Artificial Intelligence Tools In Radiation Oncology In Low- and Middle-Income Countries: A Systematic Review
Abstract
Purpose
Access to radiotherapy in low- and middle-income countries (LMICs) remains severely limited due to workforce shortages, infrastructure constraints, and workflow inefficiencies. Artificial intelligence (AI) has the potential to alleviate these pressures by supporting key components of the radiotherapy workflow; however, transferring AI tools developed in high-income countries (HICs) without local validation may introduce safety risks and widen inequities. This systematic review evaluated patterns of AI deployment, validation, performance, and implementation within LMIC radiation oncology settings.
Methods
A systematic search of publications from January 2000 to 2025 was conducted following PRISMA guidelines to identify studies reporting AI applications in LMIC radiotherapy. Of 1,196 screened records, 19 studies met inclusion criteria and underwent full-text analysis. Two reviewers independently extracted the data using Rayyan, and a third reviewer resolved any disagreements. Risk of bias was evaluated using the ROBINS‑I framework.
Results
Most studies originated from Latin America (25), Africa (12), and Asia (5). Web‑ or cloud-based platforms were the most common deployment strategy (12), while seven studies did not specify a model. Ten studies used local LMIC datasets for validation. AI tools were mainly applied to treatment planning (13) and contouring (6), with cervical (10) and head and neck cancers (10) most frequently studied, followed by brain (6), breast (4), prostate (3), and lung (1). Enablers included institutional support (14), standardized protocols (7), guideline-based templates (5), and web/cloud access (2). Major challenges involved data privacy and governance (15), model generalizability (7), limited computational/GPU capacity (3), and connectivity constraints (2).
Conclusion
AI applications in LMIC radiotherapy show promise for improving workflow efficiency and reducing workforce burden. However, inconsistent reporting of validation methods, deployment infrastructure, and quality assurance integration limits interpretability and may compromise clinical safety. Establishing implementation standards—including commissioning, end‑to‑end testing, data governance, and ongoing performance monitoring—is essential for equitable and safe adoption.