**Background:** Artificial intelligence (AI) encompasses computer science techniques that simulate human intellectual behavior, including machine learning (ML), deep learning (DL), natural language processing (NLP), support vector machines (SVM), and artificial neural networks (ANN). AI has potential advantages such as reliability, cost-effectiveness, and ability to solve complex issues, and is increasingly applied in medical care for diagnosis, therapy, and prognosis. However, its use may be limited in low- and middle-income countries (LMICs) due to infrastructure and expertise gaps. This scoping review aimed to map and synthesize available evidence on the use of AI to deliver medical care services globally and regionally.
**Methods:** The review followed the Arksey and O'Malley methodology for scoping reviews. A search was conducted in PubMed (07 March 2022) and Scopus (16 March 2022) using keywords: ("Artificial intelligence" OR AI OR "machine learning" OR "machine intelligence" OR "deep learning") AND ("health care" OR health OR "health delivery"), with no language or date restrictions. Titles and abstracts were screened independently by two researchers, then re-screened by a second pair. Full texts of potentially eligible studies were assessed, and primary studies were included if they reported on AI use, addressed a health condition, assessed effectiveness, and were published in English in a peer-reviewed journal. Reviews were excluded. Data extraction was performed independently by multiple authors using a piloted form, capturing first author, year, population, country, aim, health issue, AI method, application, and findings. WHO region and health issue category were charted. A narrative synthesis was conducted without methodological quality assessment.
**Key Results:** The search yielded 172,375 articles (11,695 from PubMed, 160,680 from Scopus). After screening, 801 abstracts were considered potentially eligible, and a random sample of 100 full texts was assessed, of which 91 met inclusion criteria. Studies were distributed across WHO regions: Region of the Americas (AMR) had 28 studies (30.8%), European Region (EUR) had 28 studies (30.8%), Eastern Mediterranean Region (EMR) had 14 studies (15.4%), Western Pacific Region (WPR) had 11 studies (12.1%), South-East Asian Region (SEAR) had 4 studies (4.3%), and African Region (AFR) had 1 study (1.1%). Two studies involved multiple countries, two were global, and one was from Taiwan. AI methods identified included: ML (46 studies), DL (15 studies), ANN (2 studies), CNN (1 study), AIOM-based technology (1 study), Bayesian network (1 study), DNN (2 studies), Fuzzy K-means clustering algorithm (1 study), and COVID Inception-ResNet model (1 study). Combined approaches included DL+ML (3 studies), ML+ANN (1 study), DL+NLP (1 study), and others. Health conditions addressed included infectious diseases (21 studies, mostly COVID-19), cardiovascular diseases (13 studies), cancers (6 studies), mental health (8 studies), and assorted conditions (21 studies). AI applications focused on detection, diagnosis, classification, treatment, prognosis, and management.
**Clinical Implications:** The review demonstrates that AI, particularly ML and DL, is widely used in high-income countries for various health conditions, showing positive outcomes in detection, diagnosis, and management. However, the scarcity of studies from LMICs, especially Africa, indicates a critical gap. For AI to benefit global health, infrastructure, digital literacy, and training for healthcare workers in LMICs are essential. The findings support the need for systematic reviews and further research in under-represented regions to ensure equitable access to AI-driven healthcare innovations.