 statement [,], was registered on PROSPERO (Prospective Register of Systematic Reviews; ID CRD42023453611) [,].
The systematic review drew upon 3 primary academic search systems and online databases: Scopus, PubMed, and the Association for Computing Machinery (ACM) Digital Library. These platforms are widely recognized as pivotal sources for comprehensive literature reviews []. The searches encompassed all published literature up to July 25, 2023.
Ethical Considerations
As this systematic literature review focused on retrospective studies, no ethical approval was required.
Search Strategy
For our search strategy, we used 3 primary keywords: “artificial intelligence,” “tuberculosis,” and “detect*.” Additionally, we incorporated several alternative terms for each main keyword to ensure comprehensive coverage during the search process, resulting in the findings outlined in 
Table 1. Initially, we identified a total of 1146 records; however, certain records were not available from the respective databases (n=58, 5.1%). The full search queries can be seen in .
| Search query | Scopus, n/N (%) | PubMed, n/N (%) | ACMa, n/N (%) |
| (TITLE-ABS-KEY (“Machine Learning” OR “Predictive Analytics” OR “Statistical Learning” OR “Deep Learning” OR “Artificial Intelligence” OR “AI”) AND TITLE-ABS-KEY (“Tuberculosis” OR “TB”) AND TITLE-ABS-KEY (“Early detection” OR “detect*” OR “Early diagnosis” OR “diagnosis”)) AND DOCTYPE (ar OR cp) AND ( LIMIT-TO (LANGUAGE , “English”)) AND (LIMIT-TO (SRCTYPE , “j”) OR LIMIT-TO (SRCTYPE , “p”)) | 857/1146 (74.8) | 284/1146 (24.8) | 5/1146 (0.4) |
| Not available for download (n=58) | 53/58 (91.4) | 5/58 (8.6) | 0 |
| Available for download (n=1088) | 804/1088 (73.9) | 279/1088 (25.6) | 5/1088 (0.5) |
aACM: Association for Computing Machinery.
Inclusion and Exclusion Criteria, Data Extraction, and Storage
All downloaded records underwent scrutiny against the inclusion and exclusion criteria outlined for this review study. There were 4 criteria that all included studies had to meet: (1) full-text papers reporting original data published in peer-reviewed journals or proceedings, (2) cross-sectional diagnostic test evaluations, (3) papers focused on TB detection using AI, and (4) papers written in English.
The information extracted from each included study comprised the title, authors, publication year, journal title, objectives, and outcome measures. Dataset characteristics, including their source and total number of data items, were documented, along with details regarding the modality and type of data used. The AI methods applied, evaluation techniques used, performance metrics, best results obtained, and any comparisons with other studies were recorded. Additionally, outcome types, citation counts, and information regarding sponsors or funding sources, were also gathered as part of the comprehensive extraction process.
Each review author was actively engaged in the conceptualization and execution of all phases outlined in the PRISMA 2020 guidelines. At least 2 review authors were involved in each phase, including identification, screening, and assessment of eligibility for inclusion. To ensure accuracy and transparency, we documented the results of each phase using a standardized spreadsheet.
Outcomes Assessed
The first outcome from this review was the compilation of diverse AI methods used for TB detection. The second outcome encompassed diagnostic performance exhibited by various AI methods in TB detection. This summary encompassed key metrics, such as accuracy, area under the receiver operating characteristic (ROC) curve (AUROC), sensitivity, and specificity, providing valuable insights into the efficacy and reliability of these methods in clinical practice.
Strategy for Data Analysis and Synthesis
We adopted a narrative synthesis approach, integrating information extracted from the included studies within the text. The narrative synthesis method facilitated an in-depth exploration of the relationships and insights both within and across included studies, stratified by various attributes, such as modality, biomarker types, and AI methods. Descriptive statistics, primarily using box-whisker and violin plots, alongside tables and figures, presented the quantitative outcomes of this systematic review. Given the pronounced heterogeneity observed in study designs, comparators, biomarker types, evaluation techniques, and AI methods encompassed in this review study, a meta-analysis was deemed impractical.
Risk-of-Bias Assessment
In assessing the risk of bias within the included studies, we used the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies version 2) tool. This tool is widely recommended for systematic reviews as it enables the evaluation of bias and applicability in diagnostic accuracy studies []. It comprises 4 key domains: (1) patient selection, (2) index test, (3) reference standard, and (4) flow and timing. To tailor the assessment to the specific focus of this review on evaluating AI-based methods’ performance in TB detection, adjustments were made to the questions in the QUADAS-2 tool. Two review authors (SH and AW) assessed each included study, and a third opinion was sought from the author AA or the author GBM in the case of disagreements.
Results
Overview
Figure 1 delineates all the processes conducted throughout the systematic review. We adhered to the PRISMA 2020 flowchart, including the incorporation of a process timeline and the inclusion of a registered protocol status in PROSPERO.
Of the 1146 records initially identified from Scopus (n=857, 74.8%), PubMed (n=284, 24.8%), and ACM Digital Library (n=5, 0.4%), 58 (5.1%) were found to be unavailable. Upon checking and removing duplicates (n=227, 20.9%), we were left with 861 (79.1%) unique titles to be passed to the screening phase. Subsequently, titles, abstracts, and keywords underwent examination based on 4 primary selection criteria, where 413 (48%) records remained for eligibility assessment.
All 413 records underwent thorough assessment by examining their full-text content. From this pool, 261 (63.2%) papers were excluded, leaving us with a total of 152 (36.8%) papers being included as part of this systematic review. Subsequently, the data extraction process was carried out for all included studies. The entirety of the review process spanned approximately 6 months (July 2023-January 2024).
General Characteristics of Included Studies
A summary of the general characteristics of the 152 studies included in this systematic review is included in Tables 2 and 3. The majority of the included studies opted for a single modality (n=141, 92.8%), with a smaller subset using a multimodal approach (n=11, 7.2%). Among the various data types used, radiographic data emerged as the most prevalent data type (n=129, 84.9%), followed by biochemical (n=21, 13.8%) and physiological (n=16, 10.5%) data types. Several studies using a multimodal approach integrated more than 1 data type, such as a combination of radiographic and physiological data types.
| Characteristics | Studies, n (%) | |
| Modality | ||
| Single | 141 (92.8) | |
| Multimodal | 11 (7.2) | |
| Evaluation technique | ||
| Holdout | 95 (62.5) | |
| k-Fold CVa | 48 (31.6) | |
| External data | 9 (5.9) | |
| Outcome type | ||
| Model | 143 (94.1) | |
| Application/prototype | 8 (5.3) | |
| Clinical evaluation | 1 (0.7) | |
| TLb | 89 (58.6) | |
| Comparison with other studies | 68 (44.7) | |
| Funding/sponsor | 78 (51.3) | |
aCV: cross-validation.
bTL: transfer learning.
| Characteristics | Studies, n (%) | |
| AIa type | ||
| MLb | 49 (32.2) | |
| DLc | 122 (80.3) | |
| Reference standard | ||
| Human reader | 103 (67.8) | |
| Bacteriological | 43 (28.3) | |
| Not available | 24 (15.8) | |
| Type of data | ||
| Radiographic | 129 (84.9) | |
| Biochemical | 21 (13.8) | |
| Physiological and clinical | 16 (10.5) | |
aAI: artificial intelligence.
bML: machine learning.
cDL: deep learning.
In terms of evaluation techniques, the majority of studies (n=95, 62.5%) used the holdout evaluation technique, which involves dividing the entire dataset into training and test sets. In total, 48 (31.6%) studies used the k-fold cross-validation (CV) technique, dividing the dataset into k partitions and conducting the evaluation k times. Only 9 (5.9%) studies incorporated external datasets during the evaluation phase. Of the 152 included studies, 143 (94.1%) focused on AI model development, while 8 (5.3%) studies concentrated on application or prototype creation, and 1 (0.7%) study examined the clinical evaluation of AI-based CAD tools.
Regarding the AI approaches, a large number of studies (n=122, 80.3%) used various DL methods, while 49 (32.2%) studies used ML methods. We also collected information regarding various reference standards used in the included studies. We classified them into 2 groups, namely the human reader and bacteriological standards. Of the 152 included studies, 103 (67.8%) used human reader reference standards, 43 (28.3%) used bacteriological reference standards, and 24 (15.8%) did not report any reference standard. Similar to the data-types group, both AI-type and reference standard groups were not mutually exclusive.
In addition, 89 (58.6%) studies used the TL approach, a relatively recent and validated strategy for addressing domain-specific challenges [], especially prevalent in studies using DL methods. Moreover, 68 (44.7%) studies conducted comparisons with multiple prior studies. Notably, an equal proportion of studies received funding or sponsorship from various entities (n=78, 51.3%) compared to those that did not receive any funding (n=74, 48.7%). All included studies were published between 2011 and 2023.
A total of 141 distinct data sources were identified across the included studies, comprising 2,072,457 records. Among these, the most frequently used data sources were Shenzhen (SZ; n=51, 36.2%), Montgomery County (MC; n=46, 32.6%), Kaggle TB CXR (n=6, 4.3%), and TBX11K (n=5, 3.5%), all of which provided radiographic data. MC and SZ were the predominant data sources, both publicly available CXR datasets from the US National Library of Medicine. MC contains 138 images, while SZ offers 662 images []. The Kaggle TB CXR dataset contains 7000 images, although only 4200 (60%) are publicly accessible []. Conversely, TBX11K, introduced in 2020, stands out as a newer and more extensive CXR dataset for TB-related research, featuring a total of 11,200 images []. For further insights into the extracted data from all included studies, detailed information is available in .
AI Techniques for TB Detection
Among the 152 included studies, the majority were published within the past 8 years, with only 3 (2%) studies predating 2016 [-]. As of January 18, 2024, these studies collectively garnered 6070 citations on Google Scholar. Remarkably, the top 3 (2%) cited publications were identified as Lakhani and Sundaram [] with 1695 citations, Pasa et al [] with 312 citations, and Lopes and Valiati [] with 260 citations. This underscored the burgeoning interest among researchers in integrating AI-based methods into TB detection studies.
