A bibliometric analysis of artificial intelligence applications in heart failure

Introduction

Heart failure (HF) is one of the leading causes of death and disability globally (1). It is primarily categorized into three subtypes: heart failure with reduced ejection fraction [HFrEF; left ventricular ejection fraction (LVEF) ≤40%], heart failure with mildly reduced ejection fraction (HFmrEF; LVEF 41–49%), and heart failure with preserved ejection fraction (HFpEF; LVEF ≥50%). Patients presenting with overt HF symptoms usually have a poor prognosis, with the 5-year survival rate remaining around 50% even during clinically stable periods (2,3). In 2019, the global prevalence of HF was estimated to reach 56.2 million. Restricted by diagnostic limitations and data gaps, the actual disease burden is likely to be further underestimated (4). Therefore, early diagnosis and accurate risk stratification are indispensable for reducing HF-related mortality. Consistent with the global epidemiological trend, China also carries a heavy HF disease burden. A cross-sectional study of community-dwelling older adults in China revealed that the prevalence rates of HF risk and pre-HF were as high as 35.8% and 42.8%, respectively, indicating a substantial population of individuals in the pre-HF phase (5). Furthermore, HF imposes a considerable economic burden: patients are hospitalized an average of 3.3 times per year, with annual medical costs of approximately $4,406.80—an impact that significantly impairs quality of life, work capacity, and family caregiving (6). Despite major advances in HF treatment, substantial residual cardiovascular risk still persists, emphasizing the necessity of early screening and risk assessment among high-risk populations (7). However, early diagnosis of HF poses a significant challenge in elderly populations with multiple comorbidities, who often exhibit atypical or subtle clinical manifestations—such as fatigue, shortness of breath, confusion, or decreased appetite—resulting in a substantial proportion of patients remaining unidentified and untreated in the early stages of the disease. Studies indicate that approximately half of all HF patients have not received a definitive diagnosis, constituting a major barrier to early prevention and treatment (8).

In recent years, the rapid advancement of artificial intelligence (AI) in healthcare has offered new opportunities to address these challenges, demonstrating significant potential in the early screening, risk prediction, diagnostic classification, and prognosis assessment of HF (9). Among these applications, artificial intelligence-enhanced electrocardiography (AI-ECG) has been proven effective in screening for asymptomatic left ventricular dysfunction and identifying HF risk (10). Additionally, an AI-ECG model developed by an international study has demonstrated the ability to diagnose and predict future valvular heart disease, including associated mitral regurgitation, aortic regurgitation, and tricuspid regurgitation (11). Furthermore, genome-wide association studies (GWAS) and electronic health records (EHRs)-based risk scoring systems, which aims to improve HF predictive performance, have gradually become research frontiers (12).

Non-invasive remote patient monitoring (RPM) shows strong potential for reducing HF-related hospitalizations and mortality. The combination of non-invasive sensors, AI, and cardiac telerehabilitation can further unlock the clinical promise of RPM (13). Simultaneously, by integrating clinical and imaging data, AI models can predict arrhythmias causing sudden cardiac death, identify atrial fibrillation and HF risks, thus enabling more comprehensive risk assessment and personalized treatment strategies (14).

Against the backdrop of rapidly accumulating scientific literature, it is critical to systematically analyze the intellectual structure and evolutionary dynamics of AI research in HF (15). Bibliometric analysis methods enable the quantitative evaluation of massive literature and have been widely applied to reveal emerging themes, core knowledge bases, and research gaps (16). On this basis, this study proposes to use bibliometric methods to systematically assess research trends, knowledge foundations, hotspots, and frontier directions in AI applications for HF. These findings aim to clarify the evolutionary trajectory of this field and provide a reliable reference for subsequent academic research and clinical translation. We present this article in accordance with the BIBLIO reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0853/rc).

Methods

Data sources

The Web of Science Core Collection (WoSCC), curated by Clarivate Analytics, is a multidisciplinary and authoritative citation database covering a wide range of high-impact academic journals. It is well recognized for its comprehensive inclusion of citation metadata (17). In the present study, WoSCC was selected as the platform for literature retrieval and data extraction. Notably, literature data downloaded from WoSCC can be directly imported into bibliometric visualization software including CiteSpace and VOSviewer, which guarantees the reliability, accuracy, and efficiency of the subsequent bibliometric analyses (18).

Inclusion and exclusion criteria

Inclusion criteria: (I) studies focused on the intersection of AI and HF; (II) publications released between January 1, 2005, and December 31, 2025; (III) document types included “Article” or “Review”; and (IV) language restricted to English.

Exclusion criteria: literature that did not explicitly address cross-disciplinary research between AI and HF; non-core academic materials such as conference papers, dissertations, scientific achievements, patents, standards, news articles, popular science articles, and correspondence; retracted publications and literature with incomplete metadata.

Search strategy

The search strategy combined Medical Subject Headings (MeSH) and free-text terms, which were grouped into thematic blocks and integrated via Boolean logical operators. Search terms for HF were defined as MeSH1: “Heart Failure”; AI-related terms were organized as MeSH2 through MeSH8: “Artificial Intelligence”, “Machine Learning”, “Deep Learning”, “Neural Networks”, “Computer”, “Natural Language Processing”, “Large Language Models”, and “Reinforcement Machine Learning”. These blocks were combined using the strategy #9 = #2 OR #3 OR #4 OR #5 OR #6 OR #7 OR #8 and #10 = #9 AND #1. The search was limited to the period from January 1, 2005 to December 31, 2025, and restricted to article or review document types published in English. The complete search strategy is provided in the Appendix 1. The search query utilized the “Full Record and Cited References” option and was exported as a “download_XX.txt” file for subsequent processing and analysis.

