The Role of Artificial Intelligence in Enhancing Maternal and Child Health Through Digital Health Initiatives in Resource-Limited Settings: A Narrative Review

Muhammad Umar¹, Laiba Shamim² , Abdur Rehman³, Habiba Zafar², Muhammad Talha⁴, Okasha Tahir⁵, Mahtab Zafar⁶, Umama Alam⁵, Fatima Shams⁷, Zainab Syyeda Rahmat⁸ and Mirza Mohammad Ali Baig⁹
1. Khairpur Medical College, Khairpur Mirs, Pakistan
2. Jinnah Sindh Medical University, Karachi, Pakistan
3. Saidu Medical College Swat, Saidu Sharif, Pakistan
4. King Edward Medical University, Lahore, Pakistan
5. Khyber Medical College, Peshawar, Pakistan
6. Rahbar Medical and Dental College, Lahore, Pakistan
7. Nishtar Medical College, Multan, Pakistan
8. Dow Medical College, Karachi, Pakistan
9. Islamic International Medical College, Riphah International University, Rawalpindi, Pakistan
Correspondence to: Laiba Shamim, laibashamim015@gmail.com

DOI: https://doi.org/10.70389/PJAI.100021

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Muhammad Umar – Conceptualization, Data curation, Formal Analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Laiba Shamim – Conceptualization, Data curation, Formal Analysis, Visualization, Writing – original draft, Writing – review & editing. Abdur Rehman – Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Habiba Zafar – Visualizations, Writing – original draft, Writing – review & editing. Muhammad Talha – Writing – original draft, Writing – review & editing. Okasha Tahir – Validation, Visualization, Writing – original draft, Writing – review & editing. Mahtab Zafar – Validation, Visualization, Writing – original draft, Writing – review & editing. Umama Alam – Visualization, Writing – original draft, Writing – review & editing. Fatima Shams – Visualizations, Writing – original draft, Writing – review & editing. Zainab Syyeda Rahmat – Writing – original draft, Writing – review & editing. Mirza Mohammad Ali Baig – Writing – review & editing. All Authors reviewed and approved of the final version of the manuscript.
• Guarantor: Laiba Shamim
• Provenance and peer-review: Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: AI-powered maternal risk prediction, mhealth interventions in low-resource settings, Wearable prenatal monitoring sensors, Telemedi-cine-based neonatal care, Ethical challenges of AI deployment.

Peer Review
Received: 14 August 2025
Last revised: 27 November 2025
Accepted: 25 December 2025
Version accepted: 4
Published: 20 February 2026

Plain Language Summary Infographic

Introduction

Maternal and child healthcare has tremendous burdens in resource-limited parts of the world. From a case report, an estimated 800 women die daily due to pregnancy complications, mostly occurring in third-world countries due to less readily available health centers.¹ In 2020, the maternal death ratio was 13 per 100,000 in developed countries versus 430 per 100,000 in developing areas.² The World Health Organization (WHO) has reported that 95% of maternal deaths in developing countries could have been prevented if they had received proper care.² The same applies to the incidence of neonatal mortality: in countries such as Australia, Canada, and the United States, the incidence is less than 10 per 1,000 live births, but in countries such as Niger, Chad, and other developing countries, it is greater than 10 per 1,000 live births.³

UNICEF also highlights the grim reality that a child born in an underserved area with low resources is 18 times more likely to die before the age of five than a child born in a highly developed area with adequate healthcare.³ Contributing to such statistics are lower access to quality healthcare, inadequately trained staff, and late diagnosis.⁴ The challenge of an aging global population is among the central problems to be managed through early screening and tracking to relieve the burden in poor-resource settings, and artificial intelligence (AI) has been reported to deliver promising results with enhanced diagnosis and prognosis prediction.⁵ Figure 1 illustrates the issues related to maternal health and interventions. Bridging the urban-rural gap is crucial for guaranteeing an equal and enhanced healthcare system.

Digital health initiatives, such as telemedicine, telehealth, and mobile health (mHealth), could be advantageous through increased access to medical care in remote or underserved areas and improved management of medical information.⁶ Digital health and AI technologies are transforming healthcare delivery in low-resource environments. For example, technologies such as OpenMRS, an electronic medical record (EMR) platform, are closing essential gaps in care and improving health services in developing countries.⁷ In addition to EMR, wearable healthcare technologies rank among the most promising applications of AI in child and maternal health.

These technologies can track maternal health indicators, such as glucose, blood pressure, and pulse, in real time, allowing for early diagnosis.⁸ Vital monitoring via AI-facilitated wearables has the potential to detect high-risk pregnancy or neonatal illness at an early stage, allowing effective and timely intervention to save lives. Thus, the use of AI in maternal and neonatal health, such as predictive models for maternal complications and diagnostics, has the potential to improve the standards of care and avoid unnecessary deaths. By incorporating AI into digital health programs, we can set the pace for accessible healthcare so that every mother and child can have a healthy life, regardless of where they are.

Methodology

The present narrative review discusses the literature proving the capability of digital health interventions in low-resource countries to use AI for the betterment of maternity and child health outcomes. Despite its nature as a narrative review, a systematic and open search strategy was utilized to guarantee complete detection of pertinent evidence and to improve reproducibility at the same time. A comprehensive literature search was systematically performed in the databases of PubMed, Web of Science, Scopus, and Google Scholar, and in addition, conducted for grey literature targeted from the recognized sources such as WHO, UNICEF, USAID, Save the Children, Jhpiego, and other well-known international health organizations and NGOs. The Boolean operators combined the keywords like “artificial intelligence,” “machine learning,” “deep learning,” “digital health,” “mHealth,” “telemedicine,” “maternal health,” “newborn health,” “child health,” “antenatal care,” “postnatal care,” “immunization,” “nutrition,” and the words describing resource-limited settings (“low- and middle-income countries” OR LMIC OR “developing countries” OR “resource-constrained” OR “low-resource setting”). The time frame for the search was limited to publications dated January 2010 to November 2024 in order to highlight the modern AI use only.

The inclusion criteria consisted of peer-reviewed articles, conference papers, preprints, and grey literature reports that either described or assessed AI or AI-augmented digital health interventions aimed at improving maternal, newborn, or child health outcomes or processes of care in low- and middle-income countries (LMICs) or other resource-poor settings, published in English (or with English full text/abstract available) during the period from 2010 to 2024. Studies undertaken solely in high-income countries with no clear connection to resource-limited areas, the use of digital tools without AI or machine learning (ML) features, research dealing only with non-maternal/child health outcomes, and editorials or opinion pieces lacking primary data or comprehensive intervention descriptions were the main reasons for exclusion.

The selection of studies is demonstrated in a PRISMA flow diagram (Figure 2). Initially, 1,234 records were recognized following the removal of duplicates, and the 892 titles and abstracts were subjected to a two-reviewer independent screening (MU, LS). If the reviewers had differing opinions, they would resolve the issue by discussing it with a third reviewer (HZ). After that, 214 full-text articles were checked for inclusion criteria, which led to the inclusion of 80 studies, including peer-reviewed publications and grey literature reports/preprints as well. Due to the review’s narrative nature and study designs’ heterogeneity (randomized trials, observational studies, pilot projects, and implementation reports), no formal quality assessment tool was applied. Rather, a structured rubric covering five domains was used to evaluate each included study: (1) clarity and reproducibility of the AI methodology, (2) appropriateness of the evaluation design for resource-limited contexts, (3) sample size and representativeness, (4) validity and clinical relevance of outcome measures, and (5) reporting of ethical considerations and potential biases. Studies were classified into high, moderate, or low methodological rigor categories. The included studies’ quality appraisal summary table is in Table S1.

