This study aimed to quantitatively investigate the role of marginal chordae of the mitral valve during static closure by analyzing the force within these chordae and its potential correlation with the transvalvular pressure. Porcine (n = 10) and human (n = 4) mitral valves were mounted in a saddle-shaped annulus clamp, while an in vitro system was utilized to replicate a physiological systolic blood pressure by vacuum pressure applied to the atrial side of the valve. A Millar pressure catheter and force transducers (“C-gauges”) were used to acquire the data. Strut chordal force was recorded simultaneously with the marginal chordal force, to serve as a verification of the employed equipment. The marginal chordal force in porcine and human averaged to 0.012 N ± 0.008 N and 0.006 N ± 0.008 N respectively at peak systolic pressure of 120 mmHg. Additionally, a simplified finite element study was conducted and supports low marginal forces relative to intermediary chords. This low force calls into question the traditionally-described role of the marginal chordae in preventing mitral prolapse at closure. The chords do, however, still show signs of assisting with the leaflets’ positioning. The relationship between marginal chordal forces and transvalvular pressure varies significantly across experiments, further obfuscating their role in normal valvular function.
Univentricular heart defects remain one of the greatest challenges in pediatric cardiovascular surgery. Although well-established, the traditional approach consisting of three serial palliative procedures (Norwood, Glenn, and Fontan) entails cumulative risks, repeated hospitalizations, high healthcare costs, and significant psychosocial impact. This study proposes a single-stage neonatal surgical strategy integrating a tissue-engineered vascular graft and a differential flow control valve to enable adaptive and gradual modulation of cavopulmonary flow, avoiding abrupt physiological transitions. To evaluate the hemodynamic feasibility of the proposed concept, a three-dimensional computational fluid dynamics model was developed and analyzed under controlled flow conditions. A three-dimensional model of the total cavopulmonary connection was developed to evaluate the hemodynamic feasibility of the surgical proposal. The fenestration diameter was reduced from 14 to 0 mm, representing progressive hemodynamic conditions. Computational fluid dynamics simulations were conducted under standardized physiological conditions, with neonatal pulmonary vascular resistance represented by porous zones. Flow distributions, pressure gradients, and hemodynamic equilibrium points were evaluated. Progressive reduction of fenestration diameter from 14 to 0 mm promoted gradual redistribution of venous flow from predominant right atrial diversion to predominant pulmonary perfusion. Systemic venous pressure increased from 4.2 to 10.6 mmHg as fenestration diameter decreased. An intermediate functional range between 5 and 6 mm provided a balanced distribution between pulmonary perfusion and right atrial diversion, consistent with an intermediate cavopulmonary flow condition. Within the assumptions of this proof-of-concept computational model, the proposed strategy demonstrated hemodynamic feasibility for progressive cavopulmonary flow modulation. By integrating tissue engineering, adaptive flow control, and computational modeling, the proposed concept provides a foundation for future patient-specific, experimental, and translational investigations in the treatment of univentricular heart defects.
Aortic stenosis (AS) develops from calcific aortic valve disease (CAVD), which narrows the aortic valve opening as leaflet stiffness increases due to calcium deposition. This study extends the Reverse Calcification Technique (RCT) by incorporating a time dimension to develop a quantitative parametric model for patient-specific prediction of CAVD progression from sequential CT scans. Seventeen pre-transcatheter aortic valve replacement (TAVR) patients underwent sequential CT scans (1.2–6.5 years); baseline aortic valve calcification (AVC) volumes: ≈ 250 to ≈ 1,600 mm3 (cohort mean ≈ 730 mm3 at the first scan and ≈ 920 mm3 at the follow-up scan). A parametric model was developed using two approaches: forward prediction (mild to severe stages) and backward reconstruction (from severe to moderate stages). 34 test cases were assessed through alternating calibration and verification, with performance evaluated using Bland–Altman analysis, paired t-tests, and relative error calculations. For scan intervals under 3 years, forward prediction achieved a mean absolute error of 77 mm3 (7.0