Data cleaning and standardization

A preliminary screening of literature titles and abstracts was performed, with full-text evaluation conducted when necessary. Prior to the bibliometric analysis, data standardization was carried out, including the manual standardization of elements such as author names and keywords. The two researchers independently completed all data standardization tasks, including the normalization of author names, institutional affiliations, and keywords. Inter-rater reliability was quantitatively assessed using percentage agreement, a widely recognized standard indicator, calculated as (number of identically standardized items ÷ total number of items requiring standardization) × 100%. The initial percentage agreement exceeded 90% across all standardized items. All discrepancies were resolved through discussion or consultation with a third researcher, until full consensus on 100% was reached (19).

Data analysis

All relevant literature was imported into Zotero for management. CiteSpace 6.4.R1 was employed to generate and visualize knowledge maps through citation analysis across multiple dimensions, including cited journals, cited documents, institutions, and countries. The time slicing was set from January 2005 to December 2025, with 1 year per slice. The g-index criterion was set as k =25. Additionally, keyword salience and keyword clustering analyses were conducted to identify research trends and emerging themes in this field (20); cluster labels were extracted using the log-likelihood ratio (LLR) algorithm. Keyword co-occurrence analysis and heatmap generation were carried out using VOSviewer 1.6.18. The full counting method was adopted. To focus on core research hotspots, a minimum keyword occurrence threshold of 70 was applied, and the clustering resolution was set to the default value of 1.00. Statistical analysis and visualization were performed using Microsoft Office Excel 2024, while SCImago Graphica Beta 1.0.51 was used to generate geographic distribution maps of publications by country, with node sizes mapped to publication volume. To assess the robustness of the bibliometric findings, we conducted a parallel analysis using both VOSviewer and CiteSpace to cross-validate the consistency of the identified research hotspots and cluster structures. In addition, the threshold was adjusted the minimum keyword occurrence thresholds within a plausible range, and the core results remained stable.

Results

Literature retrieval and screening results

A preliminary search identified 4,133 records. After excluding 6 supplementary publications and 17 retracted articles, a total of 4,110 records were enrolled in the present study, comprising 3,457 original research articles and 653 review articles. The literature search and screening process was illustrated in Figure 1.

Figure 1 Literature search and screening flowchart.

Publication trend analysis

Publication trends in this field from 2005 to 2025 are depicted in Figure 2. From 2005 to 2018, the number of publications increased gradually, with an average of fewer than 80 papers per year. Since 2019, the publication output grew significantly, reaching 781 in 2024 and further rising to 1,067 in 2025. This accelerated growth trend indicates that the application of AI in HF research has become a global research hotspot.

Figure 2 Annual publication trends in AI research on HF. AI, artificial intelligence; HF, heart failure.

Journal distribution analysis

The enrolled publications spanned 1,158 academic journals. Among them, 674 journals published only one relevant article, 172 published two, 128 published three to four, and 184 published five or more. Frontiers in Cardiovascular Medicine had the highest publication volume, with 138 articles, followed by Scientific Reports, which published 119 articles. Table 1 lists the top 15 journals by publication volume, along with their countries of publication, impact factors, Journal Citation Reports (JCR) quartiles, and article counts. In terms of citation frequency, the three most cited journals were Circulation (2,089 citations), the Journal of the American College of Cardiology (1,760 citations), and the European Heart Journal (1,475 citations). Detailed information is shown in Table 2 and Figure 3A.

Figure 3 The analysis of journal. (A) Journal co-citation distribution map. (B) Overlay of journal distribution dual-chart.

This study adopted a journal dual-map overlay analysis to reveal the distribution characteristics and citation patterns of source and target journals, as presented in Figure 3B. In this visualization, the left map represents the citing journals, while the right map corresponds to the cited journals. Each journal is labeled with its subject category, and colored curves denote citation links extending from the left (citing journals) to the right (cited journals). The overlay reveals two primary citation pathways. The first pathway originates from journals in the field of “Medicine, Medical, Clinical” and leads to those in “Health, Nursing, Medicine” (z =9.387, f =11,675). The second pathway also starts from “Medicine, Medical, Clinical” but cites journals in “Molecular, Biology, Genetics” (z =4.750, f =6,063). This pattern reflects the diffusion of knowledge from basic clinical research into areas such as healthcare, applied technology, and mechanistic studies.

Co-cited literature analysis

Given that citation frequency is a key metric for evaluating the impact of academic research, co-citation analysis was performed on the enrolled publications to further explore citation patterns within this field. The top 10 most frequently cited articles each have more than 60 citations (Table 3). The most highly cited work is the “2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure”, published in the European Heart Journal in 2021, with a total of 222 citations. It is followed by “Screening for Cardiac Contractile Dysfunction Using an Artificial Intelligence-Enabled Electrocardiogram”, published in Nature Medicine in 2019, which was cited 164 times. The results of the citation cluster analysis identified five major clusters: #0 Learning model, #1 Heart disease prediction, #2 Artificial intelligence, #3 Digital solutions, and #4 Cardiovascular medicine (Figure 4A). The co-citation cluster analysis demonstrates the multidisciplinary integration of AI, cardiovascular clinical medicine, and digital health solutions, reflecting the core technological foundations and clinical application orientation of current research in this field.

Figure 4 Comprehensive analysis of references and authors. (A) Reference co-citation cluster analysis. (B) Core author publication volume statistics. (C) Author collaboration network analysis. (D) Author co-citation network map.

