Algorithmic design, value trade-offs, and public preferences for autonomous buses: Insights from a conjoint experiment

Dataset of perceptions of waiting in a virtual reality (VR) doctor's office receptionist queue with and without notifications on the reason for a delay

High-performance intelligent and robust emergency braking mechanism for autonomous vehicles

Towards automated optimization algorithm design with LLM: An exploratory study in multi-objective weather routing

Efficient artifact removal for adaptive deep brain stimulation and a temporal event localization analysis.

Multidimensional chromatographic fingerprint fusion with machine learning: Entropy-based feature evaluation for TCM quality marker discovery.

Large language model assisted meta-evolution for automated constrained optimization evolutionary algorithm design

A framework for reputation aware uninorm-driven consensus algorithms for blockchain networks

ObjectiveThis retrospective, case-control study with internal validation evaluates the performance of machine learning (ML) and deep learning (DL) models in classifying pediatric patients at risk for anxiety disorders using structured electronic health records (EHRs) and area-based measures of health (ABMH). The aim is to enable proactive care by monitoring potential anxiety onset across developmental stages.MethodsWe trained a series of ML models (Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, XGBoost) and DL models (LSTM, GRU, RETAIN, Dipole) using structured EHR data from 30-day windows prior to diagnosis. Two datasets were used per age group: one with structured EHR data only, and another including both EHR and ABMH data. ML models were trained using short-term cross-sectional features, while DL models leveraged full longitudinal patient histories. Performance was assessed using AUROC, AUPRC, PPV, NPV, F1 score, and accuracy. Due to differences in input scope, model performance reflects both algorithmic and temporal design differences and is not intended as a direct comparison between ML and DL.ResultsML models offered strong baseline performance, with XGBoost achieving AUROC scores of 0.817 (EHR) and 0.816 (EHR+ABMH) for 8-year-olds. Adding ABMH features did not significantly improve performance. DL models, particularly RETAIN and Dipole, achieved the highest AUROC values (e.g., Dipole: 0.853 with EHR, 0.857 with EHR+ABMH for 8-year-olds), outperforming other DL and ML models within their respective design constraints.ConclusionBoth ML and DL models successfully identified likely anxiety onset using structured EHR data. DL models using longitudinal data achieved the highest performance, while XGBoost provided a robust ML baseline. The minimal impact of ABMH features highlights integration challenges, and performance variation across ages emphasizes the need for age-stratified modeling approaches.

Liberation from prolonged mechanical ventilation is challenging and its outcomes are poor. Patients who failed at least three spontaneous breathing trials, often referred to as prolonged weaning patients, are usually weaned with protocolized programs in specialized weaning units, but there are no standardized strategies to facilitate their ventilator liberation. The objective of this study was to compare the ventilator liberation rate of two common ventilator weaning programs. Tracheostomized patients with ongoing invasive mechanical ventilation for at least 21day who were admitted to Barlow Respiratory Hospital for ventilator weaning were studied. Patients who passed spontaneous breathing trial on admission were excluded. In a prospective parallel group, non-blinded clinical study, patients were randomized to receive either the Pressure Support Ventilation (PSV) weaning program or the Therapist-Implemented Patient-Specific (TIPS) weaning program. Randomization was performed using a computer algorithm of block design. The primary outcome was ventilator liberation success. The secondary outcomes were hospital length of stay, physical recovery, discharge disposition and mortality. Significant hospital events were also compared between the groups. N = 25 patients were studied in PSV and N = 26 in the TIPS group. Outcomes were reported for all patients. The liberation success rate at 30 days was 37.5% (standard error, SE = 9.9%) in the PSV and 46.2% (SE = 9.8%) in the TIPS group (p = 0.58, odds ratio, OR 1.42, RD 8.7%, 95% confidence interval, CI=-18.6-35.9). The liberation rate at discharge was 44% (SE = 9.9%) in the PSV group and 53.8% (SE = 9.8%) in the TIPS group (p = 0.54, OR:1.48, RD 9.8%, CI=-17.2-37.2%). The inpatient mortality was: PSV = 24% (SE 8.5%) and TIPS = 11.5% (SE 6.3%), p = 0.291, OR 0.413, RD=-12.5%, CI=-33.2-8.3%. We did not find a significant difference between the two ventilator weaning programs in any of our outcomes, but our study describes a very sick patient population. Continued weaning beyond 30 days had improved liberation success. Both weaning paths are equally beneficial for prolonged mechanical ventilation patients who undergo prolonged weaning. The trial was registered retroactively at ClinicalTrials.gov, NCT06976554.