The credibility of grey literature and preprints was judged with an adapted version of the AACODS checklist (Authority, Accuracy, Coverage, Objectivity, Date, Significance). Reports from well-known worldwide organizations or implementing partners with evident authorship, open funding, and detailed methodology were kept. Preprints were considered only if they presented novel real-world insights that were not yet accessible in peer-reviewed literature. Data were narratively synthesized and organized thematically around AI applications, ethical issues, implementation difficulties, and gaps in evidence.

To provide a transparent view of the review process, several limitations and possible biases are acknowledged: the limitation of the review to only English publications might have resulted in language bias; a large number of the studies that were included were small-scale pilots, one-center initiatives, or based on surrogate outcomes (for example, attendance or satisfaction rates) rather than hard clinical endpoints such as mortality or severe morbidity. The use of grey literature and preprints added practical implementation evidence to the review, but these are generally not peer-reviewed and may be more prone to reporting bias. Therefore, the overall certainty of evidence is still considered low to moderate. Additionally, gray literature is not peer-reviewed for the most part, making it more susceptible to publication bias. Hence, while the initial findings look good, the certainty of evidence is low to moderate.⁹

Digital Health and AI Use in Maternal and Child Health

Maternal and child health in low-resource settings (LRS) faces persistent challenges, including limited access to services, long distances to facilities, poverty, and social stigma, which often deters women from seeking care. Fragmented care increases risks for both mothers and infants.^10–13

Digital health technologies have emerged as effective bridges across public and private healthcare divides. Over 600 mHealth projects in developing countries demonstrate efforts to reach vulnerable populations.¹⁴ Mobile health (mHealth) apps reorient pregnant and postpartum women’s engagement with care by providing timely reminders, education, and support.¹⁵ The Adaptive Multi-Care (AM-Care) system integrates web- and mobile-based maternal and child health services, ensuring essential information accompanies the patient.¹⁶ Similarly, e-health systems enable computerized record storage and transfer, allowing continuous and coordinated care.¹⁷ These interventions have improved prenatal care attendance, expanded immunization coverage, and empowered women with actionable information.¹⁸ Table 1 summarizes digital health interventions, including impacts, challenges, and regions of implementation.

Table 1: Overview of digital health interventions in maternal and child healthcare in resource-limited settings.
Intervention	Description	Impact/Outcomes	Challenges	Regions/Settings
Mobile Health (m-Health)	Use of mobile applications for maternal and child health education, reminders, and remote consultations.	Increased prenatal attendance, improved child immunization rates, and enhanced patient engagement.	Limited internet connectivity, infrastructure issues, and cultural stigma.	Widely implemented in LMICs.
AM-Care Application	Comprehensive healthcare services via net and mobile technologies, integrating maternal and child health management.	Facilitates holistic healthcare plans and better patient care.	It requires continuous internet access and technical training for users.	Developing countries with some digital capacity.
Remote Consultations	Telehealth services give access to healthcare professionals.	Improved access to prenatal and postnatal care, reduced travelling burden.	Internet connectivity issues and the need for trained healthcare providers.	Rural and underserved areas.
e-Health Systems	Digital storage and transfer of patient information to ensure continuity of care.	Enhanced tracking of patient history and improved care continuity.	Data exchange limitations and non-integrated health information systems.	Resource-limited settings need structured

Despite these advancements, infrastructure limitations, unreliable Internet connectivity, disjointed health information systems, and insufficient technical training constrain their full potential.^19,20 Strengthening digital infrastructure, workforce training, and public-private partnerships (PPPs) can enhance equitable access. The Rwandan Babyl platform effectively connects rural communities using AI-powered chatbots and telemedicine, whereas similar efforts in Nigeria faced connectivity and linguistic challenges (Figure 3).

Building on these early digital successes, AI represents the next major opportunity to transform maternal and child health. AI systems have the potential to process large datasets, recognize subtle patterns of clinical importance, and facilitate timely decision-making.²¹ These systems are particularly relevant for use in maternal healthcare when high-risk pregnancies need to be managed, for which early intervention can be lifesaving. Nevertheless, research in this area remains limited.²²

ML, a component of AI, facilitates enhanced data-based prediction and clinical surveillance.⁹ Its applications are as broad as those in diagnostic imaging, predictive analytics, and round-the-clock surveillance. Obstetric ultrasonography, for example, benefits from the application of AI, which enhances the accuracy and efficiency of fetal evaluation through the automated segmentation of biometric information.²³ Wearable sensors allow for continuous monitoring of maternal vital signs, such as heart rate, blood pressure, and glucose levels, and can alert clinicians to impending complications before emergencies arise.^24,25 Such interventional systems can prompt alerts and facilitate interventions and improve perinatal outcomes. Predictive models are increasingly being found to be useful for distinguishing high-risk from low-risk pregnancies.

AI algorithms can anticipate preterm birth, gestational diabetes, and hypertensive disorders by analyzing ultrasound data, maternal blood profiles, and metabolic markers.^26–28 Encouragingly, surveys show a growing willingness among women to embrace AI-assisted care during pregnancy.²⁹ Collectively, these developments highlight capacity of AI to strengthen diagnostics and deliver individualized care at scale.^30–32 For example, preliminary reports and ongoing research suggest that AI interventions identifying women at risk of postpartum hemorrhage and preeclampsia enable more prompt and beneficial interventions.³³ Similarly, it has been proposed that AI could predict sepsis in neonates and detect the most important variables that may lead to sepsis, such as chorioamnionitis and respiratory rate.³⁴

To predict neonatal jaundice, one study recommended the use of a smartphone application to determine the bilirubin level of a neonate, and the app results had a sensitivity of 86%.³⁴ Another study confirmed that AI can represent the relationship between socioeconomic status and childhood malnutrition and can predict which children are prone to malnutrition.³⁴ As Internet of Things (IoT) technology has been applied widely, several studies have been conducted to elaborate on its use in improving infant and maternal health. Li et al. conducted a similar study with the aid of IoT technology to provide real-time monitoring of pregnant women and fetuses in the womb, with a statistically significant (P < 0.05) difference between the experimental and control groups in parameters such as reduced medical staff workload and enhanced productivity.³⁵ Taylor et al. developed a smartphone application called BiliCam to support the estimation of bilirubin levels in the neonate. They found a high correlation of 0.91 between the calculated value of bilirubin (with BiliCam) and the neonate’s total serum bilirubin value.³⁵ Other studies with similar findings are listed in Table 2.

