Role of artificial intelligence in early identification and risk evaluation of non-communicable diseases: a bibliometric analysis of global research trends

Abstract

Objective

This study aims to shed light on the transformative potential of artificial intelligence (AI) in the early detection and risk assessment of non-communicable diseases (NCDs).

Study design

Bibliometric analysis.

Setting

Articles related to AI in early identification and risk evaluation of NCDs from 2000 to 2024 were retrieved from the Scopus database.

Methods

This comprehensive bibliometric study focuses on a single database, Scopus and employs narrative synthesis for concise yet informative summaries. Microsoft Excel V.365 and VOSviewer software (V.1.6.20) were used to summarise bibliometric features.

Results

The study retrieved 1745 relevant articles, with a notable surge in research activity in recent years. Core journals included Scientific Reports and IEEE Access, and core institutions included the Harvard Medical School and the Ministry of Education of the People’s Republic of China, while core countries comprised China, the USA, India, the UK and Saudi Arabia. Citation trends indicated substantial growth and recognition of AI’s impact on NCDs management. Frequent author keywords identified key research hotspots, including specific NCDs like Alzheimer’s disease and diabetes. Risk assessment studies demonstrated improved predictions for heart failure, cardiovascular risk, breast cancer, diabetes and inflammatory bowel disease.

Conclusion

Our findings highlight the increasing role of AI in early detection and risk prediction of NCDs, emphasising its widening research impact and future clinical potential.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The study employs comprehensive bibliometric analysis, systematically identifying research trends, hotspots and emerging topics in artificial intelligence-driven early detection and risk assessment of non-communicable diseases.

• The use of a single well-established database, Scopus, ensures high-quality and peer-reviewed sources, minimising the inclusion of predatory or non-reputable publications.

• The study’s reliance on only one database (Scopus) may exclude relevant studies indexed in other databases, such as PubMed or Web of Science, potentially introducing selection bias.

• The exclusion of non-English publications may result in language bias, potentially overlooking significant research contributions in other languages.

Background

Non-communicable diseases (NCDs) impose a substantial and escalating burden on global public health and healthcare systems.¹ This diverse group of chronic health conditions, including cardiovascular diseases, cancer, diabetes and chronic respiratory diseases, develops over extended periods and persists throughout an individual’s lifetime, primarily influenced by lifestyle factors, genetics and environmental exposures. The WHO reports that NCDs account for nearly 74% of all global deaths, underlining the urgent need for innovative strategies to address them.¹ These diseases significantly strain healthcare budgets and resources, demanding a shift in healthcare approaches, particularly focusing on early detection and risk assessment.²,⁴

Early detection of NCDs plays a pivotal role in improving patient outcomes and alleviating the substantial economic burden on healthcare systems.⁵ Timely identification allows for more effective treatment interventions, reduces complications and extends life expectancy.⁶,⁸ Additionally, early detection enables preventive measures and health education initiatives, ultimately lowering NCD incidence within communities and lightening the healthcare infrastructure’s load.^{9 10} Furthermore, early intervention leads to long-term cost savings, as managing advanced-stage NCDs becomes resource-intensive. However, late-stage diagnoses continue to pose significant challenges, often leading to limited treatment options, poorer prognoses and increased healthcare costs.^{11 12} This late-stage diagnosis results in diminished quality of life, emotional distress and higher disability rates. Addressing these challenges necessitates a comprehensive strategy, and the present paper emphasises the pivotal role of artificial intelligence (AI) in overcoming barriers to early NCD detection, ultimately fostering improved patient outcomes and cost-effective healthcare delivery.

AI stands as a transformative force reshaping healthcare by analysing vast data sets, identifying patterns and generating actionable insights, making it a critical ally in early NCD detection and risk assessment.^{13 14} AI’s applications in healthcare range from medical imaging and predictive analytics to personalised medicine.¹⁵,¹⁷ AI-driven algorithms excel in interpreting radiological scans, predicting disease trajectories and tailoring treatments to individual profiles, showcasing their potential to revolutionise healthcare delivery.¹⁸ In the context of NCDs, AI holds the promise of earlier and more accurate diagnosis, improved risk stratification and timely intervention, advancing healthcare systems toward proactive disease management, reduced morbidity and enhanced patient outcomes.^{19 20}

