Performance Assessment of Heartbeat Detection Algorithms on Photoplethysmograph and Functional NearInfrared Spectroscopy Signals.

With wearable sensors, the acquisition of physiological signals has become affordable and feasible in everyday life. Specifically, Photoplethysmography (PPG), being a low-cost and highly portable technology, has attracted notable interest for measuring and diagnosing cardiac activity, one of the most important physiological and autonomic indicators. In addition to the technological development, several specific signal-processing algorithms have been designed to enable reliable detection of heartbeats and cope with the lower quality of the signals. In this study, we compare three heartbeat detection algorithms: Derivative-Based Detection (DBD), Recursive Combinatorial Optimization (RCO), and Multi-Scale Peak and Trough Detection (MSPTD). In particular, we considered signals from two datasets, namely, the PPG-DALIA dataset (N = 15) and the FANTASIA dataset (N = 20) which differ in terms of signal characteristics (sampling frequency and length) and type of acquisition devices (wearable and medical-grade). The comparison is performed both in terms of heartbeat detection performance and computational workload required to execute the algorithms. Finally, we explore the applicability of these algorithms on the cardiac component obtained from functional Near InfraRed Spectroscopy signals (fNIRS).The results indicate that, while the MSPTD algorithm achieves a higher F1 score in cases that involve body movements, such as cycling (MSPTD: Mean = 74.7, SD = 14.4; DBD: Mean = 54.4, SD = 21.0; DBD + RCO: Mean = 49.5, SD = 22.9) and walking up and down the stairs (MSPTD: Mean = 62.9, SD = 12.2; DBD: Mean = 50.5, SD = 11.9; DBD + RCO: Mean = 45.0, SD = 14.0), for all other activities the three algorithms perform similarly. In terms of computational complexity, the computation time of the MSPTD algorithm appears to grow exponentially with the signal sampling frequency, thus requiring longer computation times in the case of high-sampling frequency signals, where the usage of the DBD and RCO algorithms might be preferable. All three algorithms appear to be appropriate candidates for exploring the applicability of heartbeat detection on fNIRS data.

Exposure to Multicultural Context Affects Neural Response to Out-Group Faces: A Functional Magnetic Resonance Imaging Study.

Recent migration and globalization trends have led to the emergence of ethnically, religiously, and linguistically diverse countries. Understanding the unfolding of social dynamics in multicultural contexts becomes a matter of common interest to promote national harmony and social cohesion among groups. The current functional magnetic resonance imaging (fMRI) study aimed to (i) explore the neural signature of the in-group bias in the multicultural context; and (ii) assess the relationship between the brain activity and people's system-justifying ideologies. A sample of 43 (22 females) Chinese Singaporeans (M = 23.36; SD = 1.41) was recruited. All participants completed the Right Wing Authoritarianism Scale and Social Dominance Orientation Scale to assess their system-justifying ideologies. Subsequently, four types of visual stimuli were presented in an fMRI task: Chinese (in-group), Indian (typical out-group), Arabic (non-typical out-group), and Caucasian (non-typical out-group) faces. The right middle occipital gyrus and the right postcentral gyrus showed enhanced activity when participants were exposed to in-group (Chinese) rather than out-group (Arabic, Indian, and Caucasian) faces. Regions having a role in mentalization, empathetic resonance, and social cognition showed enhanced activity to Chinese (in-group) rather than Indian (typical out-group) faces. Similarly, regions typically involved in socioemotional and reward-related processing showed increased activation when participants were shown Chinese (in-group) rather than Arabic (non-typical out-group) faces. The neural activations in the right postcentral gyrus for in-group rather than out-group faces and in the right caudate in response to Chinese rather than Arabic faces were in a significant positive correlation with participants' Right Wing Authoritarianism scores (p < 0.05). Furthermore, the activity in the right middle occipital gyrus for Chinese rather than out-group faces was in a significant negative correlation with participants' Social Dominance Orientation scores (p < 0.05). Results are discussed by considering the typical role played by the activated brain regions in socioemotional processes as well as the role of familiarity to out-group faces.

Probing the association between maternal anxious attachment style and mother-child brain-to-brain coupling during passive co-viewing of visual stimuli.

