As AI systems increasingly display human-like behavior, people often attribute consciousness to these machines, raising questions about anthropomorphism, ethics, and design. We report findings from an online survey (N=553), including academics from the formal sciences, natural sciences, humanities, and a heterogeneous group of other participants. Respondents evaluated perceptions of consciousness in large language models and future AI, alongside related ethical and policy considerations. The results show that, across groups, around half of the participants attributed some degree of consciousness. Individual traits such as gender, academic position, and beliefs regarding consciousness and intelligence of systems strongly shape perceptions, outweighing the effects of technical knowledge or system transparency. Beyond shaping academic discussions, these perspectives inform how AI is designed, governed, and integrated into everyday interactions.

Monitoring the propagation of mechanical cardiac signals throughout the body is crucial for assessing cardiovascular health. A common drawback of current gold standard methods for vital sign monitoring is the necessity for continuous skin contact. Radar-based sensing offers a promising alternative by enabling contactless measurement of cardiac activity, including heart sound signals. As previous research has primarily focused on deriving signals from proximal body regions, insights into heart sound propagation to peripheral areas are lacking. To address this, we systematically investigated whether radar-based heart sound detection and propagation measurement is feasible across the whole body. We recorded heart sounds in N = 22 participants sequentially at eleven locations using a custom-built continuous-wave radar system and phonocardiogram as heart sound gold standard. Additionally, an electrocardiogram was acquired as reference for overall heart activity. After synchronization and preprocessing, we manually segmented the heart sounds and extracted temporal characteristics from ensemble-averaged signals. Our findings show that heart sounds can be detected across the entire body with the radar-based as well as the gold standard system. Furthermore, the heart sounds' temporal characteristics vary between measurement locations. As the distance to the heart increases, we observed significantly increased propagation time intervals. This finding is consistent across both systems, exhibiting a strong agreement for the first heart sound ($\mathbf {r = 0.73, p < 0.001}$) and a moderate agreement for the second heart sound ($\mathbf {r = 0.56, p < 0.001}$). In conclusion, our work is the first to demonstrate that radar-based systems are feasible for contactless evaluation of heart sound propagation, offering new possibilities for research and health monitoring.

In palliative care, effective communication about anticipated death is critical for aligning therapeutic goals, managing family expectations, and ensuring dignified care. However, prognostic uncertainty - particularly regarding the time of death - remains a challenge due to the limited reliability of current methods. This study explores the potential of radar-derived motion biomarkers as a novel approach to distinguish between living and deceased patients, addressing the need for objective decision-support tools in palliative care. Using continuous-wave radar, we recorded the torso displacement (distance signal) of 16 palliative care patients during their dying phase and derived ground-truth annotations from electronic health records (EHR). Machine learning (ML) algorithms processed 5-minute segments of radar-derived motion signals for binary vital status classification. We evaluated the results with balanced accuracy, Gini gain, and SHAP values. Palliative care specialists provided qualitative feedback to ensure clinical relevance. The ML models achieved balanced accuracy of 0.92-0.98 in distinguishing vital states, demonstrating radar technology's potential as an objective monitoring tool. This study is the first to investigate continuous motion biomarkers in end-of-life patients under real-world clinical conditions, capturing physiological changes during this critical phase. Limitations include the challenges in EHR-derived annotation accuracy, as well as the inherent complexity of physiological variability near death. Our findings highlight radar technology's viability for complementary vital status monitoring in palliative care settings. By providing objective data, this approach could reduce prognostic uncertainty while maintaining patient dignity. This work bridges technological innovation with palliative care's humanistic ethos, offering new possibilities for evidence-based end-of-life management.

Efficient data compression is crucial for the storage and transmission of visual data. However, in facial expression recognition (FER) tasks, lossy compression often leads to feature degradation and reduced accuracy. To address these challenges, this study proposes an end-to-end model designed to preserve critical features and enhance both compression and recognition performance. A custom loss function is introduced to optimize the model, tailored to balance compression and recognition performance effectively. This study also examines the influence of varying loss term weights on this balance. Experimental results indicate that fine-tuning the compression model alone improves classification accuracy by 0.71 % and compression efficiency by 49.32 %, while joint optimization achieves significant gains of 4.04 % in accuracy and 89.12 % in efficiency. Moreover, the findings demonstrate that the jointly optimized classification model maintains high accuracy on both compressed and uncompressed data, while the compression model reliably preserves image details, even at high compression rates.

Random splitting of datasets in image segmentation often leads to unrepresentative test sets, resulting in biased evaluations and poor model generalization. While stratified sampling has proven effective for addressing label distribution imbalance in classification tasks, extending these ideas to segmentation remains challenging due to the multi-label structure and class imbalance typically present in such data. Building on existing stratification concepts, we introduce Iterative Pixel Stratification (IPS), a straightforward, label-aware sampling method tailored for segmentation tasks. Additionally, we present Wasserstein-Driven Evolutionary Stratification (WDES), a novel genetic algorithm designed to minimize the Wasserstein distance, thereby optimizing the similarity of label distributions across dataset splits. We prove that WDES is globally optimal given enough generations. Using newly proposed statistical heterogeneity metrics, we evaluate both methods against random sampling and find that WDES consistently produces more representative splits. Applying WDES across diverse segmentation tasks, including street scenes, medical imaging, and satellite imagery, leads to lower performance variance and improved model evaluation. Our results also highlight the particular value of WDES in handling small, imbalanced, and low-diversity datasets, where conventional splitting strategies are most prone to bias.

Abstract Explainability in artificial intelligence (XAI) remains a crucial aspect for fostering trust and understanding in machine learning models. Current visual explanation techniques, such as gradient-based or class-activation-based methods, often exhibit a strong dependence on specific model architectures. Conversely, perturbation-based methods, despite being model-agnostic, are computationally expensive as they require evaluating models on a large number of forward passes. We introduce Foveation-based Explanations (FovEx), a novel XAI method inspired by human vision, which combines biologically inspired foveation-based transformations with gradient-driven overt attention to iteratively select locations of interest. These locations are selected to maximize the performance of the model to be explained with respect to the downstream task and then combined to generate an attribution map. We provide a thorough evaluation with qualitative and quantitative assessments on established benchmarks. Our method achieves state-of-the-art performance on both transformers (on 4 out of 5 metrics) and convolutional models (on 3 out of 5 metrics), demonstrating its versatility among various architectures. Furthermore, we show the alignment between the explanation map produced by FovEx and human gaze patterns (+14% in NSS compared to RISE, +203% in NSS compared to GradCAM). This comparison enhances our confidence in FovEx’s ability to close the interpretation gap between humans and machines.

Correction: International Journal of Computer Vision (2025) 133:7437–7459

The European Health Data Space (EHDS) aims to enable the sharing of health data across Europe to improve healthcare and research. While the EHDS mandates anonymization or pseudonymization of shared health data, these techniques may still allow adversaries to re-identify individuals. Local differential privacy (LDP) has been proposed as a formal privacy guarantee that can help mitigate this issue. In this paper, we consider a common problem when analyzing health data: estimating means for different groups. We discuss a generic privacy-preserving method for approximating the means of different groups in a decentralized setting where both the group and the value are considered private. We show that four concrete instantiations of the method based on existing mean estimation methods (Laplace, Bernoulli, Piecewise, and NPRR) are locally differentially private. We evaluate their performance on synthetic and real-world medical datasets. Our results show that the proposed methods can accurately estimate the group means, while maintaining privacy. However, similar to other LDP algorithms, our approach requires a sufficient amount of data (in our case a sufficient amount of samples per group) combined with a sufficiently large privacy budget ε to produce accurate results. We discuss concrete practical issues like choosing an appropriate input range, dealing with large privacy budgets through the use of the shuffle model of differential privacy, and the need for further analysis techniques to make LDP solutions applicable to practical medical data analysis.

Machine learning has become increasingly important in biomechanics. It allows to unveil hidden patterns from large and complex data, which leads to a more comprehensive understanding of biomechanical processes and deeper insights into human movement. However, machine learning models are often trained on a single dataset with a limited number of participants, which negatively affects their robustness and generalizability. Combining data from multiple existing sources provides an opportunity to overcome these limitations without spending more time on recruiting participants and recording new data. It is furthermore an opportunity for researchers who lack the financial requirements or laboratory equipment to conduct expensive motion capture studies themselves. At the same time, subtle interlaboratory differences can be problematic in an analysis due to the bias that they introduce. In our study, we investigated differences in motion capture datasets in the context of machine learning, for which we combined overground walking trials from four existing studies. Specifically, our goal was to examine whether a machine learning model was able to predict the original data source based on marker and GRF trajectories of single strides and how different scaling methods and pooling procedures affected the outcome. Layer-wise relevance propagation was applied to understand which factors were influential to distinguish the original data sources. We found that the model could predict the original data source with a very high accuracy (up to \({\gt}\) 99%), which decreased by about 15 percentage points when we scaled every dataset individually prior to pooling. However, none of the proposed scaling methods could fully remove the dataset bias. Layer-wise relevance propagation revealed that there was not only one single factor that differed between all datasets. Instead, every dataset had its unique characteristics that were picked up by the model. These variables differed between the scaling and pooling approaches but were mostly consistent between trials belonging to the same dataset. Our results show that motion capture data is sensitive even to small deviations in marker placement and experimental setup and that small inter-group differences should not be overinterpreted during data analysis, especially when the data was collected in different labs. Furthermore, we recommend scaling datasets individually prior to pooling them which led to the lowest accuracy. We want to raise awareness that differences in datasets always exist and are recognizable by machine learning models. Researchers should thus think about how these differences might affect their results when combining data from different studies.

