Domain Experts Research Articles

Factors influencing librarian adoptions of ChatGPT technology for entrepreneurial support

PurposeLibraries, both public and academic, have expanded their roles to become crucial hubs of innovation and business support. To cater to the emerging needs of the patrons, including entrepreneurs, libraries must digitally transform to innovate their business support services. This research aims to empirically validate the factors influencing the adoption of generative AI technologies (or switching intentions), particularly ChatGPT, among librarians providing business support services. The study explores the multifaceted factors that shape librarians’ decisions to adopt these technologies or consider alternative solutions.Design/methodology/approachThe research study employed purposive or convenience sampling to select librarians actively supporting businesses and with some experience in generative AI technologies. After pilot testing the questionnaire, it was refined and administered via Google Forms, utilizing a Likert scale for responses. Informed consent was obtained from participants, with strict privacy protocols in place. Data analysis was conducted using partial least squares structural equation modeling (PLS-SEM) with SmartPLS 4 software, adhering to rigorous quality assurance protocols.FindingsThe research empirically validated key factors influencing the adoption or switching intentions toward generative AI technologies, particularly ChatGPT, among librarians supporting entrepreneurs. Identified factors encompass social influence, domain expertise, familiarity with technology, system quality, training and support, interaction convenience, anthropomorphism, perceptions of usefulness, ease of use and enjoyment. These perceptions influence librarians’ emotional responses to technology and affect their decisions regarding adoption or considering alternative options. Notably, demographic variables – age, gender, years of experience in business support and education level – do not significantly influence switching intentions, underscoring inclusivity in innovation.Practical implicationsThe study offers practical implications for various stakeholders, including librarians, patrons, technology providers and policymakers. Librarians can benefit from fostering interdisciplinary collaborations and co-creation initiatives to strengthen social networks, enhance domain expertise, evaluate technology quality and customize training support. Technology providers have the opportunity to improve user experience by enhancing interaction convenience and anthropomorphism. Policymakers can support innovation within library settings through funding initiatives and changes in employment conditions. These recommendations aim to enable the seamless integration of generative AI technologies in libraries while fostering stakeholder collaboration to drive innovation and address evolving user needs.Originality/valueThis research enriches AI technology adoption theories within library contexts by providing empirical evidence and insights into the multifaceted dynamics influencing adoption decisions among librarians. It enhances existing theories by specifically focusing on generative AI adoption and its implications for library business support services. Despite the growing interest in AI technologies within library settings, empirical research on the factors influencing librarian decisions regarding generative AI adoption has been lacking. By addressing this gap, the study contributes to the fields of library and information sciences, entrepreneurship, technology management and innovation management, offering specificity and depth to our understanding of user decision-making processes.

Beyond minutiae: inferring missing details from global structure in fingerprints

Visual inference involves using prior knowledge and contextual cues to make educated guesses about incomplete or ambiguous information. This study explores the role of visual inference as a function of expertise in the context of fingerprint examination, where professional examiners need to determine whether two fingerprints were left by the same person, or not, often based on limited or impoverished visual information. We compare expert and novice performance on two tasks: inferring the missing details of a print at an artificial blank spot (Experiment 1) and identifying the missing surrounds of a print given only a small fragment of visual detail (Experiment 2). We hypothesized that experts would demonstrate superior performance by leveraging their extensive experience with global fingerprint patterns. Consistent with our predictions, we found that while both experts and novices performed above chance, experts consistently outperformed novices. These findings suggest that expertise in fingerprint examination involves a heightened sensitivity to gist, or global image properties within a print, enabling experts to make more accurate inferences about missing details. These results align with prior research on perceptual expertise in other expert domains, such as radiology, and extend our understanding of scene and face recognition to fingerprint examination. Our findings show that expertise emerges from an ability to combine local and global visual information—experts skillfully process both the fine details and overall patterns in fingerprints. This research provides insight into how perceptual expertise supports accurate visual discrimination in a high-stakes, real-world task with broader implications for theoretical models of visual cognition.

Enhancing user protection by identifying phishing attacks using Explainable AI

In today's world, artificial intelligence (AI) is solving complex problems for various domains, and cyber security is one of them. A phishing attack is one of the most important cyber security attacks. Researchers have found that AI can detect phishing attacks much better than humans but do not give the reasons that were important in detecting it, which could be very useful to domain experts and users. So, we are using explainable artificial intelligence (XAI) to understand the reasons for Artificial Intelligence-based phishing attack detection. We have used a standard phishing attack dataset from Kaggle and applied four different AI techniques to detect the phishing attack and then applied SHAP on it to understand the reasons for Artificial Intelligence-based phishing attack detection.