The included studies implemented various AI-based algorithms, which could be broadly categorized into 2 major approaches: ML and DL. Specifically, 102 (67.1%) studies used DL methods [,,,-], while 30 (19.7%) studies used ML techniques [-,-]. Additionally, 19 (12.5%) studies used a combination of both ML and DL approaches [,-]. Furthermore, 1 (0.7%) study [] focused on the implementation of a DL-based CAD tool, namely Lunit INSIGHT v4.7.2, specifically designed for systematic TB screening in a low-prevalence setting.
Figure 2 illustrates the diverse array of ML methods used in the included studies. Notably, support vector machines (SVMs; n=33, 21.7%), random forests (RFs; n=17, 11.2%), and logistic regression (LR; n=15, 9.9%) emerged as the 3 most prevalent ML techniques.
CNNs were the predominant DL method among the included studies. Alongside custom CNNs (n=39, 25.7%), many researchers have leveraged established DL architectures, notably including Visual Geometry Group (VGG)-16 (n=37, 24.3%), ResNet-50 (n=33, 21.7%), and DenseNet-121 (n=19, 12.5%). 
Figure 3 provides an overview of the various DL methods identified in the included studies.
Regarding the diagnostic performance of AI methods used in the literature, the most frequently used evaluation metrics encompassed accuracy (n=128, 84.2%), sensitivity (n=112, 73.7%), the AUC (n=87, 57.2%), and specificity (n=77, 50.7%). Accuracy denotes the proportion of correctly predicted cases, encompassing true positives and true negatives among all cases []. Sensitivity measures a model’s ability to accurately identify individuals with a condition [,]. The AUC, derived from the ROC curve, quantitatively assesses a model’s performance []. Specificity gauges a model’s accuracy in identifying individuals without a condition [,]. Table 4 presents the distribution of various performance metrics identified in the included studies. For an in-depth exploration of the performance metrics used across the included studies, refer to the detailed description provided in .
| Metric | Studies, n (%) |
| Accuracy | 128 (84.2) |
| AUROCa | 87 (57.2) |
| Sensitivity/recall/true-positive rate | 112 (73.7) |
| Specificity/true-negative rate | 77 (50.7) |
| Precision/positive predictive value | 56 (36.8) |
| Negative predictive value | 9 (5.9) |
| F1-score | 55 (36.2) |
| Average recall | 1 (0.7) |
| Mean average precision | 2 (1.3) |
| Matthews correlation coefficient | 7 (4.6) |
| True detection rate | 1 (0.7) |
| Area under the alternative free-response ROCb curve | 1 (0.7) |
| Error loss | 1 (0.7) |
| Cohen kappa | 3 (2.0) |
| Root mean square error | 2 (1.3) |
| False positive rate | 4 (2.6) |
| False negative rate | 1 (0.7) |
| Squared error | 1 (0.7) |
| Balanced accuracy | 1 (0.7) |
| Fowlkes-Mallows index | 1 (0.7) |
aAUROC: area under the receiver operating characteristic curve.
bROC: receiver operating characteristic.
AI Diagnostic Performance
Figure 4 provides a comprehensive overview of the overall performance results across all included studies. From the depicted box plot, it is evident that accuracy ranged from 46.48% [] to 100% [], with a mean value of 91.93% (SD 8.10%, 95% CI 90.52%-93.33%) and a median of 93.59% (IQR 88.33%-98.32%). Similarly, the AUC ranged from 45.96% [] to 100% [,,,], with a mean value of 93.48% (SD 7.51%, 95% CI 91.90%-95.06%) and a median of 95.28% (IQR 91%-99%). Sensitivity spanned from 60% [] to 100% [,,,,,,,,,,], with a mean value of 92.77% (SD 7.48%, 95% CI 91.38%-94.15%) and a median of 94.05% (IQR 89%-98.87%). Meanwhile, specificity ranged from 53.2% [] to 100% [,,,], with a mean value of 92.39% (SD 9.4%, 95% CI 90.30%-94.49%) and a median of 95.38% (IQR 89.42%-99.19%). Several outliers were detected for each metric, including 4 for accuracy, at 46.48% [], 63% [], 63.9% [], and 70% []; 2 for the AUC, at 45.96% [] and 69.2% []; 3 for sensitivity, at 60% [], 65.52% [], and 69.07% []; and 5 for specificity, at 53.2% [], 60% [], 60.8% [], 69.44% [], and 73% [].
We conducted further performance comparisons based on several criteria, including whether a study used the TL method and whether it adopted a single (S) or multimodal (M) approach. 
Figure 5 presents split-grouped violin plots illustrating the comparative performance results of the included studies based on TL (yes [Y], no [N]) and the modality used.
We also examined the performance of AI methods used in all the included studies based on the AI techniques applied (ML, DL, or both [ML/DL]), the types of data or biomarkers used (radiographic [R], molecular [M], physiological [P]), and the reference standard used (human [H], bacteriological [B], or both [H/B]). 
Figure 6 showcases the box plots, offering a comparative analysis of the performance results of the included studies based on AI methods, biomarker types, and reference standards.
Specifically, the AI performance across different biomarker types showed mean accuracies of 92.45% (SD 7.83%), 89.03% (SD 8.49%), and 84.21% (SD 0%); mean AUCs of 94.47% (SD 7.32%), 88.45% (SD 8.33%), and 88.61% (SD 5.9%); mean sensitivities of 93.8% (SD 6.27%), 88.41% (SD 10.24%), and 93% (SD 0%); and mean specificities of 94.2% (SD 6.63%), 85.89% (SD 14.66%), and 95% (SD 0%) for radiographic, molecular/biochemical, and physiological biomarkers, respectively. Meanwhile, AI performance across various reference standards show mean accuracies of (SD 7.3%), 93.16% (SD 6.44%), 88.98% (SD 9.77%); mean AUCs of 90.95% (SD 7.58%), 94.89% (SD 5.18%), 92.61% (SD 6.01%); mean sensitivities of 91.76% (SD 7.02%), 93.73% (SD 6.67%), 91.34% (SD 7.71%); and mean specificities of 86.56% (SD 12.8%), 93.69% (SD 8.45%), 92.7% (SD 6.54%) for bacteriological, human reader, and combined bacteriological and human reader reference standards, respectively.
Risk-of-Bias Assessment Result
Figure 7 presents the QUADAS-2 outcomes across all included studies. Among the assessed studies, 18 (11.8%) [,-,,,,,,,,,,,,,,] were flagged for a high risk of bias, with 1 (0.7%) study [] marked as having an unclear risk concerning patient selection, largely attributable to incomplete or absent information in the data selection process. Moreover, 4 (2.6%) studies [,,,] were identified as having a high risk regarding the index test, attributed to the absence of crucial details on model architecture and parameters. Additionally, 3 (2%) studies [,,] were classified as having a high risk and 8 (5.3%) studies [,,,,,,,] as having an unclear risk concerning reference standards, due to incomplete or missing information regarding the reference standards used. Concerning flow and timing, 2 (1.3%) studies [,] were deemed to have a high risk of bias, while 8 (5.3%) studies [,,,,,,,] were assigned an unclear risk, primarily due to unclear or absent information regarding the time interval and intervention provided in those studies.
The majority of the included studies exhibited a low level of concern about applicability. Only 3 (2%) studies [,,] were identified with a high risk and 1 (0.7%) study [] with an unclear risk in terms of patient selection. Additionally, 3 (2%) studies [,,] were flagged with a high risk concerning the index test, while 1 (0.7%) study [] showed a high risk, and 8 (5.3%) studies [,,,,,,,] had an unclear risk concerning reference standards. For comprehensive QUADAS-2 results, please refer to .
Discussion
Principal Findings
This review study examined the diagnostic performance of various AI-based algorithms, including both ML and DL, for TB detection across different techniques, data modalities, and reference standards used in the included studies. Approximately 84.9% (n=129) of the included studies used radiographic data types, particularly CXRs. Notably, the SZ and MC datasets emerged as the most frequently used, with 51 and 46 studies, respectively, making use of them. However, it is noteworthy that the availability of data in SZ and MC, totaling 800 images, is relatively limited, posing challenges, especially for studies using DL methods where larger datasets are often required for optimal performance. Consequently, numerous techniques have been introduced to augment data quantity [,], including conventional methods, as well as DL-based augmentation techniques.
Few recent studies have used other data types, such as exhaled breath particles [,] and cough sounds [,,]. Although not recommended as a standard tool for TB detection, results from these noninvasive approaches could be a supplement to the standard TB detection tool using CXRs. Hence, further studies that analyze CXRs together with other data types are encouraged.
Concerning modality, the majority of the included studies adopted a single modality (n=141, 92.8%) rather than using a multimodal approach (n=11, 7.2%). Studies using a single modality tended to exhibit higher median values and narrower distribution ranges compared to those using a multimodal approach (Figure 5, right). This suggests that the single-modality approach may outperform the multimodal approach. However, more outliers were observed in all performance metrics for the single modality compared to the multimodal approach. Additionally, these outcomes were not obtained under identical scenario settings. When applied under similar scenario settings, the multimodal approach has demonstrated the potential to enhance performance results compared to the single modality [,]. This is supported by findings from several included studies that have directly compared single-modality and multimodal approaches [,,].
DL emerged as a cornerstone in the majority of the included studies (n=122, 80.3%), with CNNs taking the forefront. Among these, architectures such as VGG-16 [], ResNet-50 [,], and DenseNet-121 [] have gained notable traction. More recent DL architectures, including EfficientNets [], Vision Transformer (ViT) [], and ConvNeXts [], have gained increasing attention. These emerging models, with more efficient architectures that require fewer computational resources, show potential for application in medical image processing, particularly in TB detection tasks.
Some of the included studies explored various explainable AI (XAI) methods to bolster the interpretability and trustworthiness of their AI models’ outcomes. Originating from the need to strike a balance between interpretability and accuracy in ML and DL models [], XAI has garnered considerable attention. Among the XAI techniques used in the included studies were gradient-weighted class activation mapping (Grad-CAM), local interpretable model-agnostic explanation (LIME), and Shapley additive explanations (SHAP).
TL is a powerful technique that capitalizes on knowledge acquired from a source dataset and adapts it to a target domain, relaxing the assumption that training data must be independent and identically distributed with test data []. TL proves especially beneficial in scenarios with a limited data volume. Indeed, TL has been extensively used in the studies encompassed in this systematic review (n=89, 58.6%), particularly those using DL methods. Notably, as illustrated in Figure 5 (left), investigations integrating TL tend to exhibit a higher median accuracy compared to counterparts. Additionally, they showcase a narrower distribution range, suggesting that TL-based approaches consistently deliver superior and more consistent results.