Rheumatoid arthritis (RA) is a chronic inflammatory autoimmune disease affecting peripheral joints. Before clinical diagnosis, individuals may possess certain antibodies and experience discomfort but without specific signs of RA or inflamed joints. This stage is termed "preclinical RA," as these individuals are at risk of developing the disease. This early stage is difficult to define, necessitating the development of individual risk models. This study aims to estimate the time and risk of RA onset using various survival machine learning models. After identifying the best model, we stratify patients into risk categories and identify key risk factors. Data from 154 anonymized preclinical RA patients were collected and analyzed. Several survival analysis models were evaluated, including Survival Tree, Random Survival Forest, Extreme Gradient Boosting Survival, Linear Multi-Task Model, Neural Multi-Task Model, Support Vector Machines, and Cox Proportional Hazards. The Random Survival Forest model outperformed the others, achieving a mean C-index of 0.798. Using this model, patients were stratified into low-, medium-, and high-risk groups, facilitating personalized scheduling of clinical visits based on RA risk. To enhance model interpretability, SHapley Additive Explanations (SHAP) are employed to identify key risk factors. The baseline level of rheumatoid factor (RF) antibodies is the most significant predictor. Higher levels of anti-cyclic citrullinated peptide (anti-CCP) and RF antibodies at baseline are linked to earlier RA onset. This method provides valuable insights into key factors that might be overlooked in clinical practice and can improve patient management and quality of life for those at risk of developing RA.

Mixed-type time series (MTTS) is a bimodal data type that is common in many domains, such as healthcare, finance, environmental monitoring, and social media. It consists of regularly sampled continuous time series and irregularly sampled categorical event sequences. The integration of both modalities through multimodal fusion is a promising approach for processing MTTS. However, the question of how to effectively fuse both modalities remains open. In this paper, we present a comprehensive evaluation of several deep multimodal fusion approaches for MTTS forecasting. Our comparison includes different fusion types (early, intermediate, and late) and fusion methods (concatenation, weighted mean, weighted mean with correlation, gating, and feature sharing). We evaluate these fusion approaches on three distinct datasets, one of which was generated using a novel framework. This framework allows for the control of key data properties, such as the strength and direction of intermodal interactions, modality imbalance, and the degree of randomness in each modality, providing a more controlled environment for testing fusion approaches. Our findings show that the performance of different fusion approaches can be substantially influenced by the direction and strength of intermodal interactions. The study reveals that early and intermediate fusion approaches excel at capturing fine-grained and coarse-grained cross-modal features, respectively. These findings underscore the crucial role of intermodal interactions in determining the most effective fusion strategy for MTTS forecasting.

Certified defenses against adversarial attacks offer formal guarantees on the robustness of a model, making them more reliable than empirical methods such as adversarial training, whose effectiveness is often later reduced by unseen attacks. Still, the limited certified robustness that is currently achievable has been a bottleneck for their practical adoption. Gowal et al. and Wang et al. have shown that generating additional training data using state-of-the-art diffusion models can considerably improve the robustness of adversarial training. In this work, we demonstrate that a similar approach can substantially improve deterministic certified defenses but also reveal notable differences in the scaling behavior between certified and empirical methods. In addition, we provide a list of recommendations to scale the robustness of certified training approaches. Our approach achieves state-of-the-art deterministic robustness certificates on CIFAR-10 for the $\ell_2$ ($\epsilon = 36/255$) and $\ell_{\infty}$ ($\epsilon = 8/255$) threat models, outperforming the previous results by $+3.95$ and $+1.39$ percentage points, respectively. Furthermore, we report similar improvements for CIFAR-100.

Gait is an indicator of a person's health status and abnormal gait patterns are associated with a higher risk of falls, dementia, and mental health disorders. Wearable sensors facilitate long-term assessment of walking in the user's home environment. Earables, wearable sensors that are worn at the ear, are gaining popularity for digital health assessments because they are unobtrusive and can easily be integrated into the user's daily routine, for example, in hearing aids. A comprehensive gait analysis pipeline for an ear-worn accelerometer that includes spatial-temporal parameters is currently not existing. Therefore, we propose and compare three algorithmic approaches to estimate step length and gait speed based on ear-worn accelerometer data: a biomechanical model, feature-based machine learning (ML) models, and a convolutional neural network. We evaluated their performance on a step and walking bout level and compared it with an optical motion capture system. The feature-based ML model achieved the best performance with a precision of 4.8cm on a walking bout level. For gait speed, the machine learning approach achieved an absolute percentage error of 5.4 % ( ± 4.0 %). We find that the ML model is able to estimate step length and gait speed with clinically relevant precision. Furthermore, the model is insensitive to different age groups and sampling rates but sensitive to walking speed. To our knowledge, this work is the first contribution to estimating step length and gait speed using ear-worn accelerometers. Moreover, it lays the foundation for a comprehensive gait analysis framework for ear-worn sensors enabling continuous and long-term monitoring at home.

Outlier detection stands as a critical cornerstone in the field of data mining, with a wide range of applications spanning from fraud detection to network security. However, real-world scenarios often lack labeled data for training, necessitating unsupervised outlier detection methods. This study centers on Unsupervised Outlier Model Selection (UOMS), with a specific focus on the family of Internal, Relative Evaluation of Outlier Solutions (IREOS) algorithms. IREOS measures outlier candidate separability by evaluating multiple maximum-margin classifiers and, while effective, it is constrained by its high computational demands. We investigate the impact of several different separation methods in UOMS in terms of ranking quality and runtime. Surprisingly, our findings indicate that different separability measures have minimal impact on IREOS’ effectiveness. However, using linear separation methods within IREOS significantly reduces its computation time. These insights hold significance for real-world applications where efficient outlier detection is critical. In the context of this work, we provide the code for the IREOS algorithm and our separability techniques.

Clustering is at the very core of machine learning, and its applications proliferate with the increasing availability of data. However, as datasets grow, comparing clusterings with an adjustment for chance becomes computationally difficult, preventing unbiased ground-truth comparisons and solution selection. We propose FastAMI, a Monte Carlo-based method to efficiently approximate the Adjusted Mutual Information (AMI) and extend it to the Standardized Mutual Information (SMI). The approach is compared with the exact calculation and a recently developed variant of the AMI based on pairwise permutations, using both synthetic and real data. In contrast to the exact calculation our method is fast enough to enable these adjusted information-theoretic comparisons for large datasets while maintaining considerably more accurate results than the pairwise approach.

The widespread success of convolutional neural networks may largely be attributed to their intrinsic property of translation equivariance. However, convolutions are not equivariant to variations in scale and fail to generalize to objects of different sizes. Despite recent advances in this field, it remains unclear how well current methods generalize to unobserved scales on real-world data and to what extent scale equivariance plays a role. To address this, we propose the novel Scaled and Translated Image Recognition (STIR) benchmark based on four different domains. Additionally, we introduce a new family of models that applies many re-scaled kernels with shared weights in parallel and then selects the most appropriate one. Our experimental results on STIR show that both the existing and proposed approaches can improve generalization across scales compared to standard convolutions. We also demonstrate that our family of models is able to generalize well towards larger scales and improve scale equivariance. Moreover, due to their unique design we can validate that kernel selection is consistent with input scale. Even so, none of the evaluated models maintain their performance for large differences in scale, demonstrating that a general understanding of how scale equivariance can improve generalization and robustness is still lacking.

Indoor radio environments often consist of areas with mixed propagation conditions. In LoS-dominated areas, classic ToF methods reliably return optimal (accurate) positions, while in NLoS-dominated areas (AI-based) fingerprinting methods are required. However, these fingerprinting methods are only cost-efficient if they are used exclusively in NLoS-dominated areas due to an expensive life cycle management. Systems that are both accurate and cost-efficient in LoS- and NLoS-dominated areas require an identification of those areas to select the optimal localization method. In this paper we propose methods for uncertainty estimation of AI-based fingerprinting to determine its validity. Our experiments show that we can implicitly switch between classic and fingerprinting-based approaches to reliably estimate accurate positions, even in NLoS-dominated radio environments. Our approach works even if the AI models are only trained on radio data in certain areas of the environment. In contrast to the state-of-the-art, our approach intrinsically identifies the spatial boundaries of the AI model, and thus does not require prior area identification.

