The study, led by Xinyi Wu and Yuelong Yang alongside colleagues at Sun Yat-sen University, Guangdong Provincial People’s Hospital, Philips Healthcare, and Siemens Shenzhen Magnetic Resonance, appeared in Frontiers in Medical Technology. The team systematically searched the Web of Science Core Collection and PubMed for articles on cardiac diffusion MRI published between January 2005 and June 2024. After independent screening by two reviewers, with disagreements resolved by a third, 367 articles met the inclusion criteria. The researchers then applied a suite of bibliometric tools, including CiteSpace 6.2.R6, VOSviewer 1.6.20, and the R package bibliometrix, to construct co-citation networks, collaboration maps, keyword burst analyses, and trend topic visualizations.

The results reveal a field that, while still modest in absolute publication volume, is growing steadily and attracting increasing attention from high-impact journals. Magnetic Resonance in Medicine led productivity with 48 articles and topped the citation rankings with 2,358 total citations. But the presence of Circulation, the Journal of the American College of Cardiology, Circulation Research, and Radiology among the most-cited journals signals that cDMRI findings are resonating well beyond the technical MRI community and into mainstream cardiovascular medicine.

At the heart of the bibliometric landscape sits one dominant concept: cardiac diffusion tensor imaging, or cDTI. It emerged as the most prominent keyword, with 191 occurrences, a total link strength of 280, and a centrality of 0.98, underscoring its pivotal role in the field. The keyword also exhibited the strongest citation burst, a measure of sudden surges in scholarly attention, with a burst strength of 6.8 spanning 2022 to 2024. This recent surge suggests that cDTI has moved from a niche technical pursuit to a central focus of contemporary cardiac imaging research.

Among individual studies, one paper stood out above all others. The 2017 article by Sonia Nielles-Vallespin and colleagues, published in the Journal of the American College of Cardiology, recorded the strongest citation burst at 26.17, accumulating 163 total citations at an average of 20.38 per year. That work validated the ability of in-vivo cDTI to characterize myocardial microstructural dynamics, demonstrating for instance that sheetlet angles increase from approximately 18 degrees in diastole to 65 degrees in systole, and emphasized its clinical relevance for assessing heart disease. A companion study by the same group, implementing an optimized stimulated echo acquisition mode sequence combined with a modified prospective navigator technique and biofeedback, achieved the second-strongest burst at 20.86. Christian T. Stoeck and colleagues earned the third-strongest burst, 15.55, for pioneering second-order motion-compensated spin echo cDTI on a clinical scanner.

The most-cited paper overall, however, belongs to a different tradition. J.D. Bayer and colleagues, writing in Annals of Biomedical Engineering in 2012, ranked first in total citations with 346, citations per year at 26.62, and normalized citations per year at 4.85. Their Laplace–Dirichlet Rule Based algorithm generates cDTI-derived myocardial fiber orientations for computational heart models with high speed and precision, enabling simulations of electrical activation that have become foundational for cardiac electrophysiology research. Ming-Ting Wu’s 2006 study in Circulation, the second most-cited work with 214 citations, used in-vivo cDTI indices including apparent diffusion coefficient, fractional anisotropy, and helix angle to map fiber remodeling after myocardial infarction, correlating microstructural changes with tissue viability and wall motion.

Keyword clustering revealed four coherent thematic pillars within the field. The first concerns sequence design, with terms such as motion compensation, STEAM, spin echo, and free-breathing dominating the conversation. The second addresses image processing, encompassing mean diffusivity, fractional anisotropy, helix angle, image registration, and reconstruction. The third centers on clinical applications across diseases including heart failure, dilated cardiomyopathy, congenital heart disease, ventricular arrhythmias, and ischemia. The fourth links cDMRI to complementary cardiac MRI techniques such as late gadolinium enhancement, T1 mapping, T2 mapping, and strain imaging. Together, these clusters paint a picture of a field advancing simultaneously on technical, computational, and clinical fronts.

Yet the bibliometric analysis also exposed a significant obstacle to clinical translation: the lack of standardization. The diversity of cDMRI sequences, acquisition parameters, and MRI hardware across published studies has compromised the comparability and interpretability of cDTI indices. Factors including sequence type, acquisition resolution, b-values, cardiac phase, field strength, and coil equipment all influence the signal-to-noise ratio of cDMRI images, which in turn affects derived parameters such as ADC, FA, MD, HA, and E2A. The frequent use of in-house post-processing software has further introduced variability. The authors recommend standardized protocols adapted to different hardware systems, a unified analysis tool for cDTI, and large-scale, multi-center, multi-vendor longitudinal studies to establish normal reference ranges and confirm clinical reproducibility.

Perhaps the most striking forward-looking finding concerns deep learning. The trend topic analysis identified it as an emerging research focus since 2022, with a sustained burst strength of 3.93 beginning in 2020 and continuing through 2024. The appeal is clear: low signal-to-noise ratio and prolonged acquisition time remain major limitations of cDTI, and deep learning-based reconstruction and denoising methods offer a path forward. Phipps and colleagues developed a residual deep learning denoising convolutional neural network achieving at least a twofold acceleration for in-vivo cDTI without significant differences in SNR, structural similarity, MD, FA, or helix angle transmurality. Liu and colleagues created FG-Net, which generates high-quality cDTI parameter maps from just six raw cDWI images. Ferreira and colleagues demonstrated that a U-Net approach could reduce cDTI acquisition from approximately 12 breath-holds to just 3, a dramatic improvement in clinical feasibility.

The authors caution that deep learning in cDMRI is not without risks. Bias in reconstructing diffusion tensor maps inevitably exists, and whether such bias is acceptable remains unknown. Evaluation and validation frameworks for deep-learning-based methods have not yet been established, and no cDMRI studies employing deep learning have utilized external test sets to date, a gap the authors attribute to the lack of protocol standardization. They encourage validation of deep-learning-derived cDTI indices against histology where possible and validation in large-scale, multi-center, multi-vendor studies. As the field matures, the convergence of standardized acquisition, robust sequence design, computational electrophysiology modeling, and machine learning acceleration may finally bring cardiac microstructure imaging from the research bench into routine clinical practice.

Subject of Research: Bibliometric analysis of cardiac diffusion magnetic resonance imaging research from 2005 to 2024

Article Title: Future of cardiac diffusion magnetic resonance imaging: a bibliometric analysis

Article References: Wu, X., Yang, Y., Xu, Y., Wu, Z., Liu, H., & Chen, R. (2026). Future of cardiac diffusion magnetic resonance imaging: a bibliometric analysis. Frontiers in Medical Technology, 8, Article 1801878. https://doi.org/10.3389/fmedt.2026.1801878

Image Credits: AI Generated

DOI: 10.3389/fmedt.2026.1801878

Keywords: cardiac diffusion MRI, diffusion tensor imaging, bibliometrics, deep learning, myocardial microstructure, cardiac electrophysiology, STEAM, motion compensation, myocardial infarction, hypertrophic cardiomyopathy, signal-to-noise ratio, sequence design