Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network

Comprehensive Summary

The aim of this study was to develop an artificial intelligence using deep learning models to identify a broad range of cardiovascular issues as opposed to previous research which trained these models to detect one specific issue or pattern. The deep neural networks were programmed to detect 10 arrhythmias, sinus noise, and 12 output rhythm classes. The Ai model was trained using a dataset containing 91,232 ECG results. To assess the model’s accuracy the results from the testing set (containing 328 ECG records) and compared these results to what was concluded by a council of cardiologists. The deep neural networks had a higher specificity than the cardiologists had when detecting the majority of irregular rhythms, showing how these models can improve the accuracy of ECG analysis. The uniqueness of