Popular metrics for clustering comparison, like the Adjusted Rand Index and the Adjusted Mutual Information, are type II biased. The Standardized Mutual Information removes this bias but suffers from counterintuitive non-monotonicity and poor computational efficiency. We introduce the $p$-value adjusted Rand Index ($\operatorname{PMI}_2$), the first cluster comparison method that is type II unbiased and provably monotonous. The $\operatorname{PMI}_2$ has fast approximations that outperform the Standardized Mutual information. We demonstrate its unbiased clustering selection, approximation quality, and runtime efficiency on synthetic benchmarks. In experiments on image and social network datasets, we show how the $\operatorname{PMI}_2$ can help practitioners choose better clustering and community detection algorithms.

Abstract Progress in making neural networks more robust against adversarial attacks is mostly marginal, despite the great efforts of the research community. Moreover, the robustness evaluation is often imprecise, making it challenging to identify promising approaches. We do an observational study on the classification decisions of 19 different state-of-the-art neural networks trained to be robust against adversarial attacks. This analysis gives a new indication of the limits of the robustness of current models on a common benchmark. In addition, our findings suggest that current untargeted adversarial attacks induce misclassification toward only a limited amount of different classes. Similarly, we find that previous attacks under-explore the perturbation space during optimization. This leads to unsuccessful attacks for samples where the initial gradient direction is not a good approximation of the final adversarial perturbation direction. Additionally, we observe that both over- and under-confidence in model predictions result in an inaccurate assessment of model robustness. Based on these observations, we propose a novel loss function for adversarial attacks that consistently improves their efficiency and success rate compared to prior attacks for all 30 analyzed models.

Digital technology has had a significant impact on biomedical sciences and healthcare delivery, not only in terms of theoretical and practical contributions to involved disciplines but also due to the profound and rapid changes in medical research and in medical care applications, as well as in the management organization after the recent COVID-19 epidemic. In this regard, several elements stand out from the 11 selected contributions published in this Special Issue.

From diagnosis to therapies, the development of artificial intelligence (AI) has facilitated improvements in personalised healthcare applications. The evolution of AI in healthcare is closely related to the changes in the types and volumes of data which we need to deal with. The first generation of healthcare technologies, represented by the highly successful relational databases, are designed to handle structured data involving patient demographics, patient care, treatments, and outcomes of those treatments. Big Data platforms, which are representative of the current mainstream healthcare technologies, are built to process unstructured data from sources like electronic health records, medical imaging, genomic sequencing, and pharmaceutical research. The next generation of healthcare technologies will potentially be Edge-of-Things data, represented by massive amount of streaming data generated from Internet-of-Things frameworks, Cloud systems, and Edge computing platforms.

Falls are an eminent risk for older adults and especially patients with neurodegenerative disorders, such as Parkinson's disease (PD). Recent advancements in wearable sensor technology and machine learning may provide a possibility for an individualized prediction of fall risk based on gait recordings from standardized gait tests or from unconstrained real-world scenarios. However, the most effective aggregation of continuous real-world data as well as the potential of unsupervised gait tests recorded over multiple days for fall risk prediction still need to be investigated. Therefore, we present a data set containing real-world gait and unsupervised 4x10-Meter-Walking-Tests of 40 PD patients, continuously recorded with foot-worn inertial sensors over a period of two weeks. In this prospective study, falls were self-reported during a three-month follow-up phase, serving as ground truth for fall risk prediction. The purpose of this study was to compare different data aggregation approaches and machine learning models for the prospective prediction of fall risk using gait parameters derived either from continuous real-world recordings or from unsupervised gait tests. The highest balanced accuracy of 74.0% (sensitivity: 60.0%, specificity: 88.0%) was achieved with a Random Forest Classifier applied to the real-world gait data when aggregating all walking bouts and days of each participant. Our findings suggest that fall risk can be predicted best by merging the entire two-week real-world gait data of a patient, outperforming the prediction using unsupervised gait tests (68.0% balanced accuracy) and contribute to an improved understanding of fall risk prediction.

The reliability of neural networks is essential for their use in safety-critical applications. Existing approaches generally aim at improving the robustness of neural networks to either real-world distribution shifts (e.g., common corruptions and perturbations, spatial transformations, and natural adversarial examples) or worst-case distribution shifts (e.g., optimized adversarial examples). In this work, we propose the Decision Region Quantification (DRQ) algorithm to improve the robustness of any differentiable pre-trained model against both real-world and worst-case distribution shifts in the data. DRQ analyzes the robustness of local decision regions in the vicinity of a given data point to make more reliable predictions. We theoretically motivate the DRQ algorithm by showing that it effectively smooths spurious local extrema in the decision surface. Furthermore, we propose an implementation using targeted and untargeted adversarial attacks. An extensive empirical evaluation shows that DRQ increases the robustness of adversarially and non-adversarially trained models against real-world and worst-case distribution shifts on several computer vision benchmark datasets.

Healthcare 4.0 is a new concept that originates from the evolution of hospitals due to technological advances in medical activities. Nowadays, more and more doctors and healthcare administrators require real-time data analysis obtained from sensors and surgery monitoring. Using real-time information could make the difference between death and life in such settings. Therefore, Quality of Service (QoS) is essential in this context because, without it, the results of the applications become unreliable. Given the background, this article proposes HealthStack, a sensor middleware model for operating room facilities. HealthStack aims at improving delay and jitter from applications and, at the same time, reducing resource consumption. Our scientific contributions are twofold: 1) a middleware for operating rooms with automatic QoS support for real-time data transmission and 2) a QoS strategy based on artificial neurons to select middleware components with critical performance. We developed a prototype that uses depth cameras and ultrawideband (UWB) real-time location systems (RTLSs) to monitor workflow during surgery. The evaluation demonstrates that the strategy improves the average jitter experienced by application by 92.3%. The results also reveal a reduction of network and CPU consumption by up to 61.8% and 8.3%.

The use of machine learning (ML) with electronic health records (EHR) is growing in popularity as a means to extract knowledge that can improve the decision-making process in healthcare. Such methods require training of high-quality learning models based on diverse and comprehensive datasets, which are hard to obtain due to the sensitive nature of medical data from patients. In this context, federated learning (FL) is a methodology that enables the distributed training of machine learning models with remotely hosted datasets without the need to accumulate data and, therefore, compromise it. FL is a promising solution to improve ML-based systems, better aligning them to regulatory requirements, improving trustworthiness and data sovereignty. However, many open questions must be addressed before the use of FL becomes widespread. This article aims at presenting a systematic literature review on current research about FL in the context of EHR data for healthcare applications. Our analysis highlights the main research topics, proposed solutions, case studies, and respective ML methods. Furthermore, the article discusses a general architecture for FL applied to healthcare data based on the main insights obtained from the literature review. The collected literature corpus indicates that there is extensive research on the privacy and confidentiality aspects of training data and model sharing, which is expected given the sensitive nature of medical data. Studies also explore improvements to the aggregation mechanisms required to generate the learning model from distributed contributions and case studies with different types of medical data.

This work proposes metric learning for fast similarity-based scene retrieval of unstructured ensembles of trajectory data from large databases. We present a novel representation learning approach using Siamese Metric Learning that approximates a distance preserving low-dimensional representation and that learns to estimate reasonable solutions to the assignment problem. To this end, we employ a Temporal Convolutional Network architecture that we extend with a gating mechanism to enable learning from sparse data, leading to solutions to the assignment problem exhibiting varying degrees of sparsity. Our experimental results on professional soccer tracking data provides insights on learned features and embeddings, as well as on generalization, sensitivity, and network architectural considerations. Our low approximation errors for learned representations and the interactive performance with retrieval times several magnitudes smaller shows that we outperform previous state of the art.

Falls are among the leading causes of injuries or death for the elderly, and the prevalence is especially high for patients suffering from neurological diseases like Parkinson's disease (PD). Today, inertial measurement units (IMUs) can be integrated unobtrusively into patients' everyday lives to monitor various mobility and gait parameters, which are related to common risk factors like reduced balance or reduced lower-limb muscle strength. Although stair ambulation is a fundamental part of everyday life and is known for its unique challenges for the gait and balance system, long-term gait analysis studies have not investigated real-world stair ambulation parameters yet. Therefore, we applied a recently published gait analysis pipeline on foot-worn IMU data of 40 PD patients over a recording period of two weeks to extract objective gait parameters from level walking but also from stair ascending and descending. In combination with prospective fall records, we investigated group differences in gait parameters of future fallers compared to non-fallers for each individual gait activity. We found significant differences in stair ascending and descending parameters. Stance time was increased by up to 20 % and gait speed reduced by up to 16 % for fallers compared to non-fallers during stair walking. These differences were not present in level walking parameters. This suggests that real-world stair ambulation provides sensitive parameters for mobility and fall risk due to the challenges stairs add to the balance and control system. Our work complements existing gait analysis studies by adding new insights into mobility and gait performance during real-world gait.