Processed electroencephalography (EEG) monitors such as the bispectral index (BIS) and patient state index (PSI), are used clinically to estimate anesthetic depth, yet their algorithmic design obscures how closely these indices reflect underlying neural complexity. Entropy-based analyses, grounded in information theory, provide a quantitative framework for characterizing EEG signal irregularity and have been proposed as physiologically interpretable alternatives. However, the relationship between these commercial indices and theoretical entropy measures remains unclear. This systematic review aimed to (1) synthesize existing evidence on the relationship between commercially available processed EEG metrics and entropy-based EEG analyses, and (2) identify factors influencing their comparability, including algorithmic, demographic, and anesthetic variables. A comprehensive literature search identified experimental and clinical studies comparing BIS, PSI, and related commercial indices with theoretical entropy measures (e.g., approximate entropy, sample entropy, and permutation entropy, state entropy and response entropy) across various anesthetic agents and clinical populations. Data were extracted on study design, patient demographics, EEG metrics, analytical methods, and reported correlations or prediction probabilities. Ninety-four studies were included, encompassing participants across diverse anesthetic modalities. Overall, BIS exhibited moderate-to-strong correlations with entropy-derived measures and comparable prediction probabilities for distinguishing anesthetic depth. Entropy indices demonstrated greater resistance to certain artifacts but higher susceptibility to electromyographic contamination. Age, anesthetic type, and the use of neuromuscular blocking agents significantly influenced the relationship between indices. Across studies, heterogeneity in preprocessing, entropy algorithms, and patient selection limited direct comparability. Commercially processed EEG indices and theoretical entropy measures capture overlapping but distinct dimensions of cortical dynamics during anesthesia. While both reliably track transitions in consciousness, discrepancies arise from differences in signal filtering, algorithm design, and physiological variability. Future research should prioritize transparent algorithmic frameworks and standardized entropy computation to enhance the interpretability and cross-device comparability of EEG-derived anesthesia monitors.

Atrial fibrillation (AF) is the most common sustained arrhythmia worldwide and a major contributor to stroke and cardiovascular morbidity. Conventional diagnostics, including electrocardiograms (ECG) and ambulatory monitors, are limited by accessibility and short monitoring windows, particularly in asymptomatic or paroxysmal AF. Recently, artificial intelligence (AI)-driven technologies and wearable devices have emerged as promising tools for early AF detection. This systematic review evaluates the diagnostic performance of AI-enabled and wearable technologies for AF detection in clinical and real-world settings. A comprehensive search of PubMed, Scopus, and Web of Science identified studies reporting sensitivity, specificity, predictive values, and AUC of digital AF detection tools. Platforms included smartphone apps, smartwatches, photoplethysmography (PPG), single- and multi-lead ECGs, and machine learning algorithms. Data extraction and quality assessment used QUADAS-2. Random-effects meta-analyses and subgroup analyses synthesized findings and explored heterogeneity. Twenty-four studies met inclusion criteria. High-performing tools demonstrated sensitivity and specificity ≥94% and AUC ≥0.95, while consumer-grade devices, such as the Apple Heart Study, showed lower specificity (46%) and positive predictive value (7.6%), reflecting frequent false positives. Heterogeneity arose from device type, signal method (PPG vs. ECG), algorithm design, and population characteristics. Tools integrating explainable AI and multi-modal data generally outperformed simpler models. AI-enhanced and wearable technologies show strong potential for accurate AF detection under specific conditions. Performance variability, especially in consumer-grade devices, underscores the need for external validation, algorithm refinement, and clinical integration. Future research should assess real-world effectiveness, cost-efficiency, explainability, and long-term outcomes to support broader adoption.

