The Web of Science database is the leading platform for abstract indexing of interdisciplinary scholarly literature and contains more than 12,400 high-impact journals. It is widely used in bibliometric studies and provides important data for academic assessment. The data used in this study come from the Web of Science Core Collection, which includes citation indexes such as the Science Citation Index (SCI). The literature search was conducted on August 1, 2025. The term “TS” refers to “Topic Search”, which allows searching in titles, abstracts and keywords of publications. The search strategy based on previous research was as follows: TS = (“Demetia” OR “Senile dementia” OR “Alzheimer’s disease”) AND TS = (“Artificial intelligence” OR “Machine learning” OR “Computational intelligence” OR “Machine intelligence” OR “Deep learning” OR “Reinforcement learning” OR “Neural networks” OR “Natural language processing” OR “Large language models”) The search was limited to English-language articles and review articles covering publications from January 1, 2005, to December 31, 2024, in order to ensure accuracy and reliability. To ensure accuracy and reliability, the search process was conducted independently by two researchers. Two independent screeners conduct initial and secondary literature screening according to predetermined criteria. Disagreements are first resolved through thorough consultation. If consultation fails to reach a resolution, a third independent reviewer with relevant research background is introduced. The final consensus is reached by integrating the opinions of all three parties to ensure the rigor and effectiveness of the literature inclusion process.

The inclusion criteria were as follows: 1) The study focused on dementia. 2) The study utilized AI-related techniques (deep learning, machine learning, neural networks, natural language processing, etc.).

The exclusion criteria were as follows: 1) The study was not concerned with dementia. 2) The study does not involve AI technology.

After excluding duplicates and unpublished literature, 1443 publications were selected for further analysis. For each publication, the following basic data were collected: title, abstract, country or region, institution, author, keywords, journal and reference.

In this study, we used CiteSpace (v.6.3.R1 64-bit Advanced) and Scimago Graphica (1.0.45) for bibliometric and visual analysis.

CiteSpace, developed by Professor Chao-Mei Chen, is a specialized visual analysis software designed to identify research hotspots in the literature and predict future research trends [16]. Specifically, we used CiteSpace to accomplish the following tasks: institutional co-occurrence analysis, journal biplot overlay, citation burst analysis of references, and keyword clustering and burst detection. This paper references previous similar studies [4], and the parameters used in CiteSpace are as follows: time span: 2005-2024; slice length = 1; thresholds: g index (k = 7), LRF = 2.5, L/N = 10, LBT = 5, e = 1.0.

After performing the initial search, we screened titles and abstracts to confirm article eligibility based on predefined inclusion and exclusion criteria. The manual exclusion criteria were: 1) the study did not focus on AD, and 2) the study did not involve AI-related methods. The search process was conducted independently by two researchers to maximize the accuracy and reliability of the results. Based on these criteria, 1554 references were excluded after reviewing titles and abstracts. Duplicates and unpublished literature were then removed. Finally, 1526 articles were selected for further consideration. The literature search and screening process is shown in Figure 1Figure 1.

By statistically analyzing the number of research papers published in different time periods, we can reveal the research trends and development dynamics in specific fields. Based on the established search strategy and screening process, we successfully collected a total of 1526 research papers related to the application of AI in dementia from the WOS database over the past 20 years (Figure 1Figure 1). We then plotted publication trends over time (Figure 2Figure 2), which clearly illustrates the significant growth of research on AI applications in dementia. Specifically, prior to 2012, research in this area was still in its early stages, with only a handful of papers published. However, between 2012 and 2019, the number of publications began to gradually rise. Notably, after 2019, the field experienced exponential growth, peaking at 365 publications in 2024. Furthermore, as of the search date, these papers have been cited 36,781 times, with an average of 24.1 citations per paper, highlighting the significant impact and academic significance of research in this field.

Figure 1. Flow chart of the literature screen.

Figure 2. Annual publication trends.

A total of 93 countries and regions have published relevant articles in this field. Figure 3Figure 3 provides an overview of the global distribution of these studies, with circle size representing the number of publications per country. The figure shows that only the United States and China have published over 300 articles. Furthermore, from a collaboration perspective, we can clearly observe that the United States and China have collaborated with the majority of countries in this field. Table 1 details the top ten countries in scientific research output over the past 20 years. The table clearly shows that the United States and China have produced a significant volume of research in this field.

Figure 4Figure 4 displays a collaboration network diagram of institutions engaged in research within this field, generated using CiteSpace’s default settings. In this diagram, nodes represent the number of publications by each institution, with larger nodes indicating more articles. Links between nodes illustrate collaborative relationships among institutions. Table 2 lists the top 10 most productive institutions in this field. Harvard University ranks highest in productivity with 41 publications, followed by the Egyptian Knowledge Bank (38 publications). Institutions with high centrality include Harvard University (0.26), Sungkyunkwan University (0.18), and the University of California System (0.11).

A total of 297 authors have contributed to research in this field. Among them, 38

Figure 3. Geographical distribution of collaborations among countries regions.

