publications

2025

1. Barriers, facilitators and strategies for the implementation of artificial intelligence‐based electrocardiogram interpretation: A mixed‐methods study

   Bauke Arends, Jenna McCormick, Pim Van Der Harst, and 2 more authors

   European Journal of Clinical Investigation, Apr 2025

   Abs DOI

   Introduction: The implementation of artificial intelligence-­based electrocardiogram interpretation (AI-E­ CG) algorithms relies heavily on end-u­ ser acceptance and a well-d­ esigned implementation plan. This study aimed to identify the key barriers, facilitators and strategies for the successful adoption of AI-E­ CG in clinical practice. Methods: A sequential explanatory mixed-­methods study was conducted among future AI-­ECG end-­users in the Netherlands, including doctors, nurses, and ambulance professionals, using a clinical scenario involving chest pain. Quantitative data were collected through a three-­round Delphi survey (n = 25) to identify key barriers and facilitators. Building on these findings, qualitative data were gathered through semi-­structured interviews (n = 7) and focus groups (n = 12) to further explain the barriers and facilitators, and discuss relevant implementation strategies. Results: Participants expressed a general openness to working with AI-E­ CG. Four key barriers and twelve facilitators were identified in the quantitative phase. Participants mentioned the relative advantage of AI-E­ CG in the context of recognizing subtle, or rare, ECG abnormalities and assisting in patient triage. However, successful implementation requires end-­users to have trust in the algorithm, clear protocols, actionable model output, integration with existing clinical systems and multidisciplinary implementation teams. Several strategies were proposed to address these challenges, including conducting local consensus discussions, identifying and preparing local champions and revising professional roles. Conclusions: This mixed-­methods study grounded in established theoretical frameworks identified several barriers and facilitators to AI-E­ CG implementation and proposed strategies to address these challenges. These findings provide

2. Localization of accessory pathways in Wolff‐Parkinson‐white syndrome using ECG‐based multi‐task deep learning

   Jasper Hennecken, Bauke Arends, Thomas Mast, and 9 more authors

   European Journal of Clinical Investigation, Apr 2025

   Abs DOI

   Background: Wolff-Parkinson-White syndrome is characterized by accessory atrioventricular pathways (AP) and atrio-ventricular re-entry arrhythmias. Catheter ablation approach and success are determined by AP location. Existing rule-based algorithms based on the electrocardiogram (ECG) are time consuming, prone to inter-observer variability and use delta wave polarity as a binary variable. To overcome these challenges, we propose a model based on a deep neural network (DNN). Methods: Patients with concealed pathways, multiple antegrade conducting pathways or without any sinus rhythm ECGs were excluded. AP location was determined based on electrophysiological testing during catheter ablation and categorized into right-sided, septal and left-sided APs. Multi-task learning with auxiliary identification of the presence of pre-excitation, parahisian pathways and locations where a transseptal puncture is potentially required was used to increase usability and performance. The DNN was compared to the Milstein and Arruda algorithms. Results: Between 1997 and 2023, 645 patients who underwent catheter ablation for an AP were included in the study. The model was developed using 1.394 ECGs from 567 patients. The DNN was tested using 78 ECGs in two independent cohorts. The model outperformed both the Milstein and Arruda algorithms with an area under the receiver operating characteristic curve (AUROC) of .92 (95% confidence interval: .88–.96) compared to the Arruda algorithm (AUROC of .80; p \textless.001) and the Milstein algorithm (AUROC of .81; p \textless.001). Conclusions: Our model showed excellent discriminatory performance in predicting the location of an accessory pathway while outperforming conventional techniques. Clinically, this tool can improve preoperative planning and risk stratification.

