Early detection of depressive symptom changes is vital for timely interventions. Mobile and wearable technologies enable continuous, unobtrusive monitoring of behavioral, psychological and physiological data, offering new possibilities for digital phenotyping and just-in-time prediction of depression. This scoping review synthesized findings from 52 studies to identify commonly used features, evaluate their predictive value and assess methodological approaches. Frequently assessed features included location data, sleep metrics, physical activity, communication patterns, heart rate variability and mood self-reports. Features such as time spent at home, sleep variability and reduced mobility were strongly associated with depressive symptoms. Combining physiological, behavioral and self-report data enhanced predictive performance. Personalized models and anomaly detection approaches outperformed generalized ones in predicting individual symptom changes. Overall, mobile and wearable data show strong potential for just-in-time depression prediction. Future research should emphasize new features, diverse populations and personalized models to improve accuracy and real-world applicability.

Theta burst stimulation (TBS) is a promising form of repetitive transcranial magnetic stimulation (rTMS) capable of modulating cortical excitability and intracortical processes, offering therapeutic potential for neurological and psychiatric disorders. However, clinical translation remains limited by high inter-subject and inter-session variability in stimulation effects. To more directly capture cortical responses, there is increased interest in combining TMS with electroencephalography (EEG) to assess TMS-evoked potentials (TEPs) before and after stimulation. As an individual’s neurophysiological state influences stimulation outcomes, this study explores whether pre-stimulation resting-state EEG can predict changes in TEP component amplitudes following intermittent (iTBS) and continuous (cTBS) protocols applied to the left primary motor cortex, using data from fifteen healthy male participants in a randomized, single-blind crossover design. Linear (Lasso regression) and nonlinear (CatBoost regression) models were designed to predict changes in six TEP components (N15, P30, N45, P60, N100, P180). Both models consistently achieved lower mean absolute errors than the random guessing baseline, demonstrating their ability to capture meaningful predictive patterns in cortical responses. The best performing model varied by TEP component and TBS protocol. Incorporating feature deltas (post- vs. pre-stimulation feature difference) did not significantly enhance predictive performance. Feature importance analysis revealed the predictive value of spectral power and connectivity measures. For instance, connectivity between the stimulation site and frontal/parietal regions, together with oscillatory power in frontal and motor areas, often emerged as the top predictors. Given the limited sample size, these findings should be interpreted as exploratory and hypothesis-generating, requiring validation in larger and independent cohorts. The study establishes a framework for predicting individual TEP responses to TBS using machine learning, paving the way for personalized neuromodulation.

Zero-shot classification enables models to assign labels to unseen classes without task-specific training and has become increasingly effective with the advent of Large Language Models (LLMs). However, existing zero-shot classification approaches typically present all candidate labels as a flat, unstructured list within the prompt. As the number of classes grows, this formulation exacerbates well-known limitations of LLMs, including attention dilution and positional bias, ultimately degrading classification performance. We present Label Space Reduction (LSR), a test-time training method for transductive zero-shot LLM classification. LSR uses LLM-generated pseudo-labels to iteratively refine the classification label space by systematically ranking and reducing candidate classes, enabling the model to concentrate on the most relevant options. By leveraging unlabeled data with the statistical learning capabilities of data-driven models, LSR dynamically optimizes the label space representation at test time. Our experiments across seven benchmarks demonstrate that LSR improves macro-F1 scores by an average of 7.0