Several recent studies highlight AI’s role in various medical domains. For instance, blockchain-integrated AI solutions have been proposed for transparent and secure cancer classification, enhancing the reliability of diagnostic processes.²¹ Similarly, Machine Learning (ML)-based semantic segmentation techniques have been used to analyse microbial alterations, demonstrating their potential in microbiology and infectious disease research.²² Additionally, feature reduction techniques have been explored to enhance hepatocellular carcinoma prediction using machine learning algorithms.²³ In the realm of post-treatment monitoring, ML approaches have been successfully employed to enhance abdominal fat prediction, particularly in cavitation post-treatment scenarios.²⁴ Furthermore, deep learning frameworks have been optimised to improve epileptic seizure recognition, leveraging feature scaling and dropout layers to enhance model performance.²⁵ Beyond disease classification, convolutional neural networks (CNNs) have been applied to skin lesion classification, providing accurate detection models for infectious diseases such as monkeypox.²⁶ Furthermore, machine learning has been employed in analysing core muscle function in female sexual dysfunction.²⁷ AI has also been instrumental in optimising disease classification by analysing symptom patterns through language models.²⁸

Recognising a significant research gap in the field, the present comprehensive bibliometric study aims to synthesise the evolving landscape of AI solutions for NCDs. By systematically assessing existing research, the bibliometric study seeks to provide a comprehensive overview of AI’s current state in early NCD detection and risk assessment while identifying emerging trends. Furthermore, it evaluates the progression of AI techniques over time and their real-world applicability in NCD management. These bibliometric findings aim to inform decision-making, foster interdisciplinary collaboration and enhance patient care and resource allocation in the context of NCDs, offering insights that can shape the future of NCD management through AI’s transformative potential.

Methods

Study design

Due to the importance and the general nature of the research question addressed, a comprehensive bibliometric analysis was considered suitable for the present study. The present comprehensive bibliometric analysis is not a form of systematic or scoping review. In designing our bibliometric analysis, we intentionally made methodological choices to ensure timely completion without compromising the quality and validity of our findings. These decisions included narrowing our search to a single comprehensive database, a departure from the usual practice of using multiple databases as in systematic and scoping reviews. Additionally, we assigned specific quality checks to one author rather than following the traditional dual-reviewer process, focusing on efficiency and speed. To further accelerate the review, we opted for narrative synthesis instead of quantitative synthesis, allowing us to provide a concise yet informative summary of the available evidence. We recognise the importance of research transparency, and therefore, we are dedicated to sharing the detailed search strategy used to retrieve the data to promote reproducibility and transparency.

Database used

To carry out the study, pertinent literature was retrieved from the Scopus database. The rationale and benefits underpinning the exclusive use of Scopus (www.scopus.com) were multifaceted and included comprehensiveness, interdisciplinary coverage, convenience and time-saving.

Keywords

In the context of this study, the complete set of keywords related to AI were: “artificial intelligence”, “machine learning”, “deep learning”, “neural networks”, “cognitive computing”, “natural language processing”, “expert systems”, “computer vision”, “autonomous systems”, “robotic intelligence”, “computational intelligence”, “AI algorithms”, “intelligent systems”, “AI technologies”, “automated intelligence”, “smart systems”, “self-learning systems”, “AI applications”, “cognitive machines”, “machine intelligence”, “data-driven intelligence”, “algorithmic intelligence”, “automated decision-making”, “intelligent automation”, “AI models” or “AI frameworks”, “cognitive algorithms”, “learning algorithms”, “AI-driven systems”, “intelligent agents”, “AI-driven analytics”. Terms related to early detection and risk assessment used in the current study were: “early diagnosis”, (early AND detection), (early AND diagnosis), “pre-symptomatic diagnosis”, “early detection”, “risk of”, “risk assessment”, “risk analysis”, “risk evaluation”, “risk appraisal”, “risk estimation”, “risk profiling”, “risk determination”, “risk measurement”, “risk prediction”, “risk categorization”, “risk scoring”, “risk stratification”, “risk grading”, “risk diagnosis”, “risk identification”, “risk characterization”, “risk classification”, “risk judgment”, “risk estimation”. Terms related to NCD included a set of more than 100 NCDs including different types of cancers, diabetes mellitus and other endocrine disorders, different cardiovascular diseases and different neurological and psychiatric diseases. The list also included disease conditions such as diabetic retinopathy (online supplemental table S1).

Inclusion and exclusion criteria

In our systematic search strategy, we established specific inclusion and exclusion criteria to guide the selection of relevant studies. The inclusion criteria encompassed studies published between 2000 and 2024, written in English, and categorised as primary research articles published in journals. We focused our interest on investigations that explored the use of telemedicine/telehealth among diverse underserved groups and in various underserved areas. Excluded from our consideration were articles of a review nature, such as narrative reviews or literature reviews, as well as studies conducted in languages other than English. Additionally, studies conducted in 2025 and those primarily focused on interview methodologies, protocols, perceptions or individual cases were also excluded.