The visual depth perception is composed of monocular and binocular depth cues. Studies show that in absence of binocular depth cues the performance of visuomotor tasks like pointing to or grasping objects is limited. Thus, binocular depth cues are of great importance for motor control required in everyday life. However, binocular depth cues like retinal disparity (basis for stereopsis) might be influenced due to developmental disorders of the visual system. For example, amblyopia in which one eye's visual input is not processed leads to loss of stereopsis. The primary amblyopia treatment is occlusion of the healthy eye to force the amblyopic eye to train. However, improvements in stereopsis are poor. Therefore, binocular treatments arose that equilibrate both eyes' visual input to enable binocular vision. However, most approaches rely on divided stimuli which do not account for loss of stereopsis. We created a Virtual Reality (VR) with reduced monocular depth cues in which a stereoscopic task is shown to both eyes simultaneously, consisting of two balls jumping towards the user. One ball appears closer to the user which must be identified. To evaluate the task performance the reaction time is measured. We validated our approach with 18 participants with stereopsis under three contrast settings including one leading to monocular vision. The number of correct responses reduces from 90% under binocular vision to 52% under monocular vision corresponding to random guessing. Our results indicate that it is possible to disable monocular depth cues and create a dynamic stereoscopic task inside a VR.

The vulnerability of deep neural networks to small and even imperceptible perturbations has become a central topic in deep learning research. Although several sophisticated defense mechanisms have been introduced, most were later shown to be ineffective. However, a reliable evaluation of model robustness is mandatory for deployment in safety-critical scenarios. To overcome this problem we propose a simple yet effective modification to the gradient calculation of state-of-the-art first-order adversarial attacks. Normally, the gradient update of an attack is directly calculated for the given data point. This approach is sensitive to noise and small local optima of the loss function. Inspired by gradient sampling techniques from non-convex optimization, we propose Dynamically Sampled Nonlocal Gradient Descent (DSNGD). DSNGD calculates the gradient direction of the adversarial attack as the weighted average over past gradients of the optimization history. Moreover, distribution hyperparameters that define the sampling operation are automatically learned during the optimization scheme. We empirically show that by incorporating this nonlocal gradient information, we are able to give a more accurate estimation of the global descent direction on noisy and non-convex loss surfaces. In addition, we show that DSNGD-based attacks are on average 35% faster while achieving 0.9% to 27.1% higher success rates compared to their gradient descent-based counterparts.

The susceptibility of deep neural networks to untrustworthy predictions, including out-of-distribution (OOD) data and adversarial examples, still prevent their widespread use in safety-critical applications. Most existing methods either require a retraining of a given model to achieve robust identification of adversarial attacks or are limited to out-of-distribution sample detection only. In this work, we propose a geometric gradient analysis (GGA) to improve the identification of untrustworthy predictions without retraining of a given model. GGA analyzes the geometry of the loss landscape of neural networks based on the saliency maps of their respective input. We observe considerable differences between the input gradient geometry of trustworthy and untrustworthy predictions. Using these differences, GGA outperforms prior approaches in detecting OOD data and adversarial attacks, including state-of-the-art and adaptive attacks.

Gait supervision plays an important role in the diagnosis, analysis and rehabilitation of motor impairments and neurodegenerative disorders. For example, in Parkinson's disease, gait assessment is used for progression observation and medication guidance. Previous work has presented the potential of virtual reality (VR) supported gait applications. While virtual environments and user representation strategies are used for gait applications, the influence of appearance and context cues on gait performance is not extensively researched. In this paper, we analyzed the influence of avatar appearance, environment awareness, and camera perspective on gait parameters relevant for clinical application. Four different avatar appearances, varying in abstraction, two environmental settings, as well as an egocentric and exocentric camera perspective were compared in three walking tasks on a treadmill. Our results show that variability, as an indicator for gait stability, is significantly impacted by VR exposure in comparison to a real world (in vivo) baseline. Further, our results revealed that walking tasks influence gait behavior significantly different in VR compared to in vivo. Overall, these findings suggest that particular care has to be taken when assessing gait characteristics acquired from subjects immersed in VR and that equivalence of results with in vivo may not be blindly assumed.

The cooperation between robots and astronauts will become a core element of future space missions. This is accompanied by the demand for suitable input devices. An interface based on electromyography (EMG) represents a small, light, and wearable device to generate a continuous three-dimensional (3D) control signal from voluntarily muscle activation of the operator's arm. We analyzed the influence of microgravity on task performance during a two-dimensional (2D) task on a screen. Six subjects performed aiming and tracking tasks in parabolic flights. Three different levels of fixation-fixed feet using foot straps, semi-free by using a foot rail, and free-floating feet-are tested to investigate how much user fixation is required to operate via the interface. The user study showed that weightlessness affects the usage of the interface only to a small extent. Success rates between 89${\%}$ and 96${\%}$ are reached within all conditions during microgravity. A significant effect between 0 and 1G could not be identified for the test series of fixed and semi-free feet, while free-floating feet showed significantly worse results in fine and gross motion times in 0G compared to ground tests (with success rates of 92${\%}$ for 0G and 99${\%}$ for 1G). Further adaptation to the altered proprioception may be needed here. Hence, foot rails as already mounted in the International Space Station (ISS) would be sufficient to use the interface in weightlessness. Low impact of microgravity, high success rates, and an easy handling of the system, indicates a high potential of an EMG-based interface for teleoperation in space.

Online handwriting recognition has been studied for a long time with only few practicable results when writing on normal paper. Previous approaches using sensor-based devices encountered problems that limited the usage of the developed systems in real-world applications. This paper presents a writer-independent system that recognizes characters written on plain paper with the use of a sensor-equipped pen. This system is applicable in real-world applications and requires no user-specific training for recognition. The pen provides linear acceleration, angular velocity, magnetic field, and force applied by the user, and acts as a digitizer that transforms the analogue signals of the sensors into timeseries data while writing on regular paper. The dataset we collected with this pen consists of Latin lower-case and upper-case alphabets. We present the results of a convolutional neural network model for letter classification and show that this approach is practical and achieves promising results for writer-independent character recognition. This work aims at providing a real-time handwriting recognition system to be used for writing on normal paper.

Hereditary spastic paraplegias (HSP) represents a group of orphan neurodegenerative diseases with gait disturbance as the predominant clinical feature. Due to its rarity, research within this field is still limited. Aside from clinical analysis using established scales, gait analysis has been employed to enhance the understanding of the mechanisms behind the disease. However, state of the art gait analysis systems are often large, immobile and expensive. To overcome these limitations, this paper presents the first clinically relevant mobile gait analysis system for HSP patients. We propose an unsupervised model based on local cyclicity estimation and hierarchical hidden Markov models (LCE-hHMM). The system provides stride time, swing time, stance time, swing duration and cadence. These parameters are validated against a GAITRite system and manual sensor data labelling using a total of 24 patients within 2 separate studies. The proposed system achieves a stride time error of -0.00 ± 0.09 s (correlation coefficient, r = 1.00) and a swing duration error of -0.67 ± 3.27 % (correlation coefficient, r = 0.93) with respect to the GAITRite system. We show that these parameters are also correlated to the clinical spastic paraplegia rating scale (SPRS) in a similar manner to other state of the art gait analysis systems, as well as to supervised and general versions of the proposed model. Finally, we show a proof of concept for this system to be used to analyse alterations in the gait of individual patients. Thus, with further clinical studies, due to its automated approach and mobility, this system could be used to determine treatment effects in future clinical trials.

Mobile gait analysis using wearable inertial measurement units (IMUs) provides valuable insights for the assessment of movement impairments in different neurological and musculoskeletal diseases, for example Parkinson's disease (PD). The increase in data volume due to arising long-term monitoring requires valid, robust and efficient analysis pipelines. In many studies an upstream detection of gait is therefore applied. However, current methods do not provide a robust way to successfully reject non-gait signals. Therefore, we developed a novel algorithm for the detection of gait from continuous inertial data of sensors worn at the feet. The algorithm is focused not only on a high sensitivity but also a high specificity for gait. Sliding windows of IMU signals recorded from the feet of PD patients were processed in the frequency domain. Gait was detected if the frequency spectrum contained specific patterns of harmonic frequencies. The approach was trained and evaluated on 150 clinical measurements containing standardized gait and cyclic movement tests. The detection reached a sensitivity of 0.98 and a specificity of 0.96 for the best sensor configuration (angular rate around the medio-lateral axis). On an independent validation data set including 203 unsupervised, semi-standardized gait tests, the algorithm achieved a sensitivity of 0.97. Our algorithm for the detection of gait from continuous IMU signals works reliably and showed promising results for the application in the context of free-living and non-standardized monitoring scenarios.

