3D semantic mapping of surface geological features

Evaluation of Large Language Models for Mapping Dietary Data to Food Databases.

Medical knowledge accumulation and clinical practice form a closed loop, yet enabling effective cooperation between the two elements, namely autonomously distilling updated knowledge from dynamic data to guide practice, remains challenging, especially in the emergency department (ED). To overcome this, we developed an autonomous AI agent that integrates established medical knowledge graphs with dynamic clinical data into a hybrid graph of over 800,000 nodes. Using large language models (LLMs) for knowledge extraction and semantic mapping, the system dynamically selects the most relevant graph to power specialized tools for ED recognition, prediction, and decision-making. The agent achieves average improvements over state-of-the-art baselines of 23.13% in ED triage, 13.05% in drug-drug interaction detection, 1.58% in readmission prediction, and 5.47% in medication recommendation, demonstrating superior performance across all task categories. This demonstrates an effective framework for synergizing established medical knowledge and dynamic clinical data in emergency care.

Knowledge base question answering (KBQA) aims to answer natural language questions by reasoning over large-scale structured knowledge bases (KBs). Among existing approaches, semantic parsing-based methods have emerged as a mainstream solution, where large language models (LLMs) are employed to translate questions into structured graph queries such as logical forms (LFs). However, this paradigm faces two critical challenges: (1) The complex semantic mapping and graph retrieval operations render direct one-shot LF generation difficult; (2) LLMs suffer from inherent hallucination issues, generating semantically plausible-seeming but factually incorrect or invalid LFs, which are non-executable. To address these challenges, this paper proposes HP-Corr , a novel framework that integrates H op-wise P lanning with iterative explainable self- Corr ection for faithful knowledge reasoning. Specifically, the system utilizes a fine-tuned open-source LLM for query planning and explainable self-correction. The query planner generates reasoning paths hop-by-hop, while an explainable self-correction provides hop-wise feedback, enabling interpretable path editing based on existing reasoning paths and retrieved KB knowledge. By introducing the dual-module cooperative architecture, our system performs iterative plan-then-correct to refine query paths progressively, ensuring answer reliability and LFs executability. Experimental results demonstrate significant improvements, with our approach achieving higher accuracy while substantially reducing the search space, particularly in complex multi-hop KBQA scenarios.

Background/Objectives: Continuous quality improvement (CQI) requires timely, patient-centered evidence on how people experience healthcare delivery. Structured surveys provide important benchmarks, but their predetermined items may miss emerging or system-specific concerns. This study assesses whether unsolicited online patient reviews can serve as a scalable patient-experience data source for identifying hospital service quality priorities across contrasting healthcare systems. Methods: We analyzed 8247 Google Maps hospital reviews posted in 2024, including 5007 Chinese-language reviews from 24 Taiwanese medical centers and 3240 English-language reviews from 21 large U.S. referral hospitals. Separate language-specific preprocessing pipelines and Latent Dirichlet Allocation (LDA) topic models identified patient-salient service quality dimensions in each country. Cross-lingual semantic mapping then distinguished universal dimensions from system-specific concerns, and star-rating differences across semantically equivalent dimensions were compared. Results: Seven service quality dimensions emerged in each country: five were cross-nationally shared (emergency care, positive care experience, professional medical team, administrative process, and inpatient/treatment care), and each system had two system-specific dimensions. Taiwanese reviews foregrounded service attitude and facility/environment quality, while U.S. reviews foregrounded billing/insurance and clinic systems/access. Ratings for emergency care and administrative process were consistently low across both systems, whereas ratings for the professional medical team were substantially higher in U.S. reviews. Conclusions: Online patient reviews can complement formal patient-experience instruments by revealing actionable CQI priorities that are both universal and context dependent. Emergency care and administrative efficiency represent shared improvement needs across both systems. System-specific interventions include interpersonal training and infrastructure investment in high-utilization single-payer settings, and billing transparency and care coordination in fragmented multi-payer systems. Institutional structures appear to play a more prominent role than cultural factors in shaping which service quality dimensions emerge, though both forces contribute. Established frameworks may inadequately capture system-specific patient concerns.

