7 October, 2025

Fourteen researchers from the BSICoS group participated in the Computing in Cardiology Congress, held in São Paulo, Brazil, from September 14 to 17, where they had the opportunity to present their research on heart health.

Sofía Pérez: “The Complex Interpretation of Heart Rate Variability Components in Mechanically Ventilated Patients”

Mechanical ventilation (MV) is a critical therapeutic intervention to support patients with respiratory failure. Using a dataset of 54 adult ICU patients under MV, this study analyzes respiratory signals and ECG-derived HRV data to identify the appearance of additional frequency components, particularly at harmonic frequencies, beyond the traditional low-frequency (LF) and high-frequency (HF) bands of HRV. It is observed that, in controlled and support ventilation modes, these additional harmonics appear as a result of ventilator-induced respiratory signal modulations. This phenomenon occurs up to 93.39% of the time in the controlled ventilation mode. Specifically, when the second harmonic frequency exceeds half the heart rate, aliasing occurs, manifesting as components at FHR − 2FR. These findings underscore the importance of considering these additional frequency components for a more comprehensive understanding of HRV and its role in assessing the success of weaning from mechanical ventilation.

Maikel Noriega: “P-Wave Morphological Variability Exacerbation Prior to Atrial Fibrillation Episodes”

P-wave morphological variability (PWMV) has been shown to be a marker of atrial electrical instability and may reflect instabilities that precede atrial fibrillation (AF). This study analyzes the temporal evolution of PWMV in sinus rhythm during the hour prior to AF onset characterizing changes in PWMV, potentially signaling a vulnerable atrial substrate, and identifying significant turning times preceding the AF episodes onset.

Methods: We analyzed 32 ambulatory ECG recordings from subjects with paroxysmal AF (mean[range] duration: 104[21–156] hours). For each i-th subject and j-th AF episode, the one-hour, pre-AF PWMV is characterized in consecutive k-th 2-min windows, as: 1) the median absolute deviation (MAD) of the beat P waves   is computed; 2) the ratio of this median energy to the median of the P wave energy in each subject, is computed; and 3) The subject median across AF episodes, is computed as well and the median across patients, characterizing the cohort PWMV dynamics. A two-segment linear function fit by least squares error (LSE) minimization, is used to quantify the cohort dynamics.

Results: In 56% of patients the MAD energy ratio tracings present noticeable higher values in the last 30 min before AF event. Analyzing the cohort dynamic of the ratio, we observe that it remains relatively stable until 8 minutes before the AF episode, when its value starts increasing, see Figure 1 in pdf abstract.

Conclusion: P-wave morphological variability can be monitored by quantified their short-time MAD relative energy, showing an increasing P wave morphological variability trend from 8 minutes before AF onset.

Neurys Gómez: “T-Wave Peak-to-End Morphology Restitution as Biomarker in Brugada Syndrome”

Brugada syndrome (BrS) is a genetic disorder that affects the ion channels of the cardiac myocites, resulting in a higher predisposition to malignant ventricular arrhythmias and sudden cardiac death (SCD). Risk stratification and management of BrS patients remains a major clinical challenge.

In this work, we use a time-warping-based index of the T-wave peak-to-end morphological restitution (TPEMR) to quantify the restitution properties of the late phase of ventricular repolarization and evaluate its ability to distinguish between control and BrS patients and between symptomatic (BrS-S) and asymptomatic patients (BrS-A). 24-hour Holter ECG recordings from 89 BrS patients (29 asymptomatic, BrS-A, and 60 symptomatic, BrS-S) and 32 healthy control patients were analyzed. The original, unweighted TPEMRO as well as weighted versions with two different strategies TPEMRW1 and TPEMRW2 are computed. TPEMR derives from a time-warping, image.png, index computed from T waves at two distant RR bins normalized by the RR range.

All the three TPEMR indices were significantly higher for BrS group than for the control group. BrS-S group presented higher TPEMR indexes than BrS-A group for all indices, though significant differences were only shown in the weighted TPEMRW2 index.