This paper presents an approach for natural language question answering over a knowledge base generated by a medical texts information extraction process. The primary objective is to present a solution to help practitioners in oncology healthcare clinical environment with an intuitive method to access stored data. We identify health professional's needs in terms of information and interface with EHR systems. After that, we demonstrate a proposal to allow the integration of information extraction from clinical notes, knowledge base generation, and natural language question answering. The primary contributions are the identification of a solution to health professionals needs regarding usability in information access, and the demonstration of advantages obtained in representing health contents in a knowledge base.

The Stroop Test is a well known and regularly employed stressor in laboratory research. In contrast to other methods, it is not based on fear of physical harm or social shame. Consequently, it is more likely accepted by a wide population. In our always-on, technology-driven, social-media centered world, large-scale in-field stress research will need adequate experimental tools to explore the increasing prevalence of stress-related diseases without bringing subjects into laboratories. This is why we designed the Stroop Room: A virtual reality-based adaptation of the Stroop Test using elements of the virtual world to extend the demands of the original test and at the same time make it easily accessible. It is open source and can be used and improved by anyone as an in-the-wild, repeatable, laboratory-quality stressor. In this work, the method is presented and an evaluation study described, to demonstrate its effectiveness in provoking cognitive stress.

Parkinson's disease is a neurodegenerative disorder characterized by a variety of motor symptoms. Particularly, difficulties to start/stop movements have been observed in patients. From a technical/diagnostic point of view, these movement changes can be assessed by modeling the transitions between voiced and unvoiced segments in speech, the movement when the patient starts or stops a new stroke in handwriting, or the movement when the patient starts or stops the walking process. This study proposes a methodology to model such difficulties to start or to stop movements considering information from speech, handwriting, and gait. We used those transitions to train convolutional neural networks to classify patients and healthy subjects. The neurological state of the patients was also evaluated according to different stages of the disease (initial, intermediate, and advanced). In addition, we evaluated the robustness of the proposed approach when considering speech signals in three different languages: Spanish, German, and Czech. According to the results, the fusion of information from the three modalities is highly accurate to classify patients and healthy subjects, and it shows to be suitable to assess the neurological state of the patients in several stages of the disease. We also aimed to interpret the feature maps obtained from the deep learning architectures with respect to the presence or absence of the disease and the neurological state of the patients. As far as we know, this is one of the first works that considers multimodal information to assess Parkinson's disease following a deep learning approach.

We explore motion parameters, more specifically gait parameters, as an objective indicator to assess simulator sickness in Virtual Reality (VR). We discuss the potential relationships between simulator sickness, immersion, and presence. We used two different camera pose (position and orientation) estimation methods for the evaluation of motion tasks in a large-scale VR environment: a simple model and an optimized model that allows for a more accurate and natural mapping of human senses. Participants performed multiple motion tasks (walking, balancing, running) in three conditions: a physical reality baseline condition, a VR condition with the simple model, and a VR condition with the optimized model. We compared these conditions with regard to the resulting sickness and gait, as well as the perceived presence in the VR conditions. The subjective measures confirmed that the optimized pose estimation model reduces simulator sickness and increases the perceived presence. The results further show that both models affect the gait parameters and simulator sickness, which is why we further investigated a classification approach that deals with non-linear correlation dependencies between gait parameters and simulator sickness. We argue that our approach could be used to assess and predict simulator sickness based on human gait parameters and we provide implications for future research.

Cognitive skills and their assessment gain increasing importance in soccer. In the past, athletes' perceptual-cognitive capabilities were only assessed using 2D media with its limitations. We used a virtual reality 360° video environment, where 15 high-skilled and 15 low-skilled soccer players (age 24 ± 3 years) experienced nine real-life soccer scenes from varying perspectives. The experience was frozen at crucial time points where they had to decide for one of three soccer actions. The recognition-action time and risk-level for decision were determined. Furthermore, six different interaction concepts were evaluated with respect to user experience, presence and immersion to find the most adequate and appealing one for assessment and training in soccer. Results show that high-skillers had a significantly lower overall recognition-action time. Risk-values for decisions did not differ significantly between skill levels.

The ability to relax is sometimes challenging to achieve, nevertheless it is extremely important for mental and physical health, particularly to effectively manage stress and anxiety. We propose a virtual reality experience that integrates a wearable, low-cost EEG headband and an olfactory necklace that passively promotes relaxation. The physiological response was measured from the EEG signal. Relaxation scores were computed from EEG frequency bands associated with a relaxed mental state using an entropy-based signal processing approach. The subjective perception of relaxation was determined using a questionnaire. A user study involving 12 subjects showed that the subjective perception of relaxation increased by 26.1 % when using a VR headset with the olfactory necklace, compared to not being exposed to any stimulus. Similarly, the physiological response also increased by 25.0 %. The presented work is the first Virtual Reality Therapy system that uses scent in a wearable manner and proves its effectiveness to increase relaxation in everyday life situations.

Stress or anxiety disorders are a growing problem in industrialized countries. Those can be effectively countered by several relaxation techniques which are more effective using biofeedback. Modern virtual reality hardware provides a high level of immersion to its users. This directly affects the feeling of presence. An increased feeling of presence may allow biofeedback mechanisms to work more effectively. We build on this idea and explore how different visualizations of a user's cardiac activity in a virtual environment can be used in biofeedback scenarios - and how effective they are. Using a state-of-the-art virtual reality headset, 14 participants were subjected to four different visualizations of their own heart rate (one control visualization and three experimental visualizations). In different experiments, we examined whether they were able to estimate their heart rate based on the visualization and whether we could influence it subconsciously. Furthermore, we used the AttrakDiff questionnaire to assess the usability and attractiveness of each of the four visualizations. For the three non-reference ones, we observed significant positively correlating changes in the heart rate between the real-time true representation of the heart rate and a simulated increased heart rate visualization with a mean magnitude of 1.96 ± 0.39 beats per minute. The results from the source and number estimation experiments and the questionnaire led to the conclusion that the most appealing and best working visualization for biofeedback is a synchronized modulation/modification of the virtual environment itself.

An unobtrusive and low-cost landing analysis could not only support sports science research and ski jumping training but also decrease the risk of overuse and injuries. Although ski jumping biomechanics has been extensively researched, there is no known study on an unobtrusive analysis of the landing phase of actual ski jumps. In this work, we propose a landing momentum determination with inertial-magnetic measurement units (IMMUs) attached to the skis. We evaluate the calculated momenta against a mobile force plate and achieve accuracies of more than 90 % for three out of four jumps. Although the robustness of the measurement process can still be improved, our proposed algorithm builds the first step towards an IMMU-based landing analysis in ski jumping.

The knee flexion-extension angle is an important criterion for the rehabilitation process after a rupture of the anterior cruciate ligament (ACL), which is a typical sports injury. This paper describes the development and evaluation of a smart knee sleeve, which is capable of calculating the knee flexion-extension angle based on inertial sensors. The output is to be used to supervise and speed up rehabilitation by guiding patients through exercises and giving feedback about the execution. The accuracy of the knee angle calculation was evaluated in a study with twelve healthy subjects performing six different exercises with an optical system as reference. The effects of different functional alignment movements, knee sleeve material and recalibration on the final knee angle were evaluated. The mean absolute error (MAE) of all maxima and minima was calculated and averaged over all subjects and exercises. An MAE of 5.2° compared to the gold standard was achieved while the Pearson correlation was 0.96 for 30 minutes training without recalibration or restart of the system.

We present a system capable of computing two major kinematic parameters that are necessary for a running shoe recommendation. The system consists of one inertial measurement unit located in each sole of a pair of running shoes. This unobtrusive integration allows for a long-term and objective assessment of the strike type and the pronation of the foot, which can be characterized by the sole angle and the range of motion respectively. An algorithm for computing these parameters is presented, which includes a sensor to shoe alignment, a step segmentation and a quaternion based angle calculation. A study including 5112 ground contacts from 27 subjects was conducted to evaluate the accuracy of the presented algorithm. The best results compared to a motion capture system can be achieved with a subject dependent sensor to shoe alignment with a mean absolute error of 2.8° ± 2.5° for the sole angle and 2.0° ± 1.9° for the range of motion.