Microservice applications are omnipresent due to their advantages, such as scalability, flexibility and consequentially resource cost efficiency. The loosely-coupled microservices can be easily added, replicated, updated and/or removed to address the changing workload. However, the distributed and dynamic nature of microservice architectures introduces a complexity with regard to monitoring and observability, which is paramount to ensure reliability, especially in critical domains. Anomaly detection has become an important tool to automate microservice monitoring and detect system failures. Nevertheless, state-of-the-art solutions assume the topology of the monitored application to remain static over time and fail to account for the dynamic changes the application, and the infrastructure it is deployed on, undergoes. This paper tackles these shortcomings by introducing a context-aware anomaly detection methodology using dynamic knowledge graphs to capture contextual features which describe the evolving state of the monitored system. Our methodology leverages resource and network monitoring to capture dependencies between microservices, and the infrastructure they are running on. In addition to the methodology for anomaly detection, this paper presents an open-source benchmark framework for context-aware anomaly detection that includes monitoring, fault injection and data collection. The evaluation on this benchmark shows that our methodology consistently outperforms the non-contextual baselines. These results underscore the importance of contextual awareness for robust anomaly detection in complex, topology-driven systems. Beyond these achieved improvements, our benchmark establishes a reproducible and extensible foundation for future research, facilitating the experimentation with broader ranges of models and a continued advancement in context-aware anomaly detection.

Conformal prediction has emerged as a principled framework for uncertainty quantification in computer vision, offering rigorous finite-sample coverage guarantees. However, its application in object detection has remained largely confined to localization, as standard inference codebases typically yield only top-1 class scores, precluding full class-label conformalization. In this work, we bridge this gap by adapting four architecturally diverse detectors—Faster R-CNN, RetinaNet, YOLO11, and RT-DETRv2—to facilitate the extraction of comprehensive per-class score vectors and the estimation of background confidence in the absence of native background modeling. Leveraging these adapted architectures, we implement inductive conformal prediction (ICP) using five distinct nonconformity functions: Top-K, Adaptive Prediction Sets (APS), Hinge, Margin, and Brier score. Our framework is rigorously benchmarked across a curated 20-class subset of MS-COCO and two specialized parasite egg datasets (AI4NTD P1.5v2 and Chula-ParasiteEgg-11). In addition, a Naive cumulative-threshold method is included as a baseline for comparison with APS, given their comparable mathematical formulations. Across target coverage levels of 90%, 95%, and 99%, the conformalized models consistently achieved nominal coverage with only minor finite-sample deviations. Hinge and APS exhibited an optimal balance between statistical coverage and prediction-set efficiency, whereas Margin and Brier scores tended toward larger sets under high data complexity and strict coverage requirements. With empty prediction sets maintained below 0.1%, our findings establish ICP as a robust and adaptable paradigm for trustworthy class-label uncertainty estimation, particularly within safety-critical workflows such as automated parasite diagnostics.

In roll-to-roll (R2R) web processing systems, traction rollers impose precise velocity profiles on the moving web. Ideally, the web follows this trajectory without deviation, but slip can occur during rapid acceleration or deceleration, leading to tension loss and degraded product quality. Although slip can be detected directly using high-resolution encoders that track the actual web speed, such sensors are expensive and require machine downtime for installation, making them impractical for large-scale industrial deployment. To overcome this limitation, we developed a virtual slip sensor that estimates slip using existing machine signals only. A temporary encoder was used to collect ground-truth data, enabling the training of predictive models that eliminate the need for a permanent physical sensor. The proposed system employs an ensemble modeling approach: a CatBoost model captures low-slip behavior where data is abundant, while a linear model extrapolates to high-slip, out-of-distribution conditions. Targeted feature engineering ensures generalization across varying ramp times and web speeds. Despite being trained primarily on data containing limited slip, the models successfully generalized to scenarios with severe slip, demonstrating robust predictive performance. The ensemble reduces the regular CatBoost model’s MSE at 60 m/min by approximately 54% in the speed-based evaluation and by approximately 68% in the quantile-based evaluation while maintaining comparable performance in the low-speed regimes. The resulting virtual sensor enables continuous real-time slip monitoring, providing operators with timely insights to prevent quality degradation and operate at higher acceleration profiles to increase throughput, even on machines that have not previously experienced extreme slip.