Cardiovascular health depends critically on the integrity of the vascular extracellular matrix (ECM) and the behavior of vascular smooth muscle cells (VSMCs), both of which can be adversely affected in vascular diseases. This study quantitatively evaluates the therapeutic potential of conditioned medium (CM) derived from bone marrow (BM-MSCs) and adipose-derived stem cells (ADSCs) in an elastase-injured human aortic smooth muscle cell (HASMC) model. We systematically varied seeding densities (2000, 5000, and 10,000 cells/cm2) and serum conditions to optimize the SC-SMC secretome for vascular repair. Our results indicate that neither BM-SMC nor AD-SMC CM significantly enhanced lysyl oxidase (LOX) activity. In fact, serum-supplemented BM-SMC CM significantly suppressed LOX activity at seeding densities of 2000 cells/cm2 (p = 0.0378) and 10,000 cells/cm2 (p = 0.0080) compared to injured untreated controls. High-density AD-SMC CM (10,000 cells/cm2) also resulted in a significant decrease in elastin levels (p < 0.05). In addition, serum presence was critical for maintaining the reparative phenotype. Serum-free (SF) conditions for both cell sources led to widespread, statistically significant reductions (p < 0.0001) in key repair and inflammatory biomarkers, including PDGF-AA, Leptin, Lipocalin-2, Osteopontin, RBP4, MMP-1, and IL-6. IL-11 emerged as a primary discriminatory biomarker, showing significant differences between BM and AD treatments at high seeding densities, with both sources causing a significant decrease (p < 0.0001) compared to injured untreated controls. These findings demonstrate that seeding density and serum conditions are critical variables that quantitatively modulate the efficacy of SC-SMC-CM. The study highlights that BM-SMC-derived CM offers a more stable platform for elastin maintenance under serum-free conditions, providing a foundation for developing tailored, cell-free regenerative therapies for cardiovascular disease.
The Contour Neurovascular System (CNS) is a novel intrasaccular flow disruptor for the endovascular treatment of wide-necked aneurysms. While more longitudinal studies are needed to fully assess its efficacy, initial study results over the past few years have been promising. However, no studies have yet assessed the CNS’s robustness to deformation over time, and few have assessed its effect on the morphology of intracranial aneurysms. Thirteen IAs were retrospectively analyzed using co-registered 3D meshes from serial imaging (2–5 scans per patient, up to 26 months of follow-up). Morphological parameters of the ostium and CNS were compared across defined time points. Intra-patient variance was also assessed through scans taken on the same day, and morphology was correlated with treatment outcome. Significant postprocedural changes were observed in the ostium’s ellipticity index (EI; Δ = −0.07 ±0.06, p = 0.01, d = −1.18) and neck circularity index (NCI; Δ = +0.04 ±0.04, p = 0.04, d = 0.90). No statistically significant long-term changes were detected for either the ostium or CNS. Most variations were within intra-patient variance, and many measured deformations were below imaging resolution (0.16−0.47 mm). No correlation was found between recurrence and morphology. Only immediate postprocedural shape changes of the ostium were significant. No measurable long-term deformation of the CNS or ostium was observed. The CNS appears morphologically stable, though larger, high-resolution studies are needed to confirm these findings.
The minimal invasive extracorporeal circulation (MiECC) was developed as abiocompatible alternative to conventional cardiopulmonary bypass (cCPB), intending to mitigate haemodilution, lessen the systemic inflammatory response, and enhance organ protection. AQ2 Although the evidence base has grown over the past 30 years, there has been no systematic mappingof the evolution of research activity and its results. The objectives are to perform a detailed analysis of the temporal, geographical, methodological, and clinical trends within MiECC research published between 1990 and 2025. Studies involving original human clinical data (randomized controlled trials, cohort studies, observational reports, methodological or protocol papers) that reported perioperative or clinical outcomes of MiECC were included. Exclusion criteria comprised reviews, meta-analyses, editorials, letters, guidelines, animal studies, and conference abstracts lacking full text. A comprehensive search of PubMed/MEDLINE, Web of Science, Scopus, and Cochrane databases was conducted up to September 2025 to gather evidence. Reference lists of included articles and relevant reviews were also screened. Data were extracted on publication year, country, sample size, study design, surgical procedure, and reported outcomes. Two reviewers independently charted the data, and discrepancies were resolved by a third reviewer. A total of 151 studies were identified, of which 128 met eligibility criteria. Publication activity increased substantially after 2002 and peaked in 2021, with major contributions from European centres (Germany, Italy, Switzerland, Greece, and the Netherlands). Methodological/protocol papers and Randomize Controlled Trials (RCT) predominated (respectively, n = 37, n = 36). Most studies focused on coronary artery bypass grafting (CABG), while valve surgery and paediatric populations were underrepresented. Reported outcomes shifted over time: early studies emphasised transfusion and inflammatory markers, whereas more recent investigations increasingly examined renal function, neurological complications, and survival. Subgroup analyses suggested notable benefits in elderly patients, those with renal dysfunction, and individuals with low ejection fraction (EF). MiECC research has progressed from feasibility series to large multicentre trials, generating evidence for reduced transfusion requirements, attenuated inflammatory response, and better organ protection, though no clear mortality benefit has been demonstrated.
