This study investigates demographic bias in ECG-based biometric recognition systems and proposes a data augmentation approach to mitigate performance disparities across gender, age, and smoking-status subgroups. Our analysis reveals significant accuracy gaps, with male subjects outperforming females by 18

Class imbalance can significantly impact the performance of learning algorithms, often leading to prediction bias toward the majority class. This challenge is particularly critical in healthcare-related domains, as medical datasets are often imbalanced, hindering the accurate prediction of the minority class, which is commonly the class of interest. As such, this work introduces a novel resampling algorithm, designated Genetic Beta Oversampling, which integrates user-defined preferences into the synthetic data generation process, allowing fine control over the model’s inclination towards false negatives or false positives. These user preferences are encoded in the form of a parameter, β , which dictates the trade-off between recall and precision that the method should seek to achieve. This flexibility is particularly relevant in clinical settings, where prioritizing recall can enhance patient care by reducing missed diagnoses. We evaluate the proposed approach on the EMPA and AI4PAIN datasets for pain classification, a recall-critical task in which undetected pain episodes must be minimized. Experimental results show that our method consistently surpasses SMOTE, SMOTE-IPF, and four cost-sensitive classifiers in terms of F β -score across a range of β values and synthetically induced imbalance ratios. These results highlight the adaptability of the method to recall-sensitive applications, including pain assessment and broader clinical decision-support scenarios.

Assessment of respiratory volumes is crucial for the long-term management of chronic respiratory diseases. However, standard methods such as spirometry require active patient cooperation and are unsuitable for regular monitoring. This study introduces capacitive pressure sensors integrated with a signal processing algorithm for respiratory assessment/monitoring. Two sensor variants using poly(glycerol sebacate) (PGS) substrates are presented: CS1, featuring a porous structure, and CS2, incorporating a pyramidal surface pattern. Both sensors measure thoracic expansion through capacitance changes. Signals are preprocessed and statistically validated against a commercial airflow transducer in 38 healthy adult participants. Although CS1 exhibits higher sensitivity (0.09 kPa(-)(1)) than CS2 (0.015 kPa(-)(1)), both sensors demonstrate strong correlation (mean R-2 > 0.91) with the reference device across volunteers. Measurement accuracy is confirmed by low mean absolute errors across respiratory cycles: 0.122 L (95% confidence interval (CI) +/- 0.027 L) for CS1, and 0.100 L (95% CI +/- 0.018 L) for CS2. These results demonstrate that the developed capacitive sensors and signal processing algorithm effectively capture thoracic volume changes, showing potential for non-invasive and continuous respiratory monitoring.

This study presents an exploratory framework for electrocardiogram (ECG)-based biometric identification using ECG signals acquired using biocompatible, solid-state conductive electrodes. The system addresses key challenges in ECG biometrics, including user convenience, temporal stability, and minimal data requirements. ECG signals were collected from 81 participants in four sessions, with intervals of one, two, and three weeks between sessions, to evaluate the system’s performance across varying temporal conditions. To validate the feasibility of the proposed electrodes, simultaneous recordings were made using traditional Ag/AgCl electrodes, with results showing strong agreement (Pearson correlation = 0.9923, RMSE = 0.0174). Gramian Angular Field (GAF) representations were used to transform ECG signals into spatial images, which were then fed into a convolutional neural network (CNN). Results demonstrate that a short ECG segment of 10 heartbeats (approximately 10 s) can achieve competitive identification accuracy, with same-session performance reaching 94–98

This study investigates the feasibility of using electrocardiogram (ECG) signals as a biometric recognition trait, offering an alternative to traditional methods such as passwords and fingerprint scanning. It leverages ECG’s inherent liveness detection, high security, and unique physiological properties to develop a robust identification system. Using data collected from three sessions over two weeks, the ECG signals were processed through filtering and normalization before feature extraction and the application of various machine learning classifiers. The experiments evaluated the impact of temporal separation between training and testing sessions, revealing a significant drop in accuracy with increased time intervals. To mitigate this, the effectiveness of multi-session training was explored, improving the system performance. The trade-off between the number of heartbeats used for training and identification accuracy was also assessed, showing that while more heartbeats generally led to higher accuracy, the marginal gains did not justify the extended acquisition time in real-world scenarios. Linear Discriminant Analysis (LDA) emerged as the most effective classifier, consistently outperforming others in various scenarios. The obtained results underscore the potential of ECG-based biometrics in secure identification systems, particularly when leveraging data from multiple sessions.