Author contribution analysis

A total of 757 researchers participated in AI-related HF research. Of these, 281 authors contributed one publication, 160 authors published two papers, 169 authors contributed three papers, 51 authors published four papers, and 84 authors published between five and 12 papers. Table 4 lists the top 10 core authors ranked by publication output, along with their publication counts, citation numbers and institutional affiliations. The most prolific author, Friedman PA, whose team is primarily affiliated with the Mayo Clinic, has contributed 38 publications to this field. The publication volume statistics and the collaboration network analysis results for core authors in this research area are visualized in Figure 4B,4C.

To further delineate the key academic contributors within this domain, a co-citation network analysis of cited authors was performed, and the metadata of highly cited core authors was systematically compiled. Figure 4D shows four major clusters centered around Attia ZI, Shah SJ, Acharya UR, and Ponikowski P, respectively. Node size is proportional to an author’s citation frequency, with larger nodes indicating higher citation counts and greater academic influence. In this visualization, the size of each node is proportional to the author’s citation frequency—larger nodes signify higher citation counts and greater academic influence. Different colors correspond to independent research clusters, while the thickness of the edges denotes the co-citation intensity between authors. Thicker links imply more frequent co-citation relationships and closer academic collaborations.

Institutional and national analysis

A total of 580 institutions were included in the analysis. The majority of these institutions demonstrated low publication productivity; 162 institutions published only one relevant paper, 57 published two, 60 published three, and 58 published four, whereas 233 institutions contributed from 5 to 55 publications. Among the top 10 most productive institutions, eight were located in the United States and two in the United Kingdom (Table 5). Harvard Medical School ranked first with 133 publications, followed by the Mayo Clinic with 126 publications. In terms of citation impact, the Mayo Clinic received the highest total citation count, amounting to 6,532.

To visualize collaborative relationships among high-productivity institutions, a collaboration network was constructed comprising the top 30 institutions ranked by publication output (Figure 5A). In this network, node size is proportional to the total number of publications, with larger nodes indicating greater research output. Node color intensity also scales with publication volume, deepening to red as productivity increases. The thickness of connecting edges denotes the strength of collaborative ties, with thicker edges indicating more frequent or intensive co-authorship. The resulting collaboration network reveals notably dense inter-institutional partnerships among leading organizations—particularly within the United States—with the Mayo Clinic, Harvard Medical School, and Stanford University emerging as core nodes, highlighting their pivotal position in driving progress in this research field.

Figure 5 Analysis of institutions and countries. (A) Institutional publication volume analysis chart. (B) Geographic distribution map of publications by country.

The global distribution of publications on AI applications in HF is illustrated in Figure 5B. The top five countries/regions by publication volume are the United States (n=1,508), China (n=952), the United Kingdom (n=512), India (n=266), and Italy (n=251). This distribution highlights the global landscape of research in this domain and underscores regional disparities in research strength and publication productivity.

Keyword analysis and research hotspots

Keywords serve as highly condensed carriers of the core themes within academic literature. Analyzing high-frequency keywords and their co-occurrence relationships facilitates the identification of intrinsic connections among research hotspots, evolutionary trends, and thematic structures in a given field. The top 20 high-frequency keywords and their hotspot distributions in AI-related HF research are shown in Figure 6A,6B and Table 6. The three most frequently occurring keywords are “heart failure” (1,735 occurrences), “machine learning” (1,342 occurrences), and “artificial intelligence” (794 occurrences). Keyword clustering analysis using the LLR algorithm identified nine major clusters, among which clusters #0 (machine learning), #1 (echocardiography), and #2 (feature extraction) emerged as the three largest and most representative (Figure 6C). Table 7 presents the silhouette scores, cluster sizes, and mean publication years for all keyword clusters containing more than 50 nodes.

Figure 6 Analysis of keywords. (A) Keyword co-occurrence network map. (B) Keyword hotspot distribution map. (C) Keyword clustering network diagram. (D) Top 25 keyword emergence.

Keyword burst detection is a powerful bibliometric approach to identify research topics that gain significant attention within specific time windows. The keyword “congestive heart failure” demonstrated the longest duration of emergence, spanning from 2005 to 2019, underscoring its sustained relevance as a core research theme within this interdisciplinary domain (Figure 6D). In comparison, the keyword “classification” exhibited the highest burst strength (13.92), with an active period from 2007 to 2021, indicating sustained focus on AI-driven HF diagnosis and risk stratification utilizing classification algorithms. In recent years, keywords such as “diastolic function” and “convolutional neural network” have emerged, highlighting the growing research frontiers in the assessment of diastolic function and AI-enabled medical image analysis for HF.

Discussion

To comprehensively characterize the landscape and developmental trajectory of AI applications in HF research, this study conducted a systematic bibliometric analysis of 4,110 publications focusing on AI in HF. Through a rigorous examination of leading contributing countries/regions, key research institutions, prolific authors, and collaboration patterns within this domain, we not only summarized the fundamental developmental features of this rapidly evolving field but also clarified the clinical value of current research hotspots in the diagnosis, treatment, and management of HF. Furthermore, in accordance with the acquired bibliometric findings, we provided perspectives on potential future research directions and emerging trends that may shape the advancement of AI in clinical practice.

Key findings

Since 2005, the number of AI-related research publications in the field of HF exhibited a sustained upward trend, transitioning into a phase of accelerated growth beginning in 2019. Notably, the research group led by Zachi I. Attia has contributed two landmark studies published in Nature Medicine and The Lancet. The first study demonstrated the efficacy of AI-assisted ECG in screening for asymptomatic left ventricular dysfunction, while the second highlighted its utility in identifying individuals at risk for atrial fibrillation (21,22). Hannun et al. reported in Nature Medicine that deep learning-based ECG interpretation methods can reduce diagnostic errors in arrhythmia detection and improve the efficiency of clinical ECG analysis, thereby offering valuable insights for the early identification of complications associated with HF (23). Collectively, these influential studies substantiate the critical role of AI in facilitating early HF diagnosis and supporting clinical decision-making, while also accelerating the adoption of AI-driven tools in essential areas such as early screening, medical imaging analysis, and risk stratification, underscoring that AI has become a central research focus within the domain of HF.