Validation of data

To ensure the accuracy and reliability of the data we extracted, we implemented a meticulous validation process. Initially, we conducted a title-abstract-based search to confirm the absence of false-positive results, enlisting the assistance of two colleagues in the pharmacy and medicine fields who volunteered to assist in this validation process. Subsequently, to guarantee the comprehensiveness of our data set, we subjected the collected articles to a rigorous cross-validation process against a selection of highly cited articles obtained from Google Scholar. The presence of all articles in the retrieved metadata affirmed the inclusiveness of our search strategy. Additionally, the concurrent alignment of active journals and prolific authors with the field further reinforced the credibility of our search approach.

Analytical mapping and research topics

We performed a comprehensive analysis and cartographic representation of the research landscape using Microsoft Excel V.365 and VOSviewer.²⁹ VOSviewer is a widely recognised software tool known for its proficiency in mapping research domains. We seamlessly integrated relevant data sourced from the Scopus database, including publication metadata, into VOSviewer to analyse frequent author keywords and terminologies. The resulting visualisations allowed us to gain valuable insights into the intricate network of research topics and relationships within the field. Interpreting VOSviewer maps is akin to deciphering a colourful and interconnected web of knowledge. Each term or keyword in the data set is visualised as a point on the map, represented by a circle or node. These nodes come in various sizes and colours, and they are connected by lines of different thicknesses. The node size is an indicator of a term’s importance or prevalence in the data set. Larger nodes signify that a particular term is frequently discussed or plays a significant role in the body of research. On the flip side, smaller nodes represent less commonly mentioned concepts. The colours of these nodes serve a dual purpose. First, they help categorise terms into thematic groups. Terms of the same colour usually belong to the same cluster or share a common theme. Second, they assist in identifying different research clusters or topic groups within the data set. For instance, if there is a cluster of blue nodes, it might indicate that these terms are all related to a specific area of research. The distance between nodes is a measure of their closeness in meaning or concept. Nodes positioned closer together share a stronger semantic or contextual connection. They are likely to be mentioned together in research articles or share a similar theme. Conversely, nodes placed farther apart have less in common in terms of their usage in literature. The lines connecting nodes represent the relationships between terms. The thickness of these lines tells us how strong or frequent these connections are. Thick lines indicate that the terms they link are often discussed together or share a robust thematic association. Thinner lines imply weaker or less frequent connections. In essence, VOSviewer maps provide a visual narrative of the underlying structure and relationships within your data set. By examining node size and colour, you can pinpoint important terms and thematic clusters. Meanwhile, the distance between nodes and line thickness unveils the semantic closeness and strength of associations between terms. These visual insights are invaluable for researchers seeking to uncover key concepts, discover research clusters and identify trends within their field of study.

Results

Bibliometric analysis of publication output (main information)

Figure 1 and table 1 show the flow chart and the main bibliometric information regarding AI-enhanced early detection and risk assessment of NCD literature included in the analysis. The retrieved publications have accumulated 37 194 citations thus far.

Figure 1. Flow chart of the role in early identification and risk evaluation of non-communicable disease-related literature search and screening.

Table 1. Main bibliometric indicator for research field.

Indicator	Total
Productivity
Number of total publications (all years)	1745
Number of active years of publication	25
Productivity per active year	69.8
Annual growth rate %	29.6
Impact
Total citations	37 194
Average citations per publication, %	21.3
Number of cited publications	1472
Citations per cited publication	25.3
h-index	88
g-index	137
Authorship
Coauthored publications	1717
Sole-authored publications	28
Number of contributing authors	10 130
Authors of sole-authored publications	28
Coauthors per publication	7.86
Collaboration
Annual collaboration index	4.8
Collaboration index	5.8
Collaboration coefficient	0.8
International coauthorships %	30.8

In the course of our systematic search, we retrieved a total of 1745 articles relevant to our research inquiry. Over the years, there were few or no publications. It was in the most recent years that we noted a substantial surge in research activity, indicating a thriving and dynamic research landscape in the area of investigation (figure 2). Several core journals had a pivotal role in the realm of AI-enhanced early detection and risk assessment of NCD. ‘Scientific Reports’ led the pack with 65 articles, accounting for 3.7% of the publications in the field. Right on its heels is ‘IEEE Access’ contributing significantly with 49 articles, making up 2.8% of the literature. The ‘Computers in Biology and Medicine’ takes a noteworthy position, featuring 31 articles (1.8%), while the ‘Plos One’ journal follows closely with 30 articles (1.7%). In the analysis of core institutions contributing to research in the field, the Harvard Medical School leads the way with a substantial 37 (2.1%) articles. Ministry of Education of the People’s Republic of China follows with 28 (1.6%) articles, followed by Brigham and Women’s Hospital (n=27; 1.5%).

