Owing to their nanoscale dimensions, well-defined atomic structure, and elevated specific surface area, clusters have emerged as a novel therapeutic platform for neurological disorders. However, efficiently and rationally designing functionalized clusters capable of specifically recognizing and modulating key disease targets remains a major difficulty. The rapid advancement of artificial intelligence (AI) technology offers a revolutionary solution to this bottleneck. By integrating deep learning, generative models, and multi-omics big data, AI can mine vast amounts of biomedical and chemical information with unprecedented speed and precision. It will drive transformative innovations in rational design of clusters, precise modulation of enzymatic activity, and high-throughput screening of therapeutic targets for neurological disorders.

Speech brain–computer interfaces (BCIs) represent an interdisciplinary neural engineering innovation enabling communication rehabilitation for individuals with anarthria or severe dysarthria. By decoding cortical activity into text or synthetic speech via nonmuscular pathways, this technology provides critical communication alternatives for patients with amyotrophic lateral sclerosis, poststroke aphasia, or locked-in syndrome. Since early demonstrations of computer cursor control, significant advancements have been achieved in real-time decoding of limited lexical sets, though challenges persist in system compatibility, decoding accuracy, transmission speed, and ethical governance. This review systematically evaluates state-of-the-art speech decoding models, describes recent technological breakthroughs, and identifies unresolved challenges in clinical translation, while assessing cross-disciplinary applications of speech BCIs in healthcare and assistive technologies.

Colonoscopy underpins colorectal cancer prevention, yet global perspectives on quality-control research are fragmented. We provide a systematic, AI-focused bibliometric overview and map thematic shifts in this field. We retrieved 1390 English-language articles (January 2004–March 2025) from the Web of Science Core Collection, selected for curated citation data and compatibility with bibliometric mapping tools. Relevance was verified manually using prespecified criteria, and duplicates were removed. CiteSpace, VOSviewer, SCImago Graphica, Biblioshiny, and Microsoft Excel were used to analyze co-authorship networks, geographic and institutional productivity, journal influence, keyword evolution, and thematic clusters. Publication output increased steadily, accelerating after 2015. The United States led in volume and citation influence, whereas China ranked among the top five with the fastest post-2015 growth but lower citation density and collaboration centrality. Hotspots shifted from procedural metrics toward AI-assisted polyp detection, optimization of adenoma detection rate, and real-time quality monitoring. Author and institutional networks were moderately centralized with limited international integration. Colonoscopy quality control is moving from conventional, single-parameter indicators to integrated, AI-enabled quality frameworks. Regional imbalances, fragmented collaborations, and gaps in clinical translation persist. Coordinated, multicenter research across regions has the potential to accelerate quality improvement in routine endoscopy.

Stroke often leads to upper limb motor impairments, underscoring the need for precise assessment to guide personalized rehabilitation. Conventional clinical scales are limited by subjectivity and the absence of detailed kinematic analysis. To address this, we propose a novel assessment framework that integrates gamified virtual reality tasks with inertial measurement unit (IMU)–based kinematic analysis, enabling fine-grained and autonomous evaluation of upper limb movements in stroke patients. Specifically, we introduce a region-based motion normalcy index (rMNI) to quantify motor deficits across five spatial regions, offering a more nuanced characterization of movement impairments. Regression models, including elastic net, ridge, and least absolute shrinkage and selection operator regression, were trained on regional rMNI features to predict Fugl–Meyer assessment upper extremity (FMA-UE) scores. Experiments with 12 stroke patients and 8 healthy controls demonstrated strong correlations between rMNI and both FMA-UE total and subscale scores (|r| > 0.70), highlighting the ability of rMNI to spatially resolve motor dysfunction and identify impaired limbs. The best-performing regression model achieved an R² of 0.90 and a Pearson's correlation coefficient of 0.95, indicating excellent predictive validity. These results suggest that the proposed framework is a promising tool for personalized rehabilitation, providing both fine-grained spatial assessment and patient-specific insights.