Table 2: Case examples of ai applications in antenatal and postnatal care, with associated outcomes and benefits.
Reference	Summary	Outcomes and Benefits
Corrêa et al.36	Formation of Smart Chatbot ‘Lhia’ to educate mothers about breastfeeding	93% accuracy for 1,851 interactions, improved conversational flow
Li et al.35	Wearable devices for monitoring and modes of administration	38.1% of the respondents were inclined to wear the device
Taylor et al.37	Checking bilirubin levels in neonates using a smartphone-based app	The overall correlation of estimated bilirubin levels was 0.91. Sensitivities = 84.6% and 100%
Masino et al.38	Infant sepsis should be recognized at least 4 hours before clinically recognizing it. Eight models considered	Of the eight models, six were well-performing, and no statistically significant pairwise difference was noted in AUC on the CPOnly and CP+Clinical dataset

Critical Appraisal of Evidence and Limitations

Although there are case illustrations of high-promise utility and accuracy, they are primarily small pilot studies with methodological limitations. To concisely summarize these constraints, the key limitations of the primary cited AI applications are listed in Table 3. For example, the BiliCam study involved fewer than 100 neonates,³⁷ and the Lhia chatbot was based on literate, self-selected patients.³⁶ IoT maternal monitoring studies have been restricted to urban settings,³⁵ and neonatal sepsis ML models have been retrospective with overfitting risk.³⁸ Overall, the quality of evidence was assessed as low to moderate. Consequently, the generalizability of these findings to broader rural settings in LMICs remains unclear.

Table 3: Mini risk-of-bias summary table.
Study/Tool	Sample Size	Study Design	Risk of Bias	Generalizability
BiliCam37	100 neonates	Pilot, single center	High (small, non-random sample)	Limited to settings with smartphone access
Lhia chatbot36	1,851 interactions	Simulation/user test	Moderate (self-selected participants, literacy bias)	Medium; assumes literacy and phone access
IoT monitoring35	200 women	Observational study	Moderate (non-random, infrastructure-dependent)	Limited to urban areas with stable networks
ML models38	NICU datasets (retrospective)	Retrospective ML modeling	High (risk of overfitting, no prospective validation)	Low; mostly tertiary NICU context

Medical education is being increasingly imparted to the general population by AI chatbots, which can help mothers take better care of their infants. An intelligent chatbot was designed in Brazil to educate mothers on the benefits of breastfeeding.³⁶ The conversation was carried out over four rounds, and the study found that the number of interactions and participants increased with each round.

The conversation flow of the chatbot improved with every round.³⁶ The most accurate pipeline was 93% accurate, with an average fallback index of 15%.³⁶ One study described the use of an ML algorithm to predict the risk of postpartum depression from data extracted from electronic health records.³⁹ The performance measures were evaluated by AUC, the area under the receiver operating characteristic curve (AUC), which was 0.9 and 0.8 for the development and validation sets, respectively. The available evidence is constrained by small sample sizes, urban or tertiary care settings, and reliance on surrogate endpoints, such as attendance rates, rather than maternal or neonatal mortality. Many studies lacked randomization and external validation, limiting their generalizability to rural or community-based populations. Furthermore, publication bias and restricted gray literature appraisal may have skewed the representation of the results toward positive outcomes. Consequently, the overall certainty of the evidence remains moderate, emphasizing the need for larger multicenter evaluations in low-resource environments.

Barriers to AI Implementation

Despite the promising applications outlined, the integration of AI into maternal and child health systems in LRS faces complex barriers. These challenges extend beyond mere technical limitations, permeating infrastructure, ethics, and financial structures, and collectively determine the ultimate impact of AI-driven solutions (Figure 4).

Technical and Infrastructure Barriers

Successful AI deployment requires a robust infrastructure, including reliable Internet connectivity, modern digital hardware, and secure data storage facilities. Unfortunately, these foundational elements are often absent or inconsistent in settings that stand to benefit the most. Geographic inaccessibility, particularly in rural areas, restricts e-health services through high costs and underdeveloped systems. Poor infrastructure presents significant technical challenges that prevent the widespread use of digital technology in healthcare. This includes the need for more stable services, such as electricity and reliable Internet, which constrain the effective implementation and use of digital health solutions.⁴⁰ This has been achieved in countries such as Brazil and India.⁴¹ Comparative experience demonstrates that success depends on the synchronization of AI with preparedness within community systems. For example, India’s ARMMAN program worked because messaging powered by AI was embedded in ongoing outreach and supported with training and follow-up.

Uganda’s surveillance initiatives also improved when AI tools were incorporated into regular data streams and public health activities rather than running as standalone pilots. In contrast, the Nigeria chatbot pilot struggled with language and connectivity, which dissipated uptake and trust in the chatbot. Health centers in rural areas tend to have constrained availability of key services, undermining device functionality and routine healthcare operations. Internet connectivity is pivotal in enabling data exchange, teleconsultation, and retrieval of patient records; however, slow and inconsistent connections prevent the adoption of cloud-based solutions and real-time telemedicine services, thereby compromising continuity of care. Collectively, these constraints highlight the foundational importance of infrastructure readiness for any AI strategy.

For example, a systematic review in Ethiopia showed that the lack of infrastructure was a key limitation in the use of electronic health records.⁴² AI adoption in LMICs requires better Internet, adequate capacity, and sufficiently affordable services.⁴³ Most health facilities remain under-equipped with appropriate technology and software, and clinicians’ digital literacy is limited. The WHO European Region 2023-2030 Regional Digital Health Action Plan identifies the development of digital literacy skills and capacity building in the population at large, particularly among the health workforce utilizing digital health services.^44,45

Ethical and Legal Issues

In addition to technical challenges, the integration of AI raises deep ethical and legal issues. Patient autonomy, beneficence, and non-maleficence are the pillars of medical ethics; however, these principles are challenged when decisions are shared with machines. Inadequate data protection laws and weak institutional oversight in low-resource countries heighten the risks of privacy breaches, data misuse, and algorithmic bias.⁴⁶ Sensitive health data risks being exploited in the absence of strong safeguards, leading to the erosion of public trust. The WHO guidance in 2021⁴⁷ is a significant global framework for principles, but differences in context guarantee that algorithms developed in high-income environments will not automatically scale to low-income ones. Therefore, ethical control must be localized and flexible to meet specific community requirements. Regional diversity in legal and cultural norms further complicates this adoption.

Socioeconomic and Cultural Determinants

One of the most promising applications of AI in healthcare is personalized medicine. AI can tailor treatment protocols according to the individual needs of a patient based on their genetic makeup, medical history, and lifestyle.⁴⁸ In some societies, traditional beliefs or mistrust of modern healthcare systems create skepticism regarding AI-driven recommendations. However, cultural sensitivity needs to be practiced while applying individualized plans, and there is a need to study the cultural elements influencing patients’ decisions and adherence to treatment plans.

Telemedicine, with the help of AI-powered chatbots and virtual health assistants, can potentially close healthcare access gaps, particularly in remote or underserved communities. Such technologies can offer medical advice, monitor diseases in patients, and transmit health education in multiple languages for cultural diversity. However, their reach still depends on basic infrastructure, such as affordable Internet and mobile access.⁴⁹ In areas where such resources are scarce, communities may continue to rely on traditional healers or community health workers (CHWs). Together, these interlocking factors underscore that effective AI implementation in maternal and child health depends as much on cultural alignment and financial viability as on technical competence.

Financial Sustainability

Regardless of whether culture and infrastructure support AI, the cost of its deployment remains a major barrier. Implementation requires not only an upfront investment in equipment and software but also ongoing expenses for maintenance, data storage, and training. These costs are often overlooked in pilot programs, rendering projects unsustainable once the initial funding ends.^50–52 In resource-constrained health systems, long-term viability requires consistent government commitment and strategic financing. Without it, AI risks becoming a short-lived experiment rather than an instrument of transformation. Figure 3 presents a conceptual framework that outlines the barriers to AI application in resource-poor settings.