Figure 2. Annual publication and citation trends from 2000 up to 2024. TC, total citations; TP, total number of publications.

The number of citations received by articles in this field has grown significantly over the years, starting from 14 in 2003 and reaching 6950 in 2021 (figure 2). This indicates a substantial increase in the recognition and impact of the subject over time. The number of citations accelerated as the years progressed. For example, the number of citations more than tripled from 2010 (354) to 2016 (1142), and then it increased dramatically in subsequent years. This non-linear growth suggests that the work gained increasing attention and influence as time went on. The steep increase in the number of citations, especially in the later years, resembles an exponential growth pattern. The increase from 7214 citations in 2020 to 6950 citations in 2021 is a strong indicator of exponential growth. Exponential growth is an indication of widespread recognition of the topic investigated.

Online supplemental table S2 presents the main bibliometric indicators of publications on the the research field by the year of publication.

International research collaboration

In the analysis of core countries contributing to research in the field (table 2 and figure 3), China leads the way with a substantial 445 (25.5%) articles. The USA follows with 410 (23.5%) articles, followed by India (n=265; 15.2%), the UK (n=159; 9.1%) and Saudi Arabia (n=107; 6.1%).

Table 2. A bibliometric analysis of the 20 most collaborated countries.

Productivity and impact					International collaboration
Rank	Country	TP	TC	AC	Rank	Country	TLS
1st	China	445	7207	16.20	1st	USA	407
2nd	USA	410	13 622	33.22	2nd	UK	293
3rd	India	265	5543	20.92	3rd	India	199
4th	UK	159	5201	32.71	4th	Italy	189
5th	Saudi Arabia	107	2503	23.39	5th	Saudi Arabia	166
6th	South Korea	96	2898	30.19	6th	China	155
7th	Italy	82	2417	29.48	7th	Canada	152
8th	Australia	74	2079	28.09	8th	Germany	131
9th	Canada	69	2351	34.07	9th	Spain	107
10th	Germany	62	1376	22.19	10th	Australia	106
11th	Taiwan	61	770	12.62	11th	Netherlands	102
12th	Spain	55	1229	22.35	12th	Switzerland	94
13th	Japan	41	1111	27.10	13th	France	85
13th	Netherlands	41	1628	39.71	14th	South Korea	76
15th	France	36	731	20.31	15th	Cyprus	69
16th	Bangladesh	35	946	27.03	16th	Sweden	61
16th	Malaysia	35	956	27.31	17th	Denmark	59
18th	Switzerland	32	746	23.31	17th	Belgium	55
19th	Egypt	30	680	22.67	19th	Japan	53
20th	Iran	30	462	15.40	20th	Malaysia	49

AC, average citations; TC, total citations; TLS, total link strength; TP, total number of publications.

Figure 3. Network visualisation map for coauthorship countries (international collaboration).

The network visualisation map of countries, which considers countries with a minimum contribution of 10 articles, offers a fascinating glimpse into the dynamics of global research collaboration within a specific field (figure 3). Within this map, 45 countries are represented, and several noteworthy observations emerge. First and foremost, the USA assumes a pivotal role in the map’s landscape. Not only does it boast the highest number of connections, but it also occupies a central position, signifying its substantial research contributions in the field. The prominence of the USA underscores its role as a hub for international research collaboration. One of the most significant findings is the robust research partnership between the USA and India, evident from the thickness of the connection between these two countries. This suggests a high level of engagement and knowledge exchange between researchers based in the USA and their counterparts in India, emphasising the global nature of scientific inquiry. Moreover, the UK and China also emerge as key collaborators, following closely behind India in terms of their connections with the USA. This emphasises the importance of international cooperation, with these countries actively engaging in collaborative research efforts within this specific field. Interestingly, Saudi Arabia stands out on the map with a relatively large node size. This suggests that Saudi Arabia has made substantial contributions to research within the field and likely has its network of collaborations with various countries. Notably, the map highlights a particularly strong research partnership between Saudi Arabia and India. This collaboration underscores the active exchange of knowledge sharing and expertise between researchers and institutions in these two countries, further enriching the global research landscape in this domain.