The majority of the included studies (n=143, 94.1%) focused primarily on the development of AI models for TB detection. Only 8 studies [,,,,,,,] progressed beyond model development to build prototypes or systems aimed at supporting real-world applications. Among these, 3 studies implemented their proposed models in real-world settings: eRx, a mobile health system for TB diagnosis deployed in Lima, Peru []; the multistream integration–pulmonary tuberculosis diagnosis model (MSI-PTDM), multistream integration for a TB diagnosis model implemented in China []; and AIChest4All, an automated CXR-screening system used in Thailand []. We excluded studies evaluating commercial TB detection tools, such as Lunit INSIGHT, CAD4TB, and qXR, as they typically do not provide detailed discussions of the underlying AI models. For further insights into commercial TB detection tools, interested readers are referred to the narrative review by Singh et al [].
In terms of overall performance metrics, the AI methods displayed remarkable achievements. Further analysis based on the AI methods used in the included studies revealed interesting insights. As illustrated in Figure 6 (left), DL exhibits higher accuracy with less variation compared to ML and ML/DL. However, regarding the AUC, sensitivity, and specificity, studies integrating ML/DL tend to yield superior results, followed by DL and then ML. This suggests that the fusion of ML and DL methods may enhance model performance, albeit few studies have explored this avenue.
We also examined the performance results concerning the types of data used. Radiographic data-type studies showcased superior median scores across all performance metrics (Figure 6, middle). Notably, they demonstrated a higher level of consistency, evidenced by their shorter IQRs in all performance metrics. Regarding the reference standards used in the included studies, studies that used the human reader exceled compared to those that used bacteriological and both reference standards, as indicated by the higher median and mean scores on all performance metrics in Figure 6 (right). Despite accuracy, the IQRs were also shorter for the AUC, sensitivity, and specificity. This finding might be rooted in the large number of included studies that used DL methods on radiographic data types with human reader reference standards.
Finally, it is worth noting that the majority of the included studies overlooked the evaluation of proposed solutions regarding the domain-shift problem. Remarkably, only 1 study [] addressed domain-shift analysis in the context of TB detection, as highlighted by Hansun et al []. Although our review confirmed the high performance of AI-based methods across various data types in TB detection tasks, these evaluations were predominantly conducted on in-domain datasets. In other words, the datasets used for validation and testing shared the same distribution and characteristics as the training data. However, many AI methods, particularly DL, struggle when applied to real-world settings []. This discrepancy often arises due to differences in data distribution between the training data and real-world datasets. As such, it is imperative to broaden our focus to include not only in-domain evaluations but also domain-shift analysis in future research endeavors.
Limitations and Strengths
This systematic review has several limitations that warrant acknowledgment. First, our search was limited to the literature published in English from 3 primary academic databases: Scopus, PubMed, and ACM Digital Library. Although these sources are comprehensive, it is possible that relevant studies might exist in other databases, potentially leading to a partial representation of the literature.
Second, we did not conduct further investigations into aspects with high or unclear risks from the bias assessment. Given the prevalence of radiographic datasets across most included studies, we assumed uniformity across these aspects. However, this assumption may overlook potential variations that could impact the overall assessment.
Lastly, in this review study, we opted not to perform a meta-analysis. Although a meta-analysis could provide valuable insights, the diverse nature of the included studies made it challenging to ensure meaningful comparability across studies, potentially affecting the validity and reliability of the synthesized findings. It is also important to note that the variations in data distribution and experimental settings across studies may lead to different results. Although the descriptive statistics reported in this study have been grouped based on several approaches, they should be seen as general performance results of AI for TB detection.
Despite the aforementioned limitations, our review study is among the few to systematically review the evidence available in the literature regarding the efficacy of AI-based methods for TB detection. Although most review studies focus only on a specific biomarker type for TB detection, we performed a more comprehensive review of AI-based methods across diverse biomarker types. We further assessed the AI methods’ performance based on several dominant approaches used in the included studies, including TL, multimodalities, reference standards, and ML/DL fusion methods for TB detection, which have never been explored before.
Key Takeaways
AI-based approaches, particularly DL, have been extensively used for TB detection with high-accuracy results. A single-modality approach with chest radiographs has been dominantly used. Further studies that analyze chest radiographs together with other data types (multimodality) are encouraged. AI models trained and tested on radiographic data tend to achieve higher performance compared to other data types. Emerging DL models, such as EfficientNets, ViT, and ConvNeXts, show the potential to enhance TB detection results. TL has shown a great advantage in handling a limited data volume and consistently delivering superior results. Most of the included studies evaluated their proposed solutions using in-domain datasets. Future research endeavors should prioritize conducting domain-shift analyses to better simulate real-world scenarios in TB detection.
Conclusion
This systematic review underscores the considerable promise of AI-based approaches in TB detection. Among the array of AI methods, DL emerges as the predominant choice. This preference for DL is attributed to its consistently robust performance, likely bolstered by the prevalence of studies using radiographic data. A notable observation across many studies is the use of relatively small datasets. Despite achieving commendable results, the potential of DL models could be further enhanced with larger and more diverse datasets.
Finally, although AI models demonstrate impressive performance on in-domain datasets, there is a notable gap in evaluating their robustness to domain shifts. Future research endeavors should prioritize conducting domain-shift analyses to better simulate real-world scenarios in TB detection. This approach would provide invaluable insights into the generalizability and applicability of AI-based methods beyond controlled settings.
Acknowledgments
We used the generative artificial intelligence (AI) tool Chat Generative Pretrained Transformer (ChatGPT) by OpenAI during the editing process of the initial draft, which was subsequently reviewed and revised by the review authors. The original ChatGPT transcripts are available in .
Data Availability
All data generated or analyzed during this study are included in the published paper and its Supplementary and Figure/Appendix Files.
Authors' Contributions
SH handled conceptualization, methodology, software, validation, formal analysis, investigation, resources, data curation, writing—original draft, visualization, and project administration; AA and GBM, conceptualization, methodology, validation, formal analysis, investigation, writing—review and editing, and supervision; IB and AW, investigation, resources, data curation, and writing—review and editing; HAR, software, formal analysis, resources, and writing—review and editing; GF, methodology, formal analysis, investigation, and writing—review and editing; and STL and BGC, methodology, investigation, writing—review and editing, and supervision. All authors have read and approved the final manuscript.
Conflicts of Interest
None declared.
Multimedia Appendix 1
PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) checklist.
PDF File (Adobe PDF File), 612 KBMultimedia Appendix 3
Full characteristics of the included studies.
XLSX File (Microsoft Excel File), 113 KBMultimedia Appendix 5
QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies version 2) summary.
XLSX File (Microsoft Excel File), 68 KBMultimedia Appendix 6
ChatGPT (Chat Generative Pretrained Transformer) transcript.
PDF File (Adobe PDF File), 148 KBReferences
- Global tuberculosis report 2023. World Health Organization. 2023. URL: https://www.who.int/teams/global-tuberculosis-programme/tb-reports/global-tuberculosis-report-2023 [accessed 2025-02-20]
- Lung T, Marks GB, Nhung NV, Anh NT, Hoa NLP, Anh LTN, et al. Household contact investigation for the detection of tuberculosis in Vietnam: economic evaluation of a cluster-randomised trial. Lancet Global Health. Mar 2019;7(3):e376-e384. [CrossRef]
- Dande P, Samant P. Acquaintance to artificial neural networks and use of artificial intelligence as a diagnostic tool for tuberculosis: a review. Tuberculosis (Edinb). Jan 2018;108:1-9. [CrossRef] [Medline]
- Janiesch C, Zschech P, Heinrich K. Machine learning and deep learning. Electron Markets. Apr 08, 2021;31(3):685-695. [CrossRef]
- Woschank M, Rauch E, Zsifkovits H. A review of further directions for artificial intelligence, machine learning, and deep learning in smart logistics. Sustainability. May 06, 2020;12(9):3760. [CrossRef]
- Helm JM, Swiergosz AM, Haeberle HS, Karnuta JM, Schaffer JL, Krebs VE, et al. Machine learning and artificial intelligence: definitions, applications, and future directions. Curr Rev Musculoskelet Med. Feb 25, 2020;13(1):69-76. [FREE Full text] [CrossRef] [Medline]
- World Health Organization. WHO Operational Handbook on Tuberculosis. Module 2: Screening - Systematic Screening for Tuberculosis Disease. Geneva. World Health Organization; 2021.
- David P, Onno J, Keshavjee S, Ahmad Khan F. Conditions required for the artificial-intelligence-based computer-aided detection of tuberculosis to attain its global health potential. Lancet Digital Health. Oct 2022;4(10):e702-e704. [CrossRef]
- Jimmy, Cenggoro T, Pardamean B. Systematic literature review: an intelligent pulmonary TB detection from chest X-rays. 2021. Presented at: 2021 1st International Conference on Computer Science and Artificial Intelligence (ICCSAI); October 28, 2021:136-141; Jakarta, Indonesia. [CrossRef]
- Sharma R, Gupta P, Kaur H. A deep learning approach for tuberculosis diagnosis from chest X-rays: a survey. 2021. Presented at: International Conference on Emerging Technologies: AI, IoT, and CPS for Science & Technology Applications; September 6-7, 2021; Chandigarh, India.
