D-AB25-04 - Machine Learning Successfully Discriminate ECG Responses To His Bundle Pacing (ID 1522)
Background: His bundle pacing (HBP) has emerged as an alternative to conventional ventricular pacing due to its ability to produce paced QRS complexes similar to intrinsic activation. Several different ECG responses can be observed when the His bundle is captured. The most common are selective HBP, where the His bundle alone is captured, and non-selective HBP where myocardium is captured alongside the His bundle, resulting in an apparently pre-excited QRS (pseudo-delta-wave). These responses can be difficult to discern by visual analysis of the ECG for those new to the field, but can be important for clinical decision-marking, such as selecting the pacing output and optimising lead position.
Objective: To use artificial intelligence (AI) in the form of supervised machine learning using a convolutional neural network (CNN) to automate ECG interpretation to discriminate selective from non-selective HBP.
Methods: We identified patients at Hammersmith Hospital who had undergone HBP using an EP system and extracted raw 12-lead ECG data from selective and non-selective HBP paced beats. A 1-dimensional CNN was trained on segmented QRS complexes labelled with their response-type (selective vs non-selective HBP).
Results: The CNN was trained with 1185 QRS complexes from 57 patients. The accuracy for non-selective HBP was 84.5% and for selective HBP was 76.1% and for both AI significantly more accurate than chance (P<0.0001).
Conclusion: We have demonstrated proof-of-concept that HBP ECG responses can be discriminated using AI. When a larger dataset is trained to higher accuracy, automated AI ECG analysis could facilitate HBP implantation and follow-up.