Most impactful articles on early detection/diagnosis

The top-cited articles on early disease detection using AI encompass a range of crucial healthcare applications. Recurrent neural networks have been shown to be effective in predicting initial heart failure diagnosis by analysing temporal relations in electronic health records.³⁰ Another notable theme focuses on the early diagnosis of Alzheimer’s disease (AD).³¹,³³ Other studies tackle early-stage lung cancer diagnosis by leveraging spectroscopic analysis of circulating exosomes, showcasing the potential of combining deep learning with medical imaging for liquid biopsy.³⁴ MRI biomarkers have also been identified for early AD diagnosis,³⁵ with machine learning distinguishing between AD and mild cognitive impairment stages. These articles collectively demonstrate how AI-driven approaches are revolutionising early disease detection. Another significant theme revolves around the early identification of diabetic retinopathy,^{36 37} a leading cause of vision loss. Researchers have leveraged deep CNNs, achieving remarkable accuracy of 94.5% in classifying retinal images. These studies employ deep learning alongside brain network and clinical data, resulting in improved accuracy compared with traditional classifiers. Multimodal approaches also gain attention, combining imaging, genetic and clinical data to classify patients with AD. Oral cancer detection uses deep learning algorithms applied to hyperspectral patient images, achieving high accuracy, sensitivity and specificity.³⁸ Another notable study focuses on gastric cancer detection, introducing a novel deep convolutional neural network (DCNN) for early diagnosis.³⁹ This DCNN was designed by a system to detect early gastric cancer without blind spots during esophagogastroduodenoscopy.

Most impactful articles on risk assessment

In recent years, AI has been applied to predict the risks associated with NCDs. Several notable studies have emerged, all aimed at improving risk assessment and prediction for conditions related to NCDs. One common theme across these studies is the utilisation of machine learning algorithms to outperform traditional risk prediction methods. These algorithms analyse extensive data sets, including electronic medical records, mammograms and patient information, to develop accurate risk models. For instance, in heart failure cases, deep learning techniques demonstrated superior performance in predicting 30-day readmissions, enabling healthcare teams to focus interventions on high-risk patients.^{40 41} Cardiovascular risk assessment also benefited greatly from machine learning, as it enhanced prediction accuracy by considering complex interactions between risk factors.^{42 43} Moreover, machine learning’s potential extended to breast cancer risk prediction, where deep learning models showed substantial improvements over traditional risk assessments.^{44 45} The application of unsupervised deep learning in breast density segmentation and mammographic risk scoring demonstrated its potential to automate processes and predict breast cancer risk with remarkable accuracy.⁴⁶ Diabetes risk stratification was another area of focus, and researchers optimised machine learning systems to handle missing values and outliers, achieving higher accuracy in risk prediction.⁴⁷ Furthermore, a deep neural network can predict new-onset atrial fibrillation (AF) from the 12-lead ECG in patients with no previous history of AF. This prediction may help identify patients at risk for AF-related strokes.⁴⁸ Furthermore, inflammatory bowel disease risk assessment reached new heights as machine learning harnessed vast data sets and modern techniques to build optimal predictive models.⁴⁹ These models outperformed previous risk prediction methods, offering promising advancements in patient care. In the context of coronary artery disease, machine learning integrated information from coronary CT angiography-derived plaque data, offering enhanced risk stratification compared with conventional methods.⁵⁰ In essence, these studies collectively underscore the transformative power of machine learning and AI in revolutionising the prediction of NCDs.

Emerging research topics

The 10 most recently published articles in the data set included several studies that harnessed the power of AI to advance early detection and risk prediction in the context of NCDs. Among these studies, one focused on patients with AD, using deep learning approaches and CNNs for early diagnosis of AD on MRI images.⁵¹,⁵³ Another study introduced a recently developed deep-learning network to assess oral cancer diagnosis.⁵⁴

Additionally, a study delved into the early diagnosis and classification of autism spectrum disorder (ASD) using diverse machine learning methods.⁵⁵ These methods were employed to identify crucial ASD traits, with the goal of enhancing and automating the diagnostic process.

