Elgar Fleisch

Fan Wu, Tsai-Ning Wang, Nicolas Zumarraga et al. · eth-zurich, harvard

Auscultation is a vital diagnostic tool, yet its utility is often limited by subjective interpretation. While general-purpose Audio-Language Models (ALMs) excel in general domains, they struggle with the nuances of physiological signals. We propose a framework that aligns multi-site auscultation recordings directly with a frozen Large Language Model (LLM) embedding space via gated cross-attention. By leveraging the LLM's latent world knowledge, our approach moves beyond isolated classification toward holistic, patient-level assessment. On the CaReSound benchmark, our model achieves a state-of-the-art 0.865 F1-macro and 0.952 BERTScore. We demonstrate that lightweight, domain-specific encoders rival large-scale ALMs and that multi-site aggregation provides spatial redundancy that mitigates temporal truncation. This alignment of medical acoustics with text foundations offers a scalable path for bridging signal processing and clinical assessment.

Mesut Ceylan, Alexis Tabin, Patrick Langer et al.

Wearable devices enable continuous, population-scale monitoring of physiological signals, such as photoplethysmography (PPG), creating new opportunities for data-driven clinical assessment. Time-series extrinsic regression (TSER) models increasingly leverage PPG signals to estimate clinically relevant outcomes, including heart rate, respiratory rate, and oxygen saturation. For clinical reasoning and trust, however, single point estimates alone are insufficient: clinicians must also understand whether predictions are stable under physiologically plausible variations and to what extent realistic, attainable changes in physiological signals would meaningfully alter a model's prediction. Counterfactual explanations (CFE) address these "what-if" questions, yet existing time series CFE generation methods are largely restricted to classification, overlook waveform morphology, and often produce physiologically implausible signals, limiting their applicability to continuous biomedical time series. To address these limitations, we introduce EvoMorph, a multi-objective evolutionary framework for generating physiologically plausible and diverse CFE for TSER applications. EvoMorph optimizes morphology-aware objectives defined on interpretable signal descriptors and applies transformations to preserve the waveform structure. We evaluated EvoMorph on three PPG datasets (heart rate, respiratory rate, and oxygen saturation) against a nearest-unlike-neighbor baseline. In addition, in a case study, we evaluated EvoMorph as a tool for uncertainty quantification by relating counterfactual sensitivity to bootstrap-ensemble uncertainty and data-density measures. Overall, EvoMorph enables the generation of physiologically-aware counterfactuals for continuous biomedical signals and supports uncertainty-aware interpretability, advancing trustworthy model analysis for clinical time-series applications.

George Boateng, Xiangyu Zhao, Malgorzata Speichert et al.

Couples generally manage chronic diseases together and the management takes an emotional toll on both patients and their romantic partners. Consequently, recognizing the emotions of each partner in daily life could provide an insight into their emotional well-being in chronic disease management. Currently, the process of assessing each partner's emotions is manual, time-intensive, and costly. Despite the existence of works on emotion recognition among couples, none of these works have used data collected from couples' interactions in daily life. In this work, we collected 85 hours (1,021 5-minute samples) of real-world multimodal smartwatch sensor data (speech, heart rate, accelerometer, and gyroscope) and self-reported emotion data (n=612) from 26 partners (13 couples) managing diabetes mellitus type 2 in daily life. We extracted physiological, movement, acoustic, and linguistic features, and trained machine learning models (support vector machine and random forest) to recognize each partner's self-reported emotions (valence and arousal). Our results from the best models (balanced accuracies of 63.8% and 78.1% for arousal and valence respectively) are better than chance and our prior work that also used data from German-speaking, Swiss-based couples, albeit, in the lab. This work contributes toward building automated emotion recognition systems that would eventually enable partners to monitor their emotions in daily life and enable the delivery of interventions to improve their emotional well-being.

Robin Deuber, Lanlan Yang, Michal Bechny et al.

Alcohol-impaired driving remains a major yet preventable cause of road traffic injury and death, with many drivers underestimating their level of intoxication. Compared to in-vehicle systems, mobile drunk-driving detection using consumer smartwatches offers a scalable way to trigger preventive interventions and increase awareness without additional in-vehicle hardware. We introduce a system that leverages wrist accelerometer data and heart rate variability-derived physiological signals to detect alcohol-related driving impairment. We collected data in a randomized, controlled three-arm test-track study (n=54) and trained both logistic regression models with window-aggregated features and a two-tower 1D convolutional neural network (CNN), to detect alcohol-impaired driving. The CNN achieved a participant-averaged area under the receiver operating characteristic (AUROC) of 0.88 for detecting any alcohol intoxication and 0.86 for detecting driving above the WHO-recommended limit of 0.05 g/dL. To the best of our knowledge, this is the first work to (1) demonstrate drunk-driving detection using consumer smartwatches, (2) develop and evaluate such a system in a real vehicle on a closed test track, and (3) rigorously assess generalization to unseen participants. Together, these findings highlight the potential of wearable-based sensing to support scalable, measurement-driven prevention of alcohol-related traffic harm.