Large language models: Tools for new environmental decision-making.

This study represents the first exploration of Large Language Models (LLMs) in environmental decision-making, examining their potential benefits and limitations. To address environmental issues, we propose and compare two generalizable frameworks: an LLMs-assisted framework that leverages LLMs to augment human expertise and coding in traditional decision workflows, and an LLMs-driven framework that aims to automate optimization. Through a water engineering case study focusing on PFAS control, where Environmental Fluid Dynamics Code (EFDC) was used for water environment simulation, we illustrate how to instantiate these frameworks and assess their performance. The case study reveals generalizable insights about these frameworks. Results indicate that both frameworks can contribute to environmental decision-making optimization to varying degrees, though their applicability differs significantly when facing complex decision scenarios. The LLMs-assisted framework, which effectively regulates flow rates and achieves higher PFAS interception, demonstrates how AI can enhance human decision-making while preserving the essential role of domain expertise and professional judgment. In contrast, the LLMs-driven framework faces challenges in handling complex parameter optimization tasks due to constraints such as context window and maximum output length. The findings emphasize the advantages of integrating Artificial Intelligence (AI) with conventional environmental modeling and management practices. This work confirms a crucial principle: LLMs should enhance rather than replace human expertise, with the ultimate responsibility for environmental decisions remaining with humans. The originality of this research lies in its innovative methodological approach, which leverages process design and prompt engineering to integrate cutting-edge AI with conventional environmental models, establishing a foundation for responsible human-AI collaboration in environmental decision-making. While examining current strengths and limitations, this framework robustly generates optimized environmental decision strategies, marking a new exploration in the field.

Developing a reintegration index (RI) for a closed-loop supply chain network in the automobile industry

PurposeContinuous benchmarking of a closed-loop supply chain (CLSC) network is required to achieve circular economic viability for end-of-life vehicle (ELV) recovery programs for original equipment manufacturers (OEMs). This study develops a framework to assess and benchmark CLSC networks in ELV recovery programs, addressing the absence of a standard index and enabling circular economic viability for OEMs.Design/methodology/approachThe study uses a Bayesian evidential reasoning approach (BERA) that helps decision-makers develop a reintegration index (RI) for the automobile CLSC network. To develop the index, a total of 15 factors related to the automobile CLSC are identified from the literature. Bayesian belief network (BBN) is used on those factors to generate conditional probabilities for different nodes of the BBN. With the opinion of 12 domain experts, the ERA is used to generate a score for each node. Finally, the Markovian tree is used on the scores to generate the RI for a particular CLSC network.FindingsThe analysis demonstrates that both operational and strategic actions aimed at ensuring customer satisfaction and retaining core components are quantified using a standardized index value for each alternative amidst uncertainty. Leveraging the BERA model, decision-makers can calculate RI values, providing them with the means to assess and regulate ratings for CLSC networks. This capability serves to bolster circular economic sustainability by facilitating informed decision-making within the automotive industry.Practical implicationsThis framework offers a structured approach for decision-makers to evaluate CLSC networks in ELV recovery programs, allowing for adaptability to specific organizational objectives and facilitating informed decision-making in the automotive industry.Originality/valueThe study’s integration of expert insights and probabilistic modeling fills the gap in the literature by providing a comprehensive framework for assessing CLSC networks in ELV recovery programs, contributing to circular economic viability and strategic decision-making for OEMs.

Integrating smart contracts into the modeling paradigm to harness the potential of models

Abstract Despite the increasing interest in blockchain and smart contracts, their inherent complexity has impeded widespread adoption. In order to mitigate this issue, this work introduces , a model-based framework for the development of smart contracts in Solidity that enables the treatment of contracts as models, opening up new possibilities for their enhancement and maintenance. A key benefit of is its ability to impose a development pattern, which contributes to improved code quality and reduced vulnerabilities. The framework’s effectiveness is evaluated through several case studies, showing how model-driven engineering can mitigate contracts inherent complexity and promote better collaboration between developers and domain experts. As this work will demonstrate, when smart contracts are treated as models, a vast array of possibilities unfolds.