3. Diagnosis extraction from unstructured Dutch echocardiogram reports using span- and document-level characteristic classification

   Bauke Arends, Melle Vessies, Dirk Van Osch, and 4 more authors

   BMC Medical Informatics and Decision Making, Mar 2025

   Abs DOI

   Background Clinical machine learning research and artificial intelligence driven clinical decision support models rely on clinically accurate labels. Manually extracting these labels with the help of clinical specialists is often timeconsuming and expensive. This study tests the feasibility of automatic span- and document-level diagnosis extraction from unstructured Dutch echocardiogram reports. Methods We included 115,692 unstructured echocardiogram reports from the University Medical Center Utrecht, a large university hospital in the Netherlands. A randomly selected subset was manually annotated for the occurrence and severity of eleven commonly described cardiac characteristics. We developed and tested several automatic labelling techniques at both span and document levels, using weighted and macro F1-score, precision, and recall for performance evaluation. We compared the performance of span labelling against document labelling methods, which included both direct document classifiers and indirect document classifiers that rely on span classification results. Results The SpanCategorizer and MedRoBERTa.nl models outperformed all other span and document classifiers, respectively. The weighted F1-score varied between characteristics, ranging from 0.60 to 0.93 in SpanCategorizer and 0.96 to 0.98 in MedRoBERTa.nl. Direct document classification was superior to indirect document classification using span classifiers. SetFit achieved competitive document classification performance using only 10% of the training data. Utilizing a reduced label set yielded near-perfect document classification results. Conclusion We recommend using our published SpanCategorizer and MedRoBERTa.nl models for span- and document-level diagnosis extraction from Dutch echocardiography reports. For settings with limited training data, SetFit may be a promising alternative for document classification. Future research should be aimed at training a RoBERTa based span classifier and applying English based models on translated echocardiogram reports.

4. Deep learning for electrocardiogram interpretation: Bench to bedside

   Bas Schots, Camila Pizarro, Bauke Arends, and 4 more authors

   European Journal of Clinical Investigation, Apr 2025

   Abs DOI

   Background: Recent advancements in deep learning (DL), a subset of artificial intelligence, have shown the potential to automate and improve disease recognition, phenotyping and prediction of disease onset and outcomes by analysing various sources of medical data. The electrocardiogram (ECG) is a valuable tool for diagnosing and monitoring cardiovascular conditions. Methods: The implementation of DL in ECG analysis has been used to detect and predict rhythm abnormalities and conduction abnormalities, ischemic and structural heart diseases, with performance comparable to physicians. However, despite promising development of DL algorithms for automatic ECG analysis, the integration of DL-based ECG analysis and deployment of medical devices incorporating these algorithms into routine clinical practice remains limited. Results: This narrative review highlights the applications of DL in 12-lead ECG analysis. Furthermore, we review randomized controlled trials that assess the clinical effectiveness of these DL tools. Finally, it addresses different key barriers to widespread implementation in clinical practice, including regulatory hurdles, algorithm transparency and data privacy concerns. Conclusions: By outlining both the progress and the obstacles in this field, this review aims to provide insights into how DL could shape the future of ECG analysis and enhance cardiovascular care in daily clinical practice.

5. Dexrazoxane protects against doxorubicin‐induced cardiotoxicity in susceptible human living myocardial slices: A proof‐of‐concept study

   Jort Van Der Geest, Ilse Kelters, Bauke Arends, and 13 more authors

   British Journal of Pharmacology, May 2025

   Abs DOI

   Background and Purpose: The increasing number of cancer survivors has caused growing concern over chemotherapy-induced cardiotoxicity. This study aimed to investigate a novel human model of cardiotoxicity and explore cardioprotection.

6. MyDigiTwin: A Privacy-Preserving Framework for Personalized Cardiovascular Risk Prediction and Scenario Exploration

   Héctor Cadavid, Hyunho Mo, Bauke Arends, and 5 more authors

   Jan 2025

   arXiv:2501.12193 [cs]

   Abs DOI

   Cardiovascular disease (CVD) remains a leading cause of death, and primary prevention through personalized interventions is crucial. This paper introduces MyDigiTwin, a framework that integrates health digital twins with personal health environments to empower patients in exploring personalized health scenarios while ensuring data privacy. MyDigiTwin uses federated learning to train predictive models across distributed datasets without transferring raw data, and a novel data harmonization framework addresses semantic and format inconsistencies in health data. A proof-of-concept demonstrates the feasibility of harmonizing and using cohort data to train privacy-preserving CVD prediction models. This framework offers a scalable solution for proactive, personalized cardiovascular care and sets the stage for future applications in real-world healthcare settings.