The 4,110 articles included in this study were distributed across 1,158 journals, with 58.20% of these journals publishing only one relevant paper. The highly dispersed literature distribution highlights the field’s distinct interdisciplinary and emerging nature, underscoring AI’s cross-cutting characteristics and its status as a frontier area in HF research. The inclusion of specialized journals from biomedical engineering and computational science fields reflects this domain’s interdisciplinary nature, which integrates clinical medicine, biosignal processing, and AI algorithms (24,25). For instance, research published in the Journal of Pharmacy and Bioallied Sciences demonstrates that machine learning and deep learning algorithms can achieve early detection, intelligent diagnosis, and prognosis assessment of cardiovascular diseases using multi-source data such as electrocardiograms (26). This approach effectively improves diagnostic precision and precisely identifies high-risk populations, showcasing significant clinical application value. Concurrently, the top 15 core journals by publication volume accounted for approximately 21.75% of the total literature. These core journals are mainly issued in Switzerland, the United Kingdom, and the United States, with most falling into JCR Q1 and Q2 categories, indicating that this research field has been integrated into the mainstream academic system (27,28). The high-quality research published in these journals further confirms that AI has been extensively applied in risk stratification, prognostic forecasting, and intelligent clinical management of HF, establishing itself as an indispensable research direction in the cardiovascular field.

Analysis based on national and institutional collaboration networks offers clearer insights into research hotspots and emerging trends. In terms of publication volume, the United States leads with 1,508 papers, followed by China (952 papers) and the United Kingdom (512 papers). Collectively, these three countries account for 45.06% of the total literature, underscoring their dominant role in the field of AI-based HF research. The United States maintains its leading position due to its strong research infrastructure and extensive international collaborations, whereas China has rapidly emerged as a major contributor through a steadily rising publication output. The sustained leadership of both nations can be attributed to powerful governmental research funding, well-established scientific systems, and broad engagement in international research partnerships.

Author productivity patterns in this field are consistent with Lotka’s Law, a fundamental principle governing scientific output distribution, which underscores the skewed nature of research contribution across scholars. Specifically, authors with only one published work in the domain account for 37.12% of the total 757 researchers, reflecting a large pool of occasional contributors who participate in discrete research projects. In contrast, the core author cohort—comprising 72 individuals (9.51% of all authors) with more than five publications—accounts for nearly 20% of the total literature in the field (29). Among the 10 most prolific authors, seven come from the United States. Core authors such as Friedman PA (38 papers) and Lopez-Jimenez F (28 papers) are affiliated with the Mayo Clinic, around which a dense collaborative network has formed. This pattern indicates that AI research in the field of HF has formed a mature scientific productivity structure, in which a small group of highly productive core authors dominate major academic outputs and serve as the key driving force for the continued advancement of the field.

Analysis of AI’s role in prognosis assessment and image analysis for HF based on clustering results

Keyword co-occurrence and clustering analyses, as core methodologies in bibliometric research, effectively delineate the core themes and dynamic evolutionary characteristics of AI applications in HF studies. Based on the clustering results derived from our analysis, current research hotspots in this domain can be categorized into four distinct knowledge clusters: intelligent signal diagnosis, imaging biomarkers and phenotype identification, risk prediction and prognostic assessment, and digital health management models. These clusters collectively encapsulate the multifaceted applications of AI in HF clinical practice, highlighting the field’s breadth and depth.

Keywords such as feature extraction, convolutional neural networks (CNNs), and electrocardiogram signal classification exhibit high clustering coefficients, directly reflecting AI’s central role in ECG signal analysis and intelligent cardiac disease diagnosis (30). Deep learning models, exemplified by CNNs and Long Short-Term Memory (LSTM) networks, possess the unique capability to effectively extract spatiotemporal characteristics and temporal information from ECG signals, achieving high-precision identification of HF-related complications such as arrhythmias and ventricular premature beats. Such technical capacities supply clinicians with efficient technical tools for early HF screening, addressing a critical unmet need in proactive cardiovascular care (31). Furthermore, improved feature extraction methods—based on discrete wavelet transform and multiple entropy features—can more comprehensively characterize the complex, nonlinear properties of ECG signals. When integrated with traditional machine learning classifiers, these methods facilitate precise identification of cardiac rhythm disorders, including atrial fibrillation and congestive HF, thereby further enhancing diagnostic accuracy and reliability (32). Beyond ECG signals, multi-source physiological signals and imaging data—including heart and lung sounds, and MRI sequences—have been integrated into intelligent diagnostic systems (33). Extending this trend to exercise physiology, machine learning algorithms combined with wavelet transforms have been applied to cardiopulmonary exercise testing (CPET) biosignals, achieving a multi-class classification accuracy of 95% in distinguishing HF patients from those with metabolic syndrome and healthy individuals (34). Specifically, studies have employed hybrid deep networks and multi-point tracking algorithms to achieve automated detection of organic lesions and functional abnormalities closely associated with HF, such as hypertrophic cardiomyopathy and abnormal atrioventricular annular motion, providing clinicians with objective, data-based means for early disease identification (35,36).