BackgroundThe World Health Organization (WHO) has emphasised the need for innovative diagnostic tools to support the control and elimination of neglected tropical diseases (NTDs). Microscopy-based diagnostics, the current standard, rely on trained technicians for labour-intensive processes, posing logistical challenges in the low-resource settings where NTDs are most prevalent. This study describes the technical details of an artificial intelligence-powered digital pathology (AI-DP) platform designed to support large-scale deworming programs for two NTDs, alongside its analytical performance and user experience in laboratory and field settings.Methodology/principal findingsThe AI-DP platform integrates electronic data capture tools, whole-slide imaging scanners, onboard AI analysis, and result verification software to automate microscopy-based screening. Targeting soil-transmitted helminthiasis (STH) and intestinal schistosomiasis (SCH) as initial use cases, the system was deployed in Ethiopia and Uganda, scanning 951 Kato-Katz (KK) thick smears containing 43,919 verified helminth eggs. Using 5-fold cross-validation, precision/recall/average precision were 95.4%/91.7%/97.1% for Ascaris lumbricoides, 95.9%/86.7%/94.8% for Trichuris trichiura, 84.6%/86.6%/91.4% for hookworm, and 89.1%/79.1%/89.2% for Schistosoma mansoni. Feedback from 14 field users across 30 real-world scenarios indicated the AI-DP platform's improved usability, particularly in hardware portability and software interfaces, though the average scan time of 12.5 minutes per smear was identified as a limitation.Conclusions/significanceThe AI-DP platform demonstrates potential as a tool for efficient monitoring and evaluation of STH and SCH control programs by providing near-real-time data with quality controls. However, further validation studies are needed to assess its clinical diagnostic performance, field usability, and cost-effectiveness in large-scale STH and SCH deworming programs. Given that the platform also provides a pipeline for any microscopy-based diagnosis, its potential for other NTDs also needs further attention.

Background Acute kidney injury (AKI) is a frequent, severe complication in the intensive care units (ICU). Existing machine learning models are typically inflexible, classification-based (i.e., predicting AKI occurrence as yes/no), and of limited clinical utility. This study proposes and externally validates the first multi-step, multivariate distributional regression model that directly predicts future distributions of serum creatinine (sCr) and urine output across multiple time horizons, thereby enhancing AKI risk stratification and personalized clinical decision support. Methods The model was developed using a training cohort of 4,118 adult ICU stays from the MIMIC-IV dataset and externally validated on four independent, diverse cohorts: MIMIC-IV (N=3,838), UZGent (N=4,442), eICU (N=10,760), and AmsterdamUMC (N=6,129). The model used clinical data to generate multivariate predictive distributions hourly for urine output and sCr (up to 48 hours ahead). Predictors included demographics, vital signs, laboratory results, medications, and recent urine output, with time-varying variables summarized over the preceding 72 hours (recent value, slope, minimum, maximum, variability). Performance was evaluated by comparing our predictive distributions with state-of-the-art tree-based classifiers for 24-hour ahead prediction of KDIGO stages 1-3 AKI and persistent stage 3 AKI. Results Across all external cohorts, the distributional regression model demonstrated high discrimination (mean AUC-PR 0.774 for all stages) and excellent calibration, consistently outperforming the benchmark classifiers. By jointly predicting sCr and urine output distributions, a single model successfully enables flexible risk stratification across all stages, capturing AKI onset and persistence, and allowing changes to stage definitions. Conclusion This multi-step, multivariate distributional regression model is a reliable, more flexible, transparent, and clinically interpretable approach for AKI prediction compared to traditional classification methods. It represents a necessary step toward bedside implementation of predictive models for personalized AKI management in the ICU.

Post-hoc, model-agnostic local explainability for medical time series classification is commonly achieved through saliency-based methods, which highlight where models focus, but fail to capture what underlying patterns drive predictions. This limitation is particularly problematic in clinical settings, where trustworthy and interpretable explanations are essential. To address this, we propose a model-agnostic framework for generating and visualizing explanations based on abstract (latent) features, rather than temporal segments. By mapping each time series to an interpretable latent representation and applying feature attribution methods within this space, our approach enables explanations that reflect meaningful global and local characteristics of the signal. As a proof-of-concept, we implement this framework using a variational autoencoder to learn abstract features from a synthetic and a real-world medical ECG dataset. Our findings demonstrate that this approach can reveal clinically relevant insights that cannot be found with saliency-based explanations. We also highlight key challenges for this approach, such as achieving good feature disentanglement. Overall, this study highlights the potential of abstract feature-based explanations as a promising direction for improving interpretability and trust in medical time series classification models.

