Machine Learning in Dementia Research: Trends, Biomarkers, Prediction Goals, and Data Modalities.
Authors
Affiliations (6)
Affiliations (6)
- Department of Emergency Medicine, Taipei Medical University Hospital, Taipei, Taiwan.
- Graduate Institute of Injury Prevention and Control, College of Public Health, Taipei Medical University, Taipei, Taiwan.
- School of Medicine, College of Medicine, Taipei Medical University, Taipei, Taiwan.
- Division of Outcomes and Practice Advancement, Department of Pharmacy Practice, School of Pharmacy and Pharmaceutical Sciences, University At Buffalo, Buffalo, New York, USA.
- Division of Neurosurgery, Department of Surgery, Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, New Taipei City, Taiwan.
- Interdisciplinary Program Kinesiology and Health, Chinese Culture University, Taipei, Taiwan.
Abstract
Machine learning (ML) is increasingly used in dementia research for diagnosis, early detection, prognosis, biomarker characterization, and disease phenotyping. This study aimed to provide a bibliometric and thematic overview of ML applications in dementia, emphasizing biological mechanisms, prediction goals, data modalities, collaborative networks, and emerging computational approaches. A title-field search of the Web of Science Core Collection identified 3,833 records related to ML and dementia. After restricting publications to 2000 onward and excluding non-analytic document types, 3,479 records were included in the bibliometric analysis. A subset of 3,223 original articles was used for thematic analyses. Publication trends, authorship, institutional collaboration, keywords, biomarker themes, ML objectives, data modalities, methods, shared research resources, and large language model (LLM) applications were evaluated. ML-related dementia research increased substantially over time. Diagnostic classification was the predominant ML objective (55.0%), followed by early detection or screening (10.3%), progression or cognitive-decline prediction (6.8%), and biomarker or feature discovery (4.7%). Structural magnetic resonance imaging was the most frequently identified modality (42.6%), followed by clinical or cognitive data (22.4%), positron emission tomography/single photon emission computerized tomography (10.3%), and genetic or genomic data (10.2%). Prominent biological themes included hippocampal measures, amyloid/Aβ, electroencephalogram, tau/p-tau, and cerebrospinal fluid biomarkers. Institutional networks showed distinct international collaboration clusters and frequent use of shared infrastructures. Transformer-based and LLM approaches emerged in recent years. ML research in dementia has expanded beyond diagnostic neuroimaging toward multimodal, biomarker, prognostic, and language model applications, although diagnostic classification and structural imaging remain dominant. Emphasis on biological integration, longitudinal validation, and clinical translation are needed.