Cardiac imaging technologies play an indispensable role in HF research and clinical care, enabling direct visualization of cardiac structural features, functional parameters, and hemodynamic alterations—all of which are critical for identifying HF phenotypes and assessing disease severity (37,38). The integration of AI has further advanced imaging workflows toward automation and enhanced precision. In the domain of echocardiography and cardiac function assessment, deep learning models have been successfully employed to realize efficient, automated quantification of ventricular volume, wall thickness, and LVEF—key parameters for HF diagnosis and severity grading. These models have markedly improved the reliability and reproducibility of early HF screening and functional evaluation (39,40). Extending beyond macroscopic imaging to the microscopic molecular level, the convergence of AI and bioinformatics has introduced innovative methodological frameworks for HF research. Through the integration of transcriptomic data, differential expression analysis, and gene co-expression network analysis—combined with machine learning algorithms such as least absolute shrinkage and selection operator (LASSO) regression, support vector machine-recursive feature elimination (SVM-RFE), and random forest—researchers can efficiently identify feature genes and potential biomarkers associated with HF and its related complications. This facilitates the construction of predictive models with high diagnostic performance, advancing our understanding of the molecular mechanisms underlying HF and its risk prediction (41). Beyond diagnostic accuracy, recent economic evidence suggests when performed by junior staff, AI-enabled echocardiographic assessment of LVEF has the potential to reduce healthcare costs without compromising diagnostic quality (42). Such investigations not only advance the understanding of molecular mechanisms underlying HF and its risk prediction but also offer effective molecular tools and theoretical foundations for early warning and risk stratification in high-risk populations (43).

In HF prognostic evaluation, AI and machine learning technologies demonstrate substantial application potential, providing objective evidence for early warning and clinical prediction of HF progression in patients. Leveraging deep learning and intelligent optimization algorithms, contemporary models can perform efficient feature screening and precise prediction within complex, high-dimensional clinical data, effectively uncovering latent risk associations that may be overlooked by traditional analytical methods (44). Recent studies combining EHR-based screening, deep learning–driven automated interpretation of Digital Imaging and Communications in Medicine (DICOM) echocardiography, and biobank biomarker data underscore the growing potential of multimodal AI pipelines for HF phenotyping (45). Multimodal data fusion—integrating clinical, imaging, and molecular data—further enhances the early recognition of adverse events such as sudden cardiac death, enabling timely clinical intervention and improving patient outcomes (46). Regarding clinical data application and risk stratification, machine learning algorithms excel at integrating multi-source clinical information, including demographic data, laboratory results, and imaging findings (47,48). These algorithms can construct readmission and mortality risk prediction models for large populations, and enable precise risk stratification for heterogeneous HF phenotypes, guiding personalized treatment. Additionally, AI facilitates long-term home monitoring through non-invasive modalities such as vital signs tracking and voice detection, enabling remote assessment of disease progression and early identification of decompensation (49,50).

HF, as a chronic progressive disease, necessitates long-term follow-up monitoring, for which AI technology offers digital and remote solutions. For non-invasive continuous monitoring, researchers have developed the ECG Cross-modal Network (ECGX-Net), a framework designed for deep cross-modal feature learning. By integrating single-lead ECG data from wearable devices with transthoracic bioimpedance measurements and applying transfer learning, this framework achieves high-accuracy prediction (94% precision) of acute decompensated HF (51). For remote screening, heartbeat signals captured by smartphone sensors, combined with attention-based deep residual neural networks, enable precise atrial fibrillation detection (micro F1 score 0.88) (52). Beyond remote monitoring, large language models such as ChatGPT have also shown promise in HF patient education and health information delivery, achieving an accuracy of 78% to 98% in answering related questions (53). These studies demonstrate that AI can empower ubiquitous devices, such as wearables and smartphones, to provide real-time alerts for HF decompensation and arrhythmias in out-of-hospital settings. In parallel, the integration of IoT technologies into cardiac rehabilitation for HF has become a prominent research hotspot (54). This offers crucial technical support for establishing a comprehensive remote management system covering the entire disease course.

Future development directions and challenges based on keyword burst analysis

Keyword burst analysis is a powerful bibliometric tool that identifies sharp rises in keyword popularity over specific time periods, enabling detection of emerging research frontiers and reflecting temporal shifts in scientific research hotspots. Among the identified keywords, “congestive heart failure” emerged as the dominant core term with the highest burst intensity, occupying the central focus of research. Multiple studies have centered on this core disease entity, leveraging clinical data and machine learning models to predict mortality risk in HF patients and enable early warning of cardiovascular complications such as atrial fibrillation (55,56). Additionally, wearable devices have been integrated to facilitate remote monitoring and health management of congestive episodes (57), while AI-based image analysis technologies have been employed to establish efficient, user-friendly screening systems for cardiac structural abnormalities—collectively providing a multidimensional research foundation for disease diagnosis and treatment (58). Concurrently, “pattern recognition” has progressively become a key technology supporting disease feature extraction and objective identification, with applications spanning multiple dimensions of HF research. This is evidenced by its applications in multiple dimensions: enabling automatic classification and recognition of arrhythmias and HF through ECG signal processing (59), facilitating precise assessment of left ventricular systolic patterns based on radiomics features (60), and supporting molecular-level exploration of characteristic genes and prognostic targets (61). These technological advances collectively drive the research evolution from clinical phenomenon description toward deep, data-based pattern mining—marking a critical shift in the field’s focus. “Support vector machine”, as a representative term for classic machine learning algorithms, has long been prominent, further demonstrating its multifaceted role from feature identification to model construction. This algorithm enables multimodal feature screening and long-term risk prediction (62), exhibiting outstanding classification and early warning performance within multi-algorithm fusion systems, providing solid technical guarantee for HF risk stratification and early intervention (63).