Potential for Scaling AI Solutions in Maternal and Child Health

PPPs: Role of partnerships between governments, NGOs, and private technology firms. PPPs have leveraged the abilities of governments, private technology companies, and NGOs to overcome the limitations of infrastructure and resources in developing and scaling up AI solutions. Governments worldwide tend to collaborate with these companies to augment the monitoring, infrastructure, and supply of medical facilities to their citizens.⁵³ AI has played a vital role in improving healthcare accessibility for mothers and children, particularly in rural areas where health centers are absent or difficult to access. Telemedicine solutions and mobile health applications powered by AI have extended healthcare to underserved populations.⁵⁴ Additionally, AI-driven vaccination systems could enhance immunization programs by accurately forecasting vaccine demand, monitoring stock levels, and preventing diseases such as hepatitis and meningitis among young children.⁵⁵

In maternal health, AI-based applications contribute to reducing maternal mortality through predictive analytics, automated healthcare delivery, and improved diagnostic accuracy, enabling early identification of high-risk mothers.⁵⁶ Furthermore, effective implementation of AI in maternal and child health is supported by private-public partnerships, innovative funding models, and leveraging both governmental and philanthropic resources, which together foster broader adoption and long-term sustainability of these technologies.

Case Studies

Several countries have successfully integrated mHealth and AI into maternal and child health programs, yielding significant improvements in outcomes. In India, the ARMMAN project leverages mobile health technology and AI to strengthen health systems, enhance community training, and improve access to information, leading to a notable reduction in maternal mortality, particularly in rural areas.⁵⁷ Similarly, Tanzania has adopted an mHealth system that tracks pregnancies, monitors antenatal care, and facilitates communication between health workers and mothers to ensure timely and effective healthcare services.⁵⁸ In the Philippines, AI has been applied to newborn screening by analyzing dried blood spot samples and detecting biomarkers, enabling early identification of metabolic disorders and advancing neonatal healthcare.⁵⁹ Uganda has also embraced AI-driven disease surveillance programs that cover zoonotic infections, malaria, and computerized diagnostics, contributing to better patient outcomes and improved maternal care in underserved rural areas.⁶⁰ Table 4 compares successful AI implementations in maternal and child health across different resource-limited settings.

Table 4: Comparative analysis of successful ai implementations in maternal and child health across different resource-limited settings.
Country	AI Application	Objectives	Outcomes	Challenges Faced	References
India	AI programs for health monitoring	Improve maternal and child health outcomes	Improved antenatal care; reduced infant mortality	Low literacy rate	57
Uganda	Prediction of maternal and child diseases emergence	Predict disease outbreaks	Improved management of infectious diseases	Poor infrastructure	60
Philippines	AI-powered newborn screening using mass spectrometry.	Detection of genetic disorders in neonates	Accurate identification of infectious and metabolic disorders	Scarcity of experienced medical professionals	59
Tanzania	AI-supported diagnostics for maternal and childcare	Enhance Healthcare facilities	Enhanced detection of complications and resource allocation	Geographic disparities	60
Ethiopia	Telemedicine services for maternal care	Reduce obstetric emergencies	Improved access to specialists and better critical care	Sporadic mobile network and internet services	61
Rwanda	AI-powered drone delivery of medical supplies	Reduced maternal mortality rates	Reduced pregnancy complications and mortality rates	Limited availability to specific geographic areas.	62

Integration with Local Health Systems: Principles for Local Healthcare Systems Integration of AI

Integrating AI into maternal and child healthcare requires strategic approaches that strengthen, rather than disrupt existing systems. One approach is empowering CHWs through solar-powered, AI-enabled, and privacy-protected mobile healthcare units, which can extend services to mothers in underserved regions.^63,64 Equally important is consolidating data infrastructure, as AI systems thrive on comprehensive datasets such as electronic health records, particularly when supported by real-time data and edge-computing services. Interoperability between health systems and AI platforms further enhances communication, facilitates the creation of complete patient profiles, and enables predictive analytics for improved diagnosis and management.^63,64 Pilot programs also play a critical role by allowing the testing and iterative refinement of AI algorithms in specific maternal and child health contexts.^63,64 Finally, the expansion of telemedicine services helps address specialist shortages, supports pregnancy monitoring, and improves neonatal health, offering cost-effective and time-efficient solutions for pregnant women.^63,64

Ethical Considerations and Patient Autonomy

AI applications have tremendous potential for enhancing health quality in underdeveloped regions. AI applications can facilitate early detection of conditions such as preterm labor and gestational diabetes through data analysis and consequently allow for timely intervention to be taken. It also allows telemedicine and remote monitoring to the disadvantaged communities and further enhances outcomes of care for mothers and children by analyzing the treatments to identify individual patients. It also improves personalized care through review of patient data to personalize treatment programs and thereby improve maternal and child outcomes. These innovations could close healthcare gaps, alleviate the workload on healthcare professionals, and improve efficiency in operations in delivering services.⁶⁵ Ethical leadership also appears to influence success rates. For example, the Philippines integrated AI in national newborn screening under common ethical standards, allowing for community acceptability and sustainability.⁵⁹ Meanwhile, failed pilots in some sub-Saharan African settings had inadequate consent procedures and no culturally appropriate communication, which eroded patient trust.⁶⁶

This suggests that ethical preparedness, transparent data management, consent that is culturally appropriate, and engaging in the community can be as powerful as technical competence. Although there will be widespread benefits when AI is implemented in healthcare, there is recognition of complex ethical concerns that may arise when introducing AI, especially in resource-constrained settings. Education variation, access to technology, and provision of health can act as hindrances for both the providers of AI technology and those patients on whom the technology will be applied. There are also some issues concerning data ownership, consent, and transparency, as AI use is likely to work as a black box, and this reduces trust and patient autonomy.⁶⁶ AI use must be underpinned by ethical principles to facilitate transparency, equity, and accountability. There are three basic elements algorithmic transparency, informed consent, and community engagement that must be remembered.

Transparency in AI is crucial for ethical healthcare, particularly for vulnerable groups like children and mothers. Where resources are limited, with weakened trust in healthcare, ambiguous AI algorithms can prevent patient trust and clinician uptake. Clinical decision-making AI models must be explainable so that clinicians can comprehend and explain decisions to patients. Transparency is necessary in clinical settings, where clinicians must consider whether they should follow AI advice. Achieving transparency involves providing clear language explanations of AI decision-making. Despite challenges from complex models, interpretability promotes trust and accountability and facilitates the detection of bias that can affect patient outcomes, especially for vulnerable populations.⁶⁷ Aiming towards achieving algorithmic transparency includes

Explainable XAI Techniques

XAI methods strive to enhance the interpretability and understandability of ML models through various methods, including feature importance analysis and model-agnostic techniques. The central mandate of XAI is to reveal deep models that are prone to being black boxes with obscure decision-making mechanisms. By increasing model interpretability, XAI facilitates the creation of user trust. When individuals know why a model arrives at its conclusions, they are more likely to accept and trust its findings.⁶⁸

Addressing Bias and Ethical Concerns

Eliminating bias within AI systems is essential for ensuring equitable maternal and child healthcare outcomes. AI technology has the potential to inadvertently create inequalities, particularly among marginalized communities. As an example, discriminatory data sets may result in faulty pregnancy risk assessments, potentially undermining timely clinical intervention. In the interest of fairness, AI models should emphasize transparency, be subject to frequent audit, and be developed from representative data sets that include a range of populations.⁶⁹

Regulatory Frameworks

Regulatory frameworks such as the GDPR emphasize transparency in AI so that users can view how choices are made. The regulations encourage working on transparent AI systems and maintaining ethical principles unaltered.⁷⁰ In the field of medicine, informed consent is a significant ethical standard that ensures patients receive complete understanding and agree willingly to treatment. This becomes increasingly vital when AI-based tools enter the picture. But obtaining valid consent is sometimes difficult in LRS because of language barriers, low levels of literacy, cultural disparities, and limited understanding about AI. These hurdles can undermine the capacity of patients to understand how AI tools function and their risks and benefits. To resolve these issues, consent procedures must be modified to provide open, culturally appropriate communication, employing mechanisms such as visual aids, language interpreters, or community liaisons. These practices are essential to enable well-informed decision-making regarding AI in healthcare, to instill trust, and to enhance patient outcomes.⁷¹ For making informed consent truly existent, a series of important points must be considered.

