Computer-aided interpretation of chest radiography reveals the spectrum of tuberculosis in rural South Africa

Abstract

Computer-aided digital chest radiograph interpretation (CAD) can facilitate high-throughput screening for tuberculosis (TB), but its use in population-based active case-finding programs has been limited. In an HIV-endemic area in rural South Africa, we used a CAD algorithm (CAD4TBv5) to interpret digital chest x-rays (CXR) as part of a mobile health screening effort. Participants with TB symptoms or CAD4TBv5 score above the triaging threshold were referred for microbiological sputum assessment. During an initial pilot phase, a low CAD4TBv5 triaging threshold of 25 was selected to maximize TB case finding. We report the performance of CAD4TBv5 in screening 9,914 participants, 99 (1.0%) of whom were found to have microbiologically proven TB. CAD4TBv5 was able to identify TB cases at the same sensitivity but lower specificity as a blinded radiologist, whereas the next generation of the algorithm (CAD4TBv6) achieved comparable sensitivity and specificity to the radiologist. The CXRs of people with microbiologically confirmed TB spanned a range of lung field abnormality, including 19 (19.2%) cases deemed normal by the radiologist. HIV serostatus did not impact CAD4TB’s performance. Notably, 78.8% of the TB cases identified during this population-based survey were asymptomatic and therefore triaged for sputum collection on the basis of CAD4TBv5 score alone. While CAD4TBv6 has the potential to replace radiologists for triaging CXRs in TB prevalence surveys, population-specific piloting is necessary to set the appropriate triaging thresholds. Further work on image analysis strategies is needed to identify radiologically subtle active TB.

Reimagining the status quo: How close are we to rapid sputum-free tuberculosis diagnostics for all?

“… 29 - Able to rapidly screen all facility entrants - Specimen free - Rapidly scalable on mobile phones - Abnormal sounds in the absence of TB could be used to diagnose other respiratory diseases - No external validation data - Early-stage technology - Standardised large-scale data collection and curation required for algorithm development and testing - Performance likely highly setting-specific, especially when other lung diseases are frequent - Potential utility for evaluating post TB lung damage Stethoscopes - Stethee 28 , 108 and others undergoing evaluation - No published studies - Unknown Imaging Portable dCXR CXR highly sensitive and increasingly feasible to be near POC due to advances in low-dose portable instruments and automated image reading. Triage Software products include: qXR (QURE.ai Technologies), CAD4TB (Delft Imaging Systems), and Lunit (FujiFILM), CAD4Good (EPCON) (see AI4HLTH for a complete overview) 30 - In a very large evaluation in Bangladesh, qXR and CAD4-TB have the highest specificities (both 73%) at >90% sensitivity 50 - Automated reading systems can overcome radiologist shortages and are WHO-endorsed 48 - Expensive equipment and infrastructure (including high-speed internet) required 52 , however, existing equipment may be retrofitted such that analogue images can be converted to digital ones - Market rapidly evolving and selection of the right solutions challenging - Challenges in identifying which settings for prioritisation, including case finding scenarios where pre-symptomatic TB can be detected 115 - Achieving adequate throughput to maximise utility is challenging - Unclear how these technologies will benefit diagnosis of other diseases - Open source social impact programs such as CAD4Good may reduce costs 30 POCUS Improving affordability and portability of ultrasound devices has led to interest in the use of POCUS. Triage Many (most standard point-of-care ultrasound machines can be used) - Systematic review reported sensitivities ranging from 73-100% for subpleural nodules detected and 47-80% for lung consolidation 54 Potential to expand diagnosis of extrapulmonary TB and increase diagnostic yield in populations such as children or PLHIV - Limited data, currently majority of studies in adults and for pulmonary TB - Effects of operator and machine variability unclear - Optimal imaging protocols unclear …”

“Similar performances but markedly different triaging thresholds in three CAD4TB versions risk systematic errors in TB screening programs”

“…We previously evaluated the performance of CAD4TBv5 and v6 in a community-based screening program in rural South Africa 12 and here sought to assess the performance of the most recently released version CAD4TBv7 in the same collection of chest x-rays. As previously described, between 2018 and 2019, we used CAD4TB version 5 to triage participants for sputum examination (Xpert Ultra and Mtb culture) using a threshold of CAD4TBv5 25.…”

“…One possible explanation for this is that other studies have enrolled symptomatic patients seeking TB diagnosis, whereas our study was a community-screening, which enrolled participants regardless of symptom status, which revealed more subclinical and early-stage TB. 16 Limitations of our study have been described 12 and include that only people with symptoms or a CAD4TBv5 score greater or equals 25 were selected for microbiological sputum assessment, which might have biased our performance assessment results. A limitation of this study, is that our analysis is limited to a single CAD software.…”

“…Previous studies showed that triaging thresholds are not universal and require careful adjustment to demographic characteristics, laboratory capacities, budget, healthcare settings and study goals. 2,3,12,13,[4][5][6][7][8][9][10][11] Multiple studies compared the performance among different CAD-software platforms, 11,14,15 but comparisons between software versions are lacking in the literature and are needed to support screening programs when using CAD-software and negotiating software updates.…”

“…To deal with such intra-version variation, screening programs will need to select new triaging thresholds for each new software update. Previous work 9,12 and the developer 17 suggest conducting a population-specific pilot study to finding triage thresholds that optimally serve the goals of each screening exercise. The design and conduct of such pilots, however, may result in additional costs, slow timelines and require alterations to established screening workflows.…”