Neuromodulation studies require efficient exploration of high-dimensional stimulation spaces, where heuristic tuning is often slow and suboptimal. We present OnlineNeuro, an open-source Python framework that combines active learning with neural simulators (AxonSim, Cajal, and AxonML). The package offers a unified interface for experiment setup, model training, adaptive sampling, and reporting. By prioritizing informative queries, OnlineNeuro improves sample efficiency for parameter exploration and meta-model construction. We demonstrate the framework on neural simulation use cases and benchmark tasks.

Automatic anatomical landmark localization in medical imaging requires not just accurate predictions but reliable uncertainty quantification for effective clinical decision support. Current uncertainty quantification approaches often fall short, particularly when combined with normality assumptions, systematically underestimating total predictive uncertainty. This paper introduces conformal prediction as a framework for reliable uncertainty quantification in anatomical landmark localization, addressing a critical gap in automatic landmark localization. We present two novel approaches guaranteeing finite-sample validity for multi-output prediction: Multi-output Regression-as-Classification Conformal Prediction (M-R2CCP) and its variant Multi-output Regression to Classification Conformal Prediction set to Region (M-R2C2R). Unlike conventional methods that produce axis-aligned hyperrectangular or ellipsoidal regions, our approaches generate flexible, non-convex prediction regions that better capture the underlying uncertainty structure of landmark predictions. Through extensive empirical evaluation across multiple 2D and 3D datasets, we demonstrate that our methods consistently outperform existing multi-output conformal prediction approaches in both validity and efficiency. This work represents a significant advancement in reliable uncertainty estimation for anatomical landmark localization, providing clinicians with trustworthy confidence measures for their diagnoses. While developed for medical imaging, these methods show promise for broader applications in multi-output regression problems.

Proper scoring rules provide a rigorous theoretical basis for the training and evaluation of probabilistic forecasts. However, in the presence of right censoring, the event time is only partially observed, rendering conventional scoring rules inapplicable in their standard form. We propose a framework for proper scoring of right-censored survival outcomes based on a simple idea: first, map the predictive distribution through the censoring mechanism, then apply the underlying proper score on the induced observed-data law. This yields localized scores for fixed censoring times and marginalized scores when the censoring time is random or only partially observed. The resulting construction recovers familiar right-censored likelihood and IPCW-type criteria within a coherent framework, while also yielding right-censored versions of the CRPS, pinball loss, Brier score, and energy score. We show that the marginalized score is proper under conditional independent censoring and strictly proper on the identifiable region. The same principle also leads to censored engression, a sample-based learning objective for multivariate right-censored survival modeling. In experiments, our scores correctly rank the oracle forecast across several censoring regimes, whereas forecast-dependent plug-in weighted scores can exhibit ranking reversals. Censored engression likewise substantially improves over naive training on censored outcomes.

Understanding the dose-response relation between a continuous treatment and the outcome for an individual can greatly drive decision-making, particularly in areas like personalized drug dosing and personalized healthcare interventions. Point estimates are often insufficient in these high-risk environments, highlighting the need for uncertainty quantification to support informed decisions. Conformal prediction, a distribution-free and model-agnostic method for uncertainty quantification, has seen limited application in continuous treatments or dose-response models. To address this gap, we propose a novel methodology that frames the causal dose-response problem as a covariate shift, leveraging weighted conformal prediction. By incorporating propensity estimation, conformal predictive systems, and likelihood ratios, we present a practical solution for generating prediction intervals for dose-response models. Additionally, our method approximates local coverage for every treatment value by applying kernel functions as weights in weighted conformal prediction. Finally, we use a new synthetic and semi-synthetic benchmark dataset to demonstrate the significance of covariate shift assumptions in achieving robust prediction intervals for counterfactual dose-response models.