Brain-to-brain coupling during co-viewing of video stimuli reflects similar intersubjective mentalisation processes. During an everyday joint activity of watching video stimuli (television shows) with her child, an anxiously attached mother's preoccupation with her child is likely to distract her from understanding the mental state of characters in the show. To test the hypothesis that reduced coupling in the medial prefrontal cortex (PFC) would be observed with increasing maternal attachment anxiety (MAA), we profiled mothers' MAA using the Attachment Style Questionnaire and used functional Near-infrared Spectroscopy (fNIRS) to assess PFC coupling in 31 mother-child dyads while they watched three 1-min animation videos together. Reduced coupling was observed with increasing MAA in the medial right PFC cluster which is implicated in mentalisation processes. This result did not survive control analyses and should be taken as preliminary. Reduced coupling between anxiously-attached mothers and their children during co-viewing could undermine quality of shared experiences.

Predicting Adolescent Mental Health Outcomes Across Cultures: A Machine Learning Approach.

Adolescent mental health problems are rising rapidly around the world. To combat this rise, clinicians and policymakers need to know which risk factors matter most in predicting poor adolescent mental health. Theory-driven research has identified numerous risk factors that predict adolescent mental health problems but has difficulty distilling and replicating these findings. Data-driven machine learning methods can distill risk factors and replicate findings but have difficulty interpreting findings because these methods are atheoretical. This study demonstrates how data- and theory-driven methods can be integrated to identify the most important preadolescent risk factors in predicting adolescent mental health. Machine learning models examined which of 79 variables assessed at age 10 were the most important predictors of adolescent mental health at ages 13 and 17. These models were examined in a sample of 1176 families with adolescents from nine nations. Machine learning models accurately classified 78% of adolescents who were above-median in age 13 internalizing behavior, 77.3% who were above-median in age 13 externalizing behavior, 73.2% who were above-median in age 17 externalizing behavior, and 60.6% who were above-median in age 17 internalizing behavior. Age 10 measures of youth externalizing and internalizing behavior were the most important predictors of age 13 and 17 externalizing/internalizing behavior, followed by family context variables, parenting behaviors, individual child characteristics, and finally neighborhood and cultural variables. The combination of theoretical and machine-learning models strengthens both approaches and accurately predicts which adolescents demonstrate above average mental health difficulties in approximately 7 of 10 adolescents 3-7 years after the data used in machine learning models were collected.

Acquisition and Processing of Brain Signals.

We live within a context of unprecedented opportunities for brain research, with a flourishing of novel sensing technologies and methodological approaches [...].

Evaluating reproducibility of AI algorithms in digital pathology with DAPPER.

Artificial Intelligence is exponentially increasing its impact on healthcare. As deep learning is mastering computer vision tasks, its application to digital pathology is natural, with the promise of aiding in routine reporting and standardizing results across trials. Deep learning features inferred from digital pathology scans can improve validity and robustness of current clinico-pathological features, up to identifying novel histological patterns, e.g., from tumor infiltrating lymphocytes. In this study, we examine the issue of evaluating accuracy of predictive models from deep learning features in digital pathology, as an hallmark of reproducibility. We introduce the DAPPER framework for validation based on a rigorous Data Analysis Plan derived from the FDA's MAQC project, designed to analyze causes of variability in predictive biomarkers. We apply the framework on models that identify tissue of origin on 787 Whole Slide Images from the Genotype-Tissue Expression (GTEx) project. We test three different deep learning architectures (VGG, ResNet, Inception) as feature extractors and three classifiers (a fully connected multilayer, Support Vector Machine and Random Forests) and work with four datasets (5, 10, 20 or 30 classes), for a total of 53, 000 tiles at 512 × 512 resolution. We analyze accuracy and feature stability of the machine learning classifiers, also demonstrating the need for diagnostic tests (e.g., random labels) to identify selection bias and risks for reproducibility. Further, we use the deep features from the VGG model from GTEx on the KIMIA24 dataset for identification of slide of origin (24 classes) to train a classifier on 1, 060 annotated tiles and validated on 265 unseen ones. The DAPPER software, including its deep learning pipeline and the Histological Imaging-Newsy Tiles (HINT) benchmark dataset derived from GTEx, is released as a basis for standardization and validation initiatives in AI for digital pathology.

Comparison of Wearable and Clinical Devices for Acquisition of Peripheral Nervous System Signals.