The assessment of pain intensity after surgery is important for guiding pain management. Due to the limitations of current pain evaluation tools, there is increasing interest in using objective methods to assess pain in clinical settings. This literature review aims to provide an overview of physiological methods for postoperative pain evaluation. The Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines were followed to perform a literature search on the Scopus database from 2002 to March 2024. Sixty-one studies were included in this review investigating parameters derived from six physiological signals, including Electrocardiogram (ECG), Photoplethysmogram (PPG), Skin Conductance (SC), Electroencephalogram (EEG), Electromyogram (EMG), and pupil size and reflexes. Parameters extracted from ECG, PPG, and SC signals were the most commonly studied. While there is evidence to support the use of physiological parameters as measures for postoperative pain evaluation, further research is needed to establish the reliability and generalizability of these parameters to develop an indicator that can be consistently applied across various clinical settings. For this purpose, this review describes current research findings and identifies limitations and directions for future work that should be considered in upcoming research on postoperative pain assessment.

The acquisition of high-quality electrocardiogram (ECG) signals is crucial for biomedical applications. Traditional Ag/AgCl wet electrodes face limitations in terms of comfort, long-term use, and ease of application. This study presents the development and evaluation of solid state electrodes composed of a polydimethylsiloxane (PDMS) substrate and a silver-ink conductive layer. These electrodes were designed for the acquisition of electrocardiogram (ECG) signals at the fingers. Their performance was evaluated through comparative analysis of ECG signals acquired with traditional Ag/AgCl electrodes. Results demonstrated that the developed electrodes achieved high signal quality, with consistent waveforms and minimal distortion, even in the absence of conductive gels and a reference electrode. The synchronization technique used, namely Dynamic Time Warping, enabled effective signal alignment, further validating the similarity between the ECG signals acquired by the two electrode types. The proposed PDMS-silver ink electrodes provide distinct advantages, including improved comfort, reusability, and biocompatibility, offering a practical and user-friendly alternative to traditional electrodes. Future work will explore their robustness and potential for seamless biometric recognition systems.

Objective and automated pain intensity assessment is a critical challenge in healthcare, with physiological signals offering a promising, non-invasive avenue for quantification. This paper addresses the AI4Pain Grand Challenge task of classifying pain into three levels (No Pain, Low Pain, High Pain) using multimodal physiological data, including Electrodermal Activity (EDA), Blood Volume Pulse (BVP), Respiration (RESP), and SpO2. We conduct a systematic comparison of two distinct modeling paradigms: (1) a feature-driven approach using TabPFN, a pre-trained tabular learner, which leverages extensive handcrafted features, ensembling, and stacking techniques; and (2) a hybrid deep learning model, Transformer with Features (TwF), which combines a Transformer-based encoder for raw signal analysis with a curated set of physiological features. Our experiments demonstrate that the feature-driven TabPFN approach significantly outperforms the deep temporal model. The final stacked TabPFN model achieved a test accuracy of 60.06%, whereas the TwF model reached 54.02%. These results suggest that for physiological pain classification with limited data and high inter-subject variability, robust feature engineering combined with models possessing strong inductive biases is more effective than complex, end-to-end deep learning architectures.