Real-Time Classification of Ochratoxin a Contamination in Grapes Using AI-Enhanced IoT

Ochratoxin A (OTA) contamination presents significant risks in viticulture, affecting the safety and quality of wine and grape-derived products. This study introduces a groundbreaking method for early detection and management of OTA, leveraging environmental data such as temperature and humidity. A function derived from chemical analysis was developed to estimate OTA concentrations and used to label a synthetic dataset, establishing safe thresholds. Two AI models were trained: one for the detecting of OTA presence and the other for classifying the concentration range. These models were deployed on a M5Stick C+, a microcontroller designed for real-time data processing. The inference process is optimized for rapid response, requiring minimal time to deliver results. Additionally, the low power consumption of the M5Stick C+ ensures that the device can operate throughout the harvest period on a single charge. The system is able to transmit inference data via MQTT for real-time analysis. This comprehensive approach offers a scalable, cost-effective, on-site solution that is autonomous, eliminating the need for domain experts and extensive resources. The robustness of the system was demonstrated through its consistent performance across multiple test sets, providing an effective enhancement to food safety in grape and wine production. The study also details the system architecture, describes the function used for data labeling, outlines the training and deployment processes of the models, and finally, assesses the testing of the overall system.

“Everybody knows what a pothole is”: representations of work and intelligence in AI practice and governance

Abstract In this paper, we empirically and conceptually examine how distributed human–machine networks of labour comprise a form of underlying intelligence within Artificial Intelligence (AI), considering the implications of this for Responsible Artificial Intelligence (R-AI) innovation. R-AI aims to guide AI research, development and deployment in line with certain normative principles, for example fairness, privacy, and explainability; notions implicitly shaped by comparisons of AI with individualised notions of human intelligence. However, as critical scholarship on AI demonstrates, this is a limited framing of the nature of intelligence, both of humans and AI. Furthermore, it dismisses the skills and labour central to developing AI systems, involving a distributed network of human-directed practices and reasoning. We argue that inequities in the agency and recognition of different types of practitioners across these networks of AI development have implications beyond RAI, with narrow framings concealing considerations which are important within broader discussions of AI intelligence. Drawing from interactive workshops conducted with AI practitioners, we explore practices of data acquisition, cleaning, and annotation, as the point where practitioners interface with domain experts and data annotators. Despite forming a crucial part of AI design and development, this type of data work is frequently framed as a tedious, unskilled, and low-value process. In exploring these practices, we examine the political role of the epistemic framings that underpin AI development and how these framings can shape understandings of distributed intelligence, labour practices, and annotators’ agency within data structures. Finally, we reflect on the implications of our findings for developing more participatory and equitable approaches to machine learning applications in the service of R-AI.

A Novel Hybrid Methodology Based on Transfer Learning, Machine Learning, and ReliefF for Chickpea Seed Variety Classification

Seed quality is a critical factor in crop production. Therefore, seed classification is required to obtain high-quality seeds and to enhance agricultural sustainability and productivity. This study focuses on the varietal classification of chickpeas, an important source of protein and fiber. Chickpea seed varieties can currently be identified by domain experts; their reliability and efficiency depend on the experience and skills of experts and are prone to human error. The design of classification models with high accuracy to assist in selection mechanisms is required for chickpea varieties. In this study, a novel hybrid methodology is proposed for the chickpea classification problem. This methodology combines three well-suited and robust components: feature extraction using three pre-trained models, feature selection with the ReliefF algorithm, and classification employing classical machine learning methods to enhance classification accuracy and efficiency. Various experiments have been conducted using the four hybrid models developed. Their performance has been compared in terms of accuracy, recall, F1-score, precision, and AUC. TL+SVM and TL+LDA outperformed the other models, with test accuracies of 94.4% and 94%, respectively. These results demonstrate the potential of a powerful model that will be beneficial as a component of computer vision systems in smart agriculture applications.