A key access point to the functioning of the autonomic nervous system is the investigation of peripheral signals. Wearable devices (WDs) enable the acquisition and quantification of peripheral signals in a wide range of contexts, from personal uses to scientific research. WDs have lower costs and higher portability than medical-grade devices. However, the achievable data quality can be lower, and data are subject to artifacts due to body movements and data losses. It is therefore crucial to evaluate the reliability and validity of WDs before their use in research. In this study, we introduce a data analysis procedure for the assessment of WDs for multivariate physiological signals. The quality of cardiac and electrodermal activity signals is validated with a standard set of signal quality indicators. The pipeline is available as a collection of open source Python scripts based on the pyphysio package. We apply the indicators for the analysis of signal quality on data simultaneously recorded from a clinical-grade device and two WDs. The dataset provides signals of six different physiological measures collected from 18 subjects with WDs. This study indicates the need to validate the use of WDs in experimental settings for research and the importance of both technological and signal processing aspects to obtain reliable signals and reproducible results.

Culture, sex and social context influence brain-to-brain synchrony: an fNIRS hyperscanning study.

BACKGROUND: Unique interpersonal synchrony occurs during every social interaction, and is shaped by characteristics of participating individuals in these social contexts. Additionally, depending on context demands, interpersonal synchrony is also altered. The study therefore aims to investigate culture, sex, and social context effects simultaneously in a novel role-play paradigm. Additionally, the effect of personality traits on synchrony was investigated across cultures, and a further exploratory analysis on the effects of these variables on pre- and post-session empathy changes was conducted. METHODS: 83 dyads were recruited in two waves from Singapore and Italy and took part in a within-subjects session where they interacted with each other as themselves (Naturalistic Conversation) and as others (Role-Play and Role Reversal). Big Five Inventory (administered pre-session) and Interpersonal Reactivity Index (administered pre- and post-session) were used as measures of personality and empathy respectively, while synchrony was measured using hyperscanning functional near-infrared spectroscopy in the prefrontal cortex. After data-preprocessing and preliminary analyses, a mixture of multiple linear regression and exploratory forward stepwise regression models were used to address the above study aims. RESULTS: Results revealed significant main and interaction effects of culture, sex and social context on brain-to-brain synchrony, particularly in the medial left cluster of the prefrontal cortex, and a unique contribution of extraversion and openness to experience to synchrony in the Italian cohort only. Finally, culture-driven differences in empathy changes were identified, where significant increases in empathy across sessions were generally only observed within the Singaporean cohort. CONCLUSIONS: Main findings indicate lowered brain-to-brain synchrony during role-playing activities that is moderated by the dyad's sex make-up and culture, implying differential processing of social interactions that is also influenced by individuals' background factors. Findings align with current literature that role-playing is a cognitively demanding activity requiring greater levels of self-regulation and suppression of self-related cognition as opposed to interpersonal co-regulation characterized by synchrony. However, the current pattern of results would be better supported by future studies investigating multimodal synchronies and corroboration.

Neural Processing of Sexist Comments: Associations between Perceptions of Sexism and Prefrontal Activity.

Sexism is a widespread form of gender discrimination which includes remarks based on gender stereotypes. However, little is known about the neural basis underlying the experience of sexist-related comments and how perceptions of sexism are related to these neural processes. The present study investigated whether perceptions of sexism influence neural processing of receiving sexist-related comments. Participants (N = 67) read experimental vignettes describing scenarios of comments involving gender stereotypes while near-infrared spectroscopy recordings were made to measure the hemodynamic changes in the prefrontal cortex. Results found a significant correlation between participants' perceptions of sexism and brain activation in a brain cluster including the right dorsolateral prefrontal cortex and inferior frontal gyrus. There was a significant gender difference where female participants showed a stronger negative correlation compared to male participants. Future research can expand on these initial findings by looking at subcortical structures involved in emotional processing and gender stereotype application as well as examining cultural differences in perceptions of gender stereotypes and sexism.

Assessing Computational Methods to Quantify Mother-Child Brain Synchrony in Naturalistic Settings Based on fNIRS Signals.

