Atrial fibrillation (AF) is a public health burden with increasing incidence.Reentry events are drivers of AF, and investigation of their underlying mechanisms is ongoing.The influence of the conduction velocity (CV) on reentry events has been thoroughly studied in contrast to CV restitution.We simulated electrical wave propagation with the diffusion reaction eikonal alternant model (DREAM) in tissue slabs.Scenarios with combinations of three maximal longitudinal CV (CV 0 ) values (200, 400, and 600 mm/s) and 12 CV restitution curve morphologies extracted from electrode measurements in 13 patients with persistent AF, resulting in 36 experiments, were considered.In each experiment, an S1-S2 protocol was applied to induce reentry.The vulnerable window and average duration of reentries of each experiment were compared based on the steepness of the CV restitution.The vulnerable window increased with decreasing steepness of the CV restitution curve.The maximum increase was 59.6, 40.8, and 22.8 % for the three CV values, respectively.Similarly, for the average duration of reentries, a maximum increase of 38.6, 72.5 and 22.8 % was observed for less steep CV restitution.In conclusion, shallow CV restitution curves are characteristic of more vulnerable tissue and can contribute to maintenance of longer AF episodes.

Catheter ablation has the potential to become an effective treatment for ventricular tachycardia (VT), but the current identification of ablation sites relies on the operator's judgement and experience.This study proposes a novel machine learning approach to identify ablation targets based on signal features derived from intracardiac electrograms recorded in sinus rhythm.56 substrate maps were collected during pacing and sinus rhythm using a multipolar catheter (Advisor HD grid, Ensite Precision) in 13 pigs with chronic myocardial infarction (n=31,515 mapping points).36 VTs were induced and critical components of the VT circuit including early-, mid-and late-diastolic signals, were localized.Cardiac sites within 6 mm from these critical VT sites were considered as potential ablation targets (7.3% of all cardiac sites).46 features representing signal morphology, function, spatial and spectral properties were extracted from each bipolar and unipolar signal recorded during pacing or sinus rhythm.A random forest algorithm was trained on 80% of the data to identify the 20 most important features and 10 times 10-fold cross-validation was used to identify the best model.Validation on the remaining 20% of data showed an area under the ROC curve of 77.8%, and both 70% of sensitivity and specificity, for the best model.This study demonstrates for the first time that machine learning may support clinicians in the localization of targets for VT ablation.

Despite the rise of digital electrocardiogram (ECG) technology, paper-based ECGs continue to be prevalent, especially in underrepresented and underserved communities.This paper presents the DSAIL team's participation in the George B. Moody PhysioNet Challenge 2024 to develop an open-source algorithm for classifying ECG images.We fine-tuned a pre-trained InceptionV3 model on the PTB-XL dataset, comprising 21,799 12-lead ECG recordings, supplemented with synthetic ECG images from the ECG-Image-Kit.The model was trained using 80% of these images, reserving 20% for validation.Our choice of the InceptionV3 architecture leverages its capability to effectively capture local and global features, which is crucial for the inherent variability in ECG image patterns.The model achieved a validation macro F-measure score of 0.429 on a dataset accessible only to the organizers, securing 6th place on the official classification leaderboard.However, the algorithm struggled with mobile phone images of stained, deteriorated, and cleaned ECGs, yielding a low F-score of 0.08.In contrast, it performed significantly better on color scans of clean and deteriorated paper ECGs, achieving an F-score of 0.5.Although further improvements are necessary, neural network-based algorithms demonstrate promising potential for enhancing access to ECG-based diagnosis and cardiac care.

Mobile devices like smartwatches and handheld ECG recorders which can record single-lead ECGs which could be used to detect atrial fibrillation.A key step in using these ECGs is extracting RR-intervals.We aimed to develop an algorithm to automatically: (i) derive RRintervals from ECGs; and (ii) predict whether they are accurate enough for diagnosis.The TELE ECG Database was used, containing 250 ECGs recorded using a handheld ECG device with manual QRS annotations.An algorithm was designed to: (i) detect QRS complexes and extract RR-intervals using a primary QRS detector; and (ii) assess the reliability of each RR-interval through comparison with QRS complexes detected by a secondary QRS detector.Performance was assessed using each combination of three primary and 18 secondary open-source QRS detection algorithms.The algorithm with the lowest mean absolute error (MAE) in RR-intervals deemed to be reliable used 'unsw' (primary) and 'rpeak' (secondary) detectors, achieving a mean MAE of 24ms with 33% of RRintervals deemed reliable.The algorithm with the lowest mean MAE, whilst deeming at least 90% of RR-intervals reliable, used 'unsw' (primary) and 'nk' (secondary) detectors, achieving a mean MAE of 33ms with 92% of RRintervals deemed reliable.These algorithms accurately derive RR-intervals from telehealth ECGs.

