Humanity in the loop: Reframing artificial intelligence and the global medical imaging divide.

OBJECTIVES: Access to radiology and nuclear medicine remains profoundly unequal across global health systems. Low- and middle-income countries (LMICs) face persistent shortages in imaging equipment, specialist workforce, digital infrastructure, maintenance, regulation and sustainable financing. Artificial intelligence (AI) is increasingly positioned as a mechanism to improve imaging access, but it may also reinforce existing inequities. This narrative review examines the potential for discriminative, generative and agentic AI to reduce or widen inequities in LMIC medical imaging and proposes a sociotechnical governance framework based on the concept of Humanity-in-the-Loop. KEY FINDINGS: Discriminative AI may improve image interpretation, triage, segmentation, quantification and image-quality assessment; generative AI may enhance reporting, education, translation, communication, synthetic data generation and research workflows; and agentic AI may coordinate referrals, scheduling, follow-up, supply logistics and imaging pathways. They may also exacerbate inequity through data colonialism, algorithmic under-performance, dataset shift, infrastructure dependence, commercial lock-in, second-tier automation, liability transfer and unequal participation in AI research and development. CONCLUSION: Human-in-the-loop governance is insufficient for equitable AI implementation in LMIC medical imaging because the assumptions underpinning effective human oversight are often absent in resource-constrained settings. The proposed Humanity-in-the-Loop sociotechnical framework embeds equity, local agency, contextual validity and public benefit throughout the AI lifecycle. The framework comprises six interconnected domains: patient humanity, workforce humanity, community humanity, data humanity, system humanity and global humanity. IMPLICATIONS FOR PRACTICE: Responsible implementation of AI in medical imaging requires governance frameworks that extend beyond algorithm performance to include workforce development, data sovereignty, infrastructure, community priorities, equitable partnerships and sustainable capacity building. Adoption of a Humanity-in-the-Loop approach may help ensure that AI narrows rather than widens the medical imaging divide between high-income countries and LMICs.