In the realm of breath analysis, a study sought to achieve early detection of lung cancer.⁵⁶ It used a gas sensor array and deep learning algorithm. Furthermore, a study centred on chronic kidney disease developed novel machine-learning strategies-based prediction models.⁵⁷ The models were developed using a supervised machine-learning algorithm (random forest, support vector machine and decision tree) and feature selection techniques (recursive feature elimination with cross validation and univariate feature selection). This work was crucial for understanding the disease’s progression and recommending appropriate treatments for individuals. Moreover, researchers developed and validated a prediction model based on continuous glucose monitoring (CGM) data to identify a week-to-week risk profile of excessive hypoglycaemia in patients with type 1 diabetes. They found that the prediction models based on real-world CGM data can be used to predict the risk of hypoglycaemia in the forthcoming week. The models showed good performance in both the internal and external validation cohorts.⁵⁸ In the realm of cancer and diagnosis, a study introduced machine learning models, including classical approaches and neural networks, to predict the likelihood of recurrence in patients diagnosed with well-differentiated thyroid cancer. The outcomes demonstrated the models’ efficiency in stratifying recurrence risk among these patients.⁵⁹ Lastly, researchers explored the risk prediction model of heart diseases based on dual-stage stacked machine learning approaches.⁶⁰ The results reveal that using this two-step stacking machine learning approach has the potential to provide accurate and timely diagnosis, supporting the global effort to reduce heart disease-related deaths.

Most frequent author keywords (research hotspots)

By generating a network visualisation map of author keywords, focusing on those appearing at least 16 times, we constructed a map comprising 50 distinct items (figure 4). Within this map, the nodes rendered in the largest size correspond to the most commonly recurring items found within the literature, signifying key research focal points. The following list presents these prominent research hotspots, as visually depicted in the map and based on the number of occurrences and total link strength (TLS) provided by the VOSviewer program:

Figure 4. Network visualisation map of author keywords with a minimum occurrence of 13 times. The map had 57 items. Node size is proportional to frequency of occurrence. Distance between nodes reflects relatedness. ai, artificial intelligence; ann, artificial neural network; cnn, convolutional neural network.

1. Machine learning (615 occurrences, TLS: 790).

2. Deep learning (251 occurrences, TLS: 347).

3. AI in NCD detection (205 occurrences, TLS: 314).

4. CNN (123 occurrences, TLS: 178).

5. Modelling and prediction (79 occurrences, TLS: 141).

6. Risk prediction (71 occurrences, TLS: 139).

7. Data mining and feature selection (56 occurrences, TLS: 92).

8. Classification (52 occurrences, TLS: 93).

9. Neural networks (51 occurrences, TLS: 82).

10. Specific NCDs (ailments): several specific NCDs were explored in the data set:

    • AD (59 occurrences, TLS: 116).

    • Breast cancer (45 occurrences, TLS: 74).

    • Diabetes (43 occurrences, TLS: 80).

    • Heart failure (39 occurrences, TLS: 74).

    • Stroke (34 occurrences, TLS: 58).

    • Type 2 diabetes (31 occurrences, TLS: 51).

    • Atrial fibrillation (26 occurrences, TLS: 49).

    • Diabetic retinopathy (23 occurrences, TLS: 33).

    • Coronary artery disease/myocardial infarction (21 occurrences, TLS: 31).

    • Lung cancer (20 occurrences, TLS: 36).

    • Chronic kidney disease (18 occurrences, TLS: 37).

    • Prostate cancer (16 occurrences, TLS: 27).

    • Osteoporosis (14 occurrences, TLS: 25).

    • Cervical cancer (13 occurrences, TLS: 18).

11. Early diagnosis (43 occurrences, TLS: 77).

12. Risk assessment (39 occurrences, TLS: 61).

13. Early detection (35 occurrences, TLS: 53).

14. Medical imaging (MRI) (28 occurrences, TLS: 69).

15. Prognostic models (24 occurrences, TLS: 46).

16. Healthcare (18 occurrences, TLS: 43).

17. Electronic health records (16 occurrences, TLS: 29).

Discussion

The primary objective of the present bibliometric study is to conduct a comprehensive analysis of the role of AI in the early identification and risk assessment of NCDs. The study aims to emphasise the significance of AI-driven approaches in healthcare transformation, particularly in facilitating early disease detection and risk evaluation. In pursuit of this objective, the study systematically synthesises research trends, identifies key areas of interest, reviews influential articles and explores emerging topics within the realm of AI applications in NCD management. AI-driven early detection holds promise in promoting health equity by ensuring equitable access to timely diagnosis and healthcare services for underserved and marginalised populations.⁶¹ This aligns seamlessly with broader societal objectives centred around fairness in healthcare provision.⁶² The study’s findings carry implications for health policies aimed at supporting AI-driven healthcare initiatives and guiding the judicious allocation of resources. Healthcare organisations can leverage the insights gleaned from this research to strategically allocate resources for the implementation of AI, directing investments towards AI technologies, professional training and data infrastructure based on identified research trends.

