Brazil has over 211 million people and continues to face a high burden of TB. In the same year, an estimated 103,000 people developed TB, and 13,000 people died from the disease. Despite progress, 10,815 people with TB were missed by the health system.
We introduced CAD4TB in Brazil in 2018, delivering the solution to the Foundation for Scientific and Technological Development (FIOTEC) to strengthen early detection and improve screening accuracy. In 2025, we expanded our support by donating 2 CAD4TB and 1 Delft Light to the National Council for Scientific and Technological Development (CNPq), reinforcing Brazil’s capacity for AI-enabled TB screening.
Making a difference
Research paper
Between 2017 and 2018, inmates from 3 prisons in Brazil were screened for TB by symptom assessment, chest radiography, sputum testing by Xpert MTB/RIF fourth-generation assay, and culture. Chest radiographs were scored by an automated interpretation algorithm (Computer-Aided Detection for Tuberculosis [CAD4TB]) locally calibrated to establish a positivity threshold. Four diagnostic algorithms were evaluated. The study looked at yield (percentage of total cases found) and efficiency (prevalence among those screened) for each algorithm. The study also performed unit costing to estimate the costs of each screening or diagnostic test and calculated the cost per case detected for each algorithm.
As part of the study, 5387 prisoners were screened prisoners, of whom 214 (3.9%) were diagnosed with TB. Compared to other screening strategies initiated with chest radiography or symptoms, the trial of all participants with a single Xpert MTB/RIF sputum test detected 74% of all TB cases for US$249 per case diagnosed. Performing Xpert MTB/RIF screening tests only on those with symptoms had a similar cost per case diagnosed (US$255) but missed 35% more cases (73 vs. 54) as screening all inmates.
In the prospective study, in 3 prisons in a high TB burden country, it was found that testing all inmates with sputum Xpert MTB/RIF was a sensitive approach while remaining cost-efficient. These results support the use of Xpert MTB/RIF for mass screening in TB-endemic prisons.
For more information, please read the complete called ‘Yield, Efficiency, and Costs of Mass Screening Algorithms for Tuberculosis in Brazilian Prisons’.
Moreover, in 2023, research was conducted through a B screening in three male prisons in Brazil, using data from October 2017 to December 2019. The study administered a standardised questionnaire, performed a chest x-ray in a mobile unit, and collected sputum for confirmatory testing using Xpert MTB/RIF and culture. Moreover, the X-ray images were evaluated with different AI algorithms, including the CAD4TB artificial intelligence software of Delft Imaging. Note: The study used an older version of CAD4TB, version 6, than what is currently available. The study showed that automated X-ray interpretation algorithms (CAD/AI) can effectively triage tools for TB screening in prisons. However, the specificity was found to be insufficient in individuals with previous TB.
For more information, please read the complete publication, Evaluation of Chest X-ray with Automated Interpretation Algorithms for Mass Tuberculosis Screening in Prisons: A Cross-Sectional Study.