Mother-child brain-to-brain synchrony captures the temporal similarities in brain signals between dyadic partners, and has been shown to emerge during the display of joint behaviours. Despite the rise in the number of studies that investigate synchrony in naturalistic contexts, the use of varying methodological approaches to compute synchrony remains a central problem. When dyads engage in unstructured social interactions, the wide range of behavioural cues they display contribute to the use of varying lengths of signals to compute synchrony. The present functional Near-infrared Spectroscopy (fNIRS) study investigates how different methods to quantify brain signals during joint and non-joint portions of dyadic play affect the outcome of brain-to-brain synchrony. Three strategies to cope with unstructured data are tested and different signal lengths of 15, 20, 25, 30, 35, 40, 45s were used to determine the optimal method to sensitively capture synchrony. Results showed that using all available portions of the signals generated a greater number of less conservative results compared to the other two strategies, which were to compute the average synchrony for the joint and non-joint signals portions and to compute the difference between the average synchrony of joint and non-joint portions. From the different signal durations, only length portions of 25s to 35s generated significant results. These findings demonstrate that differences in computational approaches and signal lengths affect synchrony measurements and should be considered in naturalistic synchrony studies.

A Machine Learning Perspective on fNIRS Signal Quality Control Approaches.

Despite a rise in the use of functional Near Infra-Red Spectroscopy (fNIRS) to study neural systems, fNIRS signal processing is not standardized and is highly affected by empirical and manual procedures. At the beginning of any signal processing procedure, Signal Quality Control (SQC) is critical to prevent errors and unreliable results. In fNIRS analysis, SQC currently relies on applying empirical thresholds to handcrafted Signal Quality Indicators (SQIs). In this study, we use a dataset of fNIRS signals (N = 1,340) recorded from 67 subjects, and manually label the signal quality of a subset of segments (N = 548) to investigate the pitfalls of current practices while exploring the opportunities provided by Deep Learning approaches. We show that SQIs statistically discriminate signals with bad quality, but the identification by means of empirical thresholds lacks sensitivity. Alternatively to manual thresholding, conventional machine learning models based on the SQIs have been proven more accurate, with end-to-end approaches, based on Convolutional Neural Networks, capable of further improving the performance. The proposed approach, based on machine learning, represents a more objective SQC for fNIRS and moves towards the use of fully automated and standardized procedures.

Parent-child dyads with greater parenting stress exhibit less synchrony in posterior areas and more synchrony in frontal areas of the prefrontal cortex during shared play.

Parent-child dyads who are mutually attuned to each other during social interactions display interpersonal synchrony that can be observed behaviorally and through the temporal coordination of brain signals called interbrain synchrony. Parenting stress undermines the quality of parent-child interactions. However, no study has examined synchrony in relation to parenting stress during everyday shared play. The present fNIRS study examined the association between parenting stress and interbrain synchrony in the prefrontal cortex (PFC) of 31 mother-child and 29 father-child dyads while they engaged in shared play for 10 min. Shared play was micro-analytically coded into joint and non-joint segments. Interbrain synchrony was computed using cross-correlations over 15-, 20-, 25-, 30- and 35-s fixed-length windows. Findings showed that stressed dyads exhibited less synchrony in the posterior right cluster of the PFC during joint segments of play, and, contrary to expectations, stressed dyads also showed greater synchrony in the frontal left cluster. These findings suggest that dyads with more parenting stress experienced less similarities in brain areas involved in emotional processing and regulation, whilst simultaneously requiring greater neural entrainment in brain areas that support task management and social-behavioral organization in order to sustain prolonged periods of joint interactions.

Dataset of parent-child hyperscanning functional near-infrared spectroscopy recordings.

The term "hyperscanning" refers to the simultaneous recording of multiple individuals' brain activity. As a methodology, hyperscanning allows the investigation of brain-to-brain synchrony. Despite being a promising technique, there is a limited number of publicly available functional Near-infrared Spectroscopy (fNIRS) hyperscanning recordings. In this paper, we report a dataset of fNIRS recordings from the prefrontal cortical (PFC) activity of 33 mother-child dyads and 29 father-child dyads. Data was recorded while the parent-child dyads participated in an experiment with two sessions: a passive video attention task and a free play session. Dyadic metadata, parental psychological traits, behavioural annotations of the play sessions and information about the video stimuli complementing the dataset of fNIRS signals are described. The dataset presented here can be used to design, implement, and test novel fNIRS analysis techniques, new hyperscanning analysis tools, as well as investigate the PFC activity in participants of different ages when they engage in passive viewing tasks and active interactive tasks.

Self-perceived loneliness and depression during the Covid-19 pandemic: a two-wave replication study.