To advance the implementation of AI in the management of NCD, future research should explore various avenues. First, rigorous validation of the AI-driven models and methods in real clinical settings is imperative. This validation process, involving close collaborations between AI experts and healthcare practitioners, ensures the reliability and practical applicability of these innovative technologies within the complexities of healthcare delivery. Moreover, fostering interdisciplinary research becomes pivotal. Collaboration among experts spanning diverse fields such as computer science, medicine, epidemiology and ethics is indispensable. This multidisciplinary approach can yield holistic solutions that encompass not only the technical aspects but also ethical and social considerations. The ethical dimensions of AI in healthcare merit profound exploration, especially concerning issues like patient privacy, data security and the potential biases inherent in algorithms.⁶³ The ethical concerns surrounding AI in healthcare primarily stem from algorithmic bias, which can lead to disparities in patient outcomes due to non-diverse training data sets. Data security and patient privacy are also critical, as AI systems rely on vast repositories vulnerable to breaches. Additionally, the increasing autonomy of AI in clinical decision-making raises accountability issues, especially in cases of misdiagnosis, while the opacity of machine learning models—the ‘black box’ problem—poses challenges for transparency and interpretability. Beyond these ethical and privacy concerns, AI’s growing role in diagnostics and healthcare management impacts the doctor–patient relationship and the broader healthcare system, necessitating regulatory frameworks that balance innovation with ethical responsibility. Multidisciplinary collaboration among physicians, data scientists, legal experts, ethicists and policymakers is essential to ensure AI enhances healthcare equity, accessibility and patient-centred care rather than exacerbating disparities or depersonalising medicine. A structured, pragmatic approach is needed to align AI integration with ethical principles, respect patient autonomy and improve medical service quality and efficiency.⁶³

Subsequent research endeavours ought to concentrate on enhancing the accessibility and affordability of AI-driven early detection tools. This becomes particularly crucial in resource-limited settings, where addressing health disparities can significantly impact public health outcomes. For emerging scholars and researchers entering this field, several key implications arise. First, the study underscores the paramount importance of interdisciplinary collaboration. Young scholars should actively seek opportunities to engage across diverse disciplines to harness the full potential of AI in healthcare. Second, an emphasis on data ethics is essential. Young scholars should immerse themselves in discussions and research related to ethical considerations, particularly in the context of data privacy and security within AI-driven healthcare solutions. Lastly, young scholars can make substantial contributions by focusing their research on global health issues. Given the global impact of NCDs, addressing health disparities and ensuring the accessibility of AI-driven solutions worldwide becomes a compelling avenue for impactful research endeavours.

The present study showed much emphasis on AI-enhanced early detection and risk assessment of AD. AD represents one of the most prevalent and debilitating neurodegenerative conditions worldwide, affecting millions of individuals and their families.⁶⁴,⁶⁷ The urgency surrounding Alzheimer’s research is underscored by the profound societal and economic burden it imposes. AD is a global health crisis, with its prevalence steadily rising due to ageing populations. The impact of Alzheimer’s on individuals, families and healthcare systems is profound, necessitating an intense focus on early detection to mitigate its effects. As of now, there is no cure for AD.⁶⁴,⁶⁷ However, early diagnosis can provide an opportunity for interventions and treatments that may slow its progression and improve the quality of life for affected individuals. Delaying its onset or slowing its progression can have a substantial positive impact. Advances in medical imaging, biomarker identification and AI have opened new avenues for early AD detection. These technological advancements have heightened interest in research aimed at identifying reliable and accessible methods for early diagnosis. Early diagnosis of AD aligns with broader public health strategies aimed at preventing or delaying the onset of chronic diseases.

In comparison to other bibliometric analyses exploring the application of AI in medical fields, our study offers a focused examination of AI’s role in NCDs. For instance, Xiong et al conducted a bibliometric analysis on AI in liver cancer, revealing a rapid increase in publications since 2017, with China leading in output but the USA exhibiting higher citation impact.⁶⁸ Similarly, Rahman et al mapped AI research in medical diagnosis in India, highlighting significant international collaborations and identifying prevalent themes such as machine learning and COVID-19.⁶⁹ These studies, along with others focusing on specific diseases or regional research trends,⁷⁰,⁷² provide valuable insight into the evolving landscape of AI in healthcare. Our analysis contributes to this body of knowledge by specifically addressing the application of AI across various NCDs, thereby offering a comprehensive overview that underscores both the advancements and the existing gaps in this critical area of healthcare research.