Introducing Phylogenetics in Search-based Software Engineering: Phylogenetics-aware SBSE

Phylogenetics studies the relationships, in terms of biological history and kinship, of a set of taxa (e.g., species). We argue that in Search-based Software Engineering (SBSE), the individuals of an evolutionary computation-driven population could be considered as taxa for which the leverage of Phylogenetic Inference might be beneficial. In this work, we present our Phylogenetics-aware SBSE approach. Our approach introduces a novel Phylogenetic Operation to promote results which are sufficiently aligned (in terms of lineage) with a certain reference given by the domain expert. Our approach is evaluated in two heterogeneous industrial case studies: Procedural Content Generation from Game Software Engineering, and Feature Location from Software Maintenance. The results are analyzed using quality-of-the-solution and acceptance-by-developers measurements. We performed a statistical analysis to determine whether the impact on the results is significant compared to baselines that do not leverage Phylogenetics. The results show that our approach significantly outperforms two baselines in both case studies. Furthermore, two focus groups confirmed the acceptance of our approach and stressed that solution acceptance may make the difference in industrial environments. Our work has the potential to motivate a new breed of research work on Phylogenetics awareness to produce better results in Software Engineering.

Automating alloy design and discovery with physics-aware multimodal multiagent AI

The design of new alloys is a multiscale problem that requires a holistic approach that involves retrieving relevant knowledge, applying advanced computational methods, conducting experimental validations, and analyzing the results, a process that is typically slow and reserved for human experts. Machine learning can help accelerate this process, for instance, through the use of deep surrogate models that connect structural and chemical features to material properties, or vice versa. However, existing data-driven models often target specific material objectives, offering limited flexibility to integrate out-of-domain knowledge and cannot adapt to new, unforeseen challenges. Here, we overcome these limitations by leveraging the distinct capabilities of multiple AI agents that collaborate autonomously within a dynamic environment to solve complex materials design tasks. The proposed physics-aware generative AI platform, AtomAgents, synergizes the intelligence of large language models (LLMs) and the dynamic collaboration among AI agents with expertise in various domains, including knowledge retrieval, multimodal data integration, physics-based simulations, and comprehensive results analysis across modalities. The concerted effort of the multiagent system allows for addressing complex materials design problems, as demonstrated by examples that include autonomously designing metallic alloys with enhanced properties compared to their pure counterparts. Our results enable accurate prediction of key characteristics across alloys and highlight the crucial role of solid solution alloying to steer the development of advanced metallic alloys. Our framework enhances the efficiency of complex multiobjective design tasks and opens avenues in fields such as biomedical materials engineering, renewable energy, and environmental sustainability.

A novel hybrid ViT-LSTM model with explainable AI for brain stroke detection and classification in CT images: A case study of Rajshahi region.

Computed tomography (CT) scans play a key role in the diagnosis of stroke, a leading cause of morbidity and mortality worldwide. However, interpreting these scans is often challenging, necessitating automated solutions for timely and accurate diagnosis. This research proposed a novel hybrid model that integrates a Vision Transformer (ViT) and a Long Short Term Memory (LSTM) to accurately detect and classify stroke characteristics using CT images. The ViT identifies essential features from CT images, while LSTM processes sequential information generated by the ViT, adept at capturing crucial temporal dependencies for understanding patterns and context in sequential data. Moreover, our approach addresses class imbalance issues in stroke datasets by utilizing advanced strategies to improve model robustness. To ensure clinical relevance, Explainable Artificial Intelligence (XAI) methods, including attention maps, SHAP, and LIME, were incorporated to provide reliable and interpretable predictions. The proposed model was evaluated using the primary BrSCTHD-2023 dataset, collected from Rajshahi Medical College Hospital, achieving top accuracies of 73.80%, 91.61%, 93.50%, and 94.55% with the SGD, RMSProp, Adam, and AdamW optimizers, respectively. To further validate and generalize the model, it was also tested on the Kaggle brain stroke dataset, where it achieved an impressive accuracy of 96.61%. The proposed ViT-LSTM model significantly outperformed traditional CNNs and ViT models, demonstrating superior diagnostic performance and generalizability. This study advances automated stroke diagnosis by combining deep learning innovations, domain expertise, and enhanced interpretability to support clinical decision-making, providing reliable diagnostic solutions.