The global Covid-19 pandemic has forced countries to impose strict lockdown restrictions and mandatory stay-at-home orders with varying impacts on individual's health. Combining a data-driven machine learning paradigm and a statistical approach, our previous paper documented a U-shaped pattern in levels of self-perceived loneliness in both the UK and Greek populations during the first lockdown (17 April to 17 July 2020). The current paper aimed to test the robustness of these results by focusing on data from the first and second lockdown waves in the UK. We tested a) the impact of the chosen model on the identification of the most time-sensitive variable in the period spent in lockdown. Two new machine learning models - namely, support vector regressor (SVR) and multiple linear regressor (MLR) were adopted to identify the most time-sensitive variable in the UK dataset from Wave 1 (n = 435). In the second part of the study, we tested b) whether the pattern of self-perceived loneliness found in the first UK national lockdown was generalisable to the second wave of the UK lockdown (17 October 2020 to 31 January 2021). To do so, data from Wave 2 of the UK lockdown (n = 263) was used to conduct a graphical inspection of the week-by-week distribution of self-perceived loneliness scores. In both SVR and MLR models, depressive symptoms resulted to be the most time-sensitive variable during the lockdown period. Statistical analysis of depressive symptoms by week of lockdown resulted in a U-shaped pattern between weeks 3 and 7 of Wave 1 of the UK national lockdown. Furthermore, although the sample size by week in Wave 2 was too small to have a meaningful statistical insight, a graphical U-shaped distribution between weeks 3 and 9 of lockdown was observed. Consistent with past studies, these preliminary results suggest that self-perceived loneliness and depressive symptoms may be two of the most relevant symptoms to address when imposing lockdown restrictions.

A method to soothe and promote sleep in crying infants utilizing the transport response.

Approximately 20%-30% of infants cry excessively and exhibit sleep difficulties for no apparent reason, causing parental stress and even triggering impulsive child maltreatment in a small number of cases.1-8 While several sleep training methods or parental education programs may provide long-term improvement of infant cry and sleep problems, there is yet to be a conclusive recommendation for on-site behavioral interventions.9-13 Previously we have reported that brief carrying of infants transiently reduces infant cry via the transport response, a coordinated set of vagal activation and behavioral calming conserved in altricial mammals.14-18 In this study, we disentangled complex infant responses to maternal holding and transport by combining subsecond-scale, event-locked physiological analyses with dynamic mother-infant interactions. Infant cry was attenuated either by maternal carrying or by reciprocal motion provided by a moving cot, but not by maternal holding. Five-minute carrying promoted sleep for crying infants even in the daytime when these infants were usually awake, but not for non-crying infants. Maternal laydown of sleeping infants into a cot exerted bimodal effects, either interrupting or deepening the infants' sleep. During laydown, sleeping infants were alerted most consistently by the initiation of maternal detachment, then calmed after the completion of maternal detachment in a successful laydown. Finally, the sleep outcome after laydown was associated with the sleep duration before the laydown onset. These data propose a "5-min carrying, 5- to 8- min sitting" scheme for attending to infant cry and sleep difficulties, which should be further substantiated in future studies. VIDEO ABSTRACT.

Gene-Environment Interactions in Face Categorization: Oxytocin Receptor Genotype x Childcare Experience Shortens Reaction Time.

Human faces capture attention, provide information about group belonging, and elicit automatic prepared responses. Early experiences with other-race faces play a critical role in acquiring face expertise, but the exact mechanism through which early experience exerts its influence is still to be elucidated. Genetic factors and a multi-ethnic context are likely involved, but their specific influences have not been explored. This study investigated how oxytocin receptor gene (OXTR) genotypes and childcare experience interacted to regulate face categorization in adults. Information about single nucleotide polymorphisms of OXTR (rs53576) and experiences with own- and other-race child caregivers was collected from 89 Singaporean adults, who completed a visual categorization task with own- versus other-race faces. Participants were grouped into A/A homozygotes and G carriers and assigned a score to account for their type of child caregiver experience. A multivariate linear regression model was used to estimate the effect of genetic group, child caregiver experience, and their interaction on categorization reaction time. A significant interaction of genetic group and child caregiver experience (t = 2.48, p = 0.015), as well as main effects of both genetic group (t = -2.17, p = 0.033) and child caregiver experience (t = -4.29, p < 0.001) emerged. Post-hoc analysis revealed that the correlation between categorization reaction time and child caregiver experience was significantly different between the two genetic groups. A significant gene x environment interaction on face categorization appears to represent an indirect pathway through which genes and experiences interact to shape mature social sensitivity to faces in human adults.