The article examines the pedagogical effectiveness of interactive methods in teaching monosemous and polysemous words in native language lessons. Vocabulary teaching is one of the central components of linguistic education because pupils’ ability to understand, interpret and use words correctly directly influences their communicative competence, reading comprehension and written expression. Monosemous words, which have one lexical meaning, are usually easier for learners to identify and apply, whereas polysemous words require deeper semantic awareness because their meanings change according to context, speech situation and stylistic function. The study analyzes how interactive methods such as contextual analysis, semantic mapping, problem-based tasks, role-play, group discussion and didactic games help pupils distinguish between single-meaning and multiple-meaning words. The article is structured according to IMRAD requirements and presents methodological approaches, expected learning outcomes and didactic advantages of interactive vocabulary teaching. It is argued that interactive methods transform vocabulary learning from mechanical memorization into conscious semantic analysis and communicative practice.

Visual Simultaneous Localization and Mapping (vSLAM) is fundamental to enabling robotic mobile manipulation-i.e., the seamless integration of navigation, perception, and dexterous interaction with objects in unstructured environments. Yet current vSLAM research largely lacks a principled, task-oriented framework for map classification, resulting in suboptimal map representations that hinder robustness and efficiency in dynamic indoor settings. To bridge this gap, we propose a purpose-driven taxonomy of vSLAM maps specifically designed for mobile manipulation tasks. This taxonomy comprises four complementary categories: geometric 3D maps, semantic maps, object-level maps, and hybrid maps-each distinguished by its representational granularity, functional scope, and suitability for downstream manipulation primitives. We provide a systematic comparative analysis of their construction pipelines, underlying technical assumptions, and real-world deployment contexts, evaluating them rigorously across three critical dimensions: environmental adaptability, pose estimation accuracy, and real-time computational feasibility. Finally, we synthesize key limitations in existing approaches and identify concrete, high-impact directions for future work-including tight coupling between mapping semantics and manipulation affordances, and scalable learning-based map fusion.

Artificial Intelligence (AI) offers potential to support and empower nurses, yet its development depends on the availability of high-quality, standardized data. Nursing data are often fragmented, unstructured, and semantically inconsistent, hindering their secondary use. This work aims to harmonize heterogeneous nursing data from hospitals and nursing homes to create an AI-ready data foundation. Following a generic harmonization process, example datasets from two institutions and software systems were extracted and mapped to the Observational Medical Outcomes Partnership Common Data Model (OMOP CDM). Using open-source tools, we performed dataset specification, vocabulary identification, coverage analysis, semantic and structural mapping, and initial ETL implementation. A core dataset was defined, covering key data elements such as demographics, vital signs, and medication. Initial test transfers demonstrated the feasibility of mapping and integrating nursing data, though complex nursing constructs such as care plans and assessments remain challenging. This study establishes a structured methodological approach for cross-institutional nursing data harmonization and lays the groundwork for the development of future AI applications in nursing.

Semantic annotation of study metadata elements with concept codes from medical terminologies is essential in order to make biomedical data FAIR (Findable, Accessible, Interoperable, and Reusable) and to enable cross-border data exchange as envisioned within the forthcoming European Health Data Space (EHDS). Existing annotation tools are often detached from everyday data-capture workflows, limiting their practical uptake by researchers and data managers. REDCap Ontology annotations Made Easy (ROME) addresses this gap by embedding ontology-based annotation directly into REDCap's Online Designer. REDCap is a widely utilized electronic data collection framework in clinical and epidemiological research. As an external REDCap module, ROME allows users to import minimal datasets, search large ontologies such as SNOMED CT, map answer options, and harmonize projects through reusable semantic mappings. Metadata and annotations are stored in structured JSON to ensure machine-readability and interoperability. Future developments include dedicated interfaces for major ontologies, best-practice frameworks for managing minimal datasets, and community-driven harmonization efforts. ROME thus represents a concrete step towards bridging routine data capture and FAIR, EHDS-aligned semantic data stewardship by embedding ontology annotation directly into the design workflow.