Interactive line chart visualizations greatly enhance the effective exploration of large time series. Although downsampling has emerged as a well-established approach to enable efficient interactive visualization of large datasets, it is not an inherent feature inmost visualization tools. Furthermore, there is no library offering a convenient interface for high-performance implementations of prominent downsampling algorithms. To address these shortcomings, we present tsdownsample, an open-source Python package specifically designed for CPU-based, in-memory time series downsampling. Our library focuses on performance and convenient integration, offering optimized implementations of leading downsampling algorithms. We achieve this optimization by leveraging low-level Single Instruction, Multiple Data (SIMD) instructions and multithreading capabilities in Rust. In particular, SIMD instructions were employed to optimize the argmin and argmax operations. This SIMD optimization, along with some algorithmic tricks, proved crucial in enhancing the performance of various downsampling algorithms. We evaluate the performance of tsdownsample and demonstrate its interoperability with an established visualization framework. Our performance benchmarks indicate that the algorithmic runtime of tsdownsample approximates the CPU's memory bandwidth. This work marks a significant advancement in bringing high-performance time series downsampling to the Python ecosystem, enabling scalable visualization.

Motion can exacerbate headache during a migraine attack, potentially leading to avoidance of routine physical activity. Advances in wrist-worn actigraphy facilitate objectively analyzing how headache episodes affect physical activity in everyday settings. The primary hypothesis was hypoactivity during daytime headache events. Secondary hypotheses are hypoactivity during the prodromal and postdromal hours closest to the headache event. During a 90-day prospective observational study, participants diagnosed with migraine wore an actigraphy device on their non-dominant wrist during daily life and work, while also logging migraine-related data in a dedicated smartphone application. There were no restrictions on use of acute and preventive headache treatments. Data from the wrist-worn accelerometer were used to (i) calculate activity energy expenditure, and (ii) predict types of human activities. These metrics were used to compare daytime prodromal, ictal, and postdromal phases of headache events with time-matched intervals during non-headache periods. A significant reduction in daytime physical activity was observed during the ictal phase of headache attacks, as evidenced by decreases in both activity energy expenditure and human activity recognition prediction metrics. A reduction in movement was also observed during evening hours (18:00–24:00) on headache days. However, no significant physical activity changes were noted in the prodromal and postdromal phases. Reduced physical activity was more pronounced during the ictal phase when acute treatments were ineffective. This study is the first to examine the impact of headache on physical activity levels during daytime headache events by assessing changes in daily activities and activity energy expenditure in individuals with migraine, within their habitual environments and without restrictions on acute medication use. Our findings confirm reduced movement during the ictal phase of migraine attacks, supporting the primary hypothesis. Wrist-worn actigraphy further indicated that this reduction is more pronounced when patients experience movement sensitivity. Evening hypoactivity is also observed on headache days. Furthermore, attacks with ineffective acute treatment or moderate-to-high intensity were associated with more pronounced reductions in movement. In contrast, our data did not support the secondary hypothesis that physical activity would decrease during daytime prodromal and postdromal periods. NCT04983186 ( www.ClinicalTrials.gov ).

Objective: This research aims to address the challenges of just-in-time adaptive interventions (JITAIs) in behaviour change by introducing an architecture that integrates both the tailoring of the message to the user profile and context, and the timing of the intervention by detecting the trigger of the behaviour. Methods: We designed a system that integrates trigger detection to determine optimal intervention moments and uses prompt engineering on a large language model (LLM) to give personalised support based on the detected trigger, the context, and personal information of the person. As a proof of concept, we applied this intervention to the domain of smoking cessation. We conducted an in-depth semi-structured interview with a domain expert to evaluate the correctness, relevancy and personalisation of the chatbot’s responses. Results: An expert indicated that the support given by the chatbot is correct, personal, and tailored to the trigger and circumstances. While some suggestions were provided to further enhance the chatbot, its current capabilities were deemed effective and acceptable as a supportive tool for smoking cessation. Conclusions: An LLM with prompt engineering can be used to create a chatbot that can react to a trigger in a personalised way. Integrating both trigger detection and a generative chatbot into a JITAI is possible while ensuring privacy of the individual’s personal information and circumstances.