- Zeyu D, Yaakob R, Azman A. A review of deep learning-based detection methods for tuberculosis. 2022. Presented at: 2022 IEEE International Conference on Computing (ICOCO); November 14-16, 2022:68-73; Kota Kinabalu, Malaysia. [CrossRef]
- Oloko-Oba M, Viriri S. A systematic review of deep learning techniques for tuberculosis detection from chest radiograph. Front Med (Lausanne). 2022;9:830515. [FREE Full text] [CrossRef] [Medline]
- Santosh K, Allu S, Rajaraman S, Antani S. Advances in deep learning for tuberculosis screening using chest X-rays: the last 5 years review. J Med Syst. Oct 15, 2022;46(11):82. [FREE Full text] [CrossRef] [Medline]
- Mota Carvalho TF, Santos VLA, Silva JCF, Figueredo LJDA, de Miranda SS, Duarte RDO, et al. A systematic review and repeatability study on the use of deep learning for classifying and detecting tuberculosis bacilli in microscopic images. Prog Biophys Mol Biol. Jul 2023;180-181:1-18. [CrossRef] [Medline]
- da Silva Barros MHLF, da Silva Neto SR, de Almeida Rodrigues MG, de Souza Sampaio V, Endo P. How machine learning can help the classification of treatment outcomes of tuberculosis: a systematic review. 2023. Presented at: 56th Hawaii International Conference on System Sciences; January 5-6, 2023:1386-1395; Maui, Hawaii. [CrossRef]
- Siddiqui A, Garg V. Diagnosis of pulmonary tuberculosis through intelligent techniques: a review. 2021. Presented at: 2021 International Conference on Computing Sciences (ICCS); December 4-5, 2021:189-193; Phagwara, India. [CrossRef]
- Singh M, Pujar GV, Kumar SA, Bhagyalalitha M, Akshatha HS, Abuhaija B, et al. Evolution of machine learning in tuberculosis diagnosis: a review of deep learning-based medical applications. Electronics. Aug 23, 2022;11(17):2634. [CrossRef]
- Harris M, Qi A, Jeagal L, Torabi N, Menzies D, Korobitsyn A, et al. A systematic review of the diagnostic accuracy of artificial intelligence-based computer programs to analyze chest X-rays for pulmonary tuberculosis. PLoS One. Sep 3, 2019;14(9):e0221339. [FREE Full text] [CrossRef] [Medline]
- Hansun S, Argha A, Liaw S, Celler BG, Marks GB. Machine and deep learning for tuberculosis detection on chest X-rays: systematic literature review. J Med Internet Res. Jul 03, 2023;25:e43154. [FREE Full text] [CrossRef] [Medline]
- Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Syst Rev. Mar 29, 2021;10(1):89. [FREE Full text] [CrossRef] [Medline]
- Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. PRISMA-P Group. Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015 statement. Syst Rev. Jan 01, 2015;4(1):1. [FREE Full text] [CrossRef] [Medline]
- Shamseer L, Moher D, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. PRISMA-P Group. Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015: elaboration and explanation. BMJ. Jan 02, 2015;350:g7647. [FREE Full text] [CrossRef] [Medline]
- Page MJ, Shamseer L, Tricco AC. Registration of systematic reviews in PROSPERO: 30,000 records and counting. Syst Rev. Feb 20, 2018;7(1):32. [FREE Full text] [CrossRef] [Medline]
- Hansun S, Argha A, Bakhshayeshi I, Wicaksana A, Liaw S, Celler B, et al. Artificial intelligence-based tuberculosis detection: a systematic review. PROSPERO. 2023. URL: https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42023453611 [accessed 2024-01-25]
- Gusenbauer M, Haddaway NR. Which academic search systems are suitable for systematic reviews or meta-analyses? Evaluating retrieval qualities of Google Scholar, PubMed, and 26 other resources. Res Synth Methods. Mar 28, 2020;11(2):181-217. [FREE Full text] [CrossRef] [Medline]
- Whiting PF, Rutjes AWS, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. QUADAS-2 Group. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. Oct 18, 2011;155(8):529-536. [FREE Full text] [CrossRef] [Medline]
- Hansun S, Argha A, Alinejad-Rokny H, Liaw S, Celler B, Marks G. Revisiting transfer learning method for tuberculosis diagnosis. 2023. Presented at: 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC); July 24-27, 2023:1-4; Sydney, Australia. [CrossRef]
- Jaeger S, Candemir S, Antani S, Wáng Y-XJ, Lu P, Thoma G. Two public chest X-ray datasets for computer-aided screening of pulmonary diseases. Quant Imaging Med Surg. Dec 2014;4(6):475-477. [FREE Full text] [CrossRef] [Medline]
- Rahman T, Khandakar A, Kadir MA, Islam KR, Islam KF, Mazhar R, et al. Reliable tuberculosis detection using chest X-ray with deep learning, segmentation and visualization. IEEE Access. 2020;8:191586-191601. [CrossRef]
- Liu Y, Wu Y, Ban Y, Wang H, Cheng M-M. Rethinking computer-aided tuberculosis diagnosis. 2020. Presented at: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); June 13-19, 2020:2643-2652; Seattle, WA. [CrossRef]
- Melendez J, van Ginneken B, Maduskar P, Philipsen RHHM, Reither K, Breuninger M, et al. A novel multiple-instance learning-based approach to computer-aided detection of tuberculosis on chest X-rays. IEEE Trans. Med. Imaging. Jan 2015;34(1):179-192. [CrossRef]
- Osman M, Mohd NM, Mashor M, Jaafar H. Compact single hidden layer feedforward network for Mycobacterium tuberculosis detection. 2011. Presented at: 2011 IEEE International Conference on Control System, Computing and Engineering; November 25-27, 2011:432-436; Penang, Malaysia. [CrossRef]
- Osman M, Mashor M, Jaafar H. Online sequential extreme learning machine for classification of Mycobacterium tuberculosis in Ziehl-Neelsen stained tissue. 2012. Presented at: 2012 International Conference on Biomedical Engineering (ICoBE); February 27-28, 2012:139-143; Penang, Malaysia. [CrossRef]
- Lakhani P, Sundaram B. Deep learning at chest radiography: automated classification of pulmonary tuberculosis by using convolutional neural networks. Radiology. Aug 2017;284(2):574-582. [CrossRef] [Medline]
- Pasa F, Golkov V, Pfeiffer F, Cremers D, Pfeiffer D. Efficient deep network architectures for fast chest X-ray tuberculosis screening and visualization. Sci Rep. Apr 18, 2019;9(1):6268. [FREE Full text] [CrossRef] [Medline]
- Lopes U, Valiati J. Pre-trained convolutional neural networks as feature extractors for tuberculosis detection. Comput Biol Med. Oct 01, 2017;89:135-143. [FREE Full text] [CrossRef] [Medline]
- Tasci E, Uluturk C, Ugur A. A voting-based ensemble deep learning method focusing on image augmentation and preprocessing variations for tuberculosis detection. Neural Comput Appl. Jun 07, 2021;33(22):15541-15555. [FREE Full text] [CrossRef] [Medline]
- Ul Abideen Z, Ghafoor M, Munir K, Saqib M, Ullah A, Zia T, et al. Uncertainty assisted robust tuberculosis identification with Bayesian convolutional neural networks. IEEE Access. 2020;8:22812-22825. [CrossRef]
- Morís DI, de Moura J, Novo J, Ortega M. Unsupervised contrastive unpaired image generation approach for improving tuberculosis screening using chest X-ray images. Pattern Recogn Lett. Dec 2022;164:60-66. [CrossRef]
- Mizan M, Hasan M, Hassan S. A comparative study of tuberculosis detection using deep convolutional neural network. 2020. Presented at: 2020 2nd International Conference on Advanced Information and Communication Technology (ICAICT); November 28-29, 2020:157-161; Dhaka, Bangladesh. [CrossRef]
- Elashmawy A, Elamvazuthi I, Ali S, Natarajan E, Paramasivam S. A hybridized pre-processing method for detecting tuberculosis using deep learning. 2020. Presented at: 2020 8th International Conference on Intelligent and Advanced Systems (ICIAS); July 13-15, 2021:1-5; Kuching, Malaysia. [CrossRef]
- Malik H, Anees T, Chaudhry MU, Gono R, Jasiński M, Leonowicz Z, et al. A novel fusion model of hand-crafted features with deep convolutional neural networks for classification of several chest diseases using X-ray images. IEEE Access. 2023;11:39243-39268. [CrossRef]
- Prabakaran J, Selvaraj P. Advance IoT intelligent healthcare system for lung disease classification using ensemble techniques. Comput Syst Sci Eng. 2023;46(2):2141-2157. [CrossRef]
- Thammarach P, Khaengthanyakan S, Vongsurakrai S, Phienphanich P, Pooprasert P, Yaemsuk A, et al. AI chest 4 all. 2020. Presented at: 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC); July 20-24, 2020:1229-1233; Montreal, QC. [CrossRef]
- Kim K, Lee JH, Je Oh S, Chung MJ. AI-based computer-aided diagnostic system of chest digital tomography synthesis: demonstrating comparative advantage with X-ray-based AI systems. Comput Methods Programs Biomed. Oct 2023;240:107643. [FREE Full text] [CrossRef] [Medline]
- Abdollahi J, Mahmoudi L. An artificial intelligence system for detecting the types of the epidemic from X-rays. 2022. Presented at: 2022 27th International Computer Conference, Computer Society of Iran (CSICC); February 23-24, 2022:1-6; Tehran, Iran. [CrossRef]
- Nguyen NH, Hoang V, Bui T, Truong S, Minh T, Nguyen VD, et al. An efficient approach for tuberculosis diagnosis on chest X-ray. 2022. Presented at: 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI); March 28-31, 2022; Kolkata, India. [CrossRef]
- Saha P. An ensemble CNN-Dempster Shafer based tuberculosis detection from chest X-ray images. 2022. Presented at: 2022 IEEE Calcutta Conference (CALCON); December 10-11, 2022:228-232; Kolkata, India. [CrossRef]
- Bouali Y, Boucetta I, Bekkouch I, Bouache M, Mazouzi S. An image dataset for lung disease detection and classification. 2021. Presented at: 2021 International Conference on Theoretical and Applicative Aspects of Computer Science (ICTAACS); December 15-16, 2021:1-6; Skikda, Algeriaz. [CrossRef]
- Huy VTQ, Lin C. An improved DenseNet deep neural network model for tuberculosis detection using chest X-ray images. IEEE Access. 2023;11:42839-42849. [CrossRef]
- Roopa NK, Mamatha GS. An add-on CNN based model for the detection of tuberculosis using chest X-ray images. IJACSA. 2023;14(3):113-123. [CrossRef]
- Hooda R, Mittal A, Sofat S. A novel ensemble method for PTB classification in CXRs. Wireless Pers Commun. Jan 16, 2020;112(2):809-826. [CrossRef]
- Ahmad I, Shin S. A perceptual encryption-based image communication system for deep learning-based tuberculosis diagnosis using healthcare cloud services. Electronics. Aug 11, 2022;11(16):2514. [CrossRef]
- Nijiati M, Ma J, Hu C, Tuersun A, Abulizi A, Kelimu A, et al. Artificial intelligence assisting the early detection of active pulmonary tuberculosis from chest X-rays: a population-based study. Front Mol Biosci. Apr 8, 2022;9:874475. [FREE Full text] [CrossRef] [Medline]
- Sharma A, Rani S, Gupta D. Artificial intelligence-based classification of chest X-ray images into COVID-19 and other infectious diseases. Int J Biomed Imaging. Oct 6, 2020;2020:8889023-8889010. [FREE Full text] [CrossRef] [Medline]
- Oltu B, Guney S, Dengiz B, Agildere M. Automated tuberculosis detection using pre-trained CNN and SVM. 2021. Presented at: 2021 44th International Conference on Telecommunications and Signal Processing (TSP); July 26-28, 2021:92-95; Brno, Czech Republic. [CrossRef]
- Mithra KS, Sam Emmanuel WR. Automated identification of Mycobacterium bacillus from sputum images for tuberculosis diagnosis. SIViP. Jun 5, 2019;13(8):1585-1592. [CrossRef]
- Hooda R, Mittal A, Sofat S. Automated TB classification using ensemble of deep architectures. Multimed Tools Appl. Jul 22, 2019;78(22):31515-31532. [CrossRef]
- Silva JF, Silva JM, Pinho E, Costa C. 3D-CNN in drug resistance detection and tuberculosis classification. 2017. Presented at: CLEF 2017: Conference and Labs of the Evaluation Forum; September 11-14, 2017:1-10; Dublin, Ireland.
