Abstract:

Interatrial conduction block refers to a disturbance in the propagation of electrical impulses in the conduction pathways between the right and the left atrium. It is a risk factor for atrial fibrillation, stroke, and premature death. Clin- ical diagnostic criteria comprise an increased P wave dura- tion and biphasic P waves in lead II, III and aVF due to ret- rograde activation of the left atrium. Machine learning algo- rithms could improve the diagnosis but require a large-scale, well-controlled and balanced dataset. In silico electrocardio- gram (ECG) signals, optimally obtained from a statistical shape model to cover anatomical variability, carry the poten- tial to produce an extensive database meeting the requirements for successful machine learning application. We generated the first in silico dataset including interatrial conduction block of 9,800simulated ECG signals based on a bi-atrial statistical shape model. Automated feature analysis was performed to evaluate P wave morphology, duration and P wave terminal force in lead V1. Increased P wave duration and P wave ter- minal force in lead V1 were found for models with interatrial conduction block compared to healthy models. A wide vari- ability of P wave morphology was detected for models with in- teratrial conduction block. Contrary to previous assumptions, our results suggest that a biphasic P wave morphology seems to be neither necessary nor sufficient for the diagnosis of in- teratrial conduction block. The presented dataset is ready for a classification with machine learning algorithms and can be easily extended.

Abstract:

Interatrial conduction block (IAB) refers to a disturbance in the propagation of electrical impulses from the right to the left atrium via the interatrial conduction pathways. It is a risk factor for atrial fibrillation, stroke and premature death. The clinical diagnosis is based on an increased P wave duration and biphasic P waves in lead II, III and aVF due to the retrograde activation of the left atrium. Recently, Bayés de Luna et al. presented patients with atypical IAB due to duration or morphology who did not fulfill the clinical diagnosis criteria. Instead, typical morphology patterns of IAB without a prolonged P wave duration or different morphology patterns in some of the leads were found.Machine learning (ML) algorithms such as a feedforward neural networks (FFNNs) have already been successfully applied for the detection and classification of cardiac diseases based on electrocardiogram (ECG) signals. They could therefore improve the diagnosis of IAB but should be based on a large-scale, well-controlled and balanced dataset. In silico ECG signals carry the potential to produce an extensive database meeting the requirements for a successful machine learning application.In this thesis, an in silico ECG dataset for IAB based on 98 atrial geometries derived from a bi-atrial statistical shape model (SSM) to cover the anatomical variability of the atria was generated. An automated feature analysis was performed to evaluate 100 different P wave features including P wave morphology, P wave duration, P wave dispersion, P wave terminal force in lead V1 (PTF-V1), P wave area, root mean square (RMS) voltages and P wave amplitude. Based thereon, the general feasibility of a classification of IAB with FFNNs was investigated for various combinations of input features.None of the extracted features showed distinct ranges for healthy models and models with IAB. P wave duration and PTF-V1 were increased for signals with IAB compared to healthy signals. A wide variability of P wave morphology was detected for models with IAB. Some of the models with IAB did not fulfill any of the clinical diagnosis criteria but had an atypical P wave morphology. Using the same features as in the clinical setting, a FFNN was able to distinguish between signals from healthy geometries and geometries with IAB with an accuracy of 88.38% on average. The performance could be even improved to an accuracy of 97.44% on average by using more input features. Using only a single lead as input data, the classification of IAB could be performed with an accuracy of 95.00 % on average.Contrary to previous assumptions in literature, the results suggest that a biphasic morphology in lead III seems to be neither necessary nor sufficient for the diagnosis of IAB. Moreover, no single and decisive feature could be identified for models with IAB. However, the results of this work prove that a classification of IAB with FFNNs is generally feasible. Furthermore, this work suggests that the diagnosis of IAB can be even improved with a neural network. The results further indicate that IAB could be classified using only a single lead recording. However, only a binary classification of healthy signals and signals with IAB was studied and further research is needed for the differential diagnosis of IAB. The generated dataset is ready for further studies on classification with ML algorithms, either by using a different algorithm or in a multiclass classification setup.