Computing in cardiology最新文献

The 12-lead electrocardiogram (ECG) is the most common front-line diagnosis tool for assessing cardiovascular health, yet traditional ECG analysis cannot detect many diseases. Machine learning (ML) techniques have emerged as a powerful set of techniques for producing automated and robust ECG analysis tools that can often predict diseases and conditions not detectable by traditional ECG analysis. Many contemporary ECG-ML studies have focused on utilizing the full 12-lead ECG; however, with the increased availability of single-lead ECG data from wearable devices, there is a clear motivation to explore the development of single-lead ECG-ML techniques. In this study we developed and applied a deep learning architecture for the detection of low left ventricular ejection fraction (LVEF), and compared the performance of this architecture when it was trained with individual leads of the 12-lead ECG to the performance when trained using the entire 12-lead ECG. We observed that single-lead-trained networks performed similarly to the full 12-lead-trained network. We also noted patterns of agreement and disagreement between network low LVEF predictions across the different lead-trained networks.

Individual variability in parameter settings, due to either user selection or disease states, can impact accuracy when simulating the electrical behavior of the heart. Here, we aim to test the impact of inevitable uncertainty in conduction velocities (CVs) on the output of simulations of cardiac propagation, given three stimulus locations on the left ventricular (LV) free wall. To understand the role of physiological variability in CV in simulations of cardiac activation, we generated detailed maps of the variability in propagation simulations by implementing bi-ventricular activation simulations and quantified the effects by deploying robust uncertainty quantification techniques based on polynomial chaos expansion (PCE). PCE allows efficient stochastic exploration with reduced computational demand by utilizing an emulator for the underlying forward model. Our results suggest that CV within healthy physiological ranges plays a small role in the activation times across all stimulation locations. However, we noticed differences in variation coefficients depending on the stimulation site, i.e., LV endocardium, midmyocardium, and epicardium. We observed low levels of variation in activation times near the earliest activation sites, whereas there was higher variation toward the termination sites. These results suggest that CV variability can play a role when simulating healthy and diseased states.

Predictive models and simulations of cardiac function require accurate representations of anatomy, often to the scale of local myocardial fiber structure. However, acquiring this information in a patient-specific manner is challenging. Moreover, the impact of physiological variability in fiber orientation on simulations of cardiac activation is poorly understood. To explore these effects, we implemented bi-ventricular activation simulations using rule-based fiber algorithms and robust uncertainty quantification techniques to generate detailed maps of model variability. Specifically, we utilized polynomial chaos expansion, enabling efficient exploration with reduced computational demand through an emulator function approximating the underlying forward model. Our study focused on examining the epicardial activation sequences of the heart in response to six stimuli locations and five metrics of activation. Our findings revealed that physiological variability in fiber orientation does not significantly affect the location of activation features, but it does impact the overall spread of activation. We observed low variability near the earliest activation sites, but high variability across the rest of the epicardial surface. We conclude that the level of accuracy of myocardial fiber orientation required for simulation depends on the specific goals of the model and the related research or clinical goals.

In response to the 2023 George B. Moody PhysioNet Challenge, we propose an automated, unsupervised pre-training approach to boost the performance of models that predict neurologic outcomes after cardiac arrest. Our team, (BrownBAI), developed a model architecture consisting of three parts: a pre-processor to convert raw electroencephalograms (EEGs) into two-dimensional spectrograms, a three-layer convolutional neural network (CNN) encoder for unsupervised pre-training, and a time series transformer (TST) model. We trained the CNN encoder on unlabeled five-minute EEG samples from the Temple University EEG Corpus (TUEG), which included more than 20x the patients available in the PhysioNet competition training dataset. We then incorporated the pre-trained encoder into the TST as a base layer and trained the composite model as a classifier on EEGs from the 2023 PhysioNet Challenge dataset. Our team was not able to submit an official competition entry and was therefore not scored on the test set. However, in a side-by-side comparison on the competition training dataset, our model performed better with a pretrained (competition score 0.351), rather than randomly initialized (competition score 0.211) CNN encoder layer. These results show the potential benefits of leveraging unlabeled data to boost task-specific performance of predictive EEG models.

"Drivers" are theorized mechanisms for persistent atrial fibrillation. Machine learning algorithms have been used to identify drivers, but the small size of current driver datasets limits their performance. We hypothesized that pretraining with unsupervised learning on a large dataset of unlabeled electrograms would improve classifier accuracy on a smaller driver dataset. In this study, we used a SimCLR-based framework to pretrain a residual neural network on a dataset of 113K unlabeled 64-electrode measurements and found weighted testing accuracy to improve over a non-pretrained network (78.6±3.9% vs 71.9±3.3%). This lays ground for development of superior driver detection algorithms and supports use of transfer learning for other datasets of endocardial electrograms.

