Evaluating InferVision's Computer-Aided Detection (CAD) algorithm for Tuberculosis (TB) screening, Lusaka, Zambia

• Published in: PLOS global public health Vol. 5; no. 6; p. e0003955
• Main Authors: Paul Somwe, Minyoi Maimbolwa, Kanema Chiyenu, Mwansa Lumpa, Mary Kagujje, Monde Muyoyeta
• Format: Journal Article
• Language: English
• Published: Public Library of Science (PLoS) 01.01.2025
• ISSN: 2767-3375
• DOI: 10.1371/journal.pgph.0003955

The objective of this study was to evaluate the diagnostic performance of InferRead DR Chest for tuberculosis (TB) screening in a high HIV and TB burden setting. The study assessed the performance of InferRead DR Chest using anonymized chest X-ray images from an active TB case finding study in Lusaka, Zambia, for individuals aged 15 and older. The Xpert MTB/RIF or MTB culture was the composite reference standard. Performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), and a binary classification point was selected where the sensitivity aligned with the WHO target product profile for TB screening tools. Of the 1,890 chest X-ray images that met the inclusion criteria, 91.5% of participants reported at least one TB symptom. The median age was 38 years (IQR: 29-47), and 1,186 (62.8%) were male. From the study sample, 449 participants (23.8%) reported a history of previous TB, and 704 (37.2%) were HIV positive. Among the analyzed images, 289 (15.3%) were classified as TB positive based on the composite reference standard test results. The overall area under the curve (AUC) was 0.81 (95% CI: 0.78-0.83). Among individuals with a history of previous TB and those who were HIV positive, the AUCs were 0.71 (95% CI: 0.63-0.79) and 0.77 (95% CI: 0.72-0.82), respectively. At a sensitivity of 90.3% (95% CI: 86.3%-93.5%), InferRead DR Chest achieved a specificity of 39.2% (95% CI: 36.8%-41.7%) at TB score cut point of 0.12. InferRead DR Chest had acceptable performance in our population. Additional training and piloting of InferRead DR Chest in this population is recommended.