Vascular diseases, particularly atherosclerosis, represent a leading cause of global morbidity and mortality. Endovascular stenting has emerged as a cornerstone of therapy to restore vessel patency, yet conventional stents remain obstructed by significant clinical limitations, including in-stent restenosis, thrombosis, and mechanical failure. These adverse outcomes are intrinsically linked to their fundamental structural design, which is characterized by a positive Poisson’s ratio, leading to foreshortening and a biomechanical mismatch with the native vasculature. This review critically examines auxetic stents as a next-generation solution, engineered with a structure possessing a negative Poisson’s ratio. This unique property allows them to expand axially upon radial deployment, thereby eliminating foreshortening, enhancing conformability to tortuous vessels, and distributing mechanical stress more uniformly onto the arterial wall. This paper synthesizes the robust body of in-silico/bench-top evidence from computational modeling and in-vitro experimentation that validates these superior biomechanical characteristics. Furthermore, it explores the profound and favorable biological implications, arguing that the optimized mechanical environment and improved hemodynamics are hypothesized to attenuate the primary triggers for neointimal hyperplasia and foster rapid, complete endothelialization. The review concludes by outlining the translational pathway, including challenges in structure integration and discussing the vast future horizons for auxetic structured stents in complex peripheral, carotid, and non-vascular applications. Auxetic design represents a paradigm shift from material-centric iteration to structure-driven innovation, holding the promise to significantly improve the long-term safety and efficacy of endovascular stent implants.
Branch pulmonary artery (PA) stenosis is a significant congenital heart defect causing elevated pulmonary blood pressure, trans-stenotic pressure drop, and abnormal differential blood flow to the lungs. We propose combined pressure-flow diagnostic parameters, pressure drop coefficient (CDP) and normalized energy loss (Ēloss) for improved delineation of stenosis severity. In extension to our previous benchtop experimental study, we performed in vitro experiments using phase contrast magnetic resonance imaging (PC MRI) to evaluate the diagnostic parameters and stenosis severity non-invasively. Subject-specific branch PA test sections representing the main, left, and right PAs (MPA, LPA, and RPA) with a discrete LPA stenosis were manufactured from medical images using additive manufacturing. Three clinically-relevant stenosis severities, 70 E_loss, LPA were evaluated for the three stenosis severities using MRI measurements and compared against results from the benchtop study. The CDPLPA increased with an increase in LPA stenosis severity [70 E_loss, LPA (absolute) generally increased with an increase in LPA stenosis severity, except for 90 E_loss, LPA in accurately delineating unilateral PA stenosis severity. Further, PC MRI is a comprehensive and safe clinical modality for the non-invasive diagnosis of branch PA stenosis.
Freeze-preservation of porcine aortic tissue is often necessary for experimental planning and standardization, but the impact of freezing on aortic root properties has not been systematically investigated. This study evaluated the effects of freezing and thawing on the biomechanical, fluid dynamic, and ultrasonographic characteristics of porcine aortic roots. A total of 32 porcine hearts were obtained from a local abattoir. Specimens were randomized to a fresh group (immediate testing) or a thawed group (storage at − 20 °C for 30 days followed by thawing in saline). Uniaxial tensile testing was performed on samples from the sinotubular junction (STJ) and non-coronary cusp (NCC) (n = 10 per group). Functional testing was conducted in a pulsatile flow loop (n = 6 per group), with mean aortic pressure gradient, forward and retrograde flow measured and the regurgitant fraction calculated. Ultrasonography was used to assess annular diameters and coaptation lengths. Tensile testing revealed no statistically significant differences in E-module, maximum stress, maximum load, or stiffness between fresh and thawed specimens at either the STJ or NCC. Mean aortic gradient, retrograde flow and regurgitant fraction did not differ. Ultrasonography showed no statistically significant changes in annular geometry or leaflet coaptation length after freezing and thawing. Freezing and thawing did not result in measurable differences in the biomechanical, fluid dynamic, or geometrical properties of porcine aortic roots under the tested protocol and within the sensitivity of our measurements. These findings support the use of thawed porcine aortic roots in experimental cardiovascular research.