Transparency of AI System

They should be informed about the operation of AI systems, limitations, potential errors, and prejudice. To secure informed consent for AI-assisted treatments, the procedures involved, the role of AI in enabling such treatments, risks, and information for decision-making should be known by the persons. The patients should know the treatment required, how the treatment process would be improved by AI, the potential risks involved, and the specific details to be disclosed.⁷²

Ethical Framework for Consent

Establishing the proper circumstances to notify patients regarding AI involvement in healthcare requires a robust ethical procedure. This procedure needs to determine various factors, including potential clinical concerns, the extent of autonomous decisions being exercised by AI, and its possible effect on patient outcomes. In high-risk choice-making cases or highly autonomous AI systems, full disclosure is required to foster transparency, facilitate patient understanding, and support informed decision-making.⁷³ Choice and autonomy of patients Involving patients in discussions about using AI in their care is the key to gaining their confidence and preserving their autonomy. By explaining transparently how AI functions, as well as its advantages and restrictions, the healthcare provider can empower patients to have autonomous decision-making capabilities. One such open, participatory approach enhances the confidence of the patient and forms a relationship based on trust if AI is used within healthcare.⁷²

Visual and Verbal Explanation

Visual aids like graphs and heatmaps can assist in effectively presenting the AI predictions. Spoken descriptions also enhance perceived AI system credibility.^74,75 Engaging communities is crucial to guarantee that AI-based digital health interventions are effectively implemented, particularly in resource-poor environments. Active engagement of local populations, healthcare providers, and leaders ensures that AI solutions are adapted to meet community needs and boost acceptability and long-term viability. Research indicates that community engagement increases the adaptability and usability of AI tools. Yet, a recent scoping review reported that as little as 0.2% of AI healthcare applications now include community voices in their development. These results highlight the importance of integrating participatory approaches to enhance the impact of AI as well as establish trust with users.⁷⁵ Involving CHWs in AI tool design enhances knowledge and promotes acceptance. They are greatly trusted by the population, and accordingly, based on an understanding of local culture and healthcare requirements, CHWs extend AI applications to become more useful and applicable. Their involvement also generates trust and allows communities to embrace AI-based healthcare with greater confidence.⁷⁶

Future Directions and Research Opportunities

The trajectory of AI in maternal and child health is poised to move beyond pilot success in isolation to system-level, integrated change. This means intentional movement away from mere demonstration of technical feasibility and toward solution building for deployable, scalable, and context-specific application in LRS. The most desirable spaces for future research and implementation are where new technology and practical adaptation converge. One of the biggest bottlenecks in LRS is the lack of large, heterogeneous, and well-curated data to train robust, generalizable AI models, coupled with genuine data privacy and sovereignty concerns.⁷ Federated learning (FL) offers a paradigm shift towards addressing this by enabling collaborative model training in the absence of centralizing sensitive data, thereby aligning technological advancement with ethical imperative.⁷⁷ The run-time application of FL is to deploy an initial model to peripheral devices (e.g., phones at peripheral clinics) where it learns locally on de-identified data. Only encrypted model updates (gradients/weights) and not the raw data are transmitted periodically to a centralized aggregator when connectivity exists. This approach develops a strong, globally informed model with community data staying within the community.

This architecture is very well suited to LRS, as it minimizes the need for constant, high-bandwidth internet to the bare essentials, employs periodic connectivity for tiny packets of data, and conforms to evolving data protection regulations. Future research will need to focus on building ultra-lightweight model structures to minimize computational load on edge devices and standardization of data formats across various health facilities to enable interoperability. The way forward for continuous care is to move past expensive, consumer-grade wearables to low-cost, purpose-designed sensors integrated into IoT platforms. These systems must be constructed as long-lived, low-power, and offline-enabled to be viable in LRS.^25,78 Practical adoption is to develop solar-rechargeable biosensors for Mom’s vitals (e.g., blood pressure, glucose) that link via Bluetooth Low Energy to a smartphone of a CHW. The phone is also a local hub, running on-device AI models locally for real-time processing and notification, running offline. Data gets stored locally and synchronized to cloud-based EMRs only when the network is available, creating a robust hybrid model. Also, reimagining smartphone cameras with computer vision algorithms for uses like neonatal jaundice screening (BiliCam) turns a common device into a powerful diagnostic tool. This solution rides on the current mobile phone network penetration, even at 2G/3G levels, and is therefore extremely scalable.

It depends on co-designing the technologies with end-users to be culturally acceptable, user-friendly, and easy to use by local health workers. The last research opportunity is not in the individual technologies but in their combined use in the fragile ecosystem of LRS. This requires a multi-faceted approach. As observed by a recent scoping review, community participation during development is only included in 0.2% of AI applications for health.⁷⁵ Future efforts must adopt participatory design methods with CHWs, mothers, and local leaders from the outset so that interventions can be trusted, adopted, and maintained.⁷⁷ Scalability of AI solutions is only possible through new models of funding and implementation. New PPP structures must be crafted to encourage technology companies to develop for scalability and sustainability in LRS, rather than purely for profitable ones.^53,63 Implementation of AI must be governed by robust, context-dependent ethical frameworks that ensure algorithmic transparency, minimize bias, and safeguard patient autonomy.^46,67,71 Research is crucial for developing streamlined informed consent protocols that are culturally and linguistically tailored to diverse LRS populations.^71,73

Long-term viability of AI integration depends on facilitating the local health workforce (Figure 5). This involves the development of education programs to institute digital literacy among healthcare professionals such that they would be able to make use of, interpret, and control AI-driven tools efficiently.^44,45,79 By intentionally focusing on these potential and complementary avenues, AI can successfully democratize high-quality health care. The dream is to pave the way for a world where all children and mothers, regardless of economics or geography, have access to the predictive, preventive, and personalized care that AI can potentially offer.^4,79,80 In summary, the future of AI in maternal and child health hinges on a deliberate pivot from isolated technical triumphs to integrated, human-centered systems. By prioritizing FL, co-designing low-cost technologies with end-users, establishing innovative funding partnerships, and embedding strong ethical frameworks, these innovative tools can truly democratize high-quality healthcare. The ultimate goal is a future where every mother and child, irrespective of geography or economic status, has access to the predictive, preventive, and personalized care that AI can help deliver.

Practice Implications Box

Integrate AI tools within existing digital health programs rather than as standalone pilots. Prioritize low-cost, energy efficient IoT solutions and FL frameworks to minimize infrastructure dependency. Involve CHWs in co-design and delivery to improve trust and adoption. Align deployments with national digital health and data protection frameworks to ensure sustainability. Incorporate ethical oversight and culturally appropriate consent processes in all AI-driven initiatives. Support ongoing capacity building for healthcare professionals to enhance digital literacy.

