A multi-database systematic review and topic modeling study has mapped the research landscape of robot-assisted interventions for selected neurodegenerative disorders, showing that the published literature organizes into three distinct publication clusters. Conducted by researchers Xi Ling and Hang Yin and published in BMC Neurology, the study analyzed a final curated corpus of 160 peer-reviewed publications spanning 2016 through 2025 to determine how robotic applications are distributed across clinical domains and how their representation has developed over time.
Robotic systems are increasingly utilized in rehabilitation programs, supportive care, and precision neurosurgical procedures for neurodegenerative conditions. However, evidence captured across comprehensive academic databases spans varied diseases, distinct robotic mechanisms, clinical settings, and divergent outcome frameworks. To provide structural clarity on this diverse literature, the authors designed a prespecified search and analysis protocol to systematically evaluate the knowledge organization and temporal progression of robot-assisted treatment research.
The search strategy gathered records from the Web of Science Core Collection, Scopus, and PubMed on August 31, 2026, targeting articles and reviews published between 2016 and 2025 without language restrictions. The initial query retrieved 1,874 records, which was reduced to 806 entries following deduplication and publication-type filtering for title and abstract screening. Fifty-four records with unresolved eligibility were subsequently reassessed at full text independently by two reviewers with formal adjudication of discordant decisions, leading to the addition of 38 papers and establishing the final corpus of 160 publications.
Methodologically, the study employed non-negative matrix factorization, or NMF, as its primary topic modeling algorithm, while latent Dirichlet allocation, or LDA, served as an algorithmic sensitivity analysis. The researchers also examined publication types, journal venues, author affiliation geography, and annual topic shares. The finalized corpus consisted of 109 original articles and 51 reviews distributed across 121 academic journals, with yearly volume increasing from eight publications in 2016 to 24 in 2020 and 23 in 2025.
The primary NMF model identified three clinically recognizable publication clusters across the analyzed literature. Cognitive, social, and therapeutic care represented the largest share, comprising 65 publications or 40.6 percent of the corpus under topic T2. Motor, gait, and limb rehabilitation constituted the second cluster, designated as topic T1, with 63 publications accounting for 39.4 percent of the total. The remaining 32 publications, or 20.0 percent, formed topic T3, focusing on robot-assisted deep brain stimulation and precision neurosurgery.
The three-topic configuration remained stable across repeated model fits and 80 percent subsampling evaluations. Sensitivity testing across five LDA runs demonstrated moderate correspondence with the primary NMF model, yielding a mean adjusted Rand index of 0.691 with a standard deviation of 0.220, alongside a mean normalized mutual information score of 0.708 with a standard deviation of 0.142. Geographic participation was globally concordant, with country rankings based on full counts and fractional counts showing strong agreement at a Spearman rank correlation coefficient of 0.937.
Longitudinal assessments indicated changing relative shares across all three thematic clusters over the analyzed decade rather than a uniform monotonic trend. The authors emphasized that these publication clusters demonstrate how specific clinical workflows co-occur in the literature but do not establish comparative clinical efficacy among treatments. In addition, the researchers specified that proposed future research priorities represent interpretive conclusions drawn from topic configurations, temporal trajectories, and representative papers rather than direct mathematical outputs of the topic models.
Subject of Research: Knowledge structure and topic modeling of robot-assisted treatments in neurodegenerative disorders
Article Title: Mapping the knowledge structure of robot-assisted treatment in selected neurodegenerative disorders: a multi-database systematic review and topic modeling study using NMF and LDA
Article References: Ling, X., & Yin, H. (2026). Mapping the knowledge structure of robot-assisted treatment in selected neurodegenerative disorders: a multi-database systematic review and topic modeling study using NMF and LDA. BMC Neurology. https://doi.org/10.1186/s12883-026-05512-1
Image Credits: AI Generated
DOI: 10.1186/s12883-026-05512-1
Keywords: neurodegenerative disorders, robot-assisted treatment, topic modeling, non-negative matrix factorization, latent Dirichlet allocation, systematic review, motor rehabilitation, gait rehabilitation, limb rehabilitation, cognitive care, therapeutic care, deep brain stimulation, robot-assisted neurosurgery, BMC Neurology, publication clusters, sensitivity analysis, bibliometric evaluation
News Source: Denise Maddox. (November 12, 2026). Topic Modeling Maps Three Core Clusters in Robot-Assisted Neurodegenerative Disease Research. Scienmag.



