Tuberculosis (TB) stands out as the most widespread illness attributed to a single infectious agent, holding a place among the top 10 leading causes of death worldwide. Although TB can be prevented and treated, it impacts individuals of all age groups. In the year 2019 alone, nearly 10 million people across the globe contracted TB. Among them were 5.6 million males, 3.2 million females, and 1.2 million children (1).

Plain chest radiography remains a crucial tool in identifying early-stage pulmonary tuberculosis (TB) and monitoring the progress of treatments (2). Even when TB patients exhibit no symptoms, chest X-rays (CXRs) exhibit a high degree of sensitivity in detecting abnormalities related to pulmonary TB, particularly when interpreted by proficient radiologists. However, despite this capability, only 7.1 million out of an estimated 10 million TB cases worldwide were actually detected and reported in 2019 (3). Despite a decline in the global incidence rates of TB, these rates still fall short of the targets established by the World Health Organization's (WHO) End TB Strategy (3).

While improvements in digital radiography technology have improved the CXR image quality (4), lack of access to these facilities and skilled radiologists continues to be a problem, especially in underdeveloped regions with a high TB prevalence (5). However, the role of artificial intelligence (AI) in enhancing the accuracy of computer-aided diagnosis for pulmonary tuberculosis (TB) on chest radiography has become increasingly pivotal. AI technologies, such as deep learning algorithms, offer a unique capacity to analyze vast amounts of medical imagery with remarkable precision and speed. In the context of TB diagnosis, AI systems can swiftly and accurately detect subtle abnormalities and patterns on chest X-rays that may elude even skilled human radiologists. By providing reliable and consistent assessments, AI-driven computer-aided diagnosis has the potential to significantly expedite the identification of TB cases, especially in asymptomatic patients, ultimately leading to more timely interventions and improved treatment outcomes. This symbiotic integration of AI with medical diagnostics not only augments the overall diagnostic accuracy but also holds promise for more efficient resource utilization within healthcare systems, thus reinforcing its significance in combating pulmonary tuberculosis on a global scale (6–9). If they perform accurately, these CAD systems may facilitate CXR reading for TB screening and advance the WHO's End TB agenda (3, 10). There are only a few studies in this field, and the majority have methodological flaws, focus on a single CAD program, have scant screening data, or are industry-funded (11, 12).

Furthermore, most studies compared performance against a suboptimal reference standard of a single sputum specimen tested with Xpert MTB/RIF evaluated an online CAD processing system or shared images with the CAD vendors (13, 14). This highlights the need for independent and thorough studies. Offline and multiple AI systems have been the focus of more recent research (15–17), but there are still very few. An international meeting by WHO in 2016 concluded that more data on the effectiveness and application of CAD systems for TB screening were needed (18).

Although the Gaza Strip is a low TB burden region, it is well-known that tuberculosis is primarily a socioeconomic problem associated with overcrowding, poor hygiene, a lack of fresh water, and limited access to health care, which is typical in the Gaza Strip (19, 20). There is a lack of well-organized healthcare infrastructure, which affects the finding and treatment of TB cases, complicated disease control in the Gaza Strip, and a lack of statistics on TB after 2016.

Accordingly, this study aims to evaluate the performance of the automatic software computer-aided detection for tuberculosis (CAD4TBv6) in diagnosing pulmonary TB on chest radiography and compare the CAD4TB software reading with the results of radiologists' reports in the Gaza Strip, Palestine.

To the best of our knowledge, this is the first study to evaluate the performance of the automatic software computer-aided detection for tuberculosis (CAD4TBv6) in diagnosing pulmonary TB on chest radiography and compare the CAD4TB software reading with the results of radiologists' reports in the Gaza strip, Palestine. Sputum samples were obtained from those with a score > 40 for GeneXpert analysis and those with symptoms strongly suggestive of pulmonary TB. A certified radiologist read the obtained CXRs.