Today, radiologists diagnose three dimensional medical data using two dimensional displays. When designing environments with optimal conditions for such a process various aspects like contrast, screen reflection and background light have to be considered. As shown in previous research, applying virtual environments in combination with a Head-Mounted Display for diagnostic imaging provides potential benefits to reduce issues of bad posture and diagnostic mistakes. However, there is little research in exploring the usability and user experience of such beneficial environments. In this work we designed and evaluated different means of interaction to increase radiologists' performance. Therefore we created a virtual reality radiology reading room and employed it to evaluate three different interaction techniques. These allow a direct, semi-direct and indirect manipulation for performing scrolling- and windowing- tasks which are the most important for a radiologist. A study including nine radiologists was conducted and evaluated using the User Experience Questionnaire. Results indicate that direct manipulation is the preferred interaction technique, it outscored the other two control possibilities in attractiveness and pragmatic quality.

OBJECTIVE: Accurate estimation of spatial gait characteristics is critical to assess motor impairments resulting from neurological or musculoskeletal disease. Currently, however, methodological constraints limit clinical applicability of state-of-the-art double integration approaches to gait patterns with a clear zero-velocity phase. METHODS: We describe a novel approach to stride length estimation that uses deep convolutional neural networks to map stride-specific inertial sensor data to the resulting stride length. The model is trained on a publicly available and clinically relevant benchmark dataset consisting of 1220 strides from 101 geriatric patients. Evaluation is done in a tenfold cross validation and for three different stride definitions. RESULTS: Even though best results are achieved with strides defined from midstance to midstance with average accuracy and precision of , performance does not strongly depend on stride definition. The achieved precision outperforms state-of-the-art methods evaluated on the same benchmark dataset by . CONCLUSION: Due to the independence of stride definition, the proposed method is not subject to the methodological constrains that limit applicability of state-of-the-art double integration methods. Furthermore, it was possible to improve precision on the benchmark dataset. SIGNIFICANCE: With more precise mobile stride length estimation, new insights to the progression of neurological disease or early indications might be gained. Due to the independence of stride definition, previously uncharted diseases in terms of mobile gait analysis can now be investigated by retraining and applying the proposed method.

Wearable health sensors are about to change our health system. While several technological improvements have been presented to enhance performance and energy-efficiency, battery runtime is still a critical concern for practical use of wearable biomedical sensor systems. The runtime limitation is directly related to the battery size, which is another concern regarding practicality and customer acceptance. We introduced ULPSEK-Ultra-Low-Power Sensor Evaluation Kit-for evaluation of biomedical sensors and monitoring applications (http://ulpsek.com). ULPSEK includes a multiparameter sensor measuring and processing electrocardiogram, respiration, motion, body temperature, and photoplethysmography. Instead of a battery, ULPSEK is powered using an efficient body heat harvester. The harvester produced 171 W on average, which was sufficient to power the sensor below 25 C ambient temperature. We present design issues regarding the power supply and the power distribution network of the ULPSEK sensor platform. Due to the security aspect of self-powered health sensors, we suggest a hybrid solution consisting of a battery charged by a harvester.

Quantifying dynamic postural stability from inertial sensor data is clinically very relevant for treatment and therapy monitoring in neuromuscular diseases, e.g. Parkinson's disease (PD). We extract peak accelerations in movement direction during the loading phase and in vertical direction at ground contact from gravity-free acceleration signals captured at the patient's feet as novel markers of dynamic postural stability. The approach is tested on a dataset containing 100 idiopathic PD patients and 50 age- and weight-matched healthy controls. Experiments include group separation of the controls and PD patients with/without postural instability as assessed by the pull test and analysis of correlations to existing parameters from inertial sensor data. Both markers show significant clinical differences, specifically between the two conditions in the PD group. At least one parameter provides complementary information to the existing set of spatio-temporal gait parameters while the other one correlates highly to gait velocity but might be measurable more precisely. In conclusion, the inertial sensor derived markers can detect postural instability but further research in this domain is needed.

Measurement of stride-related, biomechanical parameters is the common rationale for objective gait impairment scoring. State-of-the-art double-integration approaches to extract these parameters from inertial sensor data are, however, limited in their clinical applicability due to the underlying assumptions. To overcome this, we present a method to translate the abstract information provided by wearable sensors to context-related expert features based on deep convolutional neural networks. Regarding mobile gait analysis, this enables integration-free and data-driven extraction of a set of eight spatio-temporal stride parameters. To this end, two modeling approaches are compared: a combined network estimating all parameters of interest and an ensemble approach that spawns less complex networks for each parameter individually. The ensemble approach is outperforming the combined modeling in the current application. On a clinically relevant and publicly available benchmark dataset, we estimate stride length, width and medio-lateral change in foot angle up to -0.15 ± 6.09 cm, -0.09 ± 4.22 cm and 0.13 ± 3.78° respectively. Stride, swing and stance time as well as heel and toe contact times are estimated up to ±0.07, ±0.05, ±0.07, ±0.07 and ±0.12 s respectively. This is comparable to and in parts outperforming or defining state of the art. Our results further indicate that the proposed change in the methodology could substitute assumption-driven double-integration methods and enable mobile assessment of spatio-temporal stride parameters in clinically critical situations as, e.g., in the case of spastic gait impairments.

Salivary markers have been proposed as noninvasive and easy-to-collect indicators of dehydrations during physical exercise. It has been demonstrated that threshold-based classifications can distinguish dehydrated from euhydrated subjects. However, considerable challenges were reported simultaneously, for example, high intersubject variabilities in these markers. Therefore, we propose a machine-learning approach to handle the intersubject variabilities and to advance from binary classifications to quantitative estimations of total body water (TBW) loss. For this purpose, salivary samples and reference values of TBW loss were collected from ten subjects during a 2-h running workout without fluid intake. The salivary samples were analyzed for previously investigated markers (osmolality, proteins) as well as additional unexplored markers (amylase, chloride, cortisol, cortisone, and potassium). Processing all these markers with a Gaussian process approach showed that quantitative TBW loss estimations are possible within an error of 0.34 l, roughly speaking, a glass of water. Furthermore, a data analysis illustrated that the salivary markers grow nonlinearly during progressive dehydration, which is in contrast to previously reported linear observations. This insight could help to develop more accurate physiological models for salivary markers and TBW loss. Such models, in turn, could facilitate even more precise TBW loss estimations in the future.

Digital motion analysis in freestyle snowboarding requires a stable trick detection and accurate classification. Freestyle snowboarding contains several trick categories that all have to be recognized for an application in training sessions or competitions. While previous work already addressed the classification of specific tricks or turns, there is no known method that contains a full pipeline for detection and classification of tricks from multiple categories. In this paper, we suggest a classification pipeline containing the detection, categorization and classification of tricks of two major freestyle trick categories. We evaluated our algorithm based on data from two different acquisitions with a total number of eleven athletes and 275 trick events. Tricks of both categories were categorized with recall results of 96.6% and 97.4%. The classification of the tricks was evaluated to an accuracy of 90.3 % for the first and 93.3% for the second category.

Arrhythmia detection algorithms require the exact and instantaneous detection of fiducial points in the ECG signal. These fiducial points (QRS-complex, P- and T-wave) correspond to distinct cardiac contraction phases. The performance evaluation of different fiducial points detection algorithms require the existence of large databases (DBs) encompassing reference annotations. Up to last year, P- and T-wave annotations were only available for the QT DB. This was addressed by Elgendi et al. who provided P- and T-wave annotations to the MIT-BIH arrhythmia DB. A variety of ECG fiducial points detection algorithms exists in literature, whereas, to the best knowledge of the authors, we could not identify any single-lead algorithm ready for instantaneous P- and T-wave detection. In this work, we present three P- and T-wave detection algorithms: a revised version for QRS detection using line fitting capable to detect P- and T-wave, an expeditious version of a wavelet based ECG delineation algorithm, and a fast naive fiducial points detection algorithm. The fast naive fiducial points detection algorithm performed best on both DBs with sensitivities ranging from 73.0% (P-wave detection, error interval of ± 40 ms) to 89.4% (T-wave detection, error interval of ± 80 ms). As this algorithm detects a wave event in every search window, it has to be investigated how this affects arrhythmia detection algorithms. The reference Matlab implementations are available for download to encourage the development of high-accurate and automated ECG processing algorithms for the integration in daily life using mobile computers.

The possibilities for wearable health care technology to improve the quality of life for chronic disease patients has been increasing within recent years. For instance, unobtrusive cardiac monitoring can be applied to people suffering from a disorder of the autonomic nervous system (ANS) which show a significantly lower heart rate variability (HRV) than healthy people. Although recent work presented solutions to analyze this relationship, they did not perform it during daily life situations. For that reason, this work presents a system for a real-time analysis of the user's HRV on an Android-based mobile device throughout the day. The system was used for the detection of an orthostatic dysregulation which can be an indicator for a disorder of the ANS. Measures for HRV analysis were computed from acquired ECG data and compared before and after a posture change. For triggering the HRV analysis, an IMU-based algorithm which detects stand up events was developed. As a proof of concept for an automatic assessment of an orthostatic dysregulation, a classification based on the derived HRV measures was performed. The performance of the stand up detection was evaluated in the first part of this study. The second part was conducted for the evaluation of the derived HRV measures and involved healthy subjects as well as patients with idiopathic Parkinson's Disease. The results of the evaluation showed a recognition rate of 90.0 % for the stand up detection algorithm. Furthermore, a clear difference in the change of HRV measures between the two groups before and after standing up was observed. The classification provided an accuracy of 96.0%, and a sensitivity of 93.3%. The results demonstrated the possibility of unobtrusive HRV monitoring during daily life situations.