<b>Objectives:</b> Automation of quantitative analysis of breast cancer (BC) immunohistochemistry (IHC) specimens is important to optimize pathologists’ workflow and improve diagnostic reproducibility. This is especially important in low- and middle-income countries where there is a shortage of highly trained pathologists. However, existing approaches face challenges in implementing fully automated quantitative IHC due to the difficulty of both delineating tumor areas, including discrete areas, especially in IHC slides with poor quality. Moreover, accurate identification of invasive carcinoma areas and accurate quantification of positive and negative cells in the specimen are critical for quantitative analysis.<br /> <b>Methods and results:</b> This study presents a method to automatically identify types of carcinoma areas in whole slide IHC images of BC, focusing on quantifying IHC images on realms of Kazakhstan. The used model is a combination of morphological characteristics and boundary features, which provides high accuracy of segmentation of tumor zones of images of mild and low quality. We used several methods includes convolutional neural network based on the Keras framework, k-nearest neighbors machine learning methods, and self-developed image analysis methods. The developed model showed high accuracy, where the results corresponded to the diagnoses of pathologists. As expected, the method proved to be ineffective when applied to severely degraded slides, such as those with insufficient staining or inadequate washing. Slides of inferior quality were excluded from analysis, which negatively affected the statistical robustness. On slides of moderate quality, the reliability of nucleus segmentation dropped significantly.<br /> <b>Conclusions:</b> The combination of models we used showed high accuracy in differentiating BC cells between the basal-like subtype of BC and its invasiveness and recurrence in Kazakhstan. However, IHC specimens with low DPI or low-quality IHC need further optimizations and improvements in algorithm design. The main issue can be considered methodological differences between the approaches of AI and humans: AI operates in a large number of cases (more than 10,000), yet its accuracy is relatively low. In contrast, humans work with a much smaller number of cases but achieve a level of precision that AI cannot currently match. This discrepancy necessitates a revision of the methodology of IHC analysis for AI, including the development of new requirements, methods, and thresholds from scratch. This approach provides analysis of the entire area of the slide, increases the speed of interpretation of IHC results, and reduces human errors in diagnosis, especially in low- and middle-income countries.

Advances in computational design have greatly acceleratedantimicrobialpeptide engineering. In this study, three Plasmodium chabaudi-derived peptides (PcDBS1R1, PcDBS1R5, and PcDBS1R9), generated usingthe Joker computational design algorithm, were synthesized and characterizedfor their structural and functional properties. Biophysical analysesrevealed that PcDBS1R5 and PcDBS1R9 predominantly adopted α-helicalstructures with high amphipathicity, whereas PcDBS1R1 exhibited greaterstructural plasticity. PcDBS1R5 and PcDBS1R9 displayed antibacterialactivity against an Acinetobacter baumannii clinicalisolate, whereas PcDBS1R1 showed pronounced antibiofilm effects. Noneof the peptides exhibited cytotoxicity toward murine macrophages,and all significantly reduced nitric oxide production in lipopolysaccharide-stimulatedmacrophages, suggesting potential anti-inflammatory activity. Overall,these findings demonstrate that computer-aided design of P.chabaudi-derived peptides can yield molecules with antibiofilm,and immunomodulatory properties, minimal cytotoxicity, and promisingtherapeutic potential as scaffolds for next-generation peptide-basedtreatments targeting biofilm-associated bacterial infections.