Class imbalance can significantly impact the performance of learning algorithms, often leading to prediction bias toward the majority class. This challenge is particularly critical in healthcare-related domains, as medical datasets are often imbalanced, hindering the accurate prediction of the minority class, which is commonly the class of interest. As such, this work introduces a novel resampling algorithm, designated Genetic Beta Oversampling, which integrates user-defined preferences into the synthetic data generation process, allowing fine control over the model’s inclination towards false negatives or false positives. These user preferences are encoded in the form of a parameter, β , which dictates the trade-off between recall and precision that the method should seek to achieve. This flexibility is particularly relevant in clinical settings, where prioritizing recall can enhance patient care by reducing missed diagnoses. We validate the approach using the EMPA dataset for the classification of pain induced by a cold stimulus, where prioritizing recall is essential to minimize missed pain detections. Experimental results demonstrate that our method outperforms SMOTE, SMOTE-IPF, and four cost-sensitive classifiers in terms of F β -score across diverse β settings. These findings underscore the method’s adaptability in recall-sensitive applications, such as pain assessment and clinical decision-making.

Postoperative pain is a subjective experience, typically assessed through self-report scales, such as the numerical rating scale. When self-reporting is not possible, healthcare providers often rely on clinical observations. However, due to its subjective nature, pain assessment remains challenging, leading to increased interest in objective methods to support assessment in clinical settings, including the use of physiological signals such as the Electrocardiogram (ECG). This study aimed to develop a method for detecting changes in physiological signals over time based on symbolic theory and compression analysis. The proposed method compares ECG segments by first transforming ECG intervals into a symbolic representation using the Symbolic Aggregate approXimation (SAX) algorithm adapted to the ECG probability distribution of the subject. Afterward, the similarity between SAX segments is assessed using the Normalized Compression Distance (NCD), a data compression metric related to the approximation of the Kolmogorov complexity. To identify changes in the ECG signal over time, the method employed two strategies: determining the similarity between each SAX segment either with the subject’s baseline SAX segment or with the preceding SAX segment. This method was applied to ECG signals collected in the recovery room after surgery, focusing specifically on intervals related to postoperative pain assessment and subsequent analgesic treatment. Although no clear association was found between variations in similarity and pain treatment, the method successfully detected changes in the ECG over time relative to the patient’s baseline.

This paper conducts an extensive review of flexible cardiac sensing devices designed for electrocardiogram (ECG) acquisitions, with emphasis on their application in cardiac health monitoring. This study focuses on characteristics crucial to these devices, including: flexibility, durability, biocompatibility, sensitivity, and stretchability. It provides a comprehensive overview of prevalent fabrication methods and materials employed for flexible electrode production, with insights from several studies that utilize these electrodes across diverse applications. Furthermore, the review highlights the significant role of machine learning (ML) in cardiac health monitoring and broader ECG analysis applications. With the most used methods being deep learning, support vector machines, random forest, and linear discriminant analysis, the paper delves into studies that leverage ML for heart disease classification as well as other applications such as emotion detection and biometric recognition. The paper culminates with an overview of studies that integrate both flexible sensing technology and ML, particularly in the domain of cardiac health monitoring. It sheds light on the important relationship between these two techniques, underscoring their impact on advancing ECG-based health monitoring methodologies. Besides reviewing the current state of these technologies, the paper also outlines future perspectives and potential directions for research in this domain.

Pain assessment is a hot topic today. As pain is perceived differently by each person and there are cases incapable of communicating it, automatic methods for pain classification based on machine learning (ML) are in constant development. Physiological signals, such as the Electrocardiogram (ECG), the Electrodermal Activity (EDA), and the Electromyogram (EMG) have been used in this context. However, there is a lack of documentation on the interpretability and explainability of its features in pain assessment. Thus, this work aims to evaluate the impact of features extracted from physiological signals in pain recognition. The data was collected during a protocol for pain induction using a Cold Pressor Task with previous emotional elicitation. Three different ML algorithms (Random Forest, XGBoost, and ADABoost) were trained using 17 features. Then, three explainable Artificial Intelligence (xAI) methods (the KernelSHAP, the SP-LIME, and the Morris Sensitivity Analyses) were applied to each of the three classifiers to select the most relevant features for each model. The ML algortihms were trained again using only the five most relevant features obtained with each xAI method. The results showed that the best accuracy is obtained by combining XGBoost + KernelSHAP. However, Random Forest + Morris Sensitivity provides fewer False Negatives. Moreover, the amplitude of the EMG of the trapezius and triceps muscles and the maximum and minimum value of the tonic component of the EDA were highlighted by the xAI methods.