Battery runtime is a critical concern for practical usage of wearable biomedical sensor systems. A long runtime requires an interdisciplinary low-power knowledge and appropriate design tools. We addressed this issue designing a toolbox in three parts: (1) Modular evaluation kit for development of wearable ultra-low-power biomedical sensors; (2) Miniaturized, wearable, and code compatible sensor system with the same properties as the development kit; (3) Web-based battery runtime calculator for our sensor systems. The purpose of the development kit is optimization of the power consumption. Once optimization is finished, the same embedded software can be transferred to the miniaturized body worn sensor. The web-based application supports development quantifying the effects of use case and design decisions on battery runtime. A sensor developer can select sensor modules, configure sensor parameters, enter use case specific requirements, and select a battery to predict the battery runtime for a specific application. Our concept adds value to development of ultra-low-power biomedical wearable sensors. The concept is effective for professional work and educational purposes.

Several research tools and projects require groups of similar code changes as input. Examples are recommendation and bug finding tools that can provide valuable information to developers based on such data. With the help of similar code changes they can simplify the application of bug fixes and code changes to multiple locations in a project. But despite their benefit, the practical value of existing tools is limited, as users need to manually specify the input data, i.e., the groups of similar code changes.

Among elderly males, benign prostate syndrome (BPS) is the most common urinary disorder. Nocturia is one of the major symptoms of BPS and has a considerable influence on the quality of life. For assessment of BPS (including nocturia), the International Prostate Symptom Score is widely used, but questionnaires are prone to bias. To date, there is no objective measurement system available for nocturia. In this study, we present an unobtrusive and nonstigmatizing device for objective measurement of nighttime micturition. In a preliminary study of six males diagnosed with BPS and nighttime micturition ≥ 2×, we showed that the device is accurate, with an average misdetection rate of 0.32 events and a mean absolute deviation of 3.8% when comparing the average number of nighttime micturition occurrences. In this extended study, an additional nine males were recorded and data from an occupancy sensor were also included. The results of the preliminary study were confirmed with an average misdetection rate of 0.33 events and a mean absolute deviation of 9.1%. The system can, therefore, be used to objectively measure nighttime micturition and, thereby, provide the basis for treatment, e.g., medication efficacy assessment.

The amount of total body water (TBW) can be estimated based on bioimpedance measurements of the human body. In sports, TBW estimations are of importance because mild water losses can impair muscular strength and aerobic endurance. Severe water losses can even be life threatening. TBW estimations based on bioimpedance, however, fail during sports because the increased body temperature corrupts bioimpedance measurements. Therefore, this paper proposes a machine learning method that eliminates the effects of increased temperature on bioimpedance and, consequently, reveals the changes in bioimpedance that are due to TBW loss. This is facilitated by utilizing changes in skin and core temperature. The method was evaluated in a study in which bioimpedance, temperature, and TBW loss were recorded every 15 min during a 2-h running workout. The evaluation demonstrated that the proposed method is able to reduce the error of TBW loss estimation by up to 71%, compared to the state of art. In the future, the proposed method in combination with portable bioimpedance devices might facilitate the development of wearable systems for continuous and noninvasive TBW loss monitoring during sports.

Athletes and their coaches aim for enhancing the sports performance. Collecting data from athletes, transforming them into useful information related to their sports performance (e.g., their type of gait), and transmitting the information to the coaches supports the enhancement. The types of gait standing, walking, and running were often examined. Lack of research remains for the two types of running, jogging and sprinting. In this work, standing, walking, jogging, and sprinting were classified with a single inertial-magnetic measurement unit that was placed at a novel position at the trunk. A comparison was made between classification systems using different combinations of accelerometer, gyroscope, and magnetometer data as well as different classifiers (Naïve Bayes, k-Nearest Neighbors, Support Vector Machine, Adaptive Boosting). After collecting data from 15 male subjects, the data were preprocessed, features were extracted and selected, and the data were classified. All classification systems were successful. With a mean true positive rate of 95.68% ±1.80%, the classification system using accelerometer and gyroscope data as well as the Naïve Bayes classifier performed best. The classification system can be used for applications in sport and sports performance analysis in particular.

A simple method for an underwater pose determination of scuba divers can provide a deeper insight in the biomechanics of scuba diving and thereby improve education and training systems. In this work, we present an inertial sensor-based approach for the pose determination of the upper body and the shank orientation during fin kicks. Accelerometer measurements of gravity and a gyroscope-based method are used to determine absolute body angles in reference to the ground and the angular change of the shanks during fin kicks. The proposed algorithms were evaluated with data acquired from ten divers and a camera-based gold standard. The results were analyzed to a mean error of 0° with a standard deviation of 10° for the upper body pose determination. The absolute angle of the shanks at the turning points between fin kicks was determined with an error of 0° ± 11°, the relative shank angle with an error of 0° ± 8°.

Far too many people are dying from stroke or other heart related diseases each year. Early detection of abnormal heart rhythm could trigger the timely presentation to the emergency department or outpatient unit. Smartphones are an integral part of everyone;s life and they form the ideal basis for mobile monitoring and real-time analysis of signals related to the human heart. In this work, we investigated the performance of arrhythmia classification systems using only features calculated from the time instances of individual heart beats. We built a sinusoidal model using N (N = 10, 15, 20) consecutive RR intervals to predict the (N+1)th RR interval. The integration of the innovative sinusoidal regression feature, together with the amplitude and phase of the proposed sinusoidal model, led to an increase in the mean class-dependent classification accuracies. Best mean class-dependent classification accuracies of 90% were achieved using a Naïve Bayes classifier. Well-performing realtime analysis arrhythmia classification algorithms using only the time instances of individual heart beats could have a tremendous impact in reducing healthcare costs and reducing the high number of deaths related to cardiovascular diseases.

This work presents a system for unobtrusive cardiac feedback in daily life. It addresses the whole pipeline from data acquisition over data processing to data visualization including wearable integration. ECG signals are recorded with a novel ECG sensor supporting Bluetooth Low Energy, which is able to transmit raw ECG data as well as estimated heart rate. ECG signals are processed in real-time on a mobile device to automatically classify the user's heart beats. A novel application for Android-based mobile devices was developed for data visualization. It offers several modes for cardiac feedback, from measuring the current heart rate to continuously monitoring the user's heart status. It also allows to store acquired data in an internal database as well as in the Google Fit platform. Further, the application provides extensions for wearables like Google Glass and smartwatches running on Android Wear. Hardware performance evaluation was performed by comparing the course of heart rate between the novel ECG sensor and a commercial ECG sensor. The mean absolute error between the two sensors was 4.83 bpm with a standard deviation of 4.46 bpm, and a Pearson correlation of 0.922. A qualitative evaluation was performed for the Android application with special emphasis on the daily usability and the wearable integration. When the Google Glass was integrated, the subjects rated the application as 2.8/5 (0 = Bad, 5 = Excellent), whereas when the application was integrated with a smartwatch the rating increased to 4.2/5.

Long battery runtime is one of the most wanted properties of wearable sensor systems. The sampling rate has an high impact on the power consumption. However, defining a sufficient sampling rate, especially for cutting edge mobile sensors is difficult. Often, a high sampling rate, up to four times higher than necessary, is chosen as a precaution. Especially for biomedical sensor applications many contradictory recommendations exist, how to select the appropriate sample rate. They all are motivated from one point of view — the signal quality. In this paper we motivate to keep the sampling rate as low as possible. Therefore we reviewed common algorithms for biomedical signal processing. For each algorithm the number of operations depending on the data rate has been estimated. The Bachmann-Landau notation has been used to evaluate the computational complexity in dependency of the sampling rate. We found linear, logarithmic, quadratic and cubic dependencies.

The papers in this special issue focus on the management of Parkinson's disease (PD). PD is the most common neurological movement disorder with a prevalence of up to 2% in the elderly. The cardinal motor symptoms bradykinesia, rigidity, tremor, and postural instability define the diagnosis of the PD.

