Medical applications of machine learning (ML) have experienced a surge in popularity in recent years. Given the abundance of available data from electronic health records, the intensive care unit (ICU) is a natural habitat for ML. Models have been proposed to address numerous ICU prediction tasks like the early detection of complications. While authors frequently report state-of-the-art performance, it is challenging to verify claims of superiority. Datasets and code are not always published, and cohort definitions, preprocessing pipelines, and training setups are difficult to reproduce. This work introduces Yet Another ICU Benchmark (YAIB), a modular framework that allows researchers to define reproducible and comparable clinical ML experiments; we offer an end-to-end solution from cohort definition to model evaluation. The framework natively supports most open-access ICU datasets (MIMIC III/IV, eICU, HiRID, AUMCdb) and is easily adaptable to future ICU datasets. Combined with a transparent preprocessing pipeline and extensible training code for multiple ML and deep learning models, YAIB enables unified model development, transfer, and evaluation. Our benchmark comes with five predefined established prediction tasks (mortality, acute kidney injury, sepsis, kidney function, and length of stay) developed in collaboration with clinicians. Adding further tasks is straightforward by design. Using YAIB, we demonstrate that the choice of dataset, cohort definition, and preprocessing have a major impact on the prediction performance — often more so than model class — indicating an urgent need for YAIB as a holistic benchmarking tool. We provide our work to the clinical ML community to accelerate method development and enable real-world clinical implementations.

Speech-based machine learning models that can distinguish between a healthy cognitive state and different stages of cognitive decline would enable a more appropriate and timely treatment of patients.However, their development is often hampered by data scarcity.Federated Learning (FL) is a potential solution that could enable entities with limited voice recordings to collectively build effective models.Motivated by this, we compare centralised, local, and federated learning for building speechbased models to discern Alzheimer's Disease, Mild Cognitive Impairment, and a healthy state.For a more realistic evaluation, we use three independently collected datasets to simulate healthcare institutions employing these strategies.Our initial analysis shows that FL may not be the best solution in every scenario, as performance improvements are not guaranteed even with small amounts of available data, and further research is needed to determine the conditions under which it is beneficial.

Multiple sclerosis (MS) is a neuroinflammatory disease that affects millions of people worldwide. Since dysarthria is prominent in people with MS (pwMS), this paper aims to identify acoustic features that differ between people with MS and healthy controls (HC). Additionally, we develop automatic classification methods to distinguish between pwMS and HC. In this work, we present a new dataset of a German-speaking cohort which contains 39 patients with low disability of relapsing MS and 16 HC. Findings suggest that certain interpretable speech features could be useful in diagnosing MS, and that machine learning methods could potentially support fast and unobtrusive screening in clinical practice. The study emphasises the importance of analysing free speech compared to read speech.

Cognitive load is frequently induced in laboratory setups to measure responses to stress, and its impact on voice has been studied in the field of computational paralinguistics. One dataset on this topic was provided in the Computational Paralinguistics Challenge (ComParE) 2014, and therefore offers great comparability. Recently, transformer-based deep learning architectures established a new state-of-the-art and are finding their way gradually into the audio domain. In this context, we investigate the performance of popular transformer architectures in the audio domain on the ComParE 2014 dataset, and the impact of different pre-training and fine-tuning setups on these models. Further, we recorded a small custom dataset, designed to be comparable with the ComParE 2014 one, to assess cross-corpus model generalisability. We find that the transformer models outperform the challenge baseline, the challenge winner, and more recent deep learning approaches. Models based on the ‘large’ architecture perform well on the task at hand, while models based on the ‘base’ architecture perform at chance level. Fine-tuning on related domains (such as ASR or emotion), before fine-tuning on the targets, yields no higher performance compared to models pre-trained only in a self-supervised manner. The generalisability of the models between datasets is more intricate than expected, as seen in an unexpected low performance on the small custom dataset, and we discuss potential ‘hidden’ underlying discrepancies between the datasets. In summary, transformer-based architectures outperform previous attempts to quantify cognitive load from voice. This is promising, in particular for healthcare-related problems in computational paralinguistics applications, since datasets are sparse in that realm.