Pain is a highly subjective phenomenon that depends on multiple factors. The common methods used to evaluate pain require the person to be awakened and cooperative, which may not always be possible. Moreover, such methods are subject to non-quantifiable influences, namely the impact of an individual’s emotional state on how pain is perceived or how negative emotions may exacerbate pain perception, while positive emotions may attenuate it. The goal of this study was to conduct a novel protocol for pain induction with emotional elicitation and assess its feasibility. In this protocol, the physiological responses were monitored, and collected, through Electrocardiogram, Electrodermal Activity, and surface Electromyogram signals. Along the protocol, the pain perception was evaluated using a 0–10 numerical rating scale and by registering the time from the pain stimulus beginning to the Pain and Tolerance Thresholds. This study comprised three emotional sessions, negative, positive, and neutral, which were performed through videos of excerpts of terror, comedy, and documentary films, respectively, followed by pain induction using the Cold Pressor Task (CPT). A total of 56 participants performed the study, with a CPT mean time of about 91.70 ± 39.64 s among all the sessions. The conducted protocol was considered feasible and safe as it allowed the collection of physiological data, pain, and questionnaires’ reports from 56 participants, without any harm to them. Moreover, the collected data can be further used to assess how emotional conditions influence pain perception and to provide better emotion-calibrated pain recognition systems based on physiological signals.

This study analyses and characterizes changes in features extracted from physiological signals in the presence of affective states, namely, neutral, stress, and amusement. With a focus on the stress condition, for this purpose a statistical analysis was performed on various features extracted from ECG (electrocardiogram), EMG (electromyography), EDA (electrodermal activity), and RESP (respiration) signals from the WESAD dataset. This dataset provides data from 15 healthy participants regarding three affective conditions. Concerning the ECG features, from the statistical analysis, it was possible to observe a significant decrease in the interval between consecutive R peaks, meaning that there was an increase in heart rate. The EMG signal showed a significant increase in muscle activation. Regarding the EDA, it was noticed that during the stress condition, the production of sweat increased, leading to greater skin conductivity. Lastly with respect to the RESP signal, although no significant difference was observed in regards to changes in the inspiration and expiration durations, the increase in the standard deviation leads inferring an increase in the irregularity of the breathing pattern of the participants during the stress condition.

Obtaining large datasets for training and evaluating models remains a challenge in the medical and healthcare fields due to patient privacy, organizational and legal constraints, and the complex, time-consuming nature of data collection from human subjects. Moreover, expert labeling of collected samples can be demanding, limiting the availability of labeled data. Training models on small datasets often results in low feasibility. Therefore, when increasing the dataset size through additional sample collection becomes unfeasible, alternative approaches should be considered. This paper offers an overview of strategies employed in the medical and healthcare domain to tackle small datasets or datasets with limited labeled samples, including data augmentation, transfer learning, and semi-supervised learning.

Pain recognition is a challenging yet crucial problem. Common methods used to measure pain are based on self-reported tools; however, not every person can communicate it. Therefore, automatic pain detection methods have emerged. Since pain is related to the emotional state of an individual, this variable must be considered. Thus, this work proposes pain prediction under different emotional contexts. For this purpose, data were collected during a protocol designed for pain induction with previous emotional elicitation. Emotions were elicited through videos composed of excerpts of documentaries, horror and comedy films, while the pain was induced through a Cold Pressor Test. Physiological signals, such as electrocardiogram, electrodermal activity and surface electromyogram, were collected during the protocol. Furthermore, several questionnaires were answered and pain reports were also registered. Two problems were addressed: pain classification and estimation of the Pain Tolerance score. The algorithm with the best performance for each problem was found using only data from the neutral session and nested cross-validation strategy. Using only physiological data from the neutral session, a F1-score of 99.32

Pain is subjective and hard to define. It is currently measured by self-reports. However, it is not ideal since not every person can report pain. Developing wearable devices that can automatically detect pain is crucial, but to achieve these devices it is necessary to consider machine learning (ML) models that can classify pain. This work aims to train ML models to perform pain recognition and another to estimate pain intensity using physiological signals. The hyperparameters of each model were tuned using Nested Cross-validation. The Random Forest Classifier was found to be the model that performed better for pain detection with an F1-score of 99.32% in the test set as the Gradient Boost Regressor was considered the best model for pain intensity estimation with a Mean Absolute Error of 0.24 in the test set. As the results obtained are promising, this work motivates further research in this area.