Although the utilization of Xpert in programmed applications has increased recently, the WHO has also advocated using screening instruments like CXR that employ more affordable diagnostic algorithms (17, 29–32). Using an automated system to analyze a chest radiograph for the presence of active pulmonary TB produces objective, repeatable results and a consistent format for reporting. Creating software that provides automated CXR interpretation is a significant step toward connecting technology advancements to mass-screening initiatives for TB (23). In addition to increasing case identification in screening programs, using CAD4TB as a triage tool to pre-screen people for Xpert may help lower program expenses (33). Those with low CAD4TB scores had a low likelihood of testing positive for TB, so they might not be prioritized for Xpert testing using this method. Employing a triage tool such as CAD4TB might encourage more judicious use of Xpert by reducing the number of cartridges used in resource-constrained environments where there is not enough money to cover testing for all individuals with presumptive TB. This also applies to settings where onsite radiologists might not always be present to review CXRs. It is important to bear in mind that the costs associated with acquiring and operating digital X-ray devices must be balanced against the potential savings resulting from a reduced need for Xpert exams. This underscores the need for a comprehensive analysis that examines both the financial implications and consequences of widespread mass-screening through chest X-rays (CXR) (25).

According to research by Gautam et al. diabetes has been proven to increase both the likelihood of contracting TB and the severity of the illness (34). Additionally, research has demonstrated that smoking contributes significantly to the development of TB and raises the severity and fatality rates (35). However, our findings were in line with Tavaziva et al. as smoking and diabetes do not seem to make TB more likely to strike or to progress more severely (22).

According to 12 single-center assessment studies, the WHO estimates that Xpert's pooled sensitivity and specificity values for the detection of TB are 92.5 and 98%, respectively (36). However, the accessibility of digital radiography is a need for CAD utilization, which is not yet available in most resource-constrained low-burden settings. Nevertheless, it has been determined to be viable. It produces chest radiography significantly superior to traditional X-ray equipment in regions with low resources, such as Gaza Strip (37).

The findings of the current study of high specificity, high negative predictive value, high sensitivity, high diagnostic accuracy, and relatively low positive predictive value were consistent with findings from other studies in different settings testing CAD4TB (24, 25, 27, 38, 39). Compared to a confirmatory test like the Xpert, triage tests should have a sensitivity of 90% and a specificity of 70%, according to a 2014 WHO consensus meeting to define targets for new TB diagnostic technologies (40).

The findings of this study point to a feasible, effective, and even cost-effective strategy for TB screening in a symptomatic group that combines CAD4TB and symptomatology. Earlier research from different contexts are consistent with our findings (41–43).

The direct comparison between computerized and radiologist reading on the same set of pictures is one of the study's strengths. The extent to which this comparison may be generalized is severely constrained by the inter-reader variance in the reading of chest X-rays and the possibility of involving only one board-certified radiologist. The fact that our short-term study was only done in a single low-burden region presents a second limit that Data gathered from patients who presented with lower respiratory tract symptoms may not accurately reflect the prevalence of illness in the general community.

It is imperative that forthcoming research endeavors direct their attention toward comprehensively examining the multifaceted ramifications encompassing the adoption of Computer-Aided Diagnosis (CAD) within diverse low-burden nations. In particular, a rigorous investigation is warranted to elucidate the intricate interplay of financial, practical, and ethical considerations inherent in the deployment of CAD within these unique contexts. Subsequent investigative trajectories should encompass an exploration of the potential synergies arising from amalgamating CAD-generated outcomes with an array of clinical parameters, encompassing symptomatic manifestations and risk profiling. Furthermore, a profound research agenda should be undertaken to systematically assess the efficacy of CAD not only within the spectrum of operational feasibility but also across various technical dimensions indispensable for its seamless integration into prevailing diagnostic frameworks. Concurrently, an evaluative lens should be directed toward pioneering CAD products that are emergent within the market landscape. This comprehensive inquiry stands to provide a robust foundation for harnessing the maximal potential of CAD, charting its trajectory toward optimized medical diagnostics within the distinctive landscape of low-burden nations.

Combining chest X-ray readings by CAD4TB and symptomatology is extremely valuable for screening a population at risk in a resource-limited country with a high burden of a neglected disease. CAD4TB is noticeably more efficient than other methods for TB screening and early diagnosis in people who would otherwise go undetected. In order to increase case finding and infection control and lower the cost of case detection within triage algorithms, CAD solutions may present an opportunity. This inspires further investigation into the best ways to utilize its potential as a support tool for clinical officers in the diagnostic interpretation of radiographs as well as a stand-alone triage test in systematic screening settings.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The study protocol was approved by the Palestinian Health Research Council (Helsinki Committee for Ethical Approval research number: PHRC/HC/1175/22). Additional approval was obtained from the included Gaza Strip hospitals. Study participants provided written informed consent for survey activities.

SA: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing — original draft, Writing — review & editing. KZ: Methodology, Project administration, Software, Validation, Visualization, Writing — original draft, Writing — review & editing.