Table 1. Top 10 countries with most publication.

authors have published three or more articles, as shown in Figure 5Figure 5. Table 3 lists the top 10 authors by number of papers published in this field. In terms of average citations, Au, Rhoda ranks first, followed by Beg, Mirza Faisal and Popuri, Karteek. Collaboration among authors from different countries appears relatively weak. Therefore, authors in this field should strive to strengthen international cooperation and promote broader exchanges with larger networks of researchers.

A total of 267 journals published relevant articles, with 243 journals publishing

Figure 4. Network of institutions related to AI in dementia.

Table 2. Top 10 institutions with most publication.

Figure 5. Network of authors related to AI in dementia.

Table 3. Top 10 authors with most publication.

more than 10 articles, as shown in Figure 6Figure 6. Table 4 lists the top 10 journals ranked by the number of published articles. In terms of citation counts, the top three journals were Alzheimer’s & Dementia, Neuro Image and Journal of Alzheimer’s Disease.

Figure 6. Network of journals related to AI in dementia.

Table 4. Top 10 most productive journals.

Keywords play a crucial role in emphasizing the core content of an article. Analyzing keywords can identify research hotspots and emerging trends in specific fields. In this study, a knowledge graph-based clustering method was adopted, and CiteSpace software was used to analyze keywords. The clustering results indicate that the research field can be divided into 12 clusters, as shown in Figure 7Figure 7. The validity and effectiveness of these clustering results were determined based on the modularity Q and Silhouette S metrics. According to previous studies [17], Q > 0.3 indicates a significant clustering structure, S > 0.5 indicates reasonable clustering results, and S > 0.7 indicates high confidence in the clustering results. The results show that Q = 0.7903 > 0.3 and S = 0.8628 > 0.7, indicating that the keyword clustering results are both significant and reasonable. The top 5 clusters include: #0 alzheimers disease, #1 mild cognitive impairment, #2 deep learning, #3 magnetic resonance imaging, #4 natural language processing.

To clarify the relationship between dementia and AI, we conducted an in-depth analysis of the initially screened keywords, focusing on the top high-frequency keywords (Figure 8Figure 8). Based on the analysis results in Figure 8Figure 8, we compiled the top 50 popular keywords, as detailed in Table 5.

Figure 7. Keyword clustering knowledge graph in the field of AI in dementia.

Figure 8. Network of keywords related to AI in dementia.

Table 5. Top 50 keywords in the field of AI for dementia.

This study conducted a comprehensive review of literature on artificial intelligence in dementia published between 2005 and 2024 using the Web of Science database. Following a rigorous screening process based on inclusion and exclusion criteria, 1526 high-quality papers were identified. These papers, authored by 297 researchers from 170 institutions across 93 countries/regions and published in 267 academic journals, provided a robust data foundation for our analysis.

The volume of scientific literature is often regarded as a primary indicator for assessing developmental dynamics within specific disciplinary fields. This study meticulously analyzed the research trends in AI applications for dementia from 2005 to 2024. Findings reveal a sustained upward trajectory in the field’s literature volume, with 88% of publications occurring within the past five years. This significant growth is largely attributed to the rapid advancement of deep learning (DL). Further analysis reveals a sharp rise in publications since 2019, primarily driven by the introduction and practical application of various novel deep learning frameworks. Currently, AI applications in dementia represent a highly prominent topic in clinical medical research, with substantial research potential and diverse applications anticipated in the future.

A review of the global distribution of artificial intelligence research reveals that the United States holds a prominent position, having published 372 related papers, accounting for 24.4% of the global total. This makes it the country with the highest number of publications in this field. China follows closely behind with 356 papers, accounting for 23.3% of the total. Together, these two nations represent 47.7% of all published papers, underscoring their leadership in the intersecting field of AI and dementia research. Notably, China has experienced the fastest growth in paper output, rising to become the world’s second-largest contributor and establishing a robust collaborative bridge with the field’s leader, the United States. As technology advances and research deepens, an increasing number of countries and researchers are expected to engage in interdisciplinary studies on dementia and artificial intelligence. This will drive progress in academic knowledge and practical applications within the field.

At the institutional level, the United States holds six of the top ten spots in research productivity. Harvard University leads the list with 41 publications, reflecting America’s significant advantage in overall publication volume. Institutions from China, the United Kingdom, South Korea, and others also feature prominently. Harvard University and Sungkyunkwan University exhibit centrality indices of 0.26 and 0.18, respectively, indicating their more pivotal positions within academic exchange networks and broader scope of external collaboration. These findings underscore the importance of strengthening global academic cooperation. Collectively, research activities from the United States, China, and the United Kingdom stand out globally. American institutions, particularly Harvard University, lead in both research output and influence. To elevate their academic standing, institutions worldwide must intensify international collaboration to further advance scientific research.