Respiratory diseases are currently monitored through traditional pulmonary function tests, such as spirometry. However, the restrictions of these procedures, particularly in the context of the COVID-19 pandemic, have underscored the need for alternative approaches to respiratory health assessment. Wearable devices have emerged as a promising solution, providing continuous data collection, and overcoming the limitations posed by conventional methods. This review explores the multifaceted field of wearable devices for respiratory monitoring, presenting the most common sensing technologies applied to pulmonary ventilation, their constituent materials, fabrication techniques, and diverse morphologies to enhance sensor performance. The role of machine learning algorithms and ethical data sharing is highlighted, contributing to the forthcoming patient-centered healthcare landscape. Ultimately, the importance of validation and calibration protocols for wearable devices is underlined. In anticipation of evolving healthcare needs, this in-depth study addresses the current challenges in wearable respiratory monitoring while laying a robust foundation for a personalized, connected, and ethically sound future for respiratory care. Beyond the limitations of traditional pulmonary function tests, wearable technology emerges as a game-changer for respiratory monitoring. This review explores the sensing technologies and materials employed in the construction of accurate physiological data acquisition devices. Leveraging artificial intelligence while adhering to strict ethical protocols, the future holds promise for wearable integration within patient-centered care. image

Anxiety is currently increasing in human daily life. Studies aimed to deepen the understanding of it, to minimize its negative impact on people's lives, have gained significant importance. In this context, the main focus of this work is to study the use of several physiological signals (electrocardiogram, electrodermal activity, and blood volume pulse) to predict the level of anxiety felt by a subject using four different approaches. These involve the use of a different number of features that are selected as more informative, and by training two classification models, with different properties. For the chosen approaches, the obtained results are compared and analysed to understand which performs better, i.e., which has a greater ability in anxiety recognition, being the anxiety level of each participant assessed through the application of the STICSA questionnaire. When selecting 20 features to train a Linear Discriminant Analysis model, an accuracy and precision of over 70% were achieved. This strategy presented the best performance as this model surpasses, for all the used metrics, the results obtained when using the models based on a decision tree. The encouraging obtained results sustained the feasibility of the use of simultaneous different physiological signals to train models for predicting the level of anxiety.

This study analyzed the utility of the Analgesia Nociception Index (ANI) in postoperative pain assessment as an objective pain indicator. Electrocardiogram (ECG) recordings from 30 patients in the Post Anesthesia Care Unit were considered, with a total of 52 pain reports registered during the recovery period, of which 32 were valid for the study. R peaks series (RR) were extracted from the ECG, and the ANI series were extracted from the RR series with a sampling frequency of 0.2 Hz. For each pain report, 11 statistical features were computed from cumulative paired windows of the ANI series before and after the pain report. Statistical tests were performed based on the distribution of ANI features, revealing that for the group of 5-minute windows, there were statistically significant differences in Median, Maximum, Mean, and Average Power. Several machine learning models to predict pain were trained using different datasets, for instance, (1) with the four above mentioned features, and (2) with all 11 features. The models resorted to the K-Nearest Neighbors (KNN) and Decision Tree (DT) algorithms. The best prediction models achieved an acceptable area under the Receiver Operating Characteristic curve of 0.74. Hence, the KNN model fitted with the dataset (1) could be considered the best due to its higher computational efficiency and greater specificity, which is preferable for avoiding unnecessary administration of analgesia. However, the model fitted with the dataset (2) presented greater sensitivity, which is preferable for avoiding the neglect of real pain cases.