Explore the vast range of titles available.

Automated travel document recognition is a key technology for digital identity verification. However, robust extraction of structured information from images captured in unconstrained conditions remains challenging due to perspective distortion, background clutter, motion blur, and heterogeneous lighting that often degrade the performance of the systems. The paper proposes a modular pipeline for automated travel document segmentation and data extraction that integrates instance segmentation, perspective rectification, optical character recognition, and rule-based field parsing. In order to avoid the use of sensitive personal data, the segmentation model is trained exclusively on a synthetic dataset generated in Blender that comprises 2500 annotated images with diverse variations in lighting, viewpoint, blur, and background. The experimental results demonstrate strong generalization from synthetic to real data with 99.50% mAP50, 99.22% mAP50-95, 90% character-level Optical Character Recognition (OCR), and 90% MRZ field extraction accuracy on synthetic data, and 88% MRZ extraction accuracy on a dataset with real documents.

In recent years, the automatic recognition of invoices has emerged as a critical component in document digitization, particularly within the financial sector, where efficient data extraction from structured documents is essential. This paper introduces an advanced method for automating invoice recognition by combining cutting-edge image preprocessing, OCR pipeline techniques, and post-processing strategies. The process begins with a robust image preprocessing stage that includes scaling, binarization, noise reduction, and skew correction based on Hough transform; learning-based angle regressors may offer adaptive alternatives for noisy or multi-aligned invoices, all of which enhance the quality of the input images, ensuring optimal performance during OCR. The core of the system is an OCR pipeline that integrates state-of-the-art text detection and recognition algorithms, effectively handling various challenges such as multi-oriented and densely packed text regions. After the initial recognition, a series of post-processing steps, including error correction, field validation, and structured data formatting, are applied to ensure that the extracted text is accurate and conforms to predefined invoice formats. Extensive experiments conducted on a dataset comprising 6349 images, including both electronic and paper invoices, demonstrate that the proposed approach achieves high recognition accuracy and reliable performance. These results highlight the system’s effectiveness in automating invoice recognition, confirming its applicability in real-world financial applications where precision and efficiency are paramount. The method provides a comprehensive solution for improving invoice digitization and streamlining financial workflows.

Premise Recently, plant science has seen transformative advances in scalable data collection for sequence and chemical data. These large datasets, combined with machine learning, have demonstrated that conducting plant metabolic research on large scales yields remarkable insights. A key next step in increasing scale has been revealed with the advent of accessible large language models, which, even in their early stages, can distill structured data from the literature. This brings us closer to creating specialized databases that consolidate virtually all published knowledge on a topic. Methods Here, we first test different combinations of prompt engineering techniques and language models in the identification of validated enzyme–product pairs. Next, we evaluate the application of automated prompt engineering and retrieval‐augmented generation to identify compound–species associations. Finally, we build and determine the accuracy of a multimodal language model–based pipeline that transcribes images of tables into machine‐readable formats. Results When tuned for each specific task, these methods perform with high (80–90%) or modest (50%) accuracies for enzyme–product pair identification and table image transcription, but with lower false‐negative rates than previous methods (decreasing from 55% to 40%) for compound–species pair identification. Discussion We enumerate several suggestions for researchers working with language models, among which is the importance of the user's domain‐specific expertise and knowledge.

Free-text surgical pathology reports hinder automated cancer registry entry and secondary analytics. This study introduces a clinically governed schema layer for interoperability, testing whether a locally-deployable Large Language Model (LLM) pipeline can deliver robust registry-grade extraction across institutions. We developed a College of American Pathologists (CAP)-aligned clinical ontology encompassing 10 cancer types, 192 per-organ scalar fields, key biomarkers, and nested structures for lymph nodes and margins. Encoded via Declarative Self-improving Python (DSPy) signatures with grammar-constrained decoding using DSPy v3.2.1, this model-agnostic pipeline was benchmarked on 893 internal reports against a pathologist-adjudicated gold standard. External validation utilized 242 The Cancer Genome Atlas (TCGA) reports. Hardware feasibility was confirmed on a single 48-gigabyte (GB) Graphics Processing Unit (GPU), ensuring suitability for privacy-preserving on-premises deployment. Using the gpt-oss-20b model, the framework achieved 92.0% macro-mean exact-match accuracy on internal data, demonstrating near-perfect run-to-run reliability. Critical prognostic indicators, including breast estrogen receptor/progesterone receptor (ER/PR) (98.7%) and margin positivity (>93%), maintained high fidelity. On the external TCGA cohort, accuracy was 77.5%, rising to 88.0% after excluding structurally silent fields absent in older narratives. Operationally, the model processed reports in 40-70 s, optimally balancing speed and accuracy. This schema-first abstraction layer successfully decouples clinical logic from specific Artificial Intelligence (AI) models. By reliably transforming narrative reports into machine-readable structures, it establishes a portable privacy-preserving foundation for automated cancer surveillance, institutional data reuse, and future multimodal clinical systems.

Evidence syntheses are valuable sources of robust and transparent knowledge that can identify gaps in research and inform evidence-based decision making. However, the process of synthesis is time consuming and costly. We investigate a new AI-based method that uses a large-language model (LLM) grounded in ontologies (i.e. structured machine-interpretable glossaries of domain terminology) to extract information from a set of 80 articles on coastal wetland restoration outcomes. We evaluated this method by comparing human-extracted data with data extracted by OntoGPT — a Python package that combines an LLM with ontologies to extract structured information. We found that OntoGPT achieved 65% average agreement with human reviewers but varied based on information type requested for extraction. The highest agreement scores were found when extracting standardized information, and lower agreement scores were found for study-specific and interpretation-heavy information. Precision and recall — two common measurements of artificial intelligence performance — were 58% and 57%. Our results highlight the potential for LLMs to save some labour in the evidence synthesis process but highlight core challenges (e.g., complex information; subjective judgments) where further development is needed. While LLMs cannot replace human reviewers, they have the potential to assist in data extraction.

