Transfer Learning and Governance Help Bridge Healthcare AI Divide
Researchers in Singapore have demonstrated that advanced artificial intelligence (AI) techniques can significantly enhance clinical diagnostics in resource-limited countries without the necessity for extensive local datasets.
Transfer Learning in Action
A team from Duke-NUS Medical School successfully applied transfer learning, a method where a model developed for one task is repurposed as a starting point for another, to predict patient outcomes after cardiac arrest. This study, published in npj Digital Medicine, addresses a critical hurdle in AI adoption within low- and middle-income countries: the scarcity of high-quality data essential for training algorithmic models from scratch.
To assess the effectiveness of transfer learning, researchers adapted a brain-recovery prediction model originally created in Japan, which used data from 46,918 out-of-hospital cardiac arrest patients. This model was then tested in Vietnam on a smaller group of 243 patients.
The results were striking. The Japanese model, when directly applied to the Vietnamese context, distinguished high-risk from low-risk patients with only 46% accuracy. In contrast, the adapted transfer learning model achieved an impressive accuracy rate of around 80%.
“The study shows AI models do not need to be rebuilt from scratch for every new setting,” stated Liu Nan, an associate professor at Duke-NUS’s Centre for Biomedical Data Science. “By adapting existing tools safely and effectively, transfer learning can lower costs, reduce development time, and help extend the benefits of AI to healthcare systems with fewer resources.”
Global AI Adoption Challenges
Despite the growing potential of AI in healthcare, the adoption of this technology remains uneven across the globe. A separate study published in Nature Health revealed that while 63% of surveyed healthcare providers utilize AI tools, adoption is more pronounced in high- and upper-middle-income countries.
The research also highlighted the potential of large language models (LLMs) to enhance access to care, diagnostics, and clinical decision-making in low- and middle-income nations facing barriers such as limited infrastructure and expertise. For instance, in Sierra Leone, community healthcare workers employ smartphone apps to detect malaria infections from blood smear samples, presenting a more cost-efficient alternative to traditional microscope-based methods. Similarly, in South Africa, chatbots provide pregnant mothers with essential prenatal advice.
“LLMs have the greatest opportunity to transform healthcare in settings where specialist physicians are scarce. However, the global health community must work collaboratively and urgently to ensure the implementation of LLMs is supported in regions facing the most significant adoption challenges,” commented Siegfried Wagner from the UCL Institute of Ophthalmology and Moorfields Eye Hospital NHS Foundation Trust.
Ning Yilin, a senior research fellow at the Centre for Biomedical Data Science at Duke-NUS, emphasized that empowering individuals should be the priority when integrating LLMs into healthcare. “Strengthening digital literacy and building confidence in using these tools will ensure AI supports, rather than disrupts, the workforce. Tailored skills-development pathways can help under-resourced workers adapt and thrive, allowing AI to enhance and add value to clinical and administrative roles,” she stated.
Call for International Governance
While AI tools hold the promise of improving healthcare delivery, governance frameworks are essential for the safe and ethical implementation of this technology. Currently, regulations governing medical technologies often overlook AI-specific risks, including privacy issues, model hallucinations, and the need for oversight of new tools.
To address these challenges, researchers at Duke-NUS have proposed establishing an international consortium known as the Partnership for Oversight, Leadership, and Accountability in Regulating Intelligent Systems-Generative Models in Medicine (Polaris-GM). This consortium aims to provide guidance for regulating new tools, monitoring their impact, and establishing safety protocols tailored for resource-limited settings.
Polaris-GM aspires to unite healthcare leaders, regulators, ethicists, and patient groups worldwide to review existing research and work toward a global consensus on AI governance in healthcare.
Jasmine Ong from Duke-NUS’s AI and Medical Sciences Initiative remarked, “With clear oversight and explicitly defined guidelines, healthcare systems can confidently leverage AI’s strengths to improve health outcomes while avoiding potential pitfalls. All stakeholders, from policymakers to patient groups, have a crucial role to play in achieving this goal.”
Future of AI in Healthcare
Experts from the pharmaceutical industry at Gitex Asia 2025 discussed the transformative impact of AI on drug discovery, including accelerating clinical trials, diagnosing rare diseases, and validating traditional medicines. Medow Health AI has introduced an AI-powered scribe platform in Singapore to assist healthcare providers in streamlining clinical documentation, alleviating administrative burdens, and enhancing care quality. Furthermore, researchers in Singapore are utilizing Google DeepMind’s AlphaFold to explore intricate protein interactions involved in Parkinson’s disease, paving the way for potential diagnostics and treatments.