Ángela Hernández: “Sex- and Menopause-Specific ECG Repolarization Patterns”

Sex and menopause influence ventricular repolarization and arrhythmia risk through hormonal effects that may be detectable on the electrocardiogram (ECG). Women have longer QT intervals and narrower QRS-T angles than men, with ECG-based sex misclassification associated with worse outcomes in females. While female hormones are cardioprotective, their decline after menopause may alter ECG patterns. This study aims to identify sex- and menopause-related ECG markers of ventricular repolarization and explore their underlying factors.

Methods: We analyzed ECG markers of ventricular repolarization (the corrected QT, QTc, and T-peak-end, T pec, intervals and the mean spatial QRS − T angle) in 55,730 healthy individuals from the UK Biobank stratified by sex, and females further subdivided into premenopausal (F-NoMP) and postmenopausal (F-MP) groups. ECG indices were computed per lead and summarized using principal component analysis. Group comparisons used one-tailed Mann–Whitney tests. Significant differences were examined with multiple linear regression to identify key demographic, clinical and cardiac structural and functional contributors.

Results: Comparison between sex-specific groups, males showed lower QTc and higher T pec and QRS − T a values than females, as well as higher R- and T-wave amplitudes in ECG waveforms. Height, body mass index (BMI) and alcohol consumption contributed most in males, whereas menopause, diastolic blood pressure and pulse rate in females. Between F-NoMP and F-MP groups, only T pec and QRS − T a differed with F-MP women showing higher values and ECG waveform amplitudes. These differences were driven by left ventricle (LV) wall thickness and LV ejection fraction in both groups and specifically in F-MP, by age, BMI and LV mass.

Conclusions: Our results suggest that both sex and menopause are reflected in ECG markers of ventricular repolarization, with F-MP showing subtle changes in morphology toward a pattern analogous to men respect to F-NoMP. These findings suggest the need for further research on hormonal influences in ECG characteristics and arrhythmogenic risk stratification.

Clara Sales: “Short- and Long-term Effects of Left Bundle Branch Area Pacing on the T wave of the ECG”

Although left bundle branch area pacing (LBBAP) provides more physiological ventricular activation than conventional pacing techniques, its impact on ventricular repolarization remains unclear. This study investigates the temporal changes of the T wave in patients with bradycardia treated with LBBAP, using vectorcardiographic analysis of 12-lead ECGs. Standard 12-lead ECGs were collected from 23 healthy subjects and 58 patients with bradycardia who underwent LBBAP. ECGs were recorded at baseline for the two cohorts and at three additional time points after LBBAP for the bradycardia group.

The principal direction of the T wave was obtained from vectorcardiographic analysis and classified into four quadrants (qx). In healthy controls, all T wave vectors were located in q1. In patients with bradycardia and narrow QRS, 80% of the T wave vectors lied in q1 at baseline. A reduction in the number of T wave vectors belonging to q1 was observed immediately after LBBAP (45%), followed by a significant increase in this number at 24 hours (75%) and again at one year post-implantation (80%) (p<0.05). Qualitatively similar results were obtained for patients with wide QRS at baseline.

These findings suggest that, although LBBAP transiently alters the direction of ventricular repolarization, the T wave vector progressively realigns to a physiological pattern over time.

 Maxi Rosales: “Cocoro: Fast Simulation of Cardiac Electrophysiology with WebGPU”

As the complexity of in silico cardiac electrophysiological models increases, so do the computational costs. To address this, we present Cocoro, a cutting-edge WebGPU-based solver for efficient computation and rendering.

We describe our implementation and benchmark its accuracy, performance, and clinical potential. Our results were compared with those of openCARP (OC) on a porcine biventricular mesh under anisotropic (A) and isotropic (I) setups. Cocoro’s execution time (ET) for 5-second simulations was measured on two GPUs with and without extra graphical information (EGI), including visualizations of pseudo-electrocardiograms and cellular data. In addition, we compared the simulated and experimental QRS complexes to showcase clinical applicability.