Objectives Embedding systematic, structured data extraction within electronic health records (EHR) is vital for improved real‐time insights into care delivery. This study evaluates the feasibility of using large language models (LLMs) to extract structured advance care planning (ACP) information from unstructured Goals of Care (GoC) clinical notes in the EHR. Materials and Methods A sample of 100 de‐identified GoC notes was manually annotated by clinicians across four ACP categories: Patient Priorities, Code Status, Decision Maker, and Documentation. Two LLMs (Mistral 24.07 and LLaMA 3.1) were prompted to extract structured outputs without domain‐specific fine‐tuning. Model outputs were compared to human annotations using cosine similarity of BioBERT embeddings. Results Mistral 24.07 achieved high semantic similarity in Code Status (0.814), Documentation (0.781), and Patient Priorities (0.770), but lower alignment in Decision Maker (0.609). Conclusions LLMs can effectively extract structured ACP information, particularly in well‐documented categories, suggesting potential for scalable, data‐driven feedback loops that improve the provision of care. However, accuracy challenges remain, and further refinement is needed for nuanced qualitative content categories.

Electronic health records (EHRs) consist of both structured data (eg, diagnostic codes) and unstructured data (eg, clinical notes). It is commonly believed that unstructured clinical narratives provide more comprehensive information. However, this assumption lacks large-scale validation and direct validation methods. This study aims to quantitatively compare the information in structured and unstructured EHR data and directly validate whether unstructured data offers more extensive information across a patient population. We analyzed both structured and unstructured data from patient records and visits in a large Dutch primary care EHR database between January 2021 and January 2024. Clinical concepts were identified from free-text notes using an extraction framework tailored for Dutch and compared with concepts from structured data. Concept embeddings were generated to measure semantic similarity between structured and extracted concepts through cosine similarity. A similarity threshold was systematically determined via annotated matches and minimized weighted Gini impurity. We then quantified the concept overlap between structured and unstructured data across various concept domains and patient populations. In a population of 1.8 million patients, only 13% of extracted concepts from patient records and 7% from individual visits had similar structured counterparts. Conversely, 42% of structured concepts in records and 25% in visits had similar matches in unstructured data. Condition concepts had the highest overlap, followed by measurements and drug concepts. Subpopulation visits, such as those with chronic conditions or psychological disorders, showed different proportions of data overlap, indicating varied reliance on structured versus unstructured data across clinical contexts. Our study demonstrates the feasibility of quantifying the information difference between structured and unstructured data, showing that the unstructured data provides important additional information in the studied database and populations. The annotated concept matches are made publicly available for the clinical natural language processing community. Despite some limitations, our proposed methodology proves versatile, and its application can lead to more robust and insightful observational clinical research.

Basketball is a popular sport worldwide, and many researchers have utilized various machine learning models to predict the outcome of basketball games. However, prior research has primarily focused on traditional machine learning models. Furthermore, models that rely on vector inputs tend to ignore the intricate interactions between teams and the spatial structure of the league. Therefore, this study aimed to apply graph neural networks to basketball game outcome prediction, by transforming structured data into unstructured graphs, to represent the interactions between teams in the 2012–2018 NBA season dataset. Initially, the study used a homogeneous network and undirected graph to build a team representation graph. The constructed graph was fed into a graph convolutional network, which yielded an average success rate of 66.90% in predicting the outcome of games. To improve the prediction success rate, feature extraction based on the random forest algorithm was combined with the model. The fused model yielded the best results, and the prediction accuracy was improved to 71.54%. Additionally, the study compared the results of the developed model with previous studies and the baseline model. Our proposed method considers the spatial structure of teams and the interaction between teams, resulting in superior performance in basketball game outcome prediction. The results of this study provide valuable insights for basketball performance prediction research.

Knowledge extraction is meant by acquiring relevant information from the unstructured document in natural language and representing them in a structured form. Enormous information in various domains, including agriculture, is available in the natural language from several resources. The knowledge needs to be represented in a structured format to understand and process by a machine for automating various applications. This paper reviews different computational approaches like rule-based and learning-based methods and explores the various techniques, features, tools, datasets, and evaluation metrics adopted for knowledge extraction from the most relevant literature.

Manually converting unstructured text pathology reports into structured pathology reports is very time-consuming and prone to errors. This study demonstrates the transformative potential of generative AI in automating the analysis of free-text pathology reports. Employing the ChatGPT Large Language Model within a Streamlit web application, we automated the extraction and structuring of information from 33 unstructured breast cancer pathology reports from Taipei Medical University Hospital. Achieving a 99.61% accuracy rate, the AI system notably reduced the processing time compared to traditional methods. This not only underscores the efficacy of AI in converting unstructured medical text into structured data but also highlights its potential to enhance the efficiency and reliability of medical text analysis. However, this study is limited to breast cancer pathology reports and was conducted using data obtained from hospitals associated with a single institution. In the future, we plan to expand the scope of this research to include pathology reports for other cancer types incrementally and conduct external validation to further substantiate the robustness and generalizability of the proposed system. Through this technological integration, we aimed to substantiate the capabilities of generative AI in improving both the speed and reliability of data processing. The outcomes of this study affirm that generative AI can significantly transform the handling of pathology reports, promising substantial advancements in biomedical research by facilitating the structured analysis of complex medical data.