- Xiong Y, Ba X, Hou A, Zhang K, Chen L, Li T. Automatic detection of Mycobacterium tuberculosis using artificial intelligence. J Thorac Dis. Mar 2018;10(3):1936-1940. [FREE Full text] [CrossRef] [Medline]
- Panicker RO, Kalmady KS, Rajan J, Sabu M. Automatic detection of tuberculosis bacilli from microscopic sputum smear images using deep learning methods. Biocybernet Biomed Eng. 2018;38(3):691-699. [CrossRef]
- Hu M, Liu Y, Zhang Y, Guan T, He Y. Automatic detection of tuberculosis bacilli in sputum smear scans based on subgraph classification. 2019. Presented at: 2019 International Conference on Medical Imaging Physics and Engineering (ICMIPE); November 22-24, 2019:1-7; Shenzhen, China. [CrossRef]
- Pahar M, Klopper M, Reeve B, Warren R, Theron G, Diacon A, et al. Automatic tuberculosis and COVID-19 cough classification using deep learning. 2022. Presented at: 2022 International Conference on Electrical, Computer and Energy Technologies (ICECET); July 20-22, 2022:1-9; Prague, Czech Republic. [CrossRef]
- Pahar M, Theron G, Niesler T. Automatic tuberculosis detection in cough patterns using NLP-style cough embeddings. 2022. Presented at: 2022 International Conference on Engineering and Emerging Technologies (ICEET); October 27-28, 2022:1-6; Kuala Lumpur, Malaysia. [CrossRef]
- Raju M, Aswath A, Kadam A, Pagidimarri V. Automatic detection of tuberculosis using deep learning methods. In: Laha A, editor. Advances in Analytics and Applications. Springer Proceedings in Business and Economics. Singapore. Springer; 2019:119-129.
- Manivannan K, Sathiamoorthy S. Bird mating optimizer with deep learning-based tuberculosis detection using chest radiographs. IJETT. Feb 28, 2023;71(2):341-348. [CrossRef]
- Karnkawinpong T, Limpiyakorn Y. Chest X-ray analysis of tuberculosis by convolutional neural networks with affine transforms. 2018. Presented at: CSAI '18: Proceedings of the 2018 2nd International Conference on Computer Science and Artificial Intelligence; December 8-10, 2018:90-93; Shenzhen, China. [CrossRef]
- Rajaraman S, Zamzmi G, Folio L, Alderson P, Antani S. Chest X-ray bone suppression for improving classification of tuberculosis-consistent findings. Diagnostics (Basel). May 07, 2021;11(5):840. [FREE Full text] [CrossRef] [Medline]
- Shelke A, Inamdar M, Shah V, Tiwari A, Hussain A, Chafekar T, et al. Chest X-ray classification using deep learning for automated COVID-19 screening. SN Comput Sci. May 26, 2021;2(4):300. [FREE Full text] [CrossRef] [Medline]
- Thakur K, Kaur M, Kumar Y. A comprehensive analysis of deep learning-based approaches for prediction and prognosis of infectious diseases. Arch Comput Methods Eng. Jun 08, 2023;30(7):1-21. [FREE Full text] [CrossRef] [Medline]
- Karnkawinpong T, Limpiyakorn Y. Classification of pulmonary tuberculosis lesion with convolutional neural networks. 2018. Presented at: 2018 11th International Conference on Computer and Electrical Engineering; October 12–14, 2018; Tokyo, Japan. [CrossRef]
- Karaca B, Guney S, Dengiz B, Agildere M. Comparative study for tuberculosis detection by using deep learning. 2021. Presented at: 2021 44th International Conference on Telecommunications and Signal Processing (TSP); July 26-28, 2021:88-91; Brno, Czech Republic. [CrossRef]
- Owais M, Arsalan M, Mahmood T, Kim YH, Park KR. Comprehensive computer-aided decision support framework to diagnose tuberculosis from chest X-ray images: data mining study. JMIR Med Inform. Dec 07, 2020;8(12):e21790. [FREE Full text] [CrossRef] [Medline]
- Jose E, Soto-Diaz R, Madera N, Soto C, Burgos-Florez F, Rodriguez AF, et al. Computer-aided diagnosis for tuberculosis classification with water strider optimization algorithm. Comput Syst Sci Eng. 2023;46(2):1337-1353. [CrossRef]
- Visuña L, Yang D, Garcia-Blas J, Carretero J. Computer-aided diagnostic for classifying chest X-ray images using deep ensemble learning. BMC Med Imaging. Oct 15, 2022;22(1):178. [FREE Full text] [CrossRef] [Medline]
- Xie Y, Wu Z, Han X, Wang H, Wu Y, Cui L, et al. Computer-aided system for the detection of multicategory pulmonary tuberculosis in radiographs. J Healthc Eng. Aug 24, 2020;2020:9205082-9205012. [FREE Full text] [CrossRef] [Medline]
- Xu T, Yuan Z. Convolution neural network with coordinate attention for the automatic detection of pulmonary tuberculosis images on chest X-rays. IEEE Access. 2022;10:86710-86717. [CrossRef]
- Mahbub MK, Biswas M, Gaur L, Alenezi F, Santosh K. Deep features to detect pulmonary abnormalities in chest X-rays due to infectious diseaseX: COVID-19, pneumonia, and tuberculosis. Inf Sci (N Y). May 2022;592:389-401. [FREE Full text] [CrossRef] [Medline]
- Heo S, Kim Y, Yun S, Lim S, Kim J, Nam C, et al. Deep learning algorithms with demographic information help to detect tuberculosis in chest radiographs in annual workers’ health examination data. Int J Environ Res Public Health. Jan 16, 2019;16(2):250. [FREE Full text] [CrossRef] [Medline]
- Nijiati M, Zhou R, Damaola M, Hu C, Li L, Qian B, et al. Deep learning based CT images automatic analysis model for active/non-active pulmonary tuberculosis differential diagnosis. Front Mol Biosci. Dec 5, 2022;9:1086047. [FREE Full text] [CrossRef] [Medline]
- Liu C, Tsai CC, Kuo L, Kuo P, Lee M, Wang J, et al. A deep learning model using chest X-ray for identifying TB and NTM-LD patients: a cross-sectional study. Insights Imaging. Apr 15, 2023;14(1):67. [FREE Full text] [CrossRef] [Medline]
- Kazemzadeh S, Yu J, Jamshy S, Pilgrim R, Nabulsi Z, Chen C, et al. Deep learning detection of active pulmonary tuberculosis at chest radiography matched the clinical performance of radiologists. Radiology. Jan 2023;306(1):124-137. [CrossRef] [Medline]
- Showkatian E, Salehi M, Ghaffari H, Reiazi R, Sadighi N. Deep learning-based automatic detection of tuberculosis disease in chest X-ray images. Pol J Radiol. Feb 28, 2022;87:e118-e124. [CrossRef] [Medline]
- Zhou W, Cheng G, Zhang Z, Zhu L, Jaeger S, Lure FYM, et al. Deep learning-based pulmonary tuberculosis automated detection on chest radiography: large-scale independent testing. Quant Imaging Med Surg. Apr 2022;12(4):2344-2355. [FREE Full text] [CrossRef] [Medline]
- Rahman M, Cao Y, Sun X, Li B, Hao Y. Deep pre-trained networks as a feature extractor with XGBoost to detect tuberculosis from chest X-ray. Comput Electr Eng. Jul 2021;93:107252. [CrossRef]
- Gozes O, Greenspan H. Deep feature learning from a hospital-scale chest X-ray dataset with application to TB detection on a small-scale dataset. 2019. Presented at: 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC); July 23-27, 2019:4076-4079; Berlin, Germany. [CrossRef]
- Nguyen Q, Nguyen B, Dao S, Unnikrishnan B, Dhingra R, Ravichandran S, et al. Deep learning models for tuberculosis detection from chest X-ray images. 2019. Presented at: 2019 26th International Conference on Telecommunications (ICT); April 8-10, 2019:381-385; Hanoi, Vietnam. [CrossRef]
- Mohammad M, Reddy M. Deep transfer learning approach for classification of chest infections in radiographic X-ray images. 2022. Presented at: 2022 Second International Conference on Computer Science, Engineering and Applications (ICCSEA); September 8, 2022:1-6; Gunupur, India. [CrossRef]
- Hooda R, Sofat S, Kaur S, Mittal A, Meriaudeau F. Deep learning: a potential method for tuberculosis detection using chest radiography. 2017. Presented at: 2017 IEEE International Conference on Signal and Image Processing Applications (ICSIPA); September 12-14, 2017:497-502; Kuching, Malaysia. [CrossRef]
- Ullah N, Marzougui M, Ahmad I, Chelloug SA. DeepLungNet: an effective DL-based approach for lung disease classification using CRIs. Electronics. Apr 14, 2023;12(8):1860. [CrossRef]
- Uçar M. Deep neural network model with Bayesian optimization for tuberculosis detection from X-ray images. Multimed Tools Appl. Apr 05, 2023;82(24):36951-36972. [CrossRef]
- Yan C, Wang L, Lin J, Xu J, Zhang T, Qi J, et al. A fully automatic artificial intelligence-based CT image analysis system for accurate detection, diagnosis, and quantitative severity evaluation of pulmonary tuberculosis. Eur Radiol. Apr 29, 2022;32(4):2188-2199. [FREE Full text] [CrossRef] [Medline]
- Singh D, Kumar V, Kaur M. Densely connected convolutional networks-based COVID-19 screening model. Appl Intell (Dordr). Feb 07, 2021;51(5):3044-3051. [FREE Full text] [CrossRef] [Medline]
- Manh TH, Thanh HL, Duc VN, Nang TD. Detecting tuberculosis from Vietnamese X-ray imaging using transfer learning approach. Comput Mater Contin. 2023;74(3):5001-5016. [CrossRef]
- Lay J, Cenggoro Binus University TW, Pardamean B, Gozali J, Tanumihardja D. Detection of pulmonary tuberculosis from chest X-ray images using multimodal ensemble method. Commun Math Biol Neurosci. 2022;126:1. [CrossRef]