Conclusion

AI holds promise to help solve some of the most intractable problems in maternal and child health, especially in resource-limited settings. As this review describes, AI can improve diagnosis, enable monitoring of full-scale digital tools, and aid in tailoring care plans to individual needs, making them more accessible and, potentially, more effective. Predictive analytics, telemedicine, and decision support systems are particularly promising venues for closing decades-old gaps in care delivery. Realizing this potential, however, is fraught with significant challenges. The hype about AI must be tempered by hard-boiled acknowledgment of the barriers that still get in the way. Issues such as unstable electricity and internet, cost of deployment, lack of skilled health professionals, and ongoing questions of privacy and fairness continue to hinder wide adoption. The real potential of AI in the future for maternal and child health rests less on sophisticated technology than on how well it is embedded in local settings. Therefore, the success of AI should be measured not by its algorithmic sophistication, but by its tangible contribution to reducing maternal and child mortality and improving health equity worldwide.

Abbreviations

• AI: Artificial intelligence
• mHealth: Mobile health
• EMR: Electronic medical records
• CHWs: Community health workers
• AM-Care: Adaptive multi-care
• ML: Machine learning
• IoT: Internet of Things

References

1. Ekwuazi E, Chigbu C, Ngene N. Reducing maternal mortality in low- and middle-income countries. Case Rep Womens Health. 2023;39:e00542. https://doi.org/10.1016/j.crwh.2023.e00542
2. World Health Organization. Maternal mortality [gray literature]. Geneva: World Health Organization; 2024. Available from: https://www.who.int/news-room/fact-sheets/detail/maternal-mortality
3. UNICEF. Levels and trends in child mortality. New York: UNICEF; 2024. Available from: https://data.unicef.org/resources/levels-and-trends-in-child-mortality-2024/
4. Khan M, Khurshid M, Vatsa M, Singh R, Duggal M, Singh K. On AI approaches for promoting maternal and neonatal health in low-resource settings: a review. Front Public Health. 2022;10:880034. https://doi.org/10.3389/fpubh.2022.880034
5. Zuhair V, Babar A, Ali R, Oduoye MO, Noor Z, Chris K, et al. Exploring the impact of artificial intelligence on global health and enhancing healthcare in developing nations. J Prim Care Community Health. 2024;15:21501319241245847. https://doi.org/10.1177/21501319241245847
6. Maaß L, Freye M, Pan CC, Dassow HH, Niess J, Jahnel T. The definitions of health apps and medical apps from the perspective of public health and law: qualitative analysis of an interdisciplinary literature overview. JMIR Mhealth Uhealth. 2022;10(10):e37980. https://doi.org/10.2196/37980
7. Wahl B, Cossy-Gantner A, Germann S, Schwalbe NR. Artificial intelligence (AI) and global health: how can AI contribute to health in resource-poor settings? BMJ Glob Health. 2018;3(4):e000798. https://doi.org/10.1136/bmjgh-2018-000798
8. Vinora A, Lloyds E, Deborah RN, Gobinath A, Sivakarthi G. Wearable sensors and AI algorithms for pregnant women. In: Artificial intelligence in healthcare. Hershey, PA: IGI Global; 2023. p. 1–15.
9. Liu Y, Chen PHC, Krause J, Peng L. How to read articles that use machine learning: users’ guides to the medical literature. JAMA. 2019;322(18):1806–16. https://doi.org/10.1001/jama.2019.16489
10. Say L, Raine R. A systematic review of inequalities in the use of maternal health care in developing countries: examining the scale of the problem and the importance of context. Bull World Health Organ. 2007;85(10):812–9. https://doi.org/10.2471/blt.06.035659
11. Black RE, Walker N, Laxminarayan R, Temmerman M. Reproductive, maternal, newborn, and child health: disease control priorities. 3rd ed. Washington, DC: World Bank Group; 2016.
12. Yuan B, Qian X, Thomsen S. Disadvantaged populations in maternal health in China—who and why? Glob Health Action. 2013;6:19542. https://doi.org/10.3402/gha.v6i0.19542
13. Koblinsky MA, Timyan J, Gay J. The health of women: a global perspective. New York: Routledge; 2018.
14. Al Dahdah M, Desgrées Du Loû A, Méadel C. Mobile health and maternal care: a winning combination for healthcare in the developing world? Health Policy Technol. 2015;4(3):225–31. https://doi.org/10.1016/j.hlpt.2015.04.002
15. Iyawa GE, Hamunyela S. mHealth apps and services for maternal healthcare in developing countries. In: 2019 IST-Africa Week Conference (IST-Africa); 2019. p. 1–10. https://doi.org/10.23919/ISTAFRICA.2019.8764878
16. Romsaiyud W, Premchaiswadi W. Adaptive multi-services system for maternal and child health care on mobile application (AM-Care). Int J Healthc Inf Syst Inform. 2010;5(3):27–43. https://doi.org/10.4018/jhisi.2010070103
17. Gupta M, Mishra A, Kumar P. Impact of digitalization in maternal healthcare: maternal healthcare with modern digitalization technique. In: Takale D, Mahalle P, Narvekar M, Mahajan R, editors. Advances in medical technologies and clinical practice. Hershey, PA: IGI Global; 2024. p. 207–27.
18. Knop MR, Nagashima-Hayashi M, Lin R, Saing CH, Ung M, Oy S, et al. Impact of mHealth interventions on maternal, newborn, and child health from conception to 24 months postpartum in low- and middle-income countries: a systematic review. BMC Med. 2024;22(1):196. https://doi.org/10.1186/s12916-024-03417-9
19. Cripps M, Scarbrough H. Making digital health “solutions” sustainable in healthcare systems: a practitioner perspective. Front Digit Health. 2022;4:727421. https://doi.org/10.3389/fdgth.2022.727421
20. Slovensky DJ, Malvey DM. Introduction to focused issue on mHealth infrastructure: issues and solutions that challenge optimal deployment of mHealth products and services. Mhealth. 2017;3:52. https://doi.org/10.21037/mhealth.2017.12.03
21. Raja R, Mukherjee I, Sarkar BK. A machine learning prediction model for preterm birth in rural India. J Healthc Eng. 2021;2021:6665573. https://doi.org/10.1155/2021/6665573
22. IMIA, Georg Thieme Verlag KG Stuttgart. Contents IMIA yearbook of medical informatics 2019. Yearb Med Inform. 2019;28:I–VIII. https://doi.org/10.1055/s-0039-1694828
23. Drukker L. Real-time identification of fetal anomalies on ultrasound using artificial intelligence: what’s next? Ultrasound Obstet Gynecol. 2022;59(3):285–7. https://doi.org/10.1002/uog.24869
24. Gulzar Ahmad S, Iqbal T, Javaid A, Ullah Munir E, Kirn N, Ullah Jan S, et al. Sensing and artificial intelligent maternal-infant health care systems: a review. Sensors. 2022;22(12):4362. https://doi.org/10.3390/s22124362
25. Yaseen I, Rather RA. A theoretical exploration of artificial intelligence’s impact on feto-maternal health from conception to delivery. Int J Womens Health. 2024;16:903–15. https://doi.org/10.2147/IJWH.S454127
26. Ramakrishnan R, Rao S, He JR. Perinatal health predictors using artificial intelligence: a review. Womens Health. 2021;17:17455065211046132. https://doi.org/10.1177/17455065211046132
27. Akazawa M, Hashimoto K. Prediction of preterm birth using artificial intelligence: a systematic review. J Obstet Gynaecol. 2022;42(7):1662–8. https://doi.org/10.1080/01443615.2022.2056828
28. Dhombres F, Bonnard J, Bailly K, Maurice P, Papageorghiou AT, Jouannic JM. Contributions of artificial intelligence reported in obstetrics and gynecology journals: systematic review. J Med Internet Res. 2022;24:e35465. https://doi.org/10.2196/35465
29. Armero W, Gray KJ, Fields KG, Cole NM, Bates DW, Kovacheva VP. A survey of pregnant patients’ perspectives on the implementation of artificial intelligence in clinical care. J Am Med Inform Assoc. 2022;30(1):46–53. https://doi.org/10.1093/jamia/ocac200
30. Kim HY, Cho GJ, Kwon HS. Applications of artificial intelligence in obstetrics. Ultrasonography. 2023;42(1):2–9. https://doi.org/10.14366/usg.22063
31. Ahn KH, Lee KS. Artificial intelligence in obstetrics. Obstet Gynecol Sci. 2022;65(2):113–24. https://doi.org/10.5468/ogs.21234
32. Emin EI, Emin E, Papalois A, Willmott F, Clarke S, Sideris M. Artificial intelligence in obstetrics and gynaecology: is this the way forward? In Vivo. 2019;33(5):1547–51. https://doi.org/10.21873/invivo.11635
33. Friesen S. Could AI predict pregnancy risk?. Salt Lake City: University of Utah; 2024. Available from: https://attheu.utah.edu/facultystaff/could-ai-predict-pregnancy-risk/
34. Cedars-Sinai. Women’s health month: artificial intelligence can improve OB-GYN care [gray literature]. Los Angeles, CA: Cedars-Sinai; 2024. Available from: https://www.cedars-sinai.org/newsroom/womens-health-month-artificial-intelligence-can-improve-ob-gyn-care/
35. Li X, Lu Y, Fu X, Qi Y. Building the Internet of Things platform for smart maternal healthcare services with wearable devices and cloud computing. Future Gener Comput Syst. 2021;118:282–96. https://doi.org/10.1016/j.future.2021.01.016
36. Corrêa JS, Neto APdA, Pinto GR, Lima LDB, Teles AS. Lhia: a smart chatbot for breastfeeding education and recruitment of human milk donors. Appl Sci. 2023;13(12):6923. https://doi.org/10.3390/app13126923
37. Taylor JA, Stout JW, de Greef L, Goel M, Patel S, Chung EK, et al. Use of a smartphone app to assess neonatal jaundice. Pediatrics. 2017;140(3):e20170312. https://doi.org/10.1542/peds.2017-0312