Breast cancer (BC) remains a major health threat, highlighting the need for accessible screening. Characteristics of the cardiovascular system can be altered in BC patients. This study explored a noninvasive method combining arterial pulse-waveform analysis and classification to compare arterial pulse-waveform characteristics between BC patients (n = 51)) and age-matched controls (n = 76)). From one-minute radial waveform recordings, 40 harmonic indices (amplitude proportions, phase angles, and their variability) were derived. Significant differences in pulse-waveform indices were observed between groups, suggesting altered pulse-wave transmission conditions in BC patients. Two classifiers were compared: machine learning and a novel pulse-distribution analysis (PDA). PDA demonstrated an acceptable discrimination with an AUC of 0.74, outperforming machine learning (AUC up to 0.58). Subgroup analysis by clinical factors such as TNM stage showed variable performance, with AUCs ranging from 0.66-0.93. Discrimination appeared to improve with disease progression (AUC 0.86 for stage IV), suggesting that tumor-induced vascular alterations may affect pulse-wave transmission. The primary contribution of this work is the identification of distinct pulse-waveform signatures in BC patients, providing a physiological basis for future development of noninvasive screening tools. The proposed method, characterized by its low cost and ease of use, warrants further validation in larger cohorts.
Restoration of physiological aortic hemodynamics is a crucial therapeutic target after transcatheter aortic valve implantation (TAVI). However, quantifying this restoration is challenging, particularly using conventional markers. This study presents a novel, unsupervised approach based on proper orthogonal decomposition (POD), suitable for holistic characterization of patient-specific aortic flow field and their association with procedural outcome. Patients (n = 8) with severe aortic valve stenosis (AS) undergoing TAVI were studied. Three metrics were evaluated: patient-specific composite cardiovascular stress signature, pre-procedural free hemoglobin levels and degree of flow restoration. Flow restoration was quantified using POD-based similarity measures (Euclidean distances and Dynamic Time Warping), and compared with conventional hemodynamic markers, i.e. shear stresses, helicity and turbulence production. Unsupervised clustering was applied to explore agreement with biochemical patterns and clinical outcomes. POD identified coherent flow structures and enabled quantification of restoration relative to healthy references. POD-based markers provided more clinically interpretable groupings than conventional measures in this cohort. Patients with low restoration exhibited more pathological composite cardiovascular stress signatures and adverse outcomes during short-term follow-up. POD-based flow analysis provides a holistic and interpretable quantification of post-TAVI aortic hemodynamics. In the given cohort, the method showed agreement with biochemical patterns and clinical outcomes, highlighting its potential to complement conventional hemodynamic and procedural assessment.
Purpose Transposition of the Great Arteries (TGA) is a congenital heart defect characterized by the abnormal positioning of the pulmonary artery over the left ventricle and aorta over the right ventricle. Systemic oxygen delivery, therefore, is dependent on the ability for blood to flow (shunt) between the "parallel" (systemic and pulmonary) circulations. Prior to the standard procedure for treatment of TGA (the arterial switch operation (ASO)), shunting is commonly inadequate, resulting in systemic hypoxemia and hypoxia. In this case the standard intervention is the balloon atrial septostomy, but it may not be sufficient and has a certain complication rate. Other adjustable anatomical and physiological factors may influence shunting but have not been adequately studied. Methods In this investigation, we develop a lumped parameter model (LPM) of TGA to explore the influence of several adjustable factors that can be altered by catheter or pharmacological interventions aimed at improving systemic arterial oxygen delivery to clinically acceptable levels in the preoperative period for newborns with TGA. This model is used to ascertain the sensitivity of oxygen saturation to these parameters and to propose optimal treatment scenarios using formal optimization techniques. Sensitivity in this study is determined using finite difference methods, and a Nelder Mead (NM) algorithm is used for optimization. Results Out of 729 tested cases in the sensitivity study, percentage systemic arterial oxygen saturation (Ssa) was most sensitive to systemic vascular resistance (SVR) in 32.77% of cases and was the most sensitive to PDA diameter in 32.46% of cases. Increasing patent ductus arteriosus (PDA) diameter and SVR both yield an increase in Ssa in nearly all cases, while decreasing pulmonary vascular resistance (PVR) yields an increase in Ssa in nearly all cases. Notably, atrial septal defect (ASD) enlargement was not always most sensitive parameter for increasing Ssa. Best results were obtained with an objective function with a weight on pulmonary-to-systemic flow ratio of 0.1 compared to the weight of unity on Ssa. Lastly, while it does not significantly impact the results of optimization, compliance ratio does have the potential to influence optimal pre-operative strategy. Conclusions This study illuminates the influence of SVR and PVR in oxygen delivery to the systemic circulation through extensive sensitivity testing. These results hold promise for applying other interventions aside from atrial septostomy for improving systemic arterial oxygen saturation. Using our methodology in conjunction with machine learning (ML) using clinical data, an optimization-based tool will be developed to guide pre-operative decision-making in TGA management.