Cocoro simulations closely aligned with those of OC, with the 90^th percentile (P₉₀) of node-wise activation time differences of 11 ms (A) and 4 ms (I). For repolarization times, P₉₀ remained below 6 ms for both A and I. The five-second simulations ran in 9.54 and 22.13 min on RTX 2070 and Titan V GPUs, respectively. Enabling EGI had a minimal impact on ETs. Furthermore, the simulated QRS complexes reproduced the experimental QRS morphologies and durations.

Thus, Cocoro enables fast, portable, and accurate fully GPU-resident cardiac electrophysiological simulations.

Hugo Hernández: “Characterization of atrial spatiotemporal dynamics for prediction of AFrecurrence after ablation”

This study aims to improve atrial substrate characterization in atrial fibrillation (AF) by identifying electrophysiological characteristics associated with AF recurrence (AFR) after pulmonary vein isolation (PVI) to support prediction and guide personalized treatment.

Methods: We analyzed preablation electroanatomical maps and EGM recordings from 45 AF patients undergoing their first PVI at Hospital Cl ́ınico Universitario Lozano Blesa. Six of these patients experienced AF recurrence within 12 months. High-density maps were obtained using the Rhythmia system and the ORION basket catheter. For localized analysis, each atrial mesh was divided into 40 to 100 regions using the K-means clustering algorithm. The medians of seven EGM markers were calculated for each region: peak-to-peak voltage (Vp−p); fractionation index; intra-region EGM similarity; and four Principal Components Analysis-based metrics that describe EGM complexity.

Results: Results are reported for 90 regions, as similar values were found in all cases. Patients with AFR showed lower voltages (0.98 vs 1.8), higher fractionation (9.29 vs 5.45), and greater EGM variability (0.53 vs 0.61), indicating structural remodeling and conduction abnormalities. PCA markers further revealed lower variance explained by the first component (55.71 vs 70.78) and a higher number of components needed to reach 90% of the variance (20.25 vs 12), suggesting increased signal heterogeneity and atrial disorganization.

Conclusions: This study demonstrates that both conventional EGM markers and PCA-derived indices based on local EGM morphology and complexity can effectively distinguish between patients with and without AFR after PVI.

Nayan Wadhwani: “Assessing Signal Quality Impact on Pulse Rate Variability Accuracy from Photoplethysmography”

Heart Rate Variability (HRV) is a valuable marker of autonomic nervous system activity, commonly derived from electrocardiogram (ECG) recordings. Pulse Rate Variability (PRV), extracted from photoplethysmography (PPG), offers a convenient alternative, particularly in wearable health technologies. However, PRV accuracy depends heavily on PPG signal quality, which may be degraded by noise, motion artifacts, or poor sensor contact. This study analyzes the relationship between the quality of the PPG signal, measured with different SQIs, and its relationship with the error of HRV variability (specifically only the RMSSD) using multiple Signal Quality Indices (SQIs).

PPG signals from 225 sessions from 120 subjects  were segmented into 10-second windows, then aggregated into 5-minute segments (n=1046). Four SQI metrics were computed: kurtosis, skewness, entropy, and perfusion index. A novel dynamic quantile-based normalization method standardized SQI values (0-1 scale). Correlation analyses assessed relationships between SQI metrics and RMSSD errors compared to reference ECG measurements.

Analysis of 1,046 five-minute segments revealed significant negative correlations between all SQI metrics and HRV errors (p < 0.001). The Perfusion Index demonstrated the strongest correlation (r = -0.189, R² = 3.6%) and maintained predictive validity in high-quality segments (SQI > 80%, r = -0.151, p < 0.001), while other metrics lost discriminative power. Quality distribution varied considerably: 84.9% of segments exceeded 80% quality for Perfusion Index, compared to 33.7% for Skewness.

Andrea Rucco: “Impact of AF Ablation Strategies on Autonomic Modulation Measured by Heart Rate Variability”

Catheter ablation is a key treatment for atrial fibrillation (AF). Yet, its impact on cardiac autonomic modulation, which contributes to AF initiation and maintenance, may vary depending on the ablation modality. This study compares the effects of cryoablation (CRYO) and pulsed field ablation (PFA) on heart rate variability (HRV) indices, which serve as markers of autonomic function.