- Duong LT, Le NH, Tran TB, Ngo VM, Nguyen PT. Detection of tuberculosis from chest X-ray images: boosting the performance with vision transformer and transfer learning. Expert Syst Appl. Dec 2021;184:115519. [CrossRef]
- Setiawan A, Rusydi M. Detection of Mycobacterium tuberculosis using residual neural network. 2022. Presented at: 2022 International Seminar on Intelligent Technology and Its Applications (ISITIA); July 20-21, 2022:7-11; Surabaya, Indonesia. [CrossRef]
- Tiwari M, Patankar M, Chaurasia V, Shandilya M, Kumar A, Potnis A. Detection of tuberculosis bacilli using deep learning. 2023. Presented at: 2023 1st International Conference on Innovations in High Speed Communication and Signal Processing (IHCSP); March 4-5, 2023:492-496; Bhopal, India. [CrossRef]
- Meraj SS, Yaakob R, Azman A, Rum SNM, Nazri ASA, Zakaria NF. Detection of pulmonary tuberculosis manifestation in chest X-rays using different convolutional neural network (CNN) models. IJEAT. Oct 30, 2019;9(1):2270-2275. [CrossRef]
- Han D, Chen Y, Li X, Li W, Zhang X, He T, et al. Development and validation of a 3D-convolutional neural network model based on chest CT for differentiating active pulmonary tuberculosis from community-acquired pneumonia. Radiol Med. Jan 27, 2023;128(1):68-80. [FREE Full text] [CrossRef] [Medline]
- Hwang EJ, Park S, Jin K, Kim JI, Choi SY, Lee JH, et al. Development and validation of a deep learning–based automatic detection algorithm for active pulmonary tuberculosis on chest radiographs. Clin Infect Dis. Aug 16, 2019;69(5):739-747. [FREE Full text] [CrossRef] [Medline]
- Nguyen H, Tran N, Le B. Diagnosing tuberculosis using graph neural network. 2022. Presented at: 2022 14th International Conference on Knowledge and Systems Engineering (KSE); October 19-21, 2022:1-6; Nha Trang, Vietnam. [CrossRef]
- Rasheed J, Alsubai S. A hybrid deep fused learning approach to segregate infectious diseases. Comput Mater Contin. 2023;74(2):4239-4259. [CrossRef]
- Han D, He T, Yu Y, Guo Y, Chen Y, Duan H, et al. Diagnosis of active pulmonary tuberculosis and community acquired pneumonia using convolution neural network based on transfer learning. Acad Radiol. Oct 2022;29(10):1486-1492. [CrossRef] [Medline]
- Park M, Lee Y, Kim S, Kim Y, Kim SY, Kim Y, et al. Distinguishing nontuberculous mycobacterial lung disease and Mycobacterium tuberculosis lung disease on X-ray images using deep transfer learning. BMC Infect Dis. Jan 19, 2023;23(1):32. [FREE Full text] [CrossRef] [Medline]
- Prasitpuriprecha C, Jantama SS, Preeprem T, Pitakaso R, Srichok T, Khonjun S, et al. Drug-resistant tuberculosis treatment recommendation, and multi-class tuberculosis detection and classification using ensemble deep learning-based system. Pharmaceuticals (Basel). Dec 22, 2022;16(1):13. [FREE Full text] [CrossRef] [Medline]
- Ayaz M, Shaukat F, Raja G. Ensemble learning based automatic detection of tuberculosis in chest X-ray images using hybrid feature descriptors. Phys Eng Sci Med. Mar 18, 2021;44(1):183-194. [FREE Full text] [CrossRef] [Medline]
- Kotei E, Thirunavukarasu R. Ensemble technique coupled with deep transfer learning framework for automatic detection of tuberculosis from chest X-ray radiographs. Healthcare (Basel). Nov 21, 2022;10(11):2335. [FREE Full text] [CrossRef] [Medline]
- Ahmad Hijazi MH, Qi Yang L, Alfred R, Mahdin H, Yaakob R. Ensemble deep learning for tuberculosis detection. IJEECS. Feb 01, 2020;17(2):1014. [CrossRef]
- Hijazi MHA, Kieu Tao Hwa S, Bade A, Yaakob R, Saffree Jeffree M. Ensemble deep learning for tuberculosis detection using chest X-ray and canny edge detected images. IJ-AI. Dec 01, 2019;8(4):429. [CrossRef]
- Alcantara MF, Cao Y, Liu B, Liu C, Zhang N, Zhang P, et al. eRx – a technological advance to speed-up TB diagnostics. Smart Health. May 2020;16:100117. [CrossRef]
- Griffin T, Chen Q, Sun X, Wang D, Brunette M, Cao Y, et al. eRxNet: a pipeline of convolutional neural networks for tuberculosis screening. 2021. Presented at: 2021 Third International Conference on Transdisciplinary AI (TransAI); September 20-22, 2021:47-56; Laguna Hills, CA. [CrossRef]
- Ameen Z, Saleh MA, Altrjman C, Alturjman S, Abdulkadir R. Explainable residual network for tuberculosis classification in the IoT era. 2021. Presented at: 2021 International Conference on Forthcoming Networks and Sustainability in AIoT Era (FoNeS-AIoT); December 27-28, 2021:9-12; Nicosia, Turkey. [CrossRef]
- Ravi V, Acharya V, Alazab M. A multichannel EfficientNet deep learning-based stacking ensemble approach for lung disease detection using chest X-ray images. Cluster Comput. Jul 19, 2023;26(2):1181-1203. [FREE Full text] [CrossRef] [Medline]
- Bhandari M, Shahi TB, Siku B, Neupane A. Explanatory classification of CXR images into COVID-19, pneumonia and tuberculosis using deep learning and XAI. Comput Biol Med. Nov 2022;150:106156. [FREE Full text] [CrossRef] [Medline]
- Ignatius JLP, Selvakumar S, Paul KGJL, Kailash AB, Keertivaas S, Akarvin Raja Prajan SAJ. Histogram matched chest X-rays based tuberculosis detection using CNN. Comput Syst Sci Eng. 2023;44(1):81-97. [CrossRef]
- Rashid R, Khawaja S, Akram M, Khan A. Hybrid RID network for efficient diagnosis of tuberculosis from chest X-rays. 2018. Presented at: 2018 9th Cairo International Biomedical Engineering Conference (CIBEC); December 20-22, 2018:167-170; Cairo, Egypt. [CrossRef]
- Okolo GI, Katsigiannis S, Ramzan N. IEViT: an enhanced vision transformer architecture for chest X-ray image classification. Comput Methods Programs Biomed. Nov 2022;226:107141. [FREE Full text] [CrossRef] [Medline]
- Munadi K, Muchtar K, Maulina N, Pradhan B. Image enhancement for tuberculosis detection using deep learning. IEEE Access. 2020;8:217897-217907. [CrossRef]
- Zainab YZS, Jameel A, Usman AM. Impact of transfer learning on chest X-ray (CXR) images for tuberculosis classification. 2021. Presented at: 2021 International Conference on Robotics and Automation in Industry (ICRAI); October 26-27, 2021:1-8; Rawalpindi, Pakistan. [CrossRef]
- Alcantara MF, Cao Y, Liu C, Liu B, Brunette M, Zhang N, et al. Improving tuberculosis diagnostics using deep learning and mobile health technologies among resource-poor communities in Perú. Smart Health. Jun 2017;1-2:66-76. [CrossRef]
- Abbas A, Abdelsamea M. Learning transformations for automated classification of manifestation of tuberculosis using convolutional neural network. 2018. Presented at: 2018 13th International Conference on Computer Engineering and Systems (ICCES); December 18-19, 2018:122-126; Cairo, Egypt. [CrossRef]
- Jasmine Pemeena Priyadarsini M, Kotecha K, Rajini GK, Hariharan K, Utkarsh Raj K, Bhargav Ram K, et al. Lung diseases detection using various deep learning algorithms. J Healthc Eng. Feb 03, 2023;2023:1-13. [CrossRef] [Medline]
- Shamrat FMJM, Azam S, Karim A, Islam R, Tasnim Z, Ghosh P, et al. LungNet22: a fine-tuned model for multiclass classification and prediction of lung disease using X-ray images. J Pers Med. Apr 24, 2022;12(5):680. [FREE Full text] [CrossRef] [Medline]
- Hamida S, El Gannour O, Cherradi B, Raihani A, Moujahid H, Ouajji H. A novel COVID-19 diagnosis support system using the stacking approach and transfer learning technique on chest X-ray images. J Healthc Eng. Nov 5, 2021;2021:1-17. [FREE Full text] [CrossRef] [Medline]
- Ravin N, Saha S, Schweitzer A, Elahi A, Dako F, Mollura D, et al. Mitigating domain shift in AI-based TB screening with unsupervised domain adaptation. IEEE Access. 2022;10:45997-46013. [CrossRef]
- Harsha Veena RN, Sreeja AN, Reddy KH, Hasmitha V, Lavanya R. Multiclass deep model for diagnosis of COVID-19 using chest X-ray. 2022. Presented at: 2022 6th International Conference on Trends in Electronics and Informatics (ICOEI); April 28-30, 2022:1767-1771; Tirunelveli, India. [CrossRef]
- Luján-García JE, Villuendas-Rey Y, López-Yáñez I, Camacho-Nieto O, Yáñez-Márquez C. NanoChest-Net: a simple convolutional network for radiological studies classification. Diagnostics (Basel). Apr 26, 2021;11(5):775. [FREE Full text] [CrossRef] [Medline]
- Liu C, Cohen I, Vishinkin R, Haick H. Nanomaterial-based sensor array signal processing and tuberculosis classification using machine learning. JLPEA. May 29, 2023;13(2):39. [CrossRef]
- Pattanasuwan C, Chongstitvatana P. Screening TB using deep transfer learning. 2021. Presented at: 2021 25th International Computer Science and Engineering Conference (ICSEC); November 18-20, 2021:330-333; Chiang Rai, Thailand. [CrossRef]
- Park S, Kim G, Oh Y, Seo JB, Lee SM, Kim JH, et al. Self-evolving vision transformer for chest X-ray diagnosis through knowledge distillation. Nat Commun. Jul 04, 2022;13(1):3848. [FREE Full text] [CrossRef] [Medline]