Chronic kidney disease (CKD) affects more than 10
The intracranial aneurysm (IA) model reconstruction is critical for pathophysiology diagnosis and computational simulations. This study aimed to quantify the impact of segmentation thresholds and software platforms on the reconstruction of IA geometry as well as the impact of inter-user variability on the assessment of morphology. 600 IA models were reconstructed from 100 patient DSA datasets using Materialise Mimics and 3D Slicer at three grey value (GV) thresholds; 1000, 1500, and 2500. Geometric measurements were performed in 3-matic by three users. Measurements included vessel diameters and aneurysm morphology parameters. Mimics, the 2500 GV threshold, and the most experienced user (R1) served as baselines for comparison. Normality was evaluated using Shapiro-Wilk tests, and statistical differences were assessed with paired t-tests and relative percent differences. All anatomical regions showed statistically significant geometric variation across software and threshold. Model evaluation showed potential statistically significant variation between users. Models from 3D Slicer were consistently smaller than those from Mimics with percent differences ranging from − 1.27 to − 4.38
BackgroundTricuspid valve replacement is the preferred method when valve repair is not feasible. Stented xenobioprostheses, which limit the growth of the fibrous ring due to their rigid titanium frame and often have a short service life, are particularly problematic in children. An alternative approach is the implantation of a mitral valve homograft in the tricuspid position, which is more resistant to calcification. However, the limited availability of small-diameter homografts from adult donors restricts their use. To address this, a new technique for downsizing an adult mitral homograft was developed and tested in a wet lab and in silico (using 3D printing technologies).MethodsTwo techniques for reducing the size of the mitral homograft were developed: one creating a single papillary muscle and the other creating two neo-papillary muscles. The hydrodynamics of the resulting prostheses were assessed using visual inspection and ultrasound evaluation.ResultsIt was demonstrated that reducing the mitral homograft by resecting the commissures, a portion of the anterior leaflet, and the posterior leaflet along with their subvalvular structures is a reliable method. This technique does not violate the congruence of the anterior and posterior leaflets and ensures optimal hydrodynamic parameters.ConclusionsThe proposed technique makes it possible to downsize an adult mitral homograft to a "pediatric" size. It can be used clinically as an intraoperative adaptation technique, performed on the "back table" in 30-40 minutes.
Establishing credibility of computational modeling of medical devices is critical when an incorrect decision could cause patient harm. Uncertainty quantification (UQ) can impart credibility by providing an estimate of the model uncertainty due to variability in the input parameters. To perform UQ, non-deterministic simulations are performed where model inputs are represented by probability density functions (PDFs). Computational modeling of medical devices, however, is typically constrained by limited sample sizes and sparse experimental data. As a result, it is generally not possible to definitively characterize the true input distributions for UQ. The sparse data must instead be fit with an assumed PDF. While the assumption of a Gaussian distribution is common, other PDFs may be more appropriate depending on the context. In this study, we investigate the influence of input PDF choice on output uncertainty from non-deterministic finite element simulations of a nitinol medical device. We first characterize the geometry, material properties, and the experimental test conditions. We then perform a screening study to determine the relative importance of each parameter on our primary quantity of interest (QOI), peak strain amplitude. Next we perform three UQ studies, each using one of three different PDF types: Gaussian, gamma, and uniform. By sampling the input parameter PDFs using a Latin hypercube approach, we perform non-deterministic simulations to predict the output distributions. Our results show that use of uniform distributions yields output predictions with the largest variance and is thus the most conservative choice unless infrequent events are of interest. In contrast, we show that for conservative predictions of extreme events, a better choice is to use Gaussian or gamma distributions with asymptotic tails of finite probability that are neglected when using uniform distributions.