Left Ventricular Assist Devices (LVADs) are increasingly used as long-term implantation therapy for advanced heart failure patients, where candidacy assessment is crucial for successful treatment and recovery. A Deep Learning system based on Electrocardiogram (ECG) diagnoses criteria to stratify candidacy is proposed, implementing multi-model processing, interpretability, and uncertainty estimation. The approach includes beat segmentation for single-lead classification, 12-lead analysis, and semantic segmentation, achieving state-of-the-art results on the classification evaluation of each model, with multilabel average AUC results of 0.9924, 0.9468, and 0.9956, respectively, presenting a novel approach for LVAD candidacy assessment, serving as an aid for decision-making.

Patients with drug-refractory ventricular tachycardia (VT) often undergo implantation of a cardiac defibrillator (ICD). While life-saving, shock from an ICD can be traumatic. To combat the need for defibrillation, ICDs come equipped with low-energy pacing protocols. These anti-tachycardia pacing (ATP) methods are conventionally delivered from a lead inserted at the apex of the right ventricle (RV) with limited success. Recent studies have shown the promise of biventricular leads placed in the left ventricle (LV) for ATP delivery. This study tested the hypothesis that stimulating ATP from multiple biventricular locations will improve termination rates in a patient-specific computational model. VT was first induced in the model, followed by ATP delivery from 1-4 biventricular stimulus sites. We found that combining stimulation sites does not alter termination success so long as a critical stimulus site is included. Combining the RV stimulus site with any combination of LV sites did not affect ATP success except for one case. Including the RV site may allow biventricular ATP to be a robust approach across different scar distributions without affecting the efficacy of other stimulation sites. Combining sites may increase the likelihood of including a critical stimulus site when such information cannot be ascertained.

Ventricular tachycardia (VT) is a life-threatening cardiac arrhythmia for which a common treatment pathway is electroanatomical mapping and ablation. Recent advances in both noninvasive ablation techniques and computational modeling have motivated the development of patient-specific computational models of VT. Such models are parameterized by a wide range of inputs, each of which is associated with an often unknown amount of error and uncertainty. Uncertainty quantification (UQ) is a technique to assess how variability in the inputs to a model affects its outputs. UQ has seen increased attention in computational cardiology as an avenue to further improve, understand, and develop patient-specific models. In this study we applied polynomial chaos-based UQ to explore the effect of varying the tissue conductivity of fibrotic border zones in a patient-specific model on the resulting VT simulation. We found that over a range of inputs, the model was most sensitive to fibrotic sheet direction, and uncertainty in fibrotic conductivity resulted in substantial variability in the VT reentry duration and cycle length. Overall, this study paves the way for future UQ applications to improve and understand VT models.

The objective of this work was to evaluate the utility of using intrapartum fetal heart rate (FHR) and uterine pressure (UP) events to detect infants at risk of hypoxic-ischemic encephalopathy (HIE). We analyzed data from 40,976 term births from three groups: 374 infants that developed HIE, 3,056 that developed fetal acidosis without HIE, and 37,546 healthy infants. We counted the transitions between FHR events and the length of FHR and UP events. Then, we used these features to train a random forest classifier to discriminate between the healthy and the pathological (acidosis or HIE) groups. Compared to the Caesarean delivery rates for each group, our system detected 6.9% more HIE cases (54.9% vs 61.8%, p<0.001) and 10.7% more acidosis cases (37.6% vs 48.3%, p<0.001), with no increase in the false positive rates in the healthy group (38.9% vs 38.8%, p=0.26). Importantly, over 3/4 of the HIE detections were made 3 hours or more before delivery. It is reasonable to expect that this would be enough lead time to permit clinical intervention to improve the outcome of birth.

Atrial fibrillation (AF) afflicts more than 33 million people worldwide. Success of therapy strategies remains poor and better understanding of the arrhythmia and how to device more effective therapies are needed. The aim of this work is to study the role of electric power distributions in rotors and AF dynamics. For this purpose, single cell and tissue simulations were performed to study the effect of ionic currents gradients and fibrosis in rotor's drifting. The root mean square of the ionic (P_ion), capacitance (P_c) and electrotonic (P_ele) power was computed over action potentials. Single cell simulations were performed for different values of I_K1 and I_CaL and number of coupled myofibroblasts. Tissue simulations were performed in presence of I_K1 and I_CaL gradients and diffused fibrosis. Single cell simulations showed that P_ion and P_c increased with I_K1, while decreased by increasing I_CaL. Increasing the number of coupled myofibroblasts reduced P_ion and P_c, whereas P_ele increased. Finally, in tissue simulations rotors drifted to regions with low power and anchored in regions with higher density of blunted ionic induced power gradients.