A Tale of Two Countries: Challenges in the Adoption of Artificial Intelligence in Medical Imaging in India and Singapore Authors: Charlene Jin Yee Liew, Swati Goyal Published: 2026 ABSTRACT This commentary discusses the perspectives of two authors working in India and Singapore in medical imaging and highlights differences in their basic structural and regulatory frameworks. Introduction Artificial intelligence (AI) has been established as a trailblazing technology in health care and medical imaging, with the ability to support diagnostic accuracy at high speed, handle high workloads, optimize workflows, and scale imaging expertise even in underserved settings. Both developed and developing countries have been exploring possibilities to develop algorithms and align existing systems with AI for the overall benefit of public health. Still, adoption of AI systems remains nonuniform due to the processing challenges of initiating research in India and implementing clinical workflow—AI integration in Singapore. A comparison between Singapore and India features two different but complementary prototypes: one with mature digital integration and governance, and the other with scale, diversity, and evolving infrastructure. This commentary discusses the perspectives of two authors working in India and Singapore in the field of medical imaging. The Indian Context India, with its population of 1.4 billion and the potential for a vast, heterogeneous data set optimal for training AI algorithms, has, in recent years, launched national initiatives such as the Ayushman Bharat Digital Mission (ABDM), which helps create longitudinal records using Ayushman Bharat Health Accounts (ABHA) IDs. Another example would be the Ayushman Bharat Pradhan Mantri Jan Arogya Yojana (PM-JAY), which generates large volumes of population-level claims data. 1 The Indian Council of Medical Research biobanks and disease registries have further paved the way for AI research. Furthermore, India is building its capacity by creating frameworks for health researchers working in the Indian Institutes of Technology (IIT), the All India Institute of Medical Sciences (AIIMS), National Institutes of Technology, and state-level medical colleges. The government has pursued this goal by conducting ethical committee trainings for AI research review and by funding various projects. Despite these initiatives, AI adoption in India has faced significant barriers due to variability in the basic structural framework and ethical issues. While tertiary care centers and corporate hospitals have more advanced systems with uninterrupted data transfer, smaller hospitals and district-level facilities—which serve the majority of patients—operate with minimal digital infrastructure. Data-related challenges further hamper the effective deployment of AI. Although India generates large volumes of imaging data, these data sets are usually fragmented, uncurated, inconsistently annotated, and stored across disparate systems. This incongruity results in nonstandardized imaging protocols and unstructured reporting practices, complicates the curation of the databases needed for training, and limits interoperability across the health care system. Without access to balanced, comprehensively sampled, diverse data sets, AI models suffer from limited generalizability across different clinical settings. As for the ethical side, regulatory issues regarding patient privacy, data ownership, and institutional approvals lead to restrictions on data sharing and multi-institutional collaborations. 2 Finances remain another major hindrance to AI adoption in India, with public health care institutions already operating with constrained budgets. The software costs, system integration, and maintenance required for AI systems must demonstrate clinical value and wider adoption in order to justify investment. Another challenge faced by India is its high rural population. Over 60% of the population resides in rural areas, creating a huge digital divide. Poor Internet connectivity, variable image quality, and extrapulmonary disease are major barriers to AI-assisted chest X-ray deployment for tuberculosis (TB) detection in mobile vans. Smaller hospitals without Picture Archiving and Communication System/Radiology Information System integration face difficulty in deploying cloud-based AI tools. Along with this, the nation's multilingual population—with 22 officially recognized languages—slows the development of natural language processing (NLP) and speech-to-text (STT) models. These obstacles are further worsened by digital illiteracy among health care workers, a shortage of clinicians (the ratio of radiologists to patients is approximately 1:100,000), and high patient volumes, all of which affect AI adoption. 3 India's epidemiological profile, with a high burden of tropical and infectious conditions such as dengue, malaria, neurocysticercosis, hydatid cyst, filariasis, leishmaniasis, and chikungunya, mandates that AI algorithms be trained and validated on local data sets. For example, India's National TB Elimination Program (NTEP) sought to create a local algorithm using AI-powered chest X-ray analysis by using mobile vans for mass screening in remote areas. However, the country's high disease burden, atypical and extrapulmonary presentations, comorbidities, and follow-up requirements pose barriers to building these data sets. In order to utilize the potential of similar public health initiatives, rigorous validation must be emphasized. Despite these challenges, India is also an important center for global teleradiology services, with the advantage of manpower and the capability to connect radiologists across different geographic regions and time zones. Integrating AI-based tools for triaging, automatic measurements, and workflow prioritization may help manage increasing imaging volumes and reduce reporting delays. However, successful reporting will require technical interoperability and clinical acceptance. 