Analysis of core authors reveals that the top 10 scholars predominantly hail from the United States, highlighting the nation’s outstanding contributions in specific research domains—a finding consistent with prior institutional analyses. Among them, Abuhmed, Tamer, Popuri, Karteek, and El-sappagh, Shaker lead in publication volume. Notably, despite publishing fewer articles, Au, Rhoda ranks first in average citations, demonstrating profound academic influence. However, Figure 5Figure 5 also reveals limitations in current research collaboration, as many teams remain confined to domestic partnerships. This underscores the urgent need to strengthen international cooperation, which is crucial for advancing research progress and innovation in disease diagnosis and treatment.

4.2.1. Research Hotspots

A keyword co-occurrence analysis provides an accurate representation of the core themes, emerging hotspots, and development trends within a specific academic field [18]. After analyzing the frequency of occurrence (Table 5), the most significant keywords linking dementia and AI can be categorized into three themes: 1) Disease Classification and Progression: Alzheimer’s disease, mild cognitive impairment, and dementia. 2) AI and ML Technologies: Machine learning, deep learning, convolutional neural networks, neural networks, transfer learning. 3) Medical imaging and auxiliary diagnosis technologies: MRI, structural MRI, functional connectivity, and computer-aided diagnosis.

Additionally, a timeline visualization based on co-occurring keyword networks (Figure 9Figure 9) illustrates the trajectory of artificial intelligence in dementia research. Initially, the focus was primarily on clinical applications [19]-[22]. However, with the continuous advancement of artificial intelligence technology and innovations in treatment methods, the focus of research has gradually shifted toward emerging fields such as early detection [23]-[26] and prognosis prediction [27][28]. Based on the analysis of the above keywords, two primary research directions have emerged in the fields of dementia and AI: 1) Early identification and diagnosis of dementia; 2) Risk assessment and prediction of disease progression in dementia.

Early Identification and Diagnosis of Dementia: The progression of dementia is complex, involving multiple stages. Individuals diagnosed with MCI have a higher risk of developing dementia, with an annual conversion rate of approximately 15% [29]. Therefore, early and accurate diagnosis is crucial for slowing disease progression and improving treatment outcomes [30]. Initially, researchers were accustomed to using traditional machine learning techniques for dementia classification

Figure 9. Network visualization graph of the cluster timeline view of keywords in the field of AI in dementia.

and diagnosis, typically with healthy individuals as controls [31]. Some have proposed a two-stage classification to distinguish MCI from dementia [32][33]. In recent years, DL models have demonstrated tremendous potential in processing high-dimensional neuroimaging data due to their powerful feature extraction and classification capabilities [34]. Researchers have begun applying DL methods to achieve precise dementia diagnosis and early prediction [35]. Notably, Ávila-Jiménez et al. proposed a deep learning model to identify AD in clinical records [36]. The neural network model they developed demonstrated excellent performance and can serve as an accurate diagnostic aid for AD. In summary, the application of artificial intelligence in dementia research is becoming increasingly sophisticated, providing substantial support for early diagnosis, precision treatment, and prognosis prediction.

Risk assessment and prediction of dementia progression: Research indicates that pathophysiological changes associated with dementia occur long before clinical symptoms emerge, underscoring the necessity of early intervention [37]. In the absence of effective drug treatments, it is crucial to reduce the risk of dementia, delay its onset, or slow its pathological progression [30]. An AI-based risk prediction model offers a novel strategy for dementia management and prevention. For example, You et al. utilized ML to develop a novel dementia prediction model, UKB-DRP, which combines 10 readily accessible predictors [38]. This model demonstrates reliable predictive capability for dementia and AD occurrence over five, ten, or even longer timeframes, enabling the identification of individuals at high risk for dementia within the general population. Li et al. developed a dementia conversion prediction model for mild cognitive impairment that integrates plasma pTau181 and structural imaging features [39], enabling the prediction of dementia-specific pathological changes.

4.2.2. Emerging Trends

Emerging keywords serve as key indicators of cutting-edge research hotspots, offering insights into emerging trends. Notably, “machine learning” is projected to become a critical focus area. Simultaneously, this reflects growing attention toward developing intelligent assessment tools capable of accurately predicting disease progression [40]-[42]. These advancements aim to provide robust data support for developing personalized treatment strategies. In summary, “machine learning” stands as a pivotal keyword in AI research related to dementia, reflecting the continuous evolution and progress of research methodologies in this field. With ongoing technological advancements and deepening research, further breakthroughs are anticipated in early disease detection, accurate diagnosis, effective treatment, predictive accuracy, and improved care and prognosis.

This study has several limitations. First, due to inherent limitations of bibliometric tools and databases, it was not possible to effectively integrate and analyze data from multiple databases or languages. Second, this research focused solely on English-language literature, excluding scholarly work in other languages. Additionally, the literature review spans from 2005 to 2024, potentially limiting consideration of earlier or most recent research. Consequently, some omissions or biases may exist in the data. To enhance the comprehensiveness and accuracy of future research, it is recommended to broaden data sources to include multilingual literature and a wider range of document types. Additionally, potential citation biases, such as self-citation inflating impact metrics and database language biases overlooking non-English works, impose further limitations on bibliometrics, thereby affecting the comprehensiveness and objectivity of results to some extent.