Patient-specific computational models exhibit strong concordance with invasively measured fractional flow reserve (FFR)-the clinical gold standard for diagnosing coronary ischemia. However, current modeling techniques frequently rely on computationally intensive assumptions such as pulsatile flow dynamics and often fail to optimally leverage patient-specific clinical data that is routinely available, limiting their practical clinical adoption. In this study, we propose a hybrid coronary angiography-based approach that reduces computational complexity through simplified steady-state flow assumptions, while simultaneously better leveraging available clinical information. Specifically, we integrate physics-based modeling with a machine learning (ML) feedback loop designed to refine and improve FFR predictions. We evaluated this hybrid framework using a retrospective two-center cohort comprising 132 patients with 132 coronary lesions. Our results demonstrate that steady-state models effectively capture essential hemodynamic patterns, closely matching pulsatile model predictions. The ML refinement step enhances diagnostic accuracy, yielding a sensitivity of 83.3%, specificity of 100.0%, positive predictive value of 100.0%, negative predictive value of 88.2%, and overall precision of 92.6%. By effectively combining efficient computational modeling with targeted ML-driven refinements, our approach represents a robust, clinically viable solution for accurate patient-specific FFR estimation.
Photoplethysmography (PPG) is a well-known technique employed to assess optically perfused bio-tissue volume changes. A PPG apparatus consists of a light emitter and a receptor. The analysis of the received light is used to infer properties of the illuminated tissues. This paper aims at presenting a novel distributed mathematical model for PPG signals, which combines a poroelastic model of tissue perfusion with a diffusion model for light absorption and scattering. We assume that the tissues undergo small deformations, allowing for a linear poroelastic description of perfusion. Since many PPG devices are applied to the fingertips (due to the rich vascularization in that area) we model the system specifically on the finger, with arterial blood pulse pressure serving as the primary perfusion driver. The numerical discretization of the governing equations is carried out using the finite element method. After calibration of the model, 216,000 simulations are performed with varying parameters (quasi-Monte Carlo approach). The aggregated results for two key biomarkers, AC/DC PPG amplitude ratio and pulse pressure, are compared against experimental PPG and pulse pressure measurements obtained from 20 volunteers. Within the prescribed parameter ranges, numerical simulations successfully reproduce the AC/DC PPG amplitude biomarker in both the red and infrared wavelengths, with a few outliers observed for the green. Furthermore, in the vicinity of the combined red and infrared measured biomarkers, there are simulated biomarkers whose corresponding pulse pressures closely match the mapped-to-finger measured pulse pressure, with a difference of less than 1 . This work demonstrates the relevance of the proposed mathematical model for simulating PPG signals and highlights its potential for estimating tissue perfusion parameters, such as arterial pulse pressure.
Electrocardiography (ECG) plays a vital role in the diagnosis of cardiovascular diseases by analyzing the electrical activity of the heart. ECG semantic segmentation is a subfield focused on sample-wise delineation of ECG waveforms by assigning a physiological label to each time sample, enabling explicit estimation of clinically meaningful onset and offset boundaries. Recent advancements in deep learning have significantly improved ECG classification accuracy; however, the same has not yet been observed in automatic ECG segmentation. Existing models often lack explainability and adaptability to patient-specific variations, thereby reducing their generalizability. This study proposes a personalized deep neural network approach for enhanced ECG processing. This method incorporates convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) networks, incorporating an attention mechanism to refine segmentation accuracy. A novel loss function is introduced to ensure smoother temporal transitions and better classification accuracy. The model was evaluated using the QT Database, demonstrating substantial improvements in P-wave and QRS delineation and in T-wave offset localization segmentation when fine-tuned for individual patients when fine-tuned and evaluated on held-out data from the same patient, demonstrating the benefit of intra-patient adaptation. Our results indicate that personalization improves delineation accuracy for challenging waveforms (notably P and T waves), supporting the potential of deep learning to better capture patient-specific morphology and providing a stronger basis for waveform-level, clinically interpretable ECG analysis.