Neural networks have been successfully applied to a wide range of human motion analysis topics in combination with wearable sensor data. However, their computation process is not readily comprehensible. Alternatively, many of the model interpretation efforts do not provide physiologically-relevant insights, thus still limiting their use in clinical settings. In this work, we take gait modifications under fatigue and cognitive task performance as a use case to present how in-depth investigations of neural networks can be performed using wearable sensor data. We collected walking data from 16 young healthy individuals in unfatigued and fatigued states and under single- (walking only) and dual-task (walking while concurrently performing a cognitive task) conditions using inertial measurement units. Convolutional neural networks were able to identify both fatigue and dual-task gait patterns with high classification accuracy. To interpret the model, the importance of each time step in the input time series was visualized using Layer-wise Relevance Propagation. The visualization revealed highly individualized gait changes among participants, as well as changes at precise time steps of the input signal that allow further investigations to infer potential underlying mechanisms. Our methods enable in-depth analysis of human movement using transparent neural networks with data collected from unobtrusive, mobile wearable sensors.

Despite remarkable advances in the development of language resources over the recent years, there is still a shortage of annotated, publicly available corpora covering (German) medical language. With the initial release of the German Guideline Program in Oncology NLP Corpus (GGPONC), we have demonstrated how such corpora can be built upon clinical guidelines, a widely available resource in many natural languages with a reasonable coverage of medical terminology. In this work, we describe a major new release for GGPONC. The corpus has been substantially extended in size and re-annotated with a new annotation scheme based on SNOMED CT top level hierarchies, reaching high inter-annotator agreement (γ=.94). Moreover, we annotated elliptical coordinated noun phrases and their resolutions, a common language phenomenon in (not only German) scientific documents. We also trained BERT-based named entity recognition models on this new data set, which achieve high performance on short, coarse-grained entity spans (F1=.89), while the rate of boundary errors increases for long entity spans. GGPONC is freely available through a data use agreement. The trained named entity recognition models, as well as the detailed annotation guide, are also made publicly available.

The quantification of subjective exertion during training is an important measurement as it has the potential to avoid injuries resulting from overtraining. In this paper, we present a method to predict the subjective exertion during resistance training using Inertial Measurement Units (IMU) and electrocardiographical data. The participants’ subjective exertion was assessed using a Rating of Perceived Exertion (RPE) scale. We obtained data from 16 participants performing squats on a flywheel training machine while being equipped with six IMU sensors and an electrocardiography (ECG) sensor. Data was analyzed using multiple regressors, such as Support Vector Regression, Random Forests, and Gradient Boosting Regression Trees, to predict the personal exertion level on the processed IMU and ECG data. The best learning model achieved a mean absolute percentage error of 7.71% with a Pearson correlation coefficient of 0.85 and a R2of 0.48. Additionally, we investigated the impact of supplementing the IMU data features with ECG-derived heart rate variability (HRV) parameters in the training stage. Our results indicate that the HRV parameters derived from ECG significantly improve prediction results, with the training impulse (TRIMP) parameter acting as the most informative feature for predicting perceived exertion.

The mental illness Obsessive-Compulsive Disorder (OCD) is characterised by obsessive thoughts and compulsive actions. The latter can occur as repetitive activities to ensure that severe fears do not come true. A diagnosis of the disease is usually very late due to a lack of knowledge and shame of the patient. Nevertheless, early detection can significantly increase the success of therapy.With the development of new wearable sensors, it is possible to recognise human activities. Accordingly, wearables can also be used to identify recurring activities that indicate an OCD. Through this form of an automatic detection system, a diagnosis can be made earlier and thus therapy can be started sooner.Since compulsive behaviour is very individual and varies from patient to patient, this paper deals with personalised federated machine learning models. We first adapt the publicly available OPPORTUNITY dataset to simulate OCD behaviour. Secondly, we evaluate two existing personalised federated learning algorithms against baseline approaches. Finally, we propose a hybrid approach that merges the two evaluated algorithms and reaches a mean area under the precision-recall curve (AUPRC) of 0.954 across clients.

With the COVID-19 pandemic, several research teams have reported successful advances in automated recognition of COVID-19 by voice. Resulting voice-based screening tools for COVID-19 could support large-scale testing efforts. While capabilities of machines on this task are progressing, we approach the so far unexplored aspect whether human raters can distinguish COVID-19 positive and negative tested speakers from voice samples, and compare their performance to a machine learning baseline. To account for the challenging symptom similarity between COVID-19 and other respiratory diseases, we use a carefully balanced dataset of voice samples, in which COVID-19 positive and negative tested speakers are matched by their symptoms alongside COVID-19 negative speakers without symptoms. Both human raters and the machine struggle to reliably identify COVID-19 positive speakers in our dataset. These results indicate that particular attention should be paid to the distribution of symptoms across all speakers of a dataset when assessing the capabilities of existing systems. The identification of acoustic aspects of COVID-19-related symptom manifestations might be the key for a reliable voice-based COVID-19 detection in the future by both trained human raters and machine learning models. Copyright ©2021 ISCA.