- Wong A, Lee JRH, Rahmat-Khah H, Sabri A, Alaref A, Liu H. TB-Net: a tailored, self-attention deep convolutional neural network design for detection of tuberculosis cases from chest X-ray images. Front Artif Intell. Apr 7, 2022;5:827299. [FREE Full text] [CrossRef] [Medline]
- Nafisah SI, Muhammad G. Tuberculosis detection in chest radiograph using convolutional neural network architecture and explainable artificial intelligence. Neural Comput Appl. Apr 19, 2022:1-21. [FREE Full text] [CrossRef] [Medline]
- Tsai A, Zhou Z, Ou Y, Wang I. Tuberculosis detection based on multiple model ensemble in chest X-ray image. 2022. Presented at: 2022 10th International Conference on Orange Technology (ICOT); November 10-11, 2022:1-4; Shanghai, China. [CrossRef]
- Tao Hwa SK, Bade A, Hijazi MHA, Saffree Jeffree M. Tuberculosis detection using deep learning and contrast-enhanced canny edge detected X-ray images. IJ-AI. Dec 01, 2020;9(4):713. [CrossRef]
- Huang Y, Darr CM, Gangopadhyay K, Gangopadhyay S, Bok S, Chakraborty S. Applications of machine learning tools for ultra-sensitive detection of lipoarabinomannan with plasmonic grating biosensors in clinical samples of tuberculosis. PLoS One. Oct 25, 2022;17(10):e0275658. [FREE Full text] [CrossRef] [Medline]
- Wu J, Bai J, Wang W, Xi L, Zhang P, Lan J, et al. ATBdiscrimination: an in silico tool for identification of active tuberculosis disease based on routine blood test and T-SPOT.TB detection results. J Chem Inf Model. Oct 14, 2019;59(11):4561-4568. [CrossRef]
- Chandra TB, Verma K, Singh BK, Jain D, Netam SS. Automatic detection of tuberculosis related abnormalities in chest X-ray images using hierarchical feature extraction scheme. Expert Syst Appl. Nov 2020;158:113514. [CrossRef]
- Wu Y, Wang H, Wu F. Automatic classification of pulmonary tuberculosis and sarcoidosis based on random forest. 2017. Presented at: 2017 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI); October 14-16, 2017:1-5; Shanghai, China. [CrossRef]
- AbuHassan K, Bakhori N, Kusnin N, Azmi U, Tania M, Evans B, et al. Automatic diagnosis of tuberculosis disease based on plasmonic ELISA and color-based image classification. 2017. Presented at: 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC); July 11-15, 2017:4512-4515; Jeju, South Korea. [CrossRef]
- Tania M, Lwin K, Shabut A, Hossain M. Clustering and classification of a qualitative colorimetric test. 2018. Presented at: 2018 International Conference on Computing, Electronics & Communications Engineering (iCCECE); August 16-17, 2018:7-11; Southend, UK. [CrossRef]
- Hu X, Wang J, Ju Y, Zhang X, Qimanguli W, Li C, et al. Combining metabolome and clinical indicators with machine learning provides some promising diagnostic markers to precisely detect smear-positive/negative pulmonary tuberculosis. BMC Infect Dis. Aug 25, 2022;22(1):707. [FREE Full text] [CrossRef] [Medline]
- Bobak CA, Titus AJ, Hill JE. Comparison of common machine learning models for classification of tuberculosis using transcriptional biomarkers from integrated datasets. Appl Soft Comput. Jan 2019;74:264-273. [CrossRef]
- Pholo MD, Hamam Y, Khalaf AB, Du C. Differentiating between COVID-19 and tuberculosis using machine learning and natural language processing. RIA. Apr 30, 2022;36(2):313-318. [CrossRef]
- Ullah R, Khan S, Ali Z, Ali H, Ahmad A, Ahmed I. Evaluating the performance of multilayer perceptron algorithm for tuberculosis disease Raman data. Photodiagn Photodyn Ther. Sep 2022;39:102924. [CrossRef] [Medline]
- Beccaria M, Bobak C, Maitshotlo B, Mellors TR, Purcaro G, Franchina FA, et al. Exhaled human breath analysis in active pulmonary tuberculosis diagnostics by comprehensive gas chromatography-mass spectrometry and chemometric techniques. J Breath Res. Nov 05, 2018;13(1):016005. [FREE Full text] [CrossRef] [Medline]
- Govindarajan S, Swaminathan R. Extreme learning machine based differentiation of pulmonary tuberculosis in chest radiographs using integrated local feature descriptors. Comput Methods Programs Biomed. Jun 2021;204:106058. [CrossRef] [Medline]
- Hu X, Liao S, Bai H, Wu L, Wang M, Wu Q, et al. Integrating exosomal microRNAs and electronic health data improved tuberculosis diagnosis. EBioMedicine. Feb 2019;40:564-573. [FREE Full text] [CrossRef] [Medline]
- Xing Z, Ding W, Zhang S, Zhong L, Wang L, Wang J, et al. Machine learning-based differentiation of nontuberculous mycobacteria lung disease and pulmonary tuberculosis using CT images. Biomed Res Int. Sep 30, 2020;2020:1-10. [FREE Full text] [CrossRef] [Medline]
- Ketchanji Mougang YC, Endale Mangamba L, Capuano R, Ciccacci F, Catini A, Paolesse R, et al. On-field test of tuberculosis diagnosis through exhaled breath analysis with a gas sensor array. Biosensors (Basel). May 22, 2023;13(5):570. [FREE Full text] [CrossRef] [Medline]
- Rashidi HH, Khan IH, Dang LT, Albahra S, Ratan U, Chadderwala N, et al. Prediction of tuberculosis using an automated machine learning platform for models trained on synthetic data. J Pathol Inform. 2022;13:10. [FREE Full text] [CrossRef] [Medline]
- Chen J, Wu L, Lv Y, Liu T, Guo W, Song J, et al. Screening of long non-coding RNAs biomarkers for the diagnosis of tuberculosis and preliminary construction of a clinical diagnosis model. Front Microbiol. Mar 3, 2022;13:774663. [FREE Full text] [CrossRef] [Medline]
- Maulana R, Kurniawan D, Ichsan M. Smart devices for self-diagnosing of lung condition based on body temperature and fingernail color. 2020. Presented at: 2020 2nd World Symposium on Artificial Intelligence (WSAI); June 27-29, 2020:113-117; Guangzhou, China. [CrossRef]
- Urooj S, Suchitra S, Krishnasamy L, Sharma N, Pathak N. Stochastic learning-based artificial neural network model for an automatic tuberculosis detection system using chest X-ray images. IEEE Access. 2022;10:103632-103643. [CrossRef]
- Fonseca A, Rocha B, Nogueira E, Vieira G, Fernandes D, Lima J, et al. Tuberculosis detection in chest radiography: a combined approach of local binary pattern features and monarch butterfly optimization algorithm. 2022. Presented at: 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC); June 27-July 1, 2022:1408-1413; Los Alamitos, CA. [CrossRef]
- Omisore MO, Samuel OW, Atajeromavwo EJ. A genetic-neuro-fuzzy inferential model for diagnosis of tuberculosis. Appl Comput Inform. Jan 2017;13(1):27-37. [CrossRef]
- Sharma A, Sharma A, Malhotra R, Singh P, Chakrabortty RK, Mahajan S, et al. An accurate artificial intelligence system for the detection of pulmonary and extra pulmonary tuberculosis. Tuberculosis (Edinb). Dec 2021;131:102143. [CrossRef] [Medline]
- Vijayaraj M, Abhinand PA, Venkatesan P, Ragunath PK. An ANN model for the differential diagnosis of tuberculosis and sarcoidosis. Bioinformation. Jul 31, 2020;16(7):539-546. [FREE Full text] [CrossRef] [Medline]
- Melendez J, Sánchez CI, Philipsen RHHM, Maduskar P, Dawson R, Theron G, et al. An automated tuberculosis screening strategy combining X-ray-based computer-aided detection and clinical information. Sci Rep. Apr 29, 2016;6(1):25265. [FREE Full text] [CrossRef] [Medline]
- Khan S, Ullah R, Shahzad S, Anbreen N, Bilal M, Khan A. Analysis of tuberculosis disease through Raman spectroscopy and machine learning. Photodiagnosis Photodyn Ther. Dec 2018;24:286-291. [CrossRef] [Medline]
- Rosli FA, Mashor MY, Noor SSM. An automated intelligent identification and counting system procedure for tuberculosis. J Phys Conf Series. Nov 2019;1372(1):012079. [CrossRef]
- Inbaraj XA, Villavicencio C, Macrohon JJ, Jeng J, Hsieh J. A novel machine learning approach for tuberculosis segmentation and prediction using chest-X-ray (CXR) images. Appl Sci. Sep 28, 2021;11(19):9057. [CrossRef]
- Xu S, Yuan H. A three-methylation-driven gene-based deep learning model for tuberculosis diagnosis in patients with and without human immunodeficiency virus co-infection. Microbiol Immunol. Jun 27, 2022;66(6):317-323. [FREE Full text] [CrossRef] [Medline]
- Saif AFM, Imtiaz T, Shahnaz C, Zhu W, Ahmad MO. Exploiting cascaded ensemble of features for the detection of tuberculosis using chest radiographs. IEEE Access. 2021;9:112388-112399. [CrossRef]
- Wang L, Zhang X, Tang J, Ma Z, Usman M, Liu Q, et al. Machine learning analysis of SERS fingerprinting for the rapid determination of Mycobacterium tuberculosis infection and drug resistance. Comput Struct Biotechnol J. 2022;20:5364-5377. [FREE Full text] [CrossRef] [Medline]
- Khatibi T, Shahsavari A, Farahani A. Proposing a novel multi-instance learning model for tuberculosis recognition from chest X-ray images based on CNNs, complex networks and stacked ensemble. Phys Eng Sci Med. Mar 22, 2021;44(1):291-311. [CrossRef] [Medline]
