Abstract:

Aims Atrial flutter (AFlut) is a common re-entrant atrial tachycardia driven by self-sustainable mechanisms that cause excitations to propagate along pathways different from sinus rhythm. Intra-cardiac electrophysiological mapping and catheter ablation are often performed without detailed prior knowledge of the mechanism perpetuating AFlut, likely prolonging the procedure time of these invasive interventions. We sought to discriminate the AFlut location [cavotricuspid isthmus-dependent (CTI), peri-mitral, and other left atrium (LA) AFlut classes] with a machine learning-based algorithm using only the non-invasive signals from the 12-lead electrocardiogram (ECG). Methods and results Hybrid 12-lead ECG dataset of 1769 signals was used (1424 in silico ECGs, and 345 clinical ECGs from 115 patients—three different ECG segments over time were extracted from each patient corresponding to single AFlut cycles). Seventy-seven features were extracted. A decision tree classifier with a hold-out classification approach was trained, validated, and tested on the dataset randomly split after selecting the most informative features. The clinical test set comprised 38 patients (114 clinical ECGs). The classifier yielded 76.3% accuracy on the clinical test set with a sensitivity of 89.7%, 75.0%, and 64.1% and a positive predictive value of 71.4%, 75.0%, and 86.2% for CTI, peri-mitral, and other LA class, respectively. Considering majority vote of the three segments taken from each patient, the CTI class was correctly classified at 92%. Conclusion Our results show that a machine learning classifier relying only on non-invasive signals can potentially identify the location of AFlut mechanisms. This method could aid in planning and tailoring patient-specific AFlut treatments.

Abstract:

Atrial Fibrillation (AF) is the most common cardiac arrhythmia in clinical practice and one of the leading causes for hospitalization and death. Rotational activities (rotors) and focal activities (focal sources) are being considered to drive and sustain AF. Identifying the presence of these mechanisms is fundamental for planning the best treatment to heal patients. Commonly AF is treated through ablation of the affected area guided by intracardiac catheters. By using a non-invasive method like the Electrocardiogram (ECG), characterization of the mechanism would be simplified and the identification process would be optimized. The ECG is widely used in clinical practice and it is therefore a perfect tool for early diagnosis and arrangement of targeted treatment. Using the ECG to determine the AF mechanism would enable physicians to directly chose the right approach for therapy and further examination. This project analyzed 12-lead ECG signals obtained by simulating rotors and focal sources driving AF on a computer model of the atria. These two mechanisms were simulated in the exact same position and with the same basic cycle length (bcl) making the results comparable and ensuring that the differences in the signals arise solely from the different mechanisms. In total 31 atrial scenarios were simulated each with a unique point of rotor stimulation and bcl. Each scenario was transformed on 8 different torso geometries resulting in a total number of 248 torso scenarios. For each of the torso scenarios the 12-lead ECG was obtained and analyzed with several signal processing methods resulting in 182 features. With alpha = 0.01, 77.47% of the features used in this work showed statistically significant differences between focal source and rotors. Furthermore, the features were analyzed using the Receiver Operator Curve (ROC), and a decision tree classifier was implemented for a multi-feature classification. These methods characterized features that perform especially well in the discrimination between rotors and focal sources. Deriving from the ROC analysis the most powerful features corresponding to AUC values between 0.9297 and 0.9111 were standard deviation of voltage amplitude of the vector combination of I, aVF, and V1, the mean organization index over 12 leads, and the standard deviation of R-value from PC4. Decision tree classification achieved 89.92% average accuracy on a pre-fixed test set over 100 trials. The methods and features presented in this thesis can potentially be used to determine the mechanism driving AF. This work demonstrated that the introduced methods can reliably distinguish the most simple cases of AF mechanisms. Based on these findings it might be possible to non-invasively characterize the mechanism of AF in patients and thus will support the physicians treating this arrhythmia.