
138 AJTCCM VOL. 30 NO. 3 2024
ABSTRACTS
Objectives. To determine whether computer-aided detection (CAD)
could accurately identify cavitary disease (proxy for infectiousness)
during community-based active case-finding (ACF) for TB.
Methods. Participants with microbiologically confirmed TB (sputum
Xpert Ultra and/or culture positivity), recruited from two community-
based ACF studies in SouthAfrica (XACT-3 and XACT- 19), underwent
point-of-care (POC) chest radiography analysed by CAD (qXR v4.0) and
two expert human readers. All participants underwent positron emission
tomography-computed tomography (PET-CT) scanning. e accuracy
of CAD in detecting cavitary disease was compared with that of PET-CT
(radiological reference standard; presence of cavitation plus SUVmax >2.5
suggesting metabolically active cavitary disease was used as a proxy for
probable infectiousness).
Results. A total of 1 455 participants were enrolled, of whom 112
(7.7%) had microbiologically confirmed TB (n=54/112 (48.2%) were
asymptomatic). Chest radiographs and PET-CT scans were available
in 82.1% (n=92/112); 61/92 (66.3%) had cavitary disease on PET-CT
(median SUVmax 5.4). At the developer-recommended cavity threshold,
CAD sensitivity and specificity were 60.7% (95% condence interval (CI)
47.3-72.9) and 84.6% (95% CI65.1-95.6), respectively. Compared with
human readers, CAD had statistically similar sensitivity (60.7% (95%
CI47.3-72.9) v. 66.7% (95% CI52.5-78.9)), specificity (84.6% (95%
CI65.1-95.6) v. 73.1% (95% CI52.2-88.4)), and positive predictive value
(90.2% (95% CI78.6-95.9) v. 83.7% (95% CI72.7- 0.9)).
Conclusion. A high proportion (two-thirds) of individuals with
microbiologically confirmed TB had evidence of metabolically active
cavitary disease, suggesting that they were probably infectious. CAD may
be a useful POC rule-in test to detect cavitary disease, and could inform
contact tracing and treatment strategies in endemic settings.
Clinical evaluation of computer-aided
digital X-ray detection of pulmonary
tuberculosis during community-based
screening/active case-finding
A J Scott, T Perumal, A Pooran, S Oelofse, T Mthiyane,
MvanderWalt, Z Z Qin, J Fehr, A D Grant, E B Wong, A Esmail,
K Dheda
Centre for Lung Infection and Immunity, Division of Pulmonology, Department
of Medicine and University of Cape Town Lung Institute, Cape Town,
SouthAfrica; Centre for the Study of Antimicrobial Resistance, SouthAfrican
Medical Research Council, Cape Town, SouthAfrica
Corresponding author: K Dheda (keertan.dheda@uct.ac.za)
Background. Computer-aided detection (CAD) has been recommended
as a tuberculosis (TB) screening tool. However, few studies have
evaluated CAD in a community-based setting.
Objectives. To determine the clinical utility of CAD during community-
based active case-finding (ACF).
Methods. Individual patient data were pooled from five community-
based ACF studies in SouthAfrica. CAD-interpreted chest radiography
(CAD4TB v7) was assessed against a microbiological reference standard
(sputum Xpert Ultra and/or culture positivity). e clinical utility of
CAD was evaluated, and a preliminary cost analysis was performed.
Results. Of 20 770 individuals enrolled across all studies, 530 (2.6%)
had microbiologically proven TB. Evaluable controls (non-TB) and
cases (TB) were randomly selected from this parent population in a 2:1
ratio (n=501 TB positive and n=938 TB negative; total N=1 439). CAD
achieved an areaunder the receiver operating curve (AUC) of 0.83 (95%
condence interval (CI) 0.80-0.85). At fixed sensitivities of 90% and
85% (thresholds of5 and 10), specificity was 44.9% (95% CI42.5-47.3)
and 54.1% (95% CI51.7-56.5), respectively. CAD (AUC) performed
worse in individuals living with HIV (v. HIV negative) (0.76 v. 0.85;
p=0.004) and in asymptomatic (v. symptomatic) individuals (0.79 v.
0.85; p=0.008). Nevertheless, CAD-directed Xpert (v.universal Xpert
testing) reduced cost by ~20% per individual with TB diagnosed, at the
detriment of a ~10% reduction in the true TB positives detected.
Conclusion. In the setting of community-based ACF, CAD did not meet
the World Health Organization screening test target product profile
(>90% sensitivity, >70% specificity) and performed more poorly in
certain subgroups. However, a context-specific CAD-directed strategy
could still be cost-saving. ese data inform community-based ACF
strategies aiming to disrupt the TB transmission cycle.
Can AI-driven computer-aided
detection optimise Xpert-orientated
community-based active case-finding
for TB (XACT-19)? An interim trial
progress report
A J Scott, A Chizema, G Chongo, T Perumal, S Jaumdally,
DMilimo, S Oelofse, A Esmail, J Mutsvangwa, H Ayles, K Dheda
Centre for Lung Infection and Immunity, Division of Pulmonology, Department
of Medicine and University of Cape Town Lung Institute, Cape Town,
SouthAfrica; Centre for the Study of Antimicrobial Resistance, SouthAfrican
Medical Research Council, Cape Town, SouthAfrica
Corresponding author: K Dheda (keertan.dheda@uct.ac.za)
Background. Approximately one-third of individuals newly ill with
tuberculosis (TB) are undiagnosed/unreported. Detecting such
individuals in endemic communities has been restricted by the lack of
sensitive, point-of-care (POC) diagnostic tools.
Objectives. To determine the impact of computer-aided detection
(CAD) during community-based active case-finding (ACF).
Methods. In an ongoing multicentre open-labelled randomised
controlled trial, individuals at risk for TB were recruited from TB/HIV-
endemic communities in SouthAfrica (SA), Zambia and Zimbabwe.
Using a low-cost mobile van staffed by three healthcare workers
and equipped with an ultra-portable X-ray and GeneXpert system,
participants were randomised into either POC ‘CAD+Xpert’ (CAD
followed by Xpert in CAD-positive participants) or POC ‘Xpert only’
(Xpert in all). e reference standard was microbiologically proven TB
(Xpert-MTB/RIF-Ultra and/or sputum culture positivity).
Results. As of March 2024, 1 667 participants had been randomised
(SA n=544 (32.6%), Zambia n=756 (45.4%), Zimbabwe n=367 (22.0%)).
ere were 714/1 667 people living with HIV (42.8%), of whom 108/714
(15.1%) were newly diagnosed. A total of 56/1 667 participants (3.4%)
tested positive for TB (SA n=37/544 (6.8%), Zambia n=18/756 (2.4%),
Zimbabwe n=1/367 (0.3%)), of whom 27/56 (48.2%) were subclinical
(i.e. asymptomatic) and 17/56 (30.4%) were smear positive. Among
the 826 participants randomised into the CAD+Xpert arm, CAD
detected 17/21 individuals with TB (81.0%). ere were 337/826 CAD
false positives (40.8%). However, CAD was truly negative in 468/826
participants (56.7%) who did not undergo Xpert testing.