Supportive AI in Five Selected Radiology Tasks: Limits and Governance for Latin America

This rapid narrative evidence brief synthesizes 27 sources across five selected radiology tasks. The main evidence cutoff was June 13, 2026, followed by targeted retrieval in July and source verification in September. Selection was neither exhaustive nor performed independently by two reviewers. The clinical and operational evidence for artificial intelligence (AI) in radiology varies across tasks. In this selected corpus, prospective evidence is most developed for mammography reading support in organized screening programs; studies also assess selected chest radiograph findings, reading time, and workload. This synthesis neither ranks the entire field nor demonstrates mortality benefit, cost savings, automatic transferability across hospitals, or replacement of radiologists. Regulatory authorization does not establish local clinical utility. Responsible evaluation must specify the tool, task, population, workflow integration, and expected human response. Evidence should be separated into retrospective development or discrimination, external validation, prospective utility, process effects, clinical outcomes, and routine implementation. In this selective map, AI-supported mammography has undergone relevant prospective evaluation in Sweden; triage and efficiency show heterogeneous process effects; and language-model support for report simplification or error detection still requires human oversight. The chest-radiograph nodule trial has a tabular correction that, according to the published erratum, does not alter its analyses or conclusions; applicability remains specific to the evaluated task and population. Regional sources retrieved by the focused search describe collaboration, implementation experience, and one Mexican retrospective evaluation, but they do not demonstrate generalizable regional clinical effectiveness. Regional frameworks support the need for institutional coordination, evaluation, monitoring, and interoperability. A narrative review identified few Latin American studies of breast screening in rural and resource-constrained settings within its own corpus. This brief does not claim exhaustive regional coverage; its recommendations describe evaluation conditions and implementation inferences rather than established local performance.