2024

1. Diagnosis extraction from unstructured Dutch echocardiogram reports using span- and document-level characteristic classification

   Bauke Arends, Melle Vessies, Dirk van Osch, and 4 more authors

   Aug 2024

   arXiv:2408.06930 [cs]

   Abs DOI

   Clinical machine learning research and AI driven clinical decision support models rely on clinically accurate labels. Manually extracting these labels with the help of clinical specialists is often time-consuming and expensive. This study tests the feasibility of automatic span- and document-level diagnosis extraction from unstructured Dutch echocardiogram reports. We included 115,692 unstructured echocardiogram reports from the UMCU a large university hospital in the Netherlands. A randomly selected subset was manually annotated for the occurrence and severity of eleven commonly described cardiac characteristics. We developed and tested several automatic labelling techniques at both span and document levels, using weighted and macro F1-score, precision, and recall for performance evaluation. We compared the performance of span labelling against document labelling methods, which included both direct document classifiers and indirect document classifiers that rely on span classification results. The SpanCategorizer and MedRoBERTa\.\nl models outperformed all other span and document classifiers, respectively. The weighted F1-score varied between characteristics, ranging from 0.60 to 0.93 in SpanCategorizer and 0.96 to 0.98 in MedRoBERTa\.\nl. Direct document classification was superior to indirect document classification using span classifiers. SetFit achieved competitive document classification performance using only 10% of the training data. Utilizing a reduced label set yielded near-perfect document classification results. We recommend using our published SpanCategorizer and MedRoBERTa\.\nl models for span- and document-level diagnosis extraction from Dutch echocardiography reports. For settings with limited training data, SetFit may be a promising alternative for document classification.

2. Leveraging FHIR in Federated Learning Environments: A Data Harmonization Framework for Cohort Studies

   Héctor Cadavid and Bauke Arends

   In Studies in Health Technology and Informatics, Aug 2024

   Abs DOI

   MyDigiTwin is a scientific initiative for the development of a platform for the early detection and prevention of cardiovascular diseases. This platform, which is supported by prediction models trained in a federated fashion to preserve data privacy, is expected to be hosted by the Dutch Personal Health Environments (PGOs). Consequently, one of the challenges for this federated learning architecture is ensuring consistency between the PGOs data and the reference datasets that will be part of it. This paper introduces a novel data harmonization framework that streamlines an efficient generation of FHIR-based representations of multiple cohort study data. Furthermore, its applicability in the integration of Lifelines’ cohort study data into the MiDigiTwin federated research infrastructure is discussed.

2019

1. Pulmonary vein anatomy addressed by computed tomography and relation to success of second‐generation cryoballoon ablation in paroxysmal atrial fibrillation

   Bart Mulder, Meelad Al‐Jazairi, Bauke Arends, and 10 more authors

   Clinical Cardiology, Apr 2019

   Abs DOI

   Background: Cryoballoon isolation is considered a safe and effective treatment for atrial fibrillation (AF). However, recurrence of AF after first cryoballoon ablation occurs in ~30% of patients. Pre-procedurally identifying patients at risk of AF recurrence could be beneficial. Hypothesis: Our aim was to determine how pulmonary vein (PV) anatomy influences the recurrence of AF using the second-generation cryoballoon in patients with paroxysmal AF. Methods: We included 88 consecutive patients with paroxysmal AF undergoing PVI procedure with a second-generation 28-mm cryoballoon. All patients were evaluated at 3, 6 and 12 months using a 12-lead ECG and 24-hour Holter monitoring. PV anatomy was assessed by creating three-dimensional models using computed tomography (CT) segmentations of the left atrium. Results: Fifty-one patients (61%) had left PVs with a shared carina, 35 patients (42%) had a shared right carina. Nine patients (11%) were classified having a right middle PV. In total 17 (20.2%) of patients had a left common PV. At 12 months, 14 patients (17%) had experienced AF recurrence. Neither PV ovality, variant anatomy, the presence of shared carina nor a common left PV was a predictor for AF recurrence. Conclusions: No specific characteristics of PV dimensions nor morphology were associated with AF recurrence after cryoballoon ablation in patients with paroxysmal AF.