38. Masino AJ, Harris MC, Forsyth D, Ostapenko S, Srinivasan L, Bonafide CP, et al. Machine learning models for early sepsis recognition in the neonatal intensive care unit using readily available electronic health record data. PLoS One. 2019;14(2):e0212665. https://doi.org/10.1371/journal.pone.0212665
39. Zhang Y, Wang S, Hermann A, Joly R, Pathak J. Development and validation of a machine learning algorithm for predicting the risk of postpartum depression among pregnant women. J Affect Disord. 2021;279:1–8. https://doi.org/10.1016/j.jad.2020.09.113
40. Parajuli R, Bohara D, Kc M, Shanmuganathan S, Mistry SK, Yadav UN. Challenges and opportunities for implementing digital health interventions in Nepal: a rapid review. Front Digit Health. 2022;4:861019. https://doi.org/10.3389/fdgth.2022.861019
41. Caetano R, Silva AB, Guedes ACCM, Paiva CCN, Ribeiro GDR, Santos DL, et al. Challenges and opportunities for telehealth during the COVID-19 pandemic: ideas on spaces and initiatives in the Brazilian context. Cad Saude Publica. 2020;36:e00088920. https://doi.org/10.1590/0102-311×00088920
42. Yehualashet DE, Seboka BT, Tesfa GA, Demeke AD, Amede ES. Barriers to the adoption of electronic medical record system in Ethiopia: a systematic review. J Multidiscip Healthc. 2020;13:2597–603. https://doi.org/10.2147/JMDH.S327539
43. Kiberu VM, Mars M, Scott RE. Barriers and opportunities to implementation of sustainable e-Health programmes in Uganda: a literature review. Afr J Prim Health Care Fam Med. 2017;9(1):1–10. https://doi.org/10.4102/phcfm.v9i1.1277
44. Regional Committee for Europe, 72nd session. Regional digital health action plan for the WHO European Region 2023-2030. Copenhagen: WHO Regional Office for Europe; 2022.
45. Nascimento IJ, Abdulazeem H, Vasanthan LT, Martinez EZ, Zucoloto ML, Østengaard L, et al. Barriers and facilitators to utilizing digital health technologies by healthcare professionals. NPJ Digit Med. 2023;6(1):28.
46. Mennella C, Maniscalco U, De Pietro G, Esposito M. Ethical and regulatory challenges of AI technologies in healthcare: a narrative review. Heliyon. 2024;10(4):e26297. https://doi.org/10.1016/j.heliyon.2024.e26297
47. World Health Organization. WHO guideline: recommendations on digital interventions for health system strengthening. Geneva: WHO; 2019.
48. Lupton D. Data selves: more-than-human perspectives. Cambridge: Polity; 2019.
49. Bajwa J, Munir U, Nori A, Williams B. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J. 2021;8(2):e188–94. https://doi.org/10.7861/fhj.2021-0095
50. Haefner N, Parida V, Gassmann O, Wincent J. Implementing and scaling artificial intelligence: a review, framework, and research agenda. Technol Forecast Soc Change. 2023;197:122878. https://doi.org/10.1016/j.techfore.2023.122878
51. Galaz V, Centeno MA, Callahan PW, Causevic A, Patterson T, Brass I, et al. Artificial intelligence, systemic risks, and sustainability. Technol Soc. 2021;67:101741. https://doi.org/10.1016/j.techsoc.2021.101741
52. Van Wynsberghe A. Sustainable AI: AI for sustainability and the sustainability of AI. AI Ethics. 2021;1(3):213–8. https://doi.org/10.1007/s43681-021-00043-6
53. Pinz A, Roudyani N, Thaler J. Public-private partnerships as instruments to achieve sustainability-related objectives: the state of the art and a research agenda. In: Boyd NM, Martin EC, editor. Sustainable public management. London: Routledge; 2021. https://doi.org/10.4324/9781003202479
54. Shaheen MY. Applications of artificial intelligence (AI) in healthcare: a review. SciOpen Preprints; 2021. http://doi.org/10.14293/S2199-1006.1.SOR-.PPVRY8K.v1
55. AI in healthcare: using cutting-edge technologies to revolutionize vaccine development and distribution. JURIHUM: Jurnal Inovasi dan Humaniora [grey literature]; 2024. Available from: https://jurnalmahasiswa.com/index.php/Jurihum/article/view/1544
56. Innovations in U.S. health care delivery to reduce disparities in maternal mortality among African American and American Indian/Alaskan Native women; 2024. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC8060040
57. ARMMAN – Helping mothers and children [grey literature]. Mumbai: ARMMAN; 2024. Available from: https://armman.org
58. Sukums F, Mzurikwao D, Sabas D, Chaula R, Mbuke J, Kabika T, et al. The use of artificial intelligence-based innovations in the health sector in Tanzania: a scoping review. Health Policy Technol. 2023;12(1):100728. https://doi.org/10.1016/j.hlpt.2023.100728
59. Padilla CD, Therrell BL, Alcausin MMLB, Chiong MAD, Abacan MAR, Reyes MEL, et al. Successful implementation of expanded newborn screening in the Philippines using tandem mass spectrometry. Int J Neonatal Screen. 2022;8(1):8. https://doi.org/10.3390/ijns8010008
60. Using artificial intelligence to predict disease emergence in Uganda. Seattle: Grand Challenges; 2024. Available from: https://gcgh.grandchallenges.org/grant/using-artificial-intelligence-predict-disease-emergence-uganda
61. Sagaro GG, Battineni G, Amenta F. Barriers to sustainable telemedicine implementation in Ethiopia: a systematic review. Telemed Rep. 2020;1(1):8–15. https://doi.org/10.1089/tmr.2020.0002
62. Nisingizwe MP, Nshimiyimana L, Karim E, et al. Effect of the use of drone-based delivery of blood products and medicines on the reliability of supply and availability at remote health facilities in Rwanda: a before-and-after study. Lancet Glob Health. 2024;12(11):e1871–9.
63. The Health Foundation. Priorities for an AI in health care strategy. London: The Health Foundation; 2024. Available from: https://www.health.org.uk/reports-and-analysis/briefings/priorities-for-an-ai-in-health-care-strategy
64. Meeting the moment: addressing barriers and facilitating clinical adoption of artificial intelligence in medical diagnosis; 2024. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC9875857/
65. Chemisto M, Gutu TJ, Kalinaki K, Mwebesa BD, Egau P, Fred K, et al. Artificial intelligence for improved maternal healthcare: a systematic literature review. In: Proceedings of the IEEE AFRICON       Conference 2023; 2023. p. 1–6.
66. Yu L-Y, Zhai X. Use of artificial intelligence to address health disparities in low- and middle-income countries: a thematic analysis of ethical issues. Public Health. 2024;234:77–83. https://doi.org/10.1016/j.puhe.2024.05.029
67. Durán JM, Jongsma KR. Who is afraid of black box algorithms? On the epistemological and ethical basis of trust in medical AI. J Med Ethics. 2021;47(11):medethics-2020-106820. https://doi.org/10.1136/medethics-2020-106820
68. Raghavan H, Devarajan D, Soibam BS. Explainable AI for cloud-based machine learning interpretable models and transparency in decision-making. TJJPT. 2024;45(2):2886–94. https://doi.org/10.52783/tjjpt.v45.i02.6376
69. Pace T, Ranes B. Bias, explainability, transparency, and trust for . In: Harguess JD, Bastian ND, Pace TL, editors. Assurance and security for AI-enabled systems. Proceedings of SPIE. Bellingham, WA: SPIE; 2024. p. 13054–11. https://doi.org/10.1117/12.3012949
70. Chaudhary G. Unveiling the black box: bringing algorithmic transparency to AI. Masaryk Univ J Law Technol. 2024;18(1):93–122.
71. Astromskė K, Peičius E, Astromskis P. Ethical and legal challenges of informed consent applying artificial intelligence in medical diagnostic consultations. AI Soc. 2021;36(2):509–20. https://doi.org/10.1007/s00146-020-01008-9
72. Iserson KV. Informed consent for artificial intelligence in emergency medicine: a practical guide. Am J Emerg Med. 2023;72:125–30. https://doi.org/10.1016/j.ajem.2023.11.022
73. Rose SL, Shapiro D. An ethically supported framework for determining patient notification and informed consent practices when using artificial intelligence in healthcare. Chest. 2024;166(3):572–8.
74. Hakes R, Zhao Y, Fang X, Tao J. Toward an explainable artificial intelligence approach to enhance medical imaging classification models. In: Proceedings of the 12th Knowledge Capture Conference 2023; 2023. p. 234–40. https://doi.org/10.1145/3587259.3627553
75. Loftus TJ, Balch JA, Abbott KL, Hu D, Ruppert MM, Shickel B, et al. Community-engaged artificial intelligence research: a scoping review. PLOS Digit Health. 2024;3(8):e0000561. https://doi.org/10.1371/journal.pdig.0000561
76. Okolo CT, Agarwal DP, Dell N, Vashistha A. “If it is easy to understand then it will have value”: examining perceptions of explainable AI with community health workers in rural India. Proc ACM Hum Comput Interact. 2024;8(CSCW1):1–28. https://doi.org/10.1145/3637348
77. Igwama GT, Olaboye JA, Maha CC, Ajegbile MD, Abdul S. Big data analytics for epidemic forecasting: policy frameworks and technical approaches. Int J Appl Res Soc Sci. 2024;6(7):1449–60. https://doi.org/10.51594/ijarss.v6i7.1334
78. Al-Turjman F, Nawaz MH, Ulusar UD. Intelligence in the Internet of Medical Things era: a systematic review of current and future trends. Comput Commun. 2020;150:644–60. https://doi.org/10.1016/j.comcom.2019.12.030
79. Fisher S, Rosella LC. Priorities for successful use of artificial intelligence by public health organizations: a literature review. BMC Public Health. 2022;22(1):2146. https://doi.org/10.1186/s12889-022-14422-z
80. World Economic Forum. How AI can aid safer births in resource-limited environments [grey literature]. Geneva: WEF; 2024. Available from: https://www.weforum.org/stories/2024/08/how-ai-can-aid-safer-births-in-resource-limited-environments/