4 The Singapore Context The adoption of AI in medical imaging is inherently context-dependent, shaped by the structural, economic, and regulatory characteristics of individual health care systems. In contrast, Singapore exemplifies a health care system with advanced digital integration and strong governance structures. National initiatives such as the National Electronic Health Record (NEHR) system facilitate interoperability and longitudinal data capture, providing a robust foundation for AI deployment. 5 6 In this context, the primary challenge is not infrastructural readiness but the integration of AI into established clinical workflows. AI tools must demonstrate clear clinical utility, seamless compatibility with existing systems, and measurable improvements in efficiency or diagnostic performance to justify adoption. Additionally, Singapore's relatively small population limits the availability of large-scale data sets, necessitating external collaborations or rigorous local validation of AI models developed elsewhere. 7 In Singapore, there have been several radiology-specific AI use cases that have been deployed and are in various stages of clinical use, from the piloting stage to mainstream daily use; these range from image acquisition optimization, detection (including triage), for example, chest radiography and mammography, quantification (measurements, volumes), reporting, and workflow prioritization. 8 9 Some of these AI use case applications are included with radiology modality vendor software, for example, Siemen's syngo.via for the quantification of lung nodules. In India, AI has been deployed mainly in the teleradiology networks, a few tertiary centers, and corporate hospitals. Current use cases comprise large-scale AI-assisted TB screening on chest radiographs using mobile vans under the NTEP, 10 intracranial hemorrhage and stroke triage in emergency settings, quantification, structured reporting, STT systems, and worklist prioritization in high-volume teleradiology settings. Regulatory considerations further distinguish between the two settings. Singapore benefits from well-defined regulatory frameworks governing AI as a medical device, with strong emphasis on patient safety, transparency, and postdeployment monitoring. 11 For example, there are various regulatory frameworks applied at all levels of governance, from hospital-level AI policies to national-level health care guidelines, for example, Singapore's Ministry of Health Artificial Intelligence in Healthcare Guidelines (AIHGLe) 12 to cross-sector governance frameworks by Singapore's Infocomm Media Development Authority (IMDA) Model AI Governance Framework. 13 Conversely, India's regulatory landscape is still evolving, with emerging policies such as the Digital Personal Data Protection Act of 2023 for managing patient data and the Central Drugs Standard Control Organization (CDSCO) recognizing software as a medical device. The heterogeneity and lack of comprehensive guidelines for data privacy safeguards, anonymization standards, secondary use consent, interinstitutional data sharing, AI validation, deployment, and liability continues to pose challenges for operational implementation by developers and health care providers in India. 14 In both countries, regulatory frameworks surrounding data privacy and consent have been implemented. For example, in Singapore, the Personal Data Protection Commission has issued advisory guidelines on use of personal data in AI recommendation and decision systems, which include consent and notification obligations, as well as data protection, which applies to the collection and use of personal data in AI systems. 15 With regards to autonomous AI tools for clinical indications, fully autonomous AI tools that make clinical decisions without human oversight are not permitted within clinical real-world use by Singapore's Ministry of Health (MOH) guidelines, 12 as Singapore's MOH adopts a “human-in-the-loop” approach, stating that AI should augment and empower health care professionals rather than replace their judgment. In India, existing applications are mainly assistive. Medicolegal accountability and liability allocation between software vendors, health care institutes, and reporting radiologists remains undefined, especially in cases with missed or delayed diagnosis due to AI-assisted interpretation. Radiologist-supervised interpretation will be predominant instead of fully autonomous AI deployment due to variability in infrastructure and medicolegal ambiguity. Shared Challenges Two persistent conundrums exist universally. First is the opaque decision-making process of AI algorithms, which makes clinicians reluctant to trust outputs without logical and transparent reasoning. Developing interpretable AI models using tools such as heatmaps and attention mechanisms will be critical for gaining approvals in compliance with the responsible use of AI in medical imaging. The lack of prospective multicenter clinical validation studies compared to retrospective algorithm development is a major challenge. The second conundrum is medicolegal accountability when conducting AI-assisted diagnosis, which remains a topic of debate. 16 Limited generalizability of model performance across institutions and populations, and the gap between algorithmic performance in research settings and real-world clinical performance also presents a challenge which is experienced equally in health care settings around the world. A recent example of this was presented in a journal article by Suleman et al, which noted that the average area under the receiver operating characteristic curve declined in external validation studies compared to internal validation studies and specificity performance was noted to decline by up to 24 percentage points. 