Federated learning allows a group of distributed clients to train a common machine learning model on private data. The exchange of model updates is managed either by a central entity or in a decentralized way, e.g. by a blockchain. However, the strong generalization across all clients makes these approaches unsuited for non-independent and identically distributed (non-IID) data.

Data privacy is a very important issue. Especially in fields like medicine, it is paramount to abide by the existing privacy regulations to preserve patients’ anonymity. However, data is required for research and training machine learning models that could help gain insight into complex correlations or personalised treatments that may otherwise stay undiscovered. Those models generally scale with the amount of data available, but the current situation often prohibits building large databases across sites. So it would be beneficial to be able to combine similar or related data from different sites all over the world while still preserving data privacy. Federated learning has been proposed as a solution for this, because it relies on the sharing of machine learning models, instead of the raw data itself. That means private data never leaves the site or device it was collected on. Federated learning is an emerging research area, and many domains have been identified for the application of those methods. This systematic literature review provides an extensive look at the concept of and research into federated learning and its applicability for confidential healthcare datasets.

Smartphones are the best personal assistants in our lives on several counts. However, their services can still be improved for a better quality of life. In this paper, we aim to determine the exact location of a smartphone in a room, i.e, on a study desk, a television table, etc. By this way, our daily settings may be automatically activated from the smartphone itself. For example, if a user puts his/her phone on the bed commode, then the phone would be able to switch itself to the night mode on its own. A successful localization in a room should be able to distinguish different corners from each other so that it can be used in various applications as a supported technology. Hence, in this work, we are proposing an indoor localization system that can distinguish different indoor places by using the smartphones' sensors and Wi-Fi services. Unlike the common location-based services, our solution is not a server-client based system. In order to enhance feasibility and availability, we only use the mobile device but no additional infrastructure. We developed two applications on Android platform. The first one allows the user to easily collect sensor data from his/her living places, such as home and office settings. The second one is a data mining application sourced by Weka. The tests were performed in different rooms of a house and office environment. We achieved 86% accuracy for room level localization.

The main contribution of this study is the design and implementation of an autonomous human robot interaction system to engage children in performing several physical exercise motions by providing real-time feedback and guidance. The system is designed after several preliminary experiments with children and exercise coaches. In order to test the feasibility and the effectiveness of the exercise system across a variety of performance and evaluation measures, an experimental study was conducted with 19 healthy children. The results of the study validate the effectiveness of the system in motivating and helping children to complete physical exercises. The children engaged in physical exercise throughout the interaction sessions and rated the interaction highly in terms of enjoyableness, and rated the robot exercise coach highly in terms of social attraction, social presence, and companionship via a questionnaire answered after each session.

Affective states play a crucial role in learning. Existing Intelligent Tutoring Systems (ITSs) fail to track affective states of learners accurately. Without an accurate detection of such states, ITSs are limited in providing truly personalized learning experience. In our longitudinal research, we have been working towards developing an empathic autonomous 'tutor' closely monitoring students in real-time using multiple sources of data to understand their affective states corresponding to emotional engagement. We focus on detecting learning related states (i.e., 'Satisfied', 'Bored', and 'Confused'). We have collected 210 hours of data through authentic classroom pilots of 17 sessions. We collected information from two modalities: (1) appearance, which is collected from the camera, and (2) context-performance, that is derived from the content platform. The learning content of the content platform consists of two section types: (1) instructional where students watch instructional videos and (2) assessment where students solve exercise questions. Since there are individual differences in expressing affective states, the detection of emotional engagement needs to be customized for each individual. In this paper, we propose a hierarchical semi-supervised model adaptation method to achieve highly accurate emotional engagement detectors. In the initial calibration phase, a personalized context-performance classifier is obtained. In the online usage phase, the appearance classifier is automatically personalized using the labels generated by the context-performance model. The experimental results show that personalization enables performance improvement of our generic emotional engagement detectors. The proposed semi-supervised hierarchical personalization method result in 89.23% and 75.20% F1 measures for the instructional and assessment sections respectively.