In recent years, the emergence intensity of keywords such as “deep learning”, “prognosis”, “care”, and “risk stratification” has continued to increase, indicating that research in this field is advancing toward more sophisticated network structures, more refined prognostic assessments, and closer alignment with real-world clinical decision support (64). In summary, future research must prioritize the development of diverse, rigorously validated, and transparent AI models. Overcoming implementation barriers through standardized evaluation frameworks, prospective clinical verification, and multicenter collaborative efforts will be critical to ensuring that AI tools achieve both clinical applicability and ethical rationality. Ultimately, such advancements will enable realization of early screening, precise classification, and personalized treatment for, clinical translation efforts, and interdisciplinary collaboration in this critical field—convert technological innovations into better clinical prognosis.

Research limitations

This study has certain limitations: first, the bibliometric data were exclusively sourced from the WoSCC database, with the exclusion of other mainstream academic databases. This may have led to the omission of some high-quality non-English literature, thereby introducing potential literature retrieval bias that could impact the comprehensiveness of our findings. Second, the article types were confined to “Article” and “Review”, with the exclusion of other publication formats such as editorials, letters, and case reports. This restriction may have impacted the comprehensive representation of research activities and perspectives in this rapidly evolving field.

Conclusions

The application of AI in the field of HF has advanced rapidly in recent years, forming a diversified research system characterized by a focus on disease diagnosis, prognostic assessment, and mechanistic exploration. The United States and China stand out as pivotal contributors to this domain, leading global scientific output and innovative research progress. Current research priorities are centered on machine learning applications and medical image analysis, whereas future directions are expected to emphasize the interpretability of AI models, multimodal data fusion, and the standardization of clinical implementation. Through a systematic bibliometric analysis, this study summarizes the evolutionary path, core characteristics, and emerging research frontiers of AI applications in HF, thereby offering a foundational reference for future investigations, clinical translation efforts, and interdisciplinary collaboration in this critical field.

Acknowledgments

We extend our heartfelt gratitude to the physicians and nurses in the Department of Cardiovascular Medicine for their unwavering support and great contribution to this project.

Footnote

Reporting Checklist: The authors have completed the BIBLIO reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0853/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0853/prf