Methods: 10-minute, 12-lead ECG recordings of 54 patients (61% PFA, 63% male) were acquired before (pre) and 3 months after (post) ablation. Time-domain (SDNN, RMSSD) and frequency-domain (PLF, PHF, PLF/HF) indices were calculated, and intra- and inter-group differences were assessed.

Results: Before ablation, HRV indices were comparable between groups. After ablation, CRYO patients exhibited a significant (p < 0.05) reduction in time- and frequency-domain indices (median reductions: -10.5 ms SDNN, -4.8 ms RMSSD, -2.3×10−4 PLF, -1.0×10−4 PHF), indicating substantial autonomic modulation (Fig.1). In contrast, the PFA group showed non-significant HRV changes from baseline (-1.62 ms SDNN, -2.80 ms RMSSD, -0.32×10−4 PLF), and a significant but lower reduction in PHF (-0.47×10−4), suggesting preserved autonomic function.

Conclusion: While CRYO appears to attenuate ganglionated plexi activity, potentially reducing post-procedural arrhythmogenic triggers, PFA largely preserves autonomic input, which can promote AF recurrence.

Sofia Romagnoli: “Temporal Variability of Ventricular Activation in Brugada Syndrome Patients”

In Brugada Syndrome (BrS), ventricular fibrillation events peak during night. This temporal pattern suggests that circadianity may modulate arrhythmogenesis. The present work evaluates the circadianity of ventricular activation in BrS patients from 24-hour ECGs with the high precordial leads. Lead-dependent propagation progression times, are computed every h-th half hour in each l-th lead as the time instant along the QRS complex when the QRS energy reaches a percentage, p, of the total QRS energy is characterized by median and standard deviation over 24 hours, as well as the MESOR and amplitude of the cosinor analysis. The median and MESOR of are longer by more than 10 ms in BrS patients than in controls. In particular, the median and standard deviation of at different energy percent can distinguish between BrS vs controls (AUC≥ 0.8), and BrS-I vs BrS-II (AUC= 0.78).

Cosinor analysis indicates that is less influenced by circadianity than heart rate. The Spearman correlation of with heart rate is lower than 0.5 in each lead for both BrS and controls. These findings suggest that intra-day variability in propagation progression times is not primarily driven by circadianity or heart rate.

Furthermore, the temporal variability of is higher in symptomatic BrS patients supporting its potential role as risk stratification marker in BrS.

Inés Noguero: “Regional Characterization of Activation Times in Spontaneous and Drug-Induced Brugada Syndrome Using ECG Imaging”

Brugada syndrome (BrS) is a cardiac channelopathy characterized by a distinctive ECG pattern, either spontaneous or drug-induced, and a high risk of sudden arrhythmic events. Linked to conduction delay in the right ventricular outflow tract (RVOT) epicardium, BrS can be assessed noninvasively using ECG imaging (ECGi). Despite clinical recognition, the pathophysiological differences between spontaneous and induced BrS patterns remain unclear, hindering risk stratification. This study investigates ECGiderived activation times (ATs) in BrS patients with spontaneous and induced patterns, compared to healthy controls.

Methods. ECGi data were acquired from 26 patients and volunteers at Hospital Clínico Universitario Lozano Blesa (Zaragoza, Spain) using the Acorys Mapping System, Corify Care SL. The population included nine healthy controls (40±13 years) and 20 BrS patients, 9 with a spontaneous (BrS1, 48±11 years) and 11 with an induced (BrS2, 58±10 years) type one pattern. Epicardial unipolar electrograms (EGMs) reconstructed from 128-electrode body surface potential maps (BSPMs) and estimated 3D biventricular meshes were processed. The QRS width (QRSw) and total AT (TAT) were calculated from BSPMs and EGMs, respectively. The ventricles were segmented into 15 regions for AT regional analysis.