- Chithra RS, Jagatheeswari P. Severity detection and infection level identification of tuberculosis using deep learning. Int J Imaging Syst Tech. Apr 06, 2020;30(4):994-1011. [CrossRef]
- Feng B, Chen X, Chen Y, Lu S, Liu K, Li K, et al. Solitary solid pulmonary nodules: a CT-based deep learning nomogram helps differentiate tuberculosis granulomas from lung adenocarcinomas. Eur Radiol. Dec 27, 2020;30(12):6497-6507. [CrossRef] [Medline]
- Kant S, Srivastava M. Towards automated tuberculosis detection using deep learning. 2018. Presented at: 2018 IEEE Symposium Series on Computational Intelligence (SSCI); November 18-21, 2018:1250-1253; Bangalore, India. [CrossRef]
- Laeli A, Rustam Z, Pandelaki J. Tuberculosis detection based on chest X-rays using ensemble method with CNN feature extraction. 2021. Presented at: 2021 International Conference on Decision Aid Sciences and Application (DASA); December 7-8, 2021:682-686; Sakheer, Bahrain. [CrossRef]
- Huang C, Wang W, Zhang X, Wang S, Zhang Y. Tuberculosis diagnosis using deep transferred EfficientNet. IEEE/ACM Trans Comput Biol Bioinf. Sep 1, 2023;20(5):2639-2646. [CrossRef]
- Chang R, Chiu Y, Lin J. Two-stage classification of tuberculosis culture diagnosis using convolutional neural network with transfer learning. J Supercomput. Jan 16, 2020;76(11):8641-8656. [CrossRef]
- Rajaraman S, Candemir S, Xue Z, Alderson PO, Kohli M, Abuya J, et al. A novel stacked generalization of models for improved TB detection in chest radiographs. 2018. Presented at: 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC); July 18-21, 2018:718-721; Honolulu, HI. [CrossRef]
- Rashidi HH, Dang LT, Albahra S, Ravindran R, Khan IH. Automated machine learning for endemic active tuberculosis prediction from multiplex serological data. Sci Rep. Sep 09, 2021;11(1):17900. [FREE Full text] [CrossRef] [Medline]
- Pahar M, Klopper M, Reeve B, Warren R, Theron G, Niesler T. Automatic cough classification for tuberculosis screening in a real-world environment. Physiol Meas. Nov 26, 2021;42(10):105014. [FREE Full text] [CrossRef] [Medline]
- Wang M, Lee C, Wei Z, Ji H, Yang Y, Yang C. Clinical assistant decision-making model of tuberculosis based on electronic health records. BioData Min. Mar 16, 2023;16(1):11. [FREE Full text] [CrossRef] [Medline]
- Fati SM, Senan EM, ElHakim N. Deep and hybrid learning technique for early detection of tuberculosis based on X-ray images using feature fusion. Appl Sci. Jul 14, 2022;12(14):7092. [CrossRef]
- Nahiduzzaman M, Faruq Goni MO, Robiul Islam M, Sayeed A, Shamim Anower M, Ahsan M, et al. Detection of various lung diseases including COVID-19 using extreme learning machine algorithm based on the features extracted from a lightweight CNN architecture. Biocybern Biomed Eng. Jun 26, 2023;43(3):528-550. [FREE Full text] [CrossRef] [Medline]
- Patel M, Das A, Pant V. Detection of tuberculosis in radiographs using deep learning-based ensemble methods. 2021. Presented at: 2021 Smart Technologies, Communication and Robotics (STCR); October 9-10, 2021:1-7; Sathyamangalam, India. [CrossRef]
- Mehrrotraa R, Ansari MA, Agrawal R, Tripathi P, Bin Heyat MB, Al-Sarem M, et al. Ensembling of efficient deep convolutional networks and machine learning algorithms for resource effective detection of tuberculosis using thoracic (chest) radiography. IEEE Access. 2022;10:85442-85458. [CrossRef]
- Lee JH, Park S, Hwang EJ, Goo JM, Lee WY, Lee S, et al. Deep learning-based automated detection algorithm for active pulmonary tuberculosis on chest radiographs: diagnostic performance in systematic screening of asymptomatic individuals. Eur Radiol. Feb 28, 2021;31(2):1069-1080. [CrossRef] [Medline]
- Zeng G. On the confusion matrix in credit scoring and its analytical properties. Commun Stat - Theory Methods. Feb 07, 2019;49(9):2080-2093. [CrossRef]
- Trevethan R. Sensitivity, specificity, and predictive values: foundations, pliabilities, and pitfalls in research and practice. Front Public Health. Nov 20, 2017;5:307. [FREE Full text] [CrossRef] [Medline]
- Lalkhen AG, McCluskey A. Clinical tests: sensitivity and specificity. Contin Educ Anaesth Crit Care Pain. Dec 12, 2008;8(6):221-223. [CrossRef]
- Islam Z, Abdel-Aty M, Cai Q, Yuan J. Crash data augmentation using variational autoencoder. Accid Anal Prev. Mar 2021;151:105950. [CrossRef] [Medline]
- Shorten C, Khoshgoftaar TM. A survey on image data augmentation for deep learning. J Big Data. Jul 6, 2019;6(1):60. [CrossRef]
- Lin N, Guo S, Xie L. TB-Transformer: integrating mouse trace with object bounding-box for image caption. 2022. Presented at: 2022 IEEE 24th International Conference on High Performance Computing & Communications; 8th International Conference on Data Science & Systems; 20th International Conference on Smart City; 8th International Conference on Dependability in Sensor, Cloud & Big Data Systems & Application (HPCC/DSS/SmartCity/DependSys); December 18-20, 2022:810-814; Hainan, China. [CrossRef]
- Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv. Preprint posted online 2014. [CrossRef]
- He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. 2016. Presented at: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); June 27-30, 2016:770-778; Las Vegas, NV. [CrossRef]
- Colombo FM, Mello GR, Andrade BF, de OL, Kritski A, Koenigkam SM, et al. Preliminary results on pulmonary tuberculosis detection in chest X-ray using convolutional neural networks. 2020. Presented at: ICCS 2020: 20th International Conference on Computational Science; June 3–5, 2020:563-576; Amsterdam, The Netherlands. [CrossRef]
- Huang G, Liu Z, Van DML, Weinberger K. Densely connected convolutional networks. 2017. Presented at: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); July 21-26, 2017:2261-2269; Honolulu, HI. [CrossRef]
- Tan M, Le QV. EfficientNet: rethinking model scaling for convolutional neural networks. arXiv. Preprint posted online 2019. [FREE Full text] [CrossRef]
- Dosovitskiy A, Beyer L, Kolesnikov A, Weissenborn D, Zhai X, Unterthiner T, et al. An image is worth 16x16 words: transformers for image recognition at scale. OpenReview. Jan 12, 2021. URL: https://openreview.net/forum?id=YicbFdNTTy [accessed 2025-02-22]
- Liu Z, Mao H, Wu C, Feichtenhofer C, Darrell T, Xie S. A ConvNet for the 2020s. 2022. Presented at: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); June 18-24, 2022:11966-11976; New Orleans, LA. [CrossRef]
- Ali S, Abuhmed T, El-Sappagh S, Muhammad K, Alonso-Moral JM, Confalonieri R, et al. Explainable artificial intelligence (XAI): what we know and what is left to attain trustworthy artificial intelligence. Inf Fusion. Nov 2023;99:101805. [CrossRef]
- Tan C, Sun F, Kong T, Zhang W, Yang C, Liu C. A survey on deep transfer learning. In: Kůrková V, Manolopoulos Y, Hammer B, Iliadis L, Maglogiannis I, editors. Lecture Notes in Computer Science (Volume 11141). Cham. Springer; 2018:270-279.
- Hansun S, Argha A, Alinejad-Rokny H, Alizadehsani R, Gorriz JM, Liaw S, et al. A new ensemble transfer learning approach with rejection mechanism for tuberculosis disease detection. IEEE Trans Radiat Plasma Med Sci. 2024:1. [CrossRef]
- Cohen JP, Hashir M, Brooks R, Bertrand H. On the limits of cross-domain generalization in automated X-ray prediction. 2020. Presented at: Third Conference on Medical Imaging with Deep Learning; July 6-8, 2020:136-155; Montreal, QC.
Abbreviations
| ACM: Association for Computing Machinery |
| AI: artificial intelligence |
| AUC: area under the curve |
| AUROC: area under the receiver operating characteristic curve |
| CAD: computer-aided detection |
| ChatGPT: Chat Generative Pretrained Transformer |
| CNN: convolutional neural network |
| CV: cross-validation |
| CXR: chest X-ray |
| DL: deep learning |
| MC: Montgomery County |
| ML: machine learning |
| PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analysis |
| PROSPERO: Prospective Register of Systematic Reviews |
| QUADAS-2: Quality Assessment of Diagnostic Accuracy Studies version 2 |
| ROC: receiver operating characteristic |
| SZ: Shenzhen |
| TB: tuberculosis |
| TL: transfer learning |
| VGG: Visual Geometry Group |
| ViT: Vision Transformer |
| WHO: World Health Organization |
| XAI: explainable AI |
Edited by A Mavragani; submitted 21.11.24; peer-reviewed by Z Zhang, C Ordun; comments to author 24.12.24; revised version received 10.01.25; accepted 07.02.25; published 07.03.25.
Copyright©Seng Hansun, Ahmadreza Argha, Ivan Bakhshayeshi, Arya Wicaksana, Hamid Alinejad-Rokny, Greg J Fox, Siaw-Teng Liaw, Branko G Celler, Guy B Marks. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 07.03.2025.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research (ISSN 1438-8871), is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.