Training healthcare professionals in AI-driven early diagnosis and risk assessment for NCDs is pivotal in modern healthcare. This entails comprehensive education for doctors, nurses and technicians, covering both AI theory and practical usage. Continuing medical education should incorporate AI to keep them updated. Trust in AI is vital since some may be initially hesitant.⁷³,⁷⁷ Trust-building involves rigorous testing, validation and clinical trials to demonstrate AI’s accuracy. AI should seamlessly integrate into clinical workflows, supporting healthcare providers rather than replacing them, freeing them from data analysis. It enhances evidence-based medicine by uncovering patterns in vast data sets, complementing clinical judgement. Ethical discussions are necessary, addressing privacy, data security and bias.⁷⁸,⁸³ Compliance with healthcare AI regulations is crucial. Effective patient communication about AI’s role is vital. In conclusion, integrating AI into healthcare requires education, trust, ethics and compliance, ensuring its role as a valuable tool for patient care while upholding ethical standards.^{84 85}

Regarding the study’s limitations, several factors deserve consideration. First, the reliance solely on the Scopus database presents a limitation. This singular source, although extensive, may not encompass the entirety of relevant articles in the field, potentially introducing bias into the selection of materials. Additionally, language and publication bias could have influenced the study’s findings. The inclusion criteria restricting articles to the English language might have inadvertently excluded valuable research published in other languages, potentially introducing language-related biases. Moreover, focusing primarily on primary research articles might have omitted valuable insights from comprehensive review articles. Temporal bias emerges as a concern as well. The choice of narrative synthesis over quantitative synthesis represents another limitation. While narrative synthesis provides a qualitative overview, it may restrict the study’s ability to draw statistically significant conclusions, potentially limiting the depth of analysis. A key limitation of bibliometric analysis is its inability to perform in-depth content evaluation, as conducted in systematic reviews. Unlike systematic reviews, which critically assess study methodologies, findings and biases, bibliometric studies primarily focus on quantitative metrics such as publication trends, citation patterns and research productivity. Consequently, this study does not examine the disadvantages, barriers or potential negative implications of AI in NCDs, nor does it explore ethical concerns or privacy protection issues. Furthermore, the absence of content analysis limits the ability to investigate conceptual frameworks, methodological rigour or research gaps in depth. Despite these constraints, bibliometric analysis provides valuable insights into the evolution and impact of research in the field.

Conclusions

In this comprehensive analysis of the role of AI in early identification and risk evaluation of NCDs, we have delved into various facets of this evolving field. The study sought to provide a thorough overview of AI’s current state in NCD detection and risk assessment, identify emerging trends and evaluate the progression of AI techniques over time.

The findings of this bibliometric study have several important implications for the future of healthcare and NCD management. First, AI-driven solutions for early disease detection and risk assessment represent transformative potential in healthcare. The increasing number of citations in recent years indicates growing recognition and impact, suggesting that AI is poised to revolutionise the field. Notably, the study identified key research hotspots, such as machine learning, deep learning and specific NCDs like AD and diabetes. These hotspots are driving innovation in AI applications for healthcare, offering valuable insights into the most promising areas for future research and development. The most impactful articles highlighted in this bibliometric study demonstrate the practical applications of AI in early disease detection and risk assessment. These studies span various NCDs, from AD to cancer and showcase how AI can outperform traditional methods, providing more accurate predictions and diagnoses.

International research collaboration emerged as a significant aspect of this field, with the USA taking a central role and collaborating closely with countries like India, the UK and China. This global collaboration enriches the research landscape and fosters knowledge exchange. Moving forward, it is crucial to validate AI-driven models in real clinical settings and address ethical considerations, patient privacy and data security. Interdisciplinary collaboration between AI experts and healthcare practitioners will be pivotal in ensuring the practical applicability of these technologies.

Additionally, the education and training of healthcare professionals in AI-driven healthcare are paramount, and building trust in AI systems is vital for their successful integration into clinical workflows.

While this study provides a comprehensive overview of AI applications in NCD early detection and risk assessment, several promising directions emerge for future research: future studies should investigate the synergistic potential of combining imaging data (MRI, CT) with omics data (genomics, proteomics) and electronic health records through advanced multimodal AI architectures. There is a critical need for large-scale prospective trials to validate AI models in diverse clinical settings, particularly in low-resource environments where early NCD detection could have a maximal impact. Parallel research is needed to establish robust ethical guidelines addressing data privacy, algorithmic bias and equitable access to AI-powered diagnostic tools. The establishment of standardised benchmarking data sets and shared computational infrastructure could accelerate innovation while facilitating direct comparison between different AI approaches.

Data availability statement

Data are available in a public, open access repository.