Funding: This work was supported by Key Special Project of Artificial and Health Science and Technology Department of Jilin Province (project No. 20240304085SF).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0853/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. van Riet EE, Hoes AW, Wagenaar KP, et al. Epidemiology of heart failure: the prevalence of heart failure and ventricular dysfunction in older adults over time. A systematic review. Eur J Heart Fail 2016;18:242-52.
2. Greene SJ, Butler J, Fonarow GC. Contextualizing Risk Among Patients With Heart Failure. JAMA 2021;326:2261-2. [Crossref] [PubMed]
3. Mericskay M, Zuurbier CJ, Heather LC, et al. Cardiac intermediary metabolism in heart failure: substrate use, signalling roles and therapeutic targets. Nat Rev Cardiol 2025;22:704-27. [Crossref] [PubMed]
4. Martin SS, Aday AW, Almarzooq ZI, et al. 2024 Heart Disease and Stroke Statistics: A Report of US and Global Data From the American Heart Association. Circulation 2024;149:e347-913. [Crossref] [PubMed]
5. Cai A, Zheng C, Qiu J, et al. Prevalence of heart failure stages in the general population and implications for heart failure prevention: reports from the China Hypertension Survey 2012-15. Eur J Prev Cardiol 2023;30:1391-400. [Crossref] [PubMed]
6. Wang H, Chai K, Du M, et al. Prevalence and Incidence of Heart Failure Among Urban Patients in China: A National Population-Based Analysis. Circ Heart Fail 2021;14:e008406. [Crossref] [PubMed]
7. Vaduganathan M, Claggett BL, Inciardi RM, et al. Estimating the Benefits of Combination Medical Therapy in Heart Failure With Mildly Reduced and Preserved Ejection Fraction. Circulation 2022;145:1741-3. [Crossref] [PubMed]
8. Barber K, Bernhardt L, McCann GP, et al. Developing core indicators for identifying people at risk of delayed heart failure diagnosis. BMC Prim Care 2025;26:316. [Crossref] [PubMed]
9. Jankauskas SS, Varzideh F, Kansakar U, et al. Artificial Intelligence in Cardiovascular Medicine: A Giant Step in Personalized Medicine? J Pers Med 2026;16:192. [Crossref] [PubMed]
10. Dhingra LS, Aminorroaya A, Sangha V, et al. Heart failure risk stratification using artificial intelligence applied to electrocardiogram images: a multinational study. Eur Heart J 2025;46:1044-53. [Crossref] [PubMed]
11. Liang Y, Sau A, Zeidaabadi B, et al. Artificial intelligence-enhanced electrocardiography to predict regurgitant valvular heart diseases: an international study. Eur Heart J 2025;46:4823-37. [Crossref] [PubMed]
12. Wu KH, Wolford BN, Du J, et al. Integrating large scale genetic and clinical information to predict cases of heart failure. Commun Med (Lond) 2025;5:493. [Crossref] [PubMed]
13. De Lathauwer ILJ, Nieuwenhuys WW, Hafkamp F, et al. Remote patient monitoring in heart failure: A comprehensive meta-analysis of effective programme components for hospitalization and mortality reduction. Eur J Heart Fail 2025;27:1670-85. [Crossref] [PubMed]
14. Ordine L, Canciello G, Borrelli F, et al. Artificial intelligence-driven electrocardiography: Innovations in hypertrophic cardiomyopathy management. Trends Cardiovasc Med 2025;35:126-34. [Crossref] [PubMed]
15. De la Vega Hernández IM, Urdaneta AS, Carayannis E. Global bibliometric mapping of the frontier of knowledge in the field of artificial intelligence for the period 1990-2019. Artif Intell Rev 2023;56:1699-729. [Crossref] [PubMed]
16. Wu Z, Chen S, Wang Y, et al. Current perspectives and trend of computer-aided drug design: a review and bibliometric analysis. Int J Surg 2024;110:3848-78. [Crossref] [PubMed]
17. Shao H, Kim G, Li Q, et al. Web of Science-Based Green Infrastructure: A Bibliometric Analysis in CiteSpace. Land (Basel) 2021;10:711. [Crossref] [PubMed]
18. Zhang X, Yi K, Xu JG, et al. Application of three-dimensional printing in cardiovascular diseases: a bibliometric analysis. Int J Surg 2024;110:1068-78. [Crossref] [PubMed]
19. McHugh ML. Interrater reliability: the kappa statistic. Biochem Med (Zagreb) 2012;22:276-82.
20. Yang C, Liu H, Feng X, et al. Research hotspots and frontiers of neoadjuvant therapy in triple-negative breast cancer: a bibliometric analysis of publications between 2002 and 2023. Int J Surg 2024;110:4976-92. [Crossref] [PubMed]
21. Attia ZI, Kapa S, Lopez-Jimenez F, et al. Screening for cardiac contractile dysfunction using an artificial intelligence-enabled electrocardiogram. Nat Med 2019;25:70-4. [Crossref] [PubMed]
22. Attia ZI, Noseworthy PA, Lopez-Jimenez F, et al. An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction. Lancet 2019;394:861-7. [Crossref] [PubMed]
23. Hannun AY, Rajpurkar P, Haghpanahi M, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med 2019;25:65-9. [Crossref] [PubMed]
24. Zhang Y, Hao X, Ma J. EF-Gated 3D Capsule Networks with Spatially Decoupled Routing for Clinical Phenotyping of Heart Failure Subtypes in Cardiac MRI. J Imaging Inform Med 2025; Epub ahead of print. [Crossref]
25. Epstein RH, Jean YK, Dudaryk R, et al. Natural Language Mapping of Electrocardiogram Interpretations to a Standardized Ontology. Methods Inf Med 2021;60:104-9. [Crossref] [PubMed]
26. Ahmad A, Ahmad S, Ahmad R, et al. Artificial Intelligence in Cardiovascular Diagnosis: Innovations and Impact on Disease Screenings. J Pharm Bioallied Sci 2025;17:S1900-3. [Crossref] [PubMed]
27. Lee SM, Kim T, Shin M, et al. Deep learning algorithm for detection of acute heart failure using standard ECG waveforms. Eur Heart J Digit Health 2026;7:ztaf132. [Crossref] [PubMed]
28. Ayano YM, Schwenker F, Dufera BD, et al. Interpretable Machine Learning Techniques in ECG-Based Heart Disease Classification: A Systematic Review. Diagnostics (Basel) 2022;13:111. [Crossref] [PubMed]
29. Kawamura M, Thomas CD, Tsurumoto A, et al. Lotka’s law and productivity index of authors in a scientific journal. J Oral Sci 2000;42:75-8. [Crossref] [PubMed]
30. Katal N, Garg H, Sharma B. A comparative analysis of CNNs and LSTMs for ECG-based diagnosis of arrythmia and congestive heart failure. Comput Methods Biomech Biomed Engin 2026;29:1608-36. [Crossref] [PubMed]
31. Ullah H, Heyat MBB, Akhtar F, et al. An Automatic Premature Ventricular Contraction Recognition System Based on Imbalanced Dataset and Pre-Trained Residual Network Using Transfer Learning on ECG Signal. Diagnostics (Basel) 2022;13:87. [Crossref] [PubMed]
32. Fuadah YN, Lim KM. Optimal Classification of Atrial Fibrillation and Congestive Heart Failure Using Machine Learning. Front Physiol 2021;12:761013. [Crossref] [PubMed]
33. Santoso RM, Huang W, Wee S, et al. Streamlining heart failure patient care with machine learning of thoracic cavity sound data. World J Cardiol 2025;17:109992. [Crossref] [PubMed]