Fan Wu, Matthias P. Nägele, Daryush D. Mehta et al.

Objective: This study aimed to evaluate which voice features can predict health deterioration in patients with chronic HF. Background: Heart failure (HF) is a chronic condition with progressive deterioration and acute decompensations, often requiring hospitalization and imposing substantial healthcare and economic burdens. Current standard-of-care (SoC) home monitoring, such as weight tracking, lacks predictive accuracy and requires high patient engagement. Voice is a promising non-invasive biomarker, though prior studies have mainly focused on acute HF stages. Methods: In a 2-month longitudinal study, 32 patients with HF collected daily voice recordings and SoC measures of weight and blood pressure at home, with biweekly questionnaires for health status. Acoustic analysis generated detailed vowel and speech features. Time-series features were extracted from aggregated lookback windows (e.g., 7 days) to predict next-day health status. Explainable machine learning with nested cross-validation identified top vocal biomarkers, and a case study illustrated model application. Results: A total of 21,863 recordings were analyzed. Acoustic vowel features showed strong correlations with health status. Time-series voice features within the lookback window outperformed corresponding standard care measures, achieving peak sensitivity and specificity of 0.826 and 0.782 versus 0.783 and 0.567 for SoC metrics. Key prognostic voice features identifying deterioration included delayed energy shift, low energy variability, and higher shimmer variability in vowels, along with reduced speaking and articulation rate, lower phonation ratio, decreased voice quality, and increased formant variability in speech. Conclusion: Voice-based monitoring offers a non-invasive approach to detect early health changes in chronic HF, supporting proactive and personalized care.

Patrick Langer, Thomas Kaar, Max Rosenblattl et al.

LLMs have emerged as powerful tools for interpreting multimodal data. In medicine, they hold particular promise for synthesizing large volumes of clinical information into actionable insights and digital health applications. Yet, a major limitation remains their inability to handle time series. To overcome this gap, we present OpenTSLM, a family of Time Series Language Models (TSLMs) created by integrating time series as a native modality to pretrained LLMs, enabling reasoning over multiple time series of any length. We investigate two architectures for OpenTSLM. The first, OpenTSLM-SoftPrompt, models time series implicitly by concatenating learnable time series tokens with text tokens via soft prompting. Although parameter-efficient, we hypothesize that explicit time series modeling scales better and outperforms implicit approaches. We thus introduce OpenTSLM-Flamingo, which integrates time series with text via cross-attention. We benchmark both variants against baselines that treat time series as text tokens or plots, across a suite of text-time-series Chain-of-Thought (CoT) reasoning tasks. We introduce three datasets: HAR-CoT, Sleep-CoT, and ECG-QA-CoT. Across all, OpenTSLM models outperform baselines, reaching 69.9 F1 in sleep staging and 65.4 in HAR, compared to 9.05 and 52.2 for finetuned text-only models. Notably, even 1B-parameter OpenTSLM models surpass GPT-4o (15.47 and 2.95). OpenTSLM-Flamingo matches OpenTSLM-SoftPrompt in performance and outperforms on longer sequences, while maintaining stable memory requirements. By contrast, SoftPrompt grows exponentially in memory with sequence length, requiring around 110 GB compared to 40 GB VRAM when training on ECG-QA with LLaMA-3B. Expert reviews by clinicians find strong reasoning capabilities exhibited by OpenTSLMs on ECG-QA. To facilitate further research, we provide all code, datasets, and models open-source.

Frederik Rajiv Manichand, Robin Deuber, Robert Jakob et al. · eth-zurich, harvard

Estimating Body Mass Index (BMI) from camera images with machine learning models enables rapid weight assessment when traditional methods are unavailable or impractical, such as in telehealth or emergency scenarios. Existing computer vision approaches have been limited to datasets of up to 14,500 images. In this study, we present a deep learning-based BMI estimation method trained on our WayBED dataset, a large proprietary collection of 84,963 smartphone images from 25,353 individuals. We introduce an automatic filtering method that uses posture clustering and person detection to curate the dataset by removing low-quality images, such as those with atypical postures or incomplete views. This process retained 71,322 high-quality images suitable for training. We achieve a Mean Absolute Percentage Error (MAPE) of 7.9% on our hold-out test set (WayBED data) using full-body images, the lowest value in the published literature to the best of our knowledge. Further, we achieve a MAPE of 13% on the completely unseen~(during training) VisualBodyToBMI dataset, comparable with state-of-the-art approaches trained on it, demonstrating robust generalization. Lastly, we fine-tune our model on VisualBodyToBMI and achieve a MAPE of 8.56%, the lowest reported value on this dataset so far. We deploy the full pipeline, including image filtering and BMI estimation, on Android devices using the CLAID framework. We release our complete code for model training, filtering, and the CLAID package for mobile deployment as open-source contributions.