AOP-networkFinder — A versatile tool for the reconstruction and visualization of adverse Outcome Pathway networks from AOP-Wiki

Abstract Motivation The AOP-Wiki, a knowledge database for Adverse Outcome Pathways (AOPs), requires an efficient way to present an overview of its content for the reconstruction of networks by experts in a given domain. We have developed the AOP-networkFinder, a user-friendly tool that retrieves AOPs of interest, allows network generation and cleaning, and finally visualizes networks built around the retrieved AOPs. Our tool constructs AOP networks by connecting AOPs that use the same Key Events (KEs) in a versatile but controlled manner. Genes related to these KEs are also displayed. The constructed networks can then be exported as images or to Cytoscape for further fine-tuning and statistical analysis. Results The AOP-networkFinder allows users to comprehensively identify relationships between KEs and visualize the overall structure of an AOP both quickly and easily. This is immensely beneficial to researchers who need to understand the complex interplay between different KEs and the overall pathway they are studying and helps them to build further networks of interest while logging relevant information about changes within the network. These efforts are in line with the FAIR principles, which are crucial attributes for any developed databases and tools for optimizing (re)use in a dynamically changing landscape of AOP-Wiki. Availability and implementation The AOP-networkFinder is an open-source application and is available online at aop-networkfinder.no, in the ‘Computational Toxicology at Norwegian Institute of Public Health’ Zenodo community at DOI 10.5281/zenodo.11068434, in the GitHub repository at github.com/folkehelseinstituttet/AOPnetworkFinder_v1, as well as in a Docker image at hub.docker.com/r/nurre123/aop_network_finder. The software is available under the GNU Affero General Public License (AGPL), v3.0. The tool uses the AOP-Wiki SPARQL endpoint to retrieve AOP data.

Large language models for data extraction from unstructured and semi-structured electronic health records: a multiple model performance evaluation

ObjectivesWe aimed to evaluate the performance of multiple large language models (LLMs) in data extraction from unstructured and semi-structured electronic health records.Methods50 synthetic medical notes in English, containing a structured...

Supporting vision-language model few-shot inference with confounder-pruned knowledge prompt.

Vision-language models are pre-trained by aligning image-text pairs in a common space to deal with open-set visual concepts. Recent works adopt fixed or learnable prompts, i.e., classification weights are synthesized from natural language descriptions of task-relevant categories, to reduce the gap between tasks during the pre-training and inference phases. However, how and what prompts can improve inference performance remains unclear. In this paper, we explicitly clarify the importance of incorporating semantic information into prompts, while existing prompting methods generate prompts without sufficiently exploring the semantic information of textual labels. Manually constructing prompts with rich semantics requires domain expertise and is extremely time-consuming. To cope with this issue, we propose a knowledge-aware prompt learning method, namely Confounder-pruned Knowledge Prompt (CPKP), which retrieves an ontology knowledge graph by treating the textual label as a query to extract task-relevant semantic information. CPKP further introduces a double-tier confounder-pruning procedure to refine the derived semantic information. Adhering to the individual causal effect principle, the graph-tier confounders are gradually identified and phased out. The feature-tier confounders are eliminated by following the maximum entropy principle in information theory. Empirically, the evaluations demonstrate the effectiveness of CPKP in few-shot inference, e.g., with only two shots, CPKP outperforms the manual-prompt method by 4.64% and the learnable-prompt method by 1.09% on average.

COTR: Efficient Job Task Recognition for Occupational Information Systems with Class-Incremental Learning

Occupation-specific job tasks (OSTs) refer to the duties, responsibilities, and activities associated with a particular occupation, which define the core functions and performance expectations for those engaged in that profession. Efficient recognition and extraction of OSTs from large-scale job description data are essential for establishing a continually updated occupational information system (OIS), such as O*NET, which serves as critical tools for advancing research in work and labor markets. However, this task presents substantial challenges due to its heavy reliance on domain experts for the labor-intensive annotation of job postings, rendering the process time-consuming and difficult to scale for large-scale implementation. To this end, in this paper, we present COTR , a novel data-driven framework designed for the efficient recognition of OSTs from job postings, capable of continually identifying new tasks through class-incremental learning. Specifically, we first employ large language models (LLMs) and prompt learning to develop a three-phase process—“expansion, translation, and generation”—that addresses the critical challenge of the absence of predefined OSTs in non-English labor market data, leveraging O*NET as a foundational reference. Subsequently, we introduce a BERT-based model for OST recognition, incorporating a uniquely designed pair-wise loss function that distills valuable insights from ChatGPT or other LLMs, thereby substantially enhancing recognition performance. In addition, to achieve cost-effective training data annotation, we develop an LLM-based coarse-to-fine candidate OSTs generation algorithm, integrating contrastive active learning to optimize the annotation process through human-machine collaboration. Notably, we design a supervised fine-tuning strategy with a novel encoding technique to optimize LLMs, improving the recall rate of the generated candidate OSTs and achieving up to a 343-fold increase in annotation efficiency compared to traditional manual expert annotation in our experiments. Afterward, we propose an efficient class-incremental learning method that incorporates an out-of-distribution (OOD) detection module for identifying potential novel OSTs and a fine-tuning module to extend the model’s recognition capabilities to include newly discovered tasks. Finally, we construct two real-world datasets using job posting data collected from the labor markets of China and the United States, respectively. Extensive experiments on the real-world datasets, along with two publicly available datasets, have demonstrated the effectiveness of the proposed COTR. Furthermore, several case studies showcase the significant benefits of COTR for various downstream applications in labor market analysis, including analyzing the evolving demand for OSTs, assessing the value of OSTs, and recognizing the relationships between OSTs and associated skills.