Current challenges demand a profound restructuration of the global healthcare system. A more efficient system is required to cope with the growing world population and increased life expectancy, which is associated with a marked prevalence of chronic neurological disorders such as Parkinson's disease (PD). One possible approach to meet this demand is a laterally distributed platform such as the Internet of Things (IoT). Real-time motion metrics in PD could be obtained virtually in any scenario by placing lightweight wearable sensors in the patient's clothes and connecting them to a medical database through mobile devices such as cell phones or tablets. Technologies exist to collect huge amounts of patient data not only during regular medical visits but also at home during activities of daily life. These data could be fed into intelligent algorithms to first discriminate relevant threatening conditions, adjust medications based on online obtained physical deficits, and facilitate strategies to modify disease progression. A major impact of this approach lies in its efficiency, by maximizing resources and drastically improving the patient experience. The patient participates actively in disease management via combined objective device- and self-assessment and by sharing information within both medical and peer groups. Here, we review and discuss the existing wearable technologies and the Internet-of-Things concept applied to PD, with an emphasis on how this technological platform may lead to a shift in paradigm in terms of diagnostics and treatment.

Insufficient physical activity is the 4th leading risk factor for mortality. The physical activity of a person is reflected in the walking behavior. Different methods for the calculation of the accurate step number exists and most of them are evaluated using different walking speeds measured on a treadmill or using a small sample size of overground walking. In this paper, we introduce the BaSA (Basic Step Activities) dataset consisting of four different step activities (walking, jogging, ascending, and descending stairs) that were performed under natural conditions. We further compare two step segmentation algorithms (a simple peak detection algorithm vs. subsequence Dynamic Time Warping (sDTW)). We calculated a multivariate Analysis of Variance (ANOVA) with repeated measures followed by multiple dependent t-tests with Bonferroni correction to test for significant differences in the two algorithms. sDTW performed equally good compared to the peak detection algorithm, but was not considerably better. In further analysis, continuous, real walking signals with transitions from one step activity to the other step activity should be considered to investigate the adaptability of these two step segmentation algorithms.

We developed an application for Android-based mobile devices that enables a real-time calculation of heart rate and cadence for biking. Therefore, both ECG and EMG data are acquired in real time by Shimmer sensors and transmitted via Bluetooth, as well as processed and evaluated on the mobile device. The ECG algorithm is based on the Pan-Tompkins algorithm for QRS-Detection and offers a heart beat detection rate of more than 94%. The EMG algorithm offers a treadle detection rate of more than 91%. The application's range of features is complemented by GPS data for the calculation of speed and location information. It is available for download and can for example be used for controlling the user's training status, for live training supervision and for the subsequent analysis of the various training runs.

Bio impedance analysis (BIA) estimates the amount of total body water (TBW) in the human body. During sports, however, the increased skin temperature distorts bio impedance measurements and, thus, prevents the application of BIA. In this paper, we propose a two-stage regression that includes temperature information in order to correct the temperature-distorted bio impedance. In detail, the first regression stage corrects temperature-distored bio impedance using information of skin and core temperature. The second regression stage estimates TBW loss on basis of the corrected bio impedance. The two-stage regression was evaluated using data of an ongoing study. The results showed that estimations of TBW loss during sports can be considerably improved if temperature information is included. However, a remaining error was still observed. Therefore, additional measurements, e.g., skin blood flow, are discussed because they also influence bio impedance and could further reduce the error.

3D displays enable immersive visual impressions but the impact on the human perception still is not fully understood. Viewing conditions like the convergence-accommodation (C-A) conflict have an unnatural influence on the visual system and might even lead to visual discomfort. As visual perception is individual we assumed the impact of simulated 3D content on the visual system to be as well. In this study we aimed to analyze the stereoscopic visual performance of 17 subjects for disparities inside and outside the in literature defined zone of comfortable viewing to provide an individual evaluation of the impact of increased disparities on the performance of the visual system. Stereoscopic stimuli were presented in a four-alternative forced choice (4AFC) setup in different disparities. The response times as well as the correct decision rates indicated the performance of stereoscopic vision. The results showed that increased disparities lead to a decline in performance. Further, the impact of the presented disparities is dependent on the difficulty of the task. The decline of performance as well as the deciding disparities for the decline were subject dependent.

The automatic classification of foot strike patterns into the three basic categories forefoot, midfoot and rearfoot striking plays an important role for applications like shoe fitting with instant feedback. This paper presents methods for this classification based on body worn accelerometers that allow giving the required direct feedback to the user. For our study, we collected data from 40 runners who had a standard accelerometer in a custom-built sensor pod attached to the laces of their running shoes. The acceleration in three axes was recorded continuously while the runners conducted their runs. Data for repeated runs at two different speed levels were collected in order to have sufficient sensor data for classification. The data was analyzed using features computed for individual steps of the runners to distinguish the three foot strike pattern classes. The labels for the strike pattern classes were established using high-speed video that was concurrently collected. We could show that the classification of the strike types based on the measured accelerations and the extracted features was up to 95.3% accurate. The established classification system can be used to support runners, for example by giving running shoe recommendations that ideally match the prevailing strike type of the runner.

The collection of kinematic data with a head-worn sensor is a promising approach for swimming data analysis in the context of athlete support systems. We present a new approach of analyzing these data and describe a system that segments the lanes of a swimming session and classifies the swimming style of each lane. Special emphasis was put on the algorithm efficiency and the analysis of the resource demands to be able to port the implementation to an embedded microcontroller. For developing the system, data of twelve subjects was collected. The data incorporated two different turn styles that mark the end of a lane as well as the four main swimming styles backstroke, breaststroke, butterfly and freestyle. All turns were successfully identified from the turn detection. Our fully automatic swimming style classification reached a classification rate of 95.0%. The results from the resource consumption analysis can be used to support the decision for the embedded target hardware of a head-worn swimming training system.

Wearable sensors are widely used for data collection in many applications. Ssensor nodes have also been applied for real-time applications, e.g. for ECG analysis or activity and fall detection. Processing of the sensor data is either done on an external device or on the node itself. While on-node processing reduces data rate and increases battery life, development and testing can be time-consuming. To allow faster implementation of such algorithms, we propose a simulation framework for the Shimmer platform using the Cooja simulator, MSPSim and the Contiki operating system. We provide the simulator and example applications compatible with the ShimmerConnect protocol, allowing streaming of raw and pre-processed sensor data to MATLAB, LabView and Android. Additionally, a simple activity and fall detection algorithm was implemented on the sensor node and evaluated using both the simulator and real hardware. In the future this will allow rapid development and testing of on-node pre-processing algorithms.

This paper presents methods for collecting and analyzing biomechanical and physiological data from several body sensors during recreational runs in order to classify an athlete's perceived fatigue state. Heart rate, heart rate variability, running speed, stride frequency and biomechanical data were recorded continuously from 431 runners during a free one-hour outdoor run. During the activity the sportsmen answered questions about their perceived fatigue state in 5 min intervals. The data were analyzed using specifically designed features computed for each of the 5 min intervals. The features were used to train different classifiers, which were able to distinguish two levels of the runner's fatigue state with an accuracy of 88.3 % across multiple study participants. Feature selection evidenced that a heart rate variability feature and two biomechanical features were best suited for classification of the perceived fatigue level. Therefore, the classification system needs the information from various sensors on the human body. The resulting classifier was implemented on an embedded microcontroller to show that it would be feasible to integrate it directly into a body sensor network. Such a wearable classification system for fatigue can be used to support sportsmen, for example by changing their training plan or by adapting their equipment to the specific needs of a fatigued athlete.

This paper presents methods for collecting and analyzing physiological and biomechanical data during recreational runs in order to classify an athletepsilas perceived fatigue state. Heart rate and its variability, running speed and stride frequency, GPS position and shoe heel compression were recorded continuously while runners moved freely outdoors. During their activity the sportsmen answered questions about their fatigue state in five-minute-intervals. Data from 84 one-hour-runs was collected for analysis. The data was analyzed using features computed for each step of the athlete to distinguish three levels of the runnerpsilas fatigue state with an accuracy of 75.3% across multiple study participants and 91.8% in the intraindividual case. The results show that for most participating runners, a heart rate variability periodogram feature and a step duration feature are best suited for classification of the perceived fatigue level. This information can be used to support sportsmen, for example by adapting their equipment to the specific needs of a fatigued athlete.

The quality of coded images is often assessed by a subjective test. Usually the viewers get as much time as they need to find a stable result. In video sequences however, the viewer has to judge the quality in a shorter time that is defined by the changing content or a following scene cut. Therefore it is desirable to know the influence of a shorter presentation time on the perceptibility of distortions. In this paper we present the results of a suitable subjective test on coded still images. The images were presented for six different durations, ranging from 200 ms to 3 s. Special care was taken to avoid the memorization effect usually present after short presentations. The results show that the viewers tend to avoid extreme votings at short durations. The variance of the votings is also discussed in detail. Based on the result of the voting for the longest presentation time, we propose a prediction model for the voting of the shorter durations using a logistic curve fit. This presentation time model (PTM) is presented and analysed in detail.