Deep Neural Networks and Transfer Learning on a Multivariate Physiological Signal Dataset.

While Deep Neural Networks (DNNs) and Transfer Learning (TL) have greatly contributed to several medical and clinical disciplines, the application to multivariate physiological datasets is still limited. Current examples mainly focus on one physiological signal and can only utilise applications that are customised for that specific measure, thus it limits the possibility of transferring the trained DNN to other domains. In this study, we composed a dataset (n=813) of six different types of physiological signals (Electrocardiogram, Electrodermal activity, Electromyogram, Photoplethysmogram, Respiration and Acceleration). Signals were collected from 232 subjects using four different acquisition devices. We used a DNN to classify the type of physiological signal and to demonstrate how the TL approach allows the exploitation of the efficiency of DNNs in other domains. After the DNN was trained to optimally classify the type of signal, the features that were automatically extracted by the DNN were used to classify the type of device used for the acquisition using a Support Vector Machine. The dataset, the code and the trained parameters of the DNN are made publicly available to encourage the adoption of DNN and TL in applications with multivariate physiological signals.

Father-child dyads exhibit unique inter-subject synchronization during co-viewing of animation video stimuli.

Inter-subject synchronization reflects the entrainment of two individuals to each other's brain signals. In parent-child dyads, synchronization indicates an attunement to each other's emotional states. Despite the ubiquity with which parents and their children watch screen media together, no study has investigated synchronization in father-child dyads during co-viewing. The present study examined whether father-child dyads would exhibit inter-subject synchronization that is unique to the dyad and hence would not be observed in control dyads (i.e., randomly paired signals). Hyperscanning fNIRS was used to record the prefrontal cortex (PFC) signals of 29 fathers and their preschool-aged children as they co-viewed children's shows. Three 1-min videos from "Brave", "Peppa Pig" and "The Incredibles" were presented to each dyad and children's ratings of video positivity and familiarity were obtained. Four PFC clusters were analyzed: medial left, medial right, frontal left and frontal right clusters. Results demonstrated that true father-child dyads showed significantly greater synchronization than control dyads in the medial left cluster during the emotionally arousing conflict scene. Dyads with older fathers displayed less synchrony and older fathers, compared to younger ones, exhibited greater activity. These findings suggest unique inter-subject synchronization in father-child dyads during co-viewing which is potentially modulated by parental age.

Normatively Irrelevant Affective Cues Affect Risk-Taking under Uncertainty: Insights from the Iowa Gambling Task (IGT), Skin Conductance Response, and Heart Rate Variability.

Being able to distinguish between safe and risky options is paramount in making functional choices. However, deliberate manipulation of decision-makers emotions can lead to risky behaviors. This study aims at understanding how affective reactions driven by normatively irrelevant affective cues can interfere with risk-taking. Good and Bad decks of the Iowa Gambling Task have been manipulated to make them unpleasant through a negative auditory manipulation. Anticipatory skin conductance response (SCR) and heart rate variability (HRV) have been investigated in line with the somatic marker hypothesis. Results showed fewer selections from Good decks when they were negatively manipulated (i.e., Incongruent condition). No effect of the manipulation was detected when Bad decks were negatively manipulated (i.e., Congruent condition). Higher anticipatory SCR was associated with Bad decks in Congruent condition. Slower heart rate was found before selections from Good decks in Control and Congruent condition and from Bad decks in Incongruent condition. Differences in heart rate between Bad and Good decks were also detected in Congruent condition. Results shed light on how normatively irrelevant affective cues can interfere with risk-taking.

Improving the Efficacy of Deep-Learning Models for Heart Beat Detection on Heterogeneous Datasets.

Deep learning (DL) has greatly contributed to bioelectric signal processing, in particular to extract physiological markers. However, the efficacy and applicability of the results proposed in the literature is often constrained to the population represented by the data used to train the models. In this study, we investigate the issues related to applying a DL model on heterogeneous datasets. In particular, by focusing on heart beat detection from electrocardiogram signals (ECG), we show that the performance of a model trained on data from healthy subjects decreases when applied to patients with cardiac conditions and to signals collected with different devices. We then evaluate the use of transfer learning (TL) to adapt the model to the different datasets. In particular, we show that the classification performance is improved, even with datasets with a small sample size. These results suggest that a greater effort should be made towards the generalizability of DL models applied on bioelectric signals, in particular, by retrieving more representative datasets.