AI-driven solutions are being employed in process monitoring and control to learn typical system behaviours under various conditions based on historical data. However, they are unable to take advantage of the rich, tacit domain expertise of experienced process engineers pertaining to these behaviours. Hybrid AI solutions are designed to fuse domain knowledge into machine learning models, but have so far been limited to specific industrial subdomains or applications and support only domain expertise in the form of equations. We propose an explainable hybrid AI methodology that can integrate any kind of tacit knowledge in an interpretable manner. First, we introduce a method to consolidate process data and domain-specific expertise in a generic fashion using Knowledge Graphs. Second, we propose a Knowledge Graph transformation technique to better capture the sequential aspects of a process and an accompanying white-box Knowledge Graph embedding technique that allows us to integrate domain knowledge directly into the feature space of a data-driven model. Third, we show how our methodology can be combined with explainability techniques, such as SHAP, to highlight directly in the graph which paths contributed most to the AI-driven decision. Our methodology has been evaluated on two real-world chemical engineering use cases. It outperforms data-driven baselines on all performance metrics, with average improvements of up to 8.57% and 10.21%.

Chronic obstructive pulmonary disease (COPD) is a leading cause of death worldwide and greatly reduces the quality of life. Utilizing remote monitoring has been shown to improve quality of life and reduce exacerbations, but remains an ongoing area of research. We introduce a novel method for estimating changes in ease of breathing for COPD patients, using obstructed breathing data collected via wearables. Physiological signals were recorded, including respiratory airflow, acceleration, audio, and bio-impedance. By comparing patient-specific measurements, this approach enables non-intrusive remote monitoring. We analyze the influence of signal selection, window parameters, feature engineering, and classification models on predictive performance, finding that acceleration signals are most effective, complemented by audio signals. The best model achieves an F1-score of 0.83. To facilitate clinical adoption, we incorporate interpretability by designing novel saliency map methods, highlighting important aspects of the signals. We adapt local explainability techniques to time series and introduce a novel imputation method for periodic signals, improving faithfulness to the data and interpretability.

Control systems for building services, such as heating and cooling, often rely on fixed timing schemes. While such approaches are convenient, they make strong assumptions about room usage, often leading to inadequate comfort and energy efficiency. This study addresses this limitation by presenting two contributions aimed at automating the configuration of building control systems. The first contribution involves the development of a presence detection model based on CO2 data, which is easy to measure and non-privacy intrusive. Unlike existing literature, which typically focuses on single-room applications, this work introduces a dataset and machine learning methodology demonstrating the generalizability of a presence detection model across various real-world rooms, even among different building types. Sliding window normalization of the sensor data is the key to achieve this unsupervised cross-room adaptability. As second contribution, we propose an occupancy profiling technique that relies on the predicted presence information. This approach facilitates the automated configuration of building control systems by using historical presence probabilities to anticipate future occupancy. In contrast to fixed timing schemes, these occupancy profiles dynamically adapt over time, accommodating changes in occupant behavior. As such, this work improves the configuration of building control systems, leading to a more comfortable and energy-efficient environment.

Migraine is a neurological disorder that affects millions of people worldwide. It is one of the most debilitating disorders which leads to many disability-adjusted life years. Conventional methods for investigating migraines, like patient interviews and diaries, suffer from self-reporting biases and intermittent tracking. This study aims to leverage smartphone-derived data as an objective tool for examining the relationship between migraines and various human behavior aspects. By utilizing built-in sensors and monitoring phone interactions, we gather data from which we derive metrics such as keyboard usage, application interaction, physical activity levels, ambient light conditions, and sleep patterns. We perform statistical analysis testing to investigate whether there is a difference in user behavioral aspects during headache and non-headache periods. Our analysis of 362 headaches reveals differences in behavioral aspects such as ambient light, use of leisure apps, and number of keystrokes during headache periods and non-headache periods. This exploratory study shows on the one hand that it is possible to monitor various human behavioral aspects using the smartphone sensors and interaction data only. On the other hand it shows that we can observe difference in human behavior between headache and non-headache periods. Our work is a step towards objectively measure the effects that migraine has on people’s lives.