17 AI models trained on data sets from high-income countries may not generalize and perform well in Indian settings with high prevalence of infectious diseases, presentation in advanced stages, and variable imaging acquisition protocols. Bias due to the underrepresentation of certain demographic, socioeconomic, and disease populations in training data sets complicates the health care inequities. Methods which improve generalizability include multicenter training and data augmentation. Future Directions Education and training are essential for responsible AI adoption in both countries. All clinicians, including radiologists, must develop a basic understanding of AI concepts, algorithm development, training, validation, deployment, and its limitations to effectively use AI in medical imaging and clinical decision-making. In India, academic institutions such as AIIMS and IIT, and professional organizations like the Indian Radiology and Imaging Association have initiated AI education through online and onsite courses and collaborative research projects. The Ministry of Health and Family Welfare recently launched a free online AI course in medical education for all health care professionals. Future progression will also rely on federated learning, multicenter data sharing collaborations, interoperable infrastructures, and development of locally validated indigenous AI models. A comparative analysis of India and Singapore highlights two contrasting paradigms of AI implementation: a resource-variable, high-demand system versus a digitally mature, highly regulated system ( Table 1 ). Table 1Comparative analysis of artificial intelligence adoption in medical imaging: India versus SingaporeDomainIndiaSingaporeDigital infrastructureUneven adoption of electronic health records; limited interoperability across institutionsAdvanced nationwide digital infrastructure with integrated electronic medical recordsData availability and qualityLarge-volume data sets, but fragmented, unstructured, and inconsistently annotatedSmaller data sets but highly curated, standardized, and of high qualityInteroperabilityLimited integration across PACS, HIS, and RIS systemsHigh interoperability supported by national platforms (e.g., NEHR)Regulatory environmentEvolving frameworks; limited clarity on validation, approval, and liabilityMature regulatory frameworks with a strong emphasis on safety, validation, and lifecycle monitoringWorkforce capacityShortage of radiologists, high workload, variable digital literacyAdequate, highly trained workforce with strong academic and clinical integrationFinancial considerationsHigh cost-sensitivity; constrained public sector funding; uncertain return on investmentStrong institutional and governmental funding; emphasis on cost-effectiveness and measurable ROIDisease profileHigh burden of infectious and tropical diseases requiring locally trained AI modelsPredominantly chronic and ageing-related diseases; facilitates adaptation of global AI modelsImplementation challengesInfrastructure gaps, data fragmentation, lack of standardization, workflow disruptionIntegration into established workflows, need for demonstrable incremental value, and clinician acceptanceValue proposition for artificial intelligenceExpanding access, reducing reporting backlog, and improving healthcare equityImproving efficiency, reducing turnaround time, enhancing diagnostic confidence, and outcomesAbbreviations: AI, artificial intelligence; HIS, Hospital Information System; NEHR, National Electronic Health Record; PACS, Picture Archiving and Communication System; RIS, Radiology Information System; ROI, return on investment. Comparative analysis of artificial intelligence adoption in medical imaging: India versus Singapore Abbreviations: AI, artificial intelligence; HIS, Hospital Information System; NEHR, National Electronic Health Record; PACS, Picture Archiving and Communication System; RIS, Radiology Information System; ROI, return on investment. Two Countries, Two Different National Priorities The divergence between the two systems is undergirded by the differential value propositions of AI in the two countries. India positions AI primarily as a tool to improve access, reducing reporting backlogs, scalability, and equity in health care delivery. Applications such as AI-assisted TB screening and triage systems are particularly relevant in addressing population-level health needs. In Singapore, AI is evaluated through the lens of optimization, or the enhancement of workflow efficiency—reducing reporting turnaround times and improving diagnostic precision within an already high-performing health care system. This distinction underscores the necessity of aligning AI implementation strategies with local health care priorities and system capabilities. From an impact-based perspective, the success of AI in medical imaging should extend beyond algorithmic performance metrics to encompass tangible clinical and operational outcomes. 18 19 Conclusion Ultimately, bridging these divergent paradigms requires a multifaceted context-aware approach. Strengthening digital infrastructure, promoting data standardization, and enabling secure data-sharing frameworks are critical in resource-variable settings. In parallel, high-resource systems must focus on rigorous validation and outcome-based evaluation of AI tools. Cross-country collaboration with bilateral learning presents a unique opportunity to leverage complementary strengths, combining India's large-scale, diverse data sets with Singapore's expertise in governance, validation, and clinical integration, to develop AI solutions that are both robust and globally applicable.