Results. Higher values of QRSw (BrS1: 137±21 ms; BrS2: 131±18 ms, Healthy: 110±13 ms) and TAT (BrS1: 86±21 ms; BrS2: 88±18 ms; Healthy: 63±13 ms) were observed in BrS patients vs controls (p<0.05). Regional analysis suggested delayed conduction in the RVOT region, with significantly higher values of 90th percentile of AT in BrS groups (BrS1: 101±25 ms; BrS2: 97±16 ms, Healthy: 71±19 ms; BrS vs Healthy, p<0.01). Differences in AT between BrS types were also observed in other left ventricular segments.

Conclusion: ECGi could noninvasively characterize ventricular activation in BrS patients compared to healthy controls and differences between BrS patients with spontaneous and drug-induced type 1 pattern were observed.

Josseline Madrid: “Unsupervised ECG Clustering Reveals Distinct Associations with Cardiac Magnetic Resonance Features”

Exploring the association between the electrocardiogram (ECG) and cardiac magnetic resonance (CMR)-derived features may enhance our understanding of cardiovascular physiology. We aimed to identify clusters of individuals without diagnosed cardiovascular disease (CVD) based on their ECG phenotypes in an unsupervised manner and evaluate their cardiac anatomical differences through CMR. Spatial and single-lead ECG markers were calculated from 10-second 12-lead ECGs from 51,974 UK Biobank individuals without diagnosed CVD. A k-means clustering model grouped individual ECG phenotypes into k clusters.

Statistical analyses were conducted to assess ECG, demographic and CMR differences across clusters. Three distinct ECG-based clusters were identified (N1=19,470, N2=22,256, N3=8,997), with significant differences in ECG morphology and CMR-derived anatomical features. The most discriminative ECG features involved ventricular repolarization in precordial leads (i.e., T- and ST-segment amplitude). Cluster-specific electro-anatomical alignment was stronger in Cluster 3.

Our findings show that ECG phenotyping through unsupervised clustering can reveal anatomical cardiac differences. Future work will evaluate the association with incident risk of each of these clusters.

Eduardo Caballero: “Comparison of Deep Learning Models based on 1D versus 2D-formatted ECG for Long-Term Prediction of Myocardial Infarction”

Risk stratification of myocardial infarction (MI) is crucial for early intervention and prevention. ECG-based models, enabled by artificial intelligence, have shown promise when predicting cardiovascular events. However, these methods are still limited to relatively short follow-up periods and may not be applicable to a general/ healthy population. We hypothesized that the performance of these AI models could improve if the input data was presented in a structured format, enabling the model to extract more relevant features from the same ECG signal.

We designed and compared the performance of two Convolutional Neural Networks, designed for MI prediction across 8 follow-up times, from 1 to 12 years. The first model was designed to take 3 seconds from a 1-lead ECG as input (1D), while the second one takes a matrix representation (2D) of the same 1-lead ECG. We trained both models using 97,382 1-Lead ECGs taken from UK Biobank from subjects without previous history of MI. Our proposed 2D model showed a slightly better performance, sustained across all predictions, with an average AUC increase of 0.047.

These results suggest that the proposed model better captures the difference between subjects with higher risk of MI. With this work, we aimed to improve the risk prediction for MI to provide healthcare professionals more accurate and opportune information for decision-making.

Diego Cajal: “Feasibility of Unsupervised Sleep Apnea Screening Using Pulse Rate Oscillations and SpO2 with a Wrist-Worn Device”

In a previous work, a method was presented to detect sleep apnea using signals commonly found in wearable devices: peripheral oxygen saturation (SpO2) and pulse photoplethysmography (PPG). However, these signals were obtained from a conventional fingertip pulse oximeter. This paper describes a pilot study applying those methods with signals acquired by a wrist-worn wearable device in a nonsupervised home environment (n = 12). A classifier was applied to differentiate normal from abnormal breathing segments. Later, the Cyclic Variation of Heart Rate Index (CVHRI) was calculated within the abnormal breathing segments. The classifier achieved an accuracy of 65.7% on the wearable data and CVHRI maintained a strong  correlation with the AHI (r = 0.85, p < 0.001), suggesting its potential for patient stratification remains viable.