Table S1: Methodological quality appraisal of included studies.
Study ID	AI Methodology Clarity	Evaluation Design Suitability	Sample Size & Representativeness	Outcome Validity	Ethics & Bias Reporting
Study 1	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 2	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 3	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 4	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 5	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 6	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 7	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 8	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 9	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 10	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 11	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 12	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 13	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 14	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 15	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 16	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 17	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 18	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 19	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 20	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 21	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 22	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 23	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 24	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 25	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 26	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 27	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 28	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 29	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 30	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 31	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 32	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 33	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 34	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 35	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 36	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 37	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 38	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 39	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 40	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 41	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 42	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 43	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 44	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 45	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 46	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 47	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 48	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 49	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 50	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 51	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 52	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 53	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 54	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 55	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 56	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 57	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 58	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 59	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 60	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 61	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 62	High/Moderate/Low	High/Moderate/ Low	High/Moderate/Low	High/Moderate/ Low	High/Moderate/Low
Study 63	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 64	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 65	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 66	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 67	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 68	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 69	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 70	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 71	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 72	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 73	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 74	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 75	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 76	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 77	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 78	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 79	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Study 80	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low	High/Moderate/Low
Summary of the quality assessment for the 80 included studies, rated across five key domains: AI Methodology Clarity, Evaluation Design Suitability, Sample Size & Representativeness, Outcome Validity, and Ethics & Bias Reporting. Ratings are High, Moderate, or Low.

Cite this article as:
Umar M, Shamim L, Rehman A, Zafar H, Talha M, Tahir O, Zafar M, Alam U, Shams F, Rahmat ZS and Baig MMA. The Role of Artificial Intelligence in Enhancing Maternal and Child Health Through Digital Health Initiatives in Resource-Limited Settings: A Narrative Review. Premier Journal of Artificial Intelligence 2026;6:100021

Export to EndNote Export to Mendeley