Explainable AI in Diagnostic Radiology for Neurological Disorders: A Systematic Review, and What Doctors Think About It.

Background: Artificial intelligence (AI) has recently made unprecedented contributions in every walk of life, but it has not been able to work its way into diagnostic medicine and standard clinical practice yet. Although data scientists, researchers, and medical experts have been working in the direction of designing and developing computer aided diagnosis (CAD) tools to serve as assistants to doctors, their large-scale adoption and integration into the healthcare system still seems far-fetched. Diagnostic radiology is no exception. Imagining techniques like magnetic resonance imaging (MRI), computed tomography (CT), and positron emission tomography (PET) scans have been widely and very effectively employed by radiologists and neurologists for the differential diagnoses of neurological disorders for decades, yet no AI-powered systems to analyze such scans have been incorporated into the standard operating procedures of healthcare systems. Why? It is absolutely understandable that in diagnostic medicine, precious human lives are on the line, and hence there is no room even for the tiniest of mistakes. Nevertheless, with the advent of explainable artificial intelligence (XAI), the old-school black boxes of deep learning (DL) systems have been unraveled. Would XAI be the turning point for medical experts to finally embrace AI in diagnostic radiology? This review is a humble endeavor to find the answers to these questions. Methods: In this review, we present the journey and contributions of AI in developing systems to recognize, preprocess, and analyze brain MRI scans for differential diagnoses of various neurological disorders, with special emphasis on CAD systems embedded with explainability. A comprehensive review of the literature from 2017 to 2024 was conducted using host databases. We also present medical domain experts' opinions and summarize the challenges up ahead that need to be addressed in order to fully exploit the tremendous potential of XAI in its application to medical diagnostics and serve humanity. Results: Forty-seven studies were summarized and tabulated with information about the XAI technology and datasets employed, along with performance accuracies. The strengths and weaknesses of the studies have also been discussed. In addition, the opinions of seven medical experts from around the world have been presented to guide engineers and data scientists in developing such CAD tools. Conclusions: Current CAD research was observed to be focused on the enhancement of the performance accuracies of the DL regimens, with less attention being paid to the authenticity and usefulness of explanations. A shortage of ground truth data for explainability was also observed. Visual explanation methods were found to dominate; however, they might not be enough, and more thorough and human professor-like explanations would be required to build the trust of healthcare professionals. Special attention to these factors along with the legal, ethical, safety, and security issues can bridge the current gap between XAI and routine clinical practice.

Nutritional intelligence in the food system: Combining food, health, data and AI expertise.

Transformative change is needed across the food system to improve health and environmental outcomes. As food, nutrition, environmental and health data are generated beyond human scale, there is an opportunity for technological tools to support multifactorial, integrated, scalable approaches to address the complexities of dietary behaviour change. Responsible technology could act as a mechanistic conduit between research, policy, industry and society, enabling timely, informed decision making and action by all stakeholders across the food system. Domain expertise in food, nutrition and health should always be integrated into both the development and continuous deployment of AI-powered nutritional intelligence (NI) to ensure it is responsible, accurate, safe, useable and effective. Dietary behaviours are complex and improving diet-related health outcomes requires socio-cultural-demographic considerations within the design and deployment of NI tools. This article describes existing examples of NI within the food system and future opportunities. Human-in-the-loop approaches with food, health and nutrition experts involved at all stages including data acquisition, processing, output validation and ongoing quality assurance are essential to ensure evidence-based practice. The same ethical considerations should apply in this domain as in any other (e.g. privacy, inclusivity, robustness, transparency and accountability) and responsible practice must encompass rigorous standards and alignment with regulatory frameworks. Critical today and in the future is accessibility to appropriate high-quality food compositional data sets, which include up-to-date information on commercially available products that reflect the constantly evolving food landscape to realise the potential of responsible AI to help address the existing food system challenges.

