HRX 2024 AbstracX Library-A Step Forward in Predictive Cardiology: AI-Driven ECG Algorithm

A Step Forward in Predictive Cardiology: AI-Driven ECG Algorithm

Pdf Summary

The document discusses a significant advancement in predictive cardiology through the development of an AI-driven algorithm designed to predict intraventricular conduction abnormalities (ICAs) using electrocardiogram (ECG) data. ICAs can be asymptomatic or emerge after heart procedures, posing risks of severe heart rhythm issues, including sudden falls and cardiac arrest. A wide QRS complex, identifiable via ECG, is a key marker of ICAs.<br /><br />The objective of the study was to create a model capable of predicting the occurrence of wide QRS complexes using current 12-lead ECG data. The study involved collecting ECG signals from patients who developed a wide QRS complex within 31 days, as well as those who did not. The ECG data were segmented into three-beat units and divided into training, validation, and test sets to ensure no patient data duplication.<br /><br />The methodological approach was two-staged. In Stage 1, a U-Net model was employed to reconstruct initial ECG measurements, achieving a mean squared error of 0.03 on the validation set. In Stage 2, the encoder from the pretrained U-Net was utilized to train a new feedforward neural network (FNN), aimed at predicting wide QRS complexes. <br /><br />The model demonstrated an F1 score of 75% on the test set, a substantial improvement over the random prediction benchmark of 49.02%. This finding underscores the potential of the model to reliably predict ICAs, enhancing early detection and intervention capabilities in cardiology. This AI-driven approach thus represents a promising step forward in managing and mitigating the risks associated with cardiac conduction abnormalities.

Keywords

predictive cardiology

AI-driven algorithm

intraventricular conduction abnormalities

electrocardiogram

wide QRS complex

U-Net model

feedforward neural network

F1 score

early detection

cardiac conduction abnormalities