This mixed-methods, quasi-experimental one-group pre–post study investigated whether a structured sequence of unplugged computational thinking (CT) activities embedded in regular primary mathematics lessons was associated with measurable changes in pupils’ CT practices and how such changes manifested in classroom interaction. Participants were 150 grade 4–5 pupils in a Vietnamese public primary school who completed eight 45-min, screen-free lessons aligned to five CT practices: decomposition, pattern recognition, abstraction, algorithm design, and evaluation/debugging. Quantitative analyses revealed statistically significant pre–post improvements across all five CT subscales (all p < .001 ), with large effect sizes (Cohen's d = 2.58 − 3.29 ). Gains were consistent across practices, indicating broad within-cohort growth rather than isolated skill shifts. Qualitative evidence from structured observations, student interviews, and artifact analysis converged with these results, documenting increased representational coordination, more explicit articulation of stepwise procedures, systematic test–revise (debugging) cycles, and distributed collaborative monitoring. Joint analysis linked subscale gains to three mechanism clusters: (i) representational fluency that stabilized invariants for inspection and generalization, (ii) iterative strategic refinement through visible debugging routines, and (iii) collaborative regulation enacted through role rotation and peer verification. Within the bounds of a classroom-based design, the findings provide convergent evidence that low-cost, unplugged tasks can be integrated into routine mathematics instruction and are associated with substantial growth in taxonomy-referenced CT practices in a resource-constrained setting.

Non-determinism in deep learning algorithm design and implementation leads to performance variation, meaning model performance is not a single value, but rather a distribution. These model performance distributions are underexplored despite their impact on robustness. We investigate the robustness of deep learning performance to sources of non-determinism, specifically focusing on how performance distributions differ across various architectures and tasks. We conducted 186 experiments on state-of-the-art image classification (ResNet, ViT) and time series forecasting (Autoformer, iTransformer, NLinear, TSMixer) architectures. Each experiment was run 100 times with different random seeds to generate performance distributions, resulting in 18,600 runs. Robustness was quantified using metrics for spread, symmetry, and tail risk. Performance distributions are frequently non-Gaussian, particularly in time series forecasting. Model size does not systematically affect robustness - larger image classification models show fewer outliers but not lower spread, while smaller time series models show lower spread but more extreme underperformers. Training duration does not scale linearly; early stopping effectively balances performance and robustness. Mean performance does not predict robustness - time series forecasting shows moderate correlation while image classification shows none. Time series models produce nearly three times more underperforming outliers than image classification models, indicating substantially higher tail risk. Tail risk poses serious concerns for Trustworthy AI in high-stakes applications. Models performing well on average may exhibit long tails and extreme outliers revealed only through distributional analysis. Mean performance alone should not guide model selection; assessment of spread, symmetry, and tail risk is essential for reliable model assessment where consistent performance is critical.

Artificial intelligence: Catalyzing a new era in pancreatic cancer cure.

The segmentation of pelvic fracture fragments in CT and X-ray images is crucial for trauma diagnosis, surgical planning, and intraoperative guidance. However, accurately and efficiently delineating the bone fragments remains a significant challenge due to complex anatomy and imaging limitations. The PENGWIN challenge, organized as a MICCAI 2024 satellite event, aimed to advance automated fracture segmentation by benchmarking state-of-the-art algorithms on these complex tasks. A diverse dataset of 150 CT scans was collected from multiple clinical centers, and a large set of simulated X-ray images was generated using the DeepDRR method. Final submissions from 16 teams worldwide were evaluated under a rigorous multi-metric testing scheme. The top-performing CT algorithm achieved an average fragment-wise intersection over union (IoU) of 0.930, demonstrating satisfactory accuracy. However, in the X-ray task, the best algorithm achieved an IoU of 0.774, which is promising but not yet sufficient for intra-operative decision-making, reflecting the inherent challenges of fragment overlap in projection imaging. Beyond the quantitative evaluation, the challenge revealed methodological diversity in algorithm design. Variations in instance representation, such as primary-secondary classification versus boundary-core separation, led to differing segmentation strategies. Despite promising results, the challenge also exposed inherent uncertainties in fragment definition, particularly in cases of incomplete fractures. These findings suggest that interactive segmentation approaches, integrating human decision-making with task-relevant information, may be essential for improving model reliability and clinical applicability.

Precision occupational lead exposure assessment through medical-informed machine learning.

Advances in iterative learning control: A recent five-year literature review.