Firefighting is a dangerous task and many research projects have aimed at supporting firefighters during missions by developing new and often costly equipment. In contrast to previous approaches, we use the smartphone to monitor firefighters during real-world missions in order to provide objective data that can be used in post-incident briefings and trainings. In this paper, we present CoenoFire, a smartphone based sensing system aimed at monitoring temporal and behavioral performance indicators of firefighting missions. We validate the performance metrics showing that they can indicate why certain teams performed faster than others in a training scenario conducted by 16 firefighting teams. Furthermore, we deployed CoenoFire over a period of six weeks in a professional fire brigade. In total, 71 firefighters participated in our study and the collected data includes 76 real-world missions totaling to over 148 hours of mission data. Additionally, we visualize real-world mission data and show how mission feedback is supported by the data.

Millions of people run. Movement scientists investigate the relationship of running kinematics to fatigue, injury, or running economy mainly using optical motion capture. It was found that running kinematics are highly individual and often cannot be summarized by single variables. We thus present a data-driven analysis of running technique using wearable technology, combining statistical features and machine learning techniques, which allows to identify non-linear, complex relationships. Wearable technology enables running kinematic analysis to a broad mass in unconstrained environments. 20 runners wore 12 sensor units during two experiments: an all out test and a fatiguing run. We used a Support Vector Machine (SVM) to distinguish skill level groups and achieved an accuracy of 76.92% with an acceleration sensor on the upper body. Sensor positions were ranked according to the movement change with fatigue using a feature selection. This ranking was consistent with visual annotations of a movement scientist. We propose a quantitative measure of movement change using a principal component analysis (PCA) and found an average correlation of 0.8369 for all runners with their perceived rating of fatigue.

The inferred cost of work-related stress call for early prevention strategies. In this, we see a new opportunity for affective and pervasive computing by detecting early warning signs. This paper goes one step toward this goal. A collective of 33 subjects underwent a laboratory stress intervention, while a set of physiological signals was collected. In this paper, we investigate whether affective information related to stress can be found in the posture channel during office work. Following more recent work in this field, we directly associate features that are derived from the pressure distribution on a chair with affective states. We found that nervous subjects reveal higher variance of movements under stress. Furthermore, we show that a person-independent discrimination of stress from cognitive load is feasible when using pressure data only. A supervised variant of a self-organizing map, which is able to adapt to different patterns of stress responses, reaches an overall accuracy of 73.75% with unknown subjects.

The inferred cost of work-related stress call for prevention strategies that aim at detecting early warning signs at the workplace. This paper goes one step towards the goal of developing a personal health system for detecting stress. We analyze the discriminative power of electrodermal activity (EDA) in distinguishing stress from cognitive load in an office environment. A collective of 33 subjects underwent a laboratory intervention that included mild cognitive load and two stress factors, which are relevant at the workplace: mental stress induced by solving arithmetic problems under time pressure and psychosocial stress induced by social-evaluative threat. During the experiments, a wearable device was used to monitor the EDA as a measure of the individual stress reaction. Analysis of the data showed that the distributions of the EDA peak height and the instantaneous peak rate carry information about the stress level of a person. Six classifiers were investigated regarding their ability to discriminate cognitive load from stress. A maximum accuracy of 82.8% was achieved for discriminating stress from cognitive load. This would allow keeping track of stressful phases during a working day by using a wearable EDA device.

We present a system for learning the 3 DOF fine-positioning task of a robot manipulator (Puma 260) using a gripper mounted camera. Small lateral gripper-target misalignments are corrected in one step. Larger ones employ a previous coarse adjustment move in order to bound the parallax effects of the close camera focus. We build object specialized, neural network-based pose estimators with a rather small set of Gabor filters. Gabor filters perform a spatially localized frequency analysis and resemble the spatial response profile of receptive fields found in visual cortex neurons. The system demonstrates efficiency w.r.t. speed and accuracy, as well as robustness against changing illumination and object conditions.

A system for learning the pre-grasp positioning task for a robot manipulator is presented. The images delivered from a gripper mounted camera are analysed using Gabor filters which resemble the spatial response profiles of receptive fields found in visual cortex neurons. Using a quite small feature set, the system demonstrated efficiency with respect to speed and accuracy, as well as robustness against changing light conditions. Furthermore, we compare it to two other approaches, aiming at the same goal: an appearance-based PCA fuzzy control and a PSOM based Hough-Transform system.