Mapping local clinical concepts to standardized terminologies such as SNOMED CT is essential for semantic interoperability and large-scale research, but manual mapping is labor-intensive and difficult to scale. We report a preliminary evaluation of a hybrid approach combining retrieval-augmented generation (RAG) and a large language model (LLM) to support the mapping of hospital datalake concepts to SNOMED CT. Using a dataset of 2,768 concepts annotated by the SIMED across five semantic categories (organisms, healthcare locations, laboratory samples, allergies, and clinical assessments), the pipeline consistently outperformed retrieval-only baselines. For organism concepts, strict accuracy reached 92.8% (weighted 98.0%, top-3 96.7%), while allergy and assessment concepts achieved 84.7% and 72.1% strict accuracy, respectively. Healthcare location and laboratory sample concepts remained more challenging, with strict accuracies of 57.5% and 57.1%, and weighted accuracy varied strongly depending on frequency distributions. Overall, strict accuracy was 72.8%, weighted accuracy 76.1%, and top-3 accuracy 83.1%. These findings suggest that combining RAG with LLM-based reasoning can reduce expert workload while maintaining accuracy, representing a step toward scalable, AI-assisted semantic interoperability in healthcare. The approach relies on a retrieval-augmented generation pipeline using embedding-based candidate retrieval followed by LLM-based disambiguation (GPT-4.1-mini), and is evaluated against a retrieval-only baseline. By improving semantic consistency in clinical data lakes, it indirectly supports interoperable access to patient-level health data.

While Open Set Semantic Mapping and 3D Semantic Scene Graphs (3DSSGs) are established paradigms in robotic perception, deploying them effectively to support high-level reasoning in large-scale, real-world environments remains a significant challenge. Most existing approaches decouple perception from representation, treating the scene graph as a derivative layer generated post hoc. This limits both consistency and scalability. In contrast, we propose a mapping architecture where the 3DSSG serves as the foundational backend, acting as the primary knowledge representation for the entire mapping process. Our approach leverages prior work on incremental scene graph prediction to infer and update the graph structure in real-time as the environment is explored. This ensures that the map remains topologically consistent and computationally efficient, even during extended operations in large-scale settings. By maintaining an explicit, spatially grounded representation that supports both flat and hierarchical topologies, we bridge the gap between sub-symbolic raw sensor data and high-level symbolic reasoning. Consequently, this provides a stable, verifiable structure that knowledge-driven frameworks, ranging from knowledge graphs and ontologies to Large Language Models (LLMs), can directly exploit, enabling agents to operate with enhanced interpretability, trustworthiness, and alignment to human concepts.

Nursing diagnoses and interventions are essential components of clinical documentation and patient-centered care, yet nursing data in German-speaking healthcare settings are commonly documented using local terminologies. This paper aims to analyze translation- and modeling-related challenges when mapping German nursing diagnoses to SNOMED CT. Nursing diagnoses from the DiZiMa® catalog were translated and mapped to SNOMED CT using a structured semantic mapping approach. Two large language models (ChatGPT and Microsoft Copilot) were used in parallel to support translation, guided by established principles of scientific translation. While 98.6% of the diagnoses could be mapped to SNOMED CT, 27.2% showed semantic precision loss, particularly for risk diagnoses and context-dependent nursing concepts, often requiring postcoordination, or remaining unmapped (1.4%). Semantic interoperability of nursing data requires more than direct translation or simple 1:1 mapping, highlighting the need for nursing-specific modeling strategies.

This study evaluated the compatibility of a Medical Informatics Initiative (MII) rehabilitation portal with the German HL7 FHIR nursing discharge standard (PIO) for interoperability between acute care and rehabilitation. Using ISO/TR 12300-based semantic mapping, 995 PIO items were analysed through seven equivalence categories: (0) not represented, (1) identical structured, (2) free-text only, (3) more detailed in PIO, (4a) informal convention-based, (4b) clearly unstructured, and (5) indeterminate. Only 21.61% of the items were represented in a structured way in PIO and the rehabilitation portal, while the majority remained unstructured or less detailed (69.95%). A small number of items (6.83%) could not be represented in the portal at all (category 0). The findings show that most nursing-relevant information in the rehabilitation portal was either unstructured or less detailed compared to the HL7 FHIR PIO standard. Only about one fifth of items were available in a fully structured and standardised format. To improve digital interoperability and support high-quality care discharges and transfers, further development of structured fields is recommended.

