The Performance of Computer-Aided Detection Digital Chest X-ray Reading Technologies for Triage of Active Tuberculosis Among Persons With a History of Previous Tuberculosis

Abstract

Background Digital Chest X-ray (dCXR) computer-aided detection (CAD) technology uses lung shape and texture analysis to determine the probability of TB. However, many patients with previously treated TB have sequelae, which also distort lung shape and texture. We evaluated the diagnostic performance of two CAD systems for triage of active TB in patients with previously treated TB. Methods We conducted a retrospective analysis of data from a cross-sectional active TB case finding study. Participants ≥15 years, with ≥1 current TB symptom, and with complete data on history of previous TB, dCXR, and TB microbiological reference (Xpert MTB/RIF) were included. dCXRs were evaluated using CAD4TB (v.7.0) and qXR (v.3.0). We determined the diagnostic accuracy of both systems, overall and stratified by history of TB, using a single threshold for each system that achieved 90% sensitivity and maximized specificity in the overall population. Results Of 1,884 participants, 452 (24.0%) had a history of previous TB. The prevalence of microbiologically confirmed TB among those with and without history of previous TB was 12.4% and 16.9% respectively. Using CAD4TB, the sensitivity and specificity was 89.3% (95%CI:78.1-96.0) and 24.0% (95%CI19.9-28.5) and 90.5% (95%CI:86.1-93.3) and 60.3% (95%CI:57.4-63.0) among those with and without previous TB, respectively. Using qXR, the sensitivity and specificity was 94.6% (95%CI:85.1-98.9) and 22.2% (95%CI:18.2-26.6) and 89.7% (95%CI:85.1-93.2) and 61.8% (95%CI:58.9-64.5) among those with and without previous TB, respectively. Conclusion The performance of CAD systems as a TB triage tool is decreased among persons previously treated for TB.

The rise of artificial intelligence reading of chest X-rays for enhanced TB diagnosis and elimination

“… 10 – 12 CAD has poorer performance with increasing age and decreasing body mass index, in smear-negative disease, in those previously diagnosed with TB and in people living with HIV. 10 , 13 – 16 This heterogeneity in the accuracy of the threshold score means that users should calibrate CAD with local data, rather than use the reported threshold scores, 10 , 11 , 13 , 15 – 17 and consider implementing different threshold scores for different populations, particularly for people living with HIV. This is one of the major challenges when implementing CAD: the burden on programmes to establish a threshold score with a study prior to implementation is high and prone to error, depending on how the study is done.…”

Diagnostic Accuracy of Computer-Aided Detection During Active Case Finding for Pulmonary Tuberculosis in Africa: A Systematic Review and Meta-analysis

“…Characteristics of included articles are presented in Table 1 . The diagnostic accuracy of CAD was evaluated retrospectively in 3 studies [ 30 , 32 , 34 ], among which 2 included data from national prevalence surveys [ 30 , 32 ]. The remaining 2 studies utilized a prospective study design [ 31 , 33 ], of which 1 reported the data of participants recruited from both community screening and health care facilities (ie, both active and passive case finding) [ 34 ].…”

“…The diagnostic accuracy of CAD was evaluated retrospectively in 3 studies [ 30 , 32 , 34 ], among which 2 included data from national prevalence surveys [ 30 , 32 ]. The remaining 2 studies utilized a prospective study design [ 31 , 33 ], of which 1 reported the data of participants recruited from both community screening and health care facilities (ie, both active and passive case finding) [ 34 ]. CAD4TB software (Delft Imaging, ‘s-Hertogenbosch, the Netherlands) was used in all studies, with 1 study including an additional evaluation using qXR software (Qure.ai, Mumbai, India) [ 34 ].…”

“…The remaining 2 studies utilized a prospective study design [ 31 , 33 ], of which 1 reported the data of participants recruited from both community screening and health care facilities (ie, both active and passive case finding) [ 34 ]. CAD4TB software (Delft Imaging, ‘s-Hertogenbosch, the Netherlands) was used in all studies, with 1 study including an additional evaluation using qXR software (Qure.ai, Mumbai, India) [ 34 ]. Predefined thresholds based on prior evidence were used by studies using prospective study designs.…”

Development and Validation of a Deep Learning Model for Detecting Signs of Tuberculosis on Chest Radiographs among US-bound Immigrants and Refugees

“…Using other available information for prediction, whether by stacking an additional model on top of the image-processing module or by altering the module itself to accept multimodal inputs, may yield diagnostic gains, not only because the approach mirrors the process by which clinical diagnoses are made and thus seems a priori sensible, but also because patient characteristics like HIV status and history of tuberculosis are known to affect model performance based on the radiographs alone. 26,46 Although some technical innovation may be required to make this approach feasible, it may well improve our ability to predict tuberculosis disease, especially in low-resourced settings where access to trained radiologists is limited, and thus seems well worth pursuing.…”