Fan Wu, Kaicheng Zhao, Elgar Fleisch et al.

AI-based voice analysis shows promise for disease diagnostics, but existing classifiers often fail to accurately identify specific pathologies because of gender-related acoustic variations and the scarcity of data for rare diseases. We propose a novel two-stage framework that first identifies gender-specific pathological patterns using ResNet-50 on Mel spectrograms, then performs gender-conditioned disease classification. We address class imbalance through multi-scale resampling and time warping augmentation. Evaluated on a merged dataset from four public repositories, our two-stage architecture with time warping achieves state-of-the-art performance (97.63\% accuracy, 95.25\% MCC), with a 5\% MCC improvement over single-stage baseline. This work advances voice pathology classification while reducing gender bias through hierarchical modeling of vocal characteristics.

George Boateng, Elgar Fleisch, Tobias Kowatsch

Couples' relationships affect the physical health and emotional well-being of partners. Automatically recognizing each partner's emotions could give a better understanding of their individual emotional well-being, enable interventions and provide clinical benefits. In the paper, we summarize and synthesize works that have focused on developing and evaluating systems to automatically recognize the emotions of each partner based on couples' interaction or conversation contexts. We identified 28 articles from IEEE, ACM, Web of Science, and Google Scholar that were published between 2010 and 2021. We detail the datasets, features, algorithms, evaluation, and results of each work as well as present main themes. We also discuss current challenges, research gaps and propose future research directions. In summary, most works have used audio data collected from the lab with annotations done by external experts and used supervised machine learning approaches for binary classification of positive and negative affect. Performance results leave room for improvement with significant research gaps such as no recognition using data from daily life. This survey will enable new researchers to get an overview of this field and eventually enable the development of emotion recognition systems to inform interventions to improve the emotional well-being of couples.

Varun Mishra, Florian Künzler, Jan-Niklas Kramer et al.

JITAI is an emerging technique with great potential to support health behavior by providing the right type and amount of support at the right time. A crucial aspect of JITAIs is properly timing the delivery of interventions, to ensure that a user is receptive and ready to process and use the support provided. Some prior works have explored the association of context and some user-specific traits on receptivity, and have built post-study machine-learning models to detect receptivity. For effective intervention delivery, however, a JITAI system needs to make in-the-moment decisions about a user's receptivity. To this end, we conducted a study in which we deployed machine-learning models to detect receptivity in the natural environment, i.e., in free-living conditions. We leveraged prior work regarding receptivity to JITAIs and deployed a chatbot-based digital coach~-- Ally~-- that provided physical-activity interventions and motivated participants to achieve their step goals. We extended the original Ally~app to include two types of machine-learning model that used contextual information about a person to predict when a person is receptive: a \textit{static model\/} that was built before the study started and remained constant for all participants and an \textit{adaptive model\/} that continuously learned the receptivity of individual participants and updated itself as the study progressed. For comparison, we included a \textit{control model\/} that sent intervention messages at random times. The app randomly selected a delivery model for each intervention message. We observed that the machine-learning models led up to a 40\% improvement in receptivity as compared to the control model. Further, we evaluated the temporal dynamics of the different models and observed that receptivity to messages from the adaptive model increased over the course of the study.

Bernhard Gahr, Shu Liu, Kevin Koch et al.

Driver identification has emerged as a vital research field, where both practitioners and researchers investigate the potential of driver identification to enable a personalized driving experience. Within recent years, a selection of studies have reported that individuals could be perfectly identified based on their driving behavior under controlled conditions. However, research investigating the potential of driver identification under naturalistic conditions claim accuracies only marginally higher than random guess. The paper at hand provides a comprehensive summary of the recent work, highlighting the main discrepancies in the design of the machine learning approaches, primarily the window length parameter that was considered. Key findings further indicate that the longitudinal vehicle control information is particularly useful for driver identification, leaving the research gap on the extent to which the lateral vehicle control can be used for reliable identification. Building upon existing work, we provide a novel approach for the design of the window length parameter that provides evidence that reliable driver identification can be achieved with data limited to the steering wheel only. The results and insights in this paper are based on data collected from the largest naturalistic driving study conducted in this field. Overall, a neural network based on GRUs was found to provide better identification performance than traditional methods, increasing the prediction accuracy from under 15\% to over 65\% for 15 drivers. When leveraging the full field study dataset, comprising 72 drivers, the accuracy of identification prediction of the approach improved a random guess approach by a factor of 25.