Obesity is commonly viewed as a reversible condition primarily driven by excess body weight. Increasing evidence, however, suggests that adipose tissue may undergo persistent immunometabolic and structural alterations that do not fully revert after weight loss, raising the hypothesis of a durable biological imprint that could hinder long-term remission. Whether such persistence is consistently reflected across the biomedical literature remains uncertain. We performed an AI-driven semantic analysis of a construct-anchored corpus of obesity-related publications retrieved from PubMed and Scopus using transformer-based biomedical embeddings (PubMedBERT). Unsupervised density-based clustering (HDBSCAN) identified coherent semantic regions, and Uniform Manifold Approximation and Projection (UMAP) enabled visualization. Core macro-domains were selected using predefined quantitative criteria (cluster stability, temporal persistence, and semantic coherence) and independently evaluated by two blinded experts. Interpretability was assessed through stratified human validation, quantifying inter-rater agreement (Cohen's κ) and AI label acceptance rates. An exportable curated core corpus of mapped publications was generated to support downstream focused screening and structured synthesis. Three mutually exclusive yet highly coherent macro-domains emerged: (1) inflammatory adipose biology, (2) adipose remodeling and chronic dysfunction, and (3) stress-triggered immune persistence. Despite document-level exclusivity, these domains showed exceptionally high semantic similarity (pairwise cosine similarity > 0.97), indicating a shared conceptual core. The semantic architecture of the corpus is consistent with obesogenic memory conceptualized as biological hysteresis in adipose tissue, although not constituting mechanistic proof. The curated corpus provides a structured foundation for subsequent conventional evidence synthesis.

This project leverages the pix2pix image-to-image translation framework to generate realistic building facades from semantic label maps using the CMP Facade dataset. The primary objective was to reproduce the original conditional GAN architecture and validate its performance on the facade-generation task. Beyond model reproduction, we developed a lightweight web-based interface using a standard JavaScript canvas that allows users to draw block-level layout sketches. These sketches are then processed by the trained model to generate corresponding facade images, providing a simpler and faster method for prototyping architectural facades compared to traditional manual sketching techniques. The interactive workflow preserves the structural and textural fidelity of the generated buildings, while making facade visualization more accessible and user-friendly. Experimental results demonstrate that the reproduced model effectively captures building features, and the interface offers a practical tool for rapid architectural concept development and visualization. Overall, this work presents an application-focused replication of pix2pix that bridges the gap between theoretical model reproduction and practical, interactive design workflows, highlighting its potential for supporting early-stage architectural prototyping and creative experimentation.

Artificial intelligence in psychiatry: A global perspective on research status, trends and clinical applications.

Multi-session perception-aware coverage path planning for active semantic SLAM and automatic change detection

As building regulations grow in complexity and digital design workflows become more integrated, the need for automated compliance checking (ACC) is intensifying. This review combines PRISMA with AI-assisted semantic retrieval and concept mapping, using Boolean and Deep Search to identify 88 peer-reviewed studies and broaden coverage across disciplines. The dual approach balances methodological rigour with the discovery capacity needed to surface studies missed by conventional keyword searches, enabling lifecycle-oriented synthesis. The paper synthesises recent advances across rule-based, ontology-driven, and AI-enhanced ACC systems, tracing a shift toward more flexible, lifecycle-aware compliance frameworks. It also introduces a conceptual map that visualise ACC processes across five lifecycle stages. Persistent barriers include interoperability gaps, limited post-construction support, and challenges in large-scale rule formalisation. Findings indicate the growing need for hybrid tools that support re-checking and traceability. The paper outlines future directions for transparent, adaptive, and jurisdiction-sensitive ACC systems in design automation and regulatory practice. • ACC systems are systematically reviewed using PRISMA and Deep Search. • A conceptual map illustrates an integrated ACC framework across the lifecycle. • Early rule-based ACC has evolved into AI, LLM, and ontology-driven systems. • Future directions focus on interoperability, re-checking, and rule scalability.

The amygdala is critical for social cognition and undergoes profound damage in frontotemporal dementia (FTD). While its atrophy is well documented, changes in its structural connectivity and their behavioral relevance remain unclear. Using fixel-based analysis and tractography, we examined amygdala connectivity in patients with behavioral variant FTD(bvFTD; n=21), semantic dementia (SD; n=19), progressive non-fluent aphasia (PNFA; n=18), and 28 controls. Associations with empathy and emotion recognition were explored using partial correlations. BvFTD and SD showed marked degeneration of amygdala-associated tracts, while PNFA exhibited subtle left temporal changes. Tractography revealed reduced amygdala connectivity with regions supporting memory, visual, language, semantic, and motor functions, to varying degrees across subtypes. Social cognitive deficits were correlated with amygdala-cerebellum connectivity in bvFTD and with amygdala-hippocampus connectivity in SD. These findings were the first to demonstrate subtype-specific patterns of amygdala white matter alteration and their relevance to social cognitive symptoms in FTD.

EOAC-LLM: An LLM-driven event ontology automatic construction system