34. Pinheiro RF, Fonseca-Pinto R. On the development of diagnostic support algorithms based on CPET biosignals data via machine learning and wavelets. PeerJ Comput Sci 2025;11:e2474. [Crossref] [PubMed]
35. V Y. T M. Convolutional Neural Network Fused With Recurrent Network for ECG-Based Detection of Hypertrophic Cardiomyopathy. Catheter Cardiovasc Interv 2025;106:2123-46. [Crossref] [PubMed]
36. Leng S, Chen J, Yang X, et al. MPTN: A video-based multi-point tracking network for atrioventricular junction detection and tracking in cardiovascular magnetic resonance imaging. Comput Methods Programs Biomed 2025;272:109048. [Crossref] [PubMed]
37. Petmezas G, Papageorgiou VE, Vassilikos V, et al. Automated HFrEF Diagnosis Using an Optimized TimeSformer Model in Echocardiography. J Imaging Inform Med 2025; Epub ahead of print. [Crossref]
38. Zhao Y, Pang G, Sun Z, et al. Block Matching Based Speckle Tracking Echocardiography: Clinical Applications and Research Outlook in a Deep Learning Context. J Imaging Inform Med 2025; Epub ahead of print. [Crossref]
39. AlRahimi JS. Multimodal diagnostic imaging for early mitral valve disease: integration of current and emerging modalities-a narrative review. Cardiovasc Diagn Ther 2025;15:1212-26. [Crossref] [PubMed]
40. Liu W, Wang F, Huang Q, et al. MFB-CBRNN: A Hybrid Network for MI Detection Using 12-Lead ECGs. IEEE J Biomed Health Inform 2020;24:503-14. [Crossref] [PubMed]
41. Wu Z, Wu Z, Sun H, et al. Identification of early warning biomarkers for type 4 cardio-renal syndrome based on bioinformatics analysis and secreted proteins. Sci Rep 2025;15:35071. [Crossref] [PubMed]
42. Kaushik A, Senanayake S, Kularatna S, et al. AI task-shifting for echocardiographic LVEF assessment in Singapore: an economic evaluation. ESC Heart Fail 2026;13:xvag069. [Crossref] [PubMed]
43. Chen S, Zhang C, Yu Y. Identification of arachidonic acid metabolism-related diagnostic markers in heart failure based on bioinformatics analysis and machine learning. Front Cardiovasc Med 2025;12:1625064. [Crossref] [PubMed]
44. Nayak MSS, Syed H. Improved Inception-Capsule deep learning model with enhanced feature selection for early prediction of heart disease. Sci Rep 2025;15:32847. [Crossref] [PubMed]
45. Oo MM, Gao C, Cole C, et al. Artificial intelligence-assisted automated heart failure detection and classification from electronic health records. ESC Heart Fail 2024;11:2769-77. [Crossref] [PubMed]
46. Telangore H, Azad V, Sharma M, et al. Early prediction of sudden cardiac death using multimodal fusion of ECG Features extracted from Hilbert-Huang and wavelet transforms with explainable vision transformer and CNN models. Comput Methods Programs Biomed 2024;257:108455. [Crossref] [PubMed]
47. Zheng B, Liang T, Mei J, et al. Prediction of 90 day readmission in heart failure with preserved ejection fraction by interpretable machine learning. ESC Heart Fail 2024;11:4267-76. [Crossref] [PubMed]
48. Fine NM, Kalmady SV, Sun W, et al. Machine Learning For Risk Prediction After Heart Failure Emergency Department Visit or Hospital Admission Using Administrative Health Data. PLOS Digit Health 2024;3:e0000636. [Crossref] [PubMed]
49. Kerexeta J, Larburu N, Escolar V, et al. Prediction and Analysis of Heart Failure Decompensation Events Based on Telemonitored Data and Artificial Intelligence Methods. J Cardiovasc Dev Dis 2023;10:48. [Crossref] [PubMed]
50. Koscova Z, Rad AB, Nasiri S, et al. From sleep patterns to heart rhythm: Predicting atrial fibrillation from overnight polysomnograms. J Electrocardiol 2024;86:153759. [Crossref] [PubMed]
51. Pan X, Wang C, Yu Y, et al. Deep cross-modal feature learning applied to predict acutely decompensated heart failure using in-home collected electrocardiography and transthoracic bioimpedance. Artif Intell Med 2023;140:102548. [Crossref] [PubMed]
52. Mehrang S, Jafari Tadi M, Knuutila T, et al. End-to-end sensor fusion and classification of atrial fibrillation using deep neural networks and smartphone mechanocardiography. Physiol Meas 2022;
53. Doyle RS, Hartnett J, Temperley HC, et al. ChatGPT Applications in Heart Failure: Patient Education, Readability Enhancement, and Clinical Utility. J Cardiovasc Dev Dis 2025;12:422. [Crossref] [PubMed]
54. Klein A, Pinheiro RF, Fonseca-Pinto R. Integrating Internet of Things into cardiac rehabilitation for heart failure: a review of emerging technologies. Front Med (Lausanne) 2025;12:1737523. [Crossref] [PubMed]
55. Askarinejad A, Bucci T, Asgharzadeh N, et al. Machine learning-based prediction of atrial fibrillation in patients with atrial high-rate episodes. Eur J Clin Invest 2026;56:e70121. [Crossref] [PubMed]
56. Yu L, Chen H, Zhang J, et al. Triglyceride-glucose index and mortality in congestive heart failure with diabetes: a machine learning predictive model. Front Endocrinol (Lausanne) 2025;16:1675152. [Crossref] [PubMed]
57. Hartmann B, Hartmann NU, Brandts J, et al. Congestion assessment using a self-supervised contrastive learning-derived risk index in patients with congestive heart failure (CONAN): protocol and design of a prospective cohort study. Eur Heart J Digit Health 2025;6:1076-83. [Crossref] [PubMed]
58. Ra J, Shin H, Park C, et al. AI-based measurement of cardiothoracic ratio in chest X-rays and prediction of echocardiographic congestive heart failure. Int J Cardiol Heart Vasc 2025;58:101678. [Crossref] [PubMed]
59. Tekin H, Kaya Y. A new approach for heart disease detection using Motif transform-based CWT’s time-frequency images with DenseNet deep transfer learning methods. Biomed Tech (Berl) 2024;69:407-17. [Crossref] [PubMed]
60. Sabouri M, Hajianfar G, Hosseini Z, et al. Myocardial Perfusion SPECT Imaging Radiomic Features and Machine Learning Algorithms for Cardiac Contractile Pattern Recognition. J Digit Imaging 2023;36:497-509. [Crossref] [PubMed]
61. Xu J, Yang Y. Integrated Gene Expression Profiling Analysis Reveals Potential Molecular Mechanisms and Candidate Biomarkers for Early Risk Stratification and Prediction of STEMI and Post-STEMI Heart Failure Patients. Front Cardiovasc Med 2021;8:736497. [Crossref] [PubMed]
62. Park JU, Urtnasan E, Kim SH, et al. A Prediction Model of Incident Cardiovascular Disease in Patients with Sleep-Disordered Breathing. Diagnostics (Basel) 2021;11:2212. [Crossref] [PubMed]
63. Shah A, Ahirrao S, Pandya S, et al. Smart Cardiac Framework for an Early Detection of Cardiac Arrest Condition and Risk. Front Public Health 2021;9:762303. [Crossref] [PubMed]
64. Boeringer T, Pardo M, Craig CJ, et al. G protein-coupled receptor digital twins for precision and personalized medicine. Comput Struct Biotechnol J 2025;28:538-51. [Crossref] [PubMed]