Semi-Supervised Learning Allows for Improved Segmentation With Reduced Annotations of Brain Metastases Using Multicenter MRI Data.

Deep learning-based segmentation of brain metastases relies on large amounts of fully annotated data by domain experts. Semi-supervised learning offers potential efficient methods to improve model performance without excessive annotation burden. This work tests the viability of semi-supervision for brain metastases segmentation. Retrospective. There were 156, 65, 324, and 200 labeled scans from four institutions and 519 unlabeled scans from a single institution. All subjects included in the study had diagnosed with brain metastases. 1.5 T and 3 T, 2D and 3D T1-weighted pre- and post-contrast, and fluid-attenuated inversion recovery (FLAIR). Three semi-supervision methods (mean teacher, cross-pseudo supervision, and interpolation consistency training) were adapted with the U-Net architecture. The three semi-supervised methods were compared to their respective supervised baseline on the full and half-sized training. Evaluation was performed on a multinational test set from four different institutions using 5-fold cross-validation. Method performance was evaluated by the following: the number of false-positive predictions, the number of true positive predictions, the 95th Hausdorff distance, and the Dice similarity coefficient (DSC). Significance was tested using a paired samples t test for a single fold, and across all folds within a given cohort. Semi-supervision outperformed the supervised baseline for all sites with the best-performing semi-supervised method achieved an on average DSC improvement of 6.3% ± 1.6%, 8.2% ± 3.8%, 8.6% ± 2.6%, and 15.4% ± 1.4%, when trained on half the dataset and 3.6% ± 0.7%, 2.0% ± 1.5%, 1.8% ± 5.7%, and 4.7% ± 1.7%, compared to the supervised baseline on four test cohorts. In addition, in three of four datasets, the semi-supervised training produced equal or better results than the supervised models trained on twice the labeled data. Semi-supervised learning allows for improved segmentation performance over the supervised baseline, and the improvement was particularly notable for independent external test sets when trained on small amounts of labeled data. Artificial intelligence requires extensive datasets with large amounts of annotated data from medical experts which can be difficult to acquire due to the large workload. To compensate for this, it is possible to utilize large amounts of un-annotated clinical data in addition to annotated data. However, this method has not been widely tested for the most common intracranial brain tumor, brain metastases. This study shows that this approach allows for data efficient deep learning models across multiple institutions with different clinical protocols and scanners. 3 TECHNICAL EFFICACY: Stage 2.

Deep Learning Models for Automatic Classification of Anatomic Location in Abdominopelvic Digital Subtraction Angiography.

To explore the information in routine digital subtraction angiography (DSA) and evaluate deep learning algorithms for automated identification of anatomic location in DSA sequences. DSA of the abdominal aorta, celiac, superior mesenteric, inferior mesenteric, and bilateral external iliac arteries was labeled with the anatomic location from retrospectively collected endovascular procedures performed between 2010 and 2020 at a tertiary care medical center. "Key" images within each sequence demonstrating the parent vessel and the first bifurcation were additionally labeled. Mode models aggregating single image predictions, trained with the full or "key" datasets, and a multiple instance learning (MIL) model were developed for location classification of the DSA sequences. Model performance was evaluated with a primary endpoint of multiclass classification accuracy and compared by McNemar's test. A total of 819 unique angiographic sequences from 205 patients and 276 procedures were included in the training, validation, and testing data and split into partitions at the patient level to preclude data leakage. The data demonstrate substantial information sparsity as a minority of the images were designated as "key" with sufficient information for localization by a domain expert. A Mode model, trained and tested with "key" images, demonstrated an overall multiclass classification accuracy of 0.975 (95% CI 0.941-1). A MIL model, trained and tested with all data, demonstrated an overall multiclass classification accuracy of 0.966 (95% CI 0.932-0.992). Both the Mode model with "key" images (p < 0.001) and MIL model (p < 0.001) significantly outperformed a Mode model trained and tested with the full dataset. The MIL model additionally automatically identified a set of top-5 images with an average overlap of 92.5% to manually labelled "key" images. Deep learning algorithms can identify anatomic locations in abdominopelvic DSA with high fidelity using manual or automatic methods to manage information sparsity.