Coronary Artery Disease (CAD) is a leading cause of mortality worldwide. This study aims to improve CAD diagnosis and facilitate tailored treatment strategies by implementing a radiomics-based machine learning algorithm for the automated evaluation of stenosis from coronary computed tomography angiography (CCTA).The study population included 220 patients undergoing CCTA for suspected CAD.Multiplanar Reconstruction images at 45 of rotation were obtained for each of the three primary coronary arteries, resulting in a total of 343 coronary artery segments.Each patient was assigned to: non-obstructive (0% stenosis), sub-obstructive (stenosis<50%) or obstructive (stenosis>50%) class.After radiomic features extraction and selection, a gradient boosting model was trained using a cascade approach, which subdivided the CAD-RADS scoring task into two easier sub-tasks: (A) no CAD vs. CAD and (B) nonobstructive vs. obstructive CAD.The dataset was divided into training and test sets (80-20%) and 5-fold cross-validation was applied to the training set to determine optimal hyperparameters.A balanced accuracy of 80% and AUC-ROC of 84% were achieved on the test set.These preliminary results are promising for automatic stenosis assessment, potentially useful in optimizing CAD management.

The prevalence of atrial fibrillation (AF) has significantly increased over the past five decades, establishing it as the leading sustained heart arrhythmia, globally.An elevated risk of cardiovascular disease was observed in patients with AF, particularly in cases of silent paroxysmal AF (PAF).Previous research had relied on ECG monitoring, heart rate variability (HRV), and premature atrial contractions (PAC) assessment to predict the risk of PAF.This study was conducted to replicate and evaluate these methods using the publicly available Icentia-11k Dataset.The data preprocessing was performed following a typical machine learning (ML) pipeline, which included noise filtering and baseline wander removal, prior to feature extraction from HRV in time, frequency, and complexity domains.Various ML classifiers were trained and tested with different sets of features, revealing that the information provided by premature beats, such as PACs, notably influenced specificity at the expense of sensitivity.

Accurate characterization of cardiac electrophysiological substrate is pivotal in cardiology for diagnosing and treating arrhythmias.The HD Grid catheter offers improved mapping capabilities, yet challenges persist in assessing tissue properties in data from swine undergone electroporation ablation.This study aims to characterize healthy and necrotic tissue using voltage amplitude and electrical propagation disorganization, using a Vector Field Heterogeneity (VFH) metric.We observed significant reductions in voltage amplitude and increases in heterogeneity post-ablation, indicating successful tissue modification.Correlations between voltage amplitude and heterogeneity underscore their complementary nature in tissue characterization, enhancing our understanding of cardiac tissue properties.This study categorized myocardial tissue using omnipolar amplitude and VFH, with healthy tissue showing high voltage and low heterogeneity, and necrotic tissue exhibiting low voltage and high heterogeneity, demonstrating their complementary nature in cardiac tissue characterization.

This study aims to explore the potential of deep transfer learning of pretrained image recognition deep neural networks (DNN) in a new diagnostic modality of atrial fibrillation (AF) using Holter ECG color maps.These maps consist of stacked color bars, transforming the ECG waveform amplitude of sequential heartbeats over 30 s into a color code, similar to ECHOView TM (Schiller AG, Switzerland).Using 13,536 training,6,624 validation and 11,400 test ECHOView color maps evenly split between AF and non-AF cases from the IRIDIA-AF Holter ECG monitoring database, we retrained, fine-tuned, and tested 18 image classification DNNs.Test accuracy of retrained DNNs (93.1-95.8%)was improved after fine-tunning (96.3-97.6%).Retrained EfficientNetV2B1 and finetuned MobileNetV2(=1.4) were both accurate and fast.ECHOView colormap images, composed of stacked heartbeat amplitudes, provide an interpretable visual modality for pretrained image classification networks that can be effectively retrained and fine-tuned to identify short AF episodes in Holter-ECG recordings.

Accurate automated segmentation of late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) images is crucial for personalized cardiac modeling, particularly for myocardial boundaries. This study aims to validate the accuracy of our proposed deep learning segmentation method for the left ventricular blood cavity (LV) and left ventricular myocardium (Myo).We propose a residual attention-based nnUNetv2 to automatically segment the LV and Myo.The residual attention mechanism enables the network to focus on more informative channels, enhancing discriminative learning and increasing the network's ability to segment the LV and Myo boundary.We also propose an image interpolation method that interpolates the voxel spacing of the original image and increases the number of image layers.Our proposed method achieved a Dice similarity coefficient (DSC) of 85.19% for LV and 72.12% for Myo.The lower DSC score for Myo is primarily due to the poor image quality of the LGE-CMR.The results demonstrate that our proposed residual attention-based nnUNetv2 network can effectively improve the network's ability to segment LV and Myo in LGE-CMR images.