This Arkangel AI whitepaper reviews what AI can and cannot do today in pulmonology and respiratory health, and proposes a validation agenda for Latin America. It is a rapid narrative synthesis of 27 sources from searches of PubMed/MEDLINE, Crossref, WHO, PAHO and SciELO run on July 22, 2026, covering six domains: thoracic imaging and nodules, tuberculosis screening, pulmonary function and obstructive disease, interstitial lung disease, sleep and procedures, and prediction and monitoring. Each domain is graded from retrospective discrimination to routine implementation.
The strongest evidence concerns chest radiographs and CT. In a randomized trial of 323 patients in respiratory outpatient clinics, AI assistance raised nonradiologists' per-radiograph AUC from 0.718 to 0.840, but clinical decisions did not change significantly. In Brazil, a retrospective validation reported internal AUCs of 0.94 for lung abnormality and tuberculosis, and a lung cancer screening study of 779 scans reported 92.5% sensitivity and 97.8% negative predictive value alongside 78.5% specificity, 50% positive predictive value and missed nodules. WHO endorses computer-aided TB detection on chest radiographs for people aged 15 or older, but a Brazilian pediatric study found 52% sensitivity against microbiological confirmation. For COPD, asthma, interstitial disease, sleep apnea and event prediction, retrospective studies predominate, and direct Latin American evidence was narrow and concentrated in Brazil. The document is a whitepaper, not peer reviewed and not a systematic review; all authors are affiliated with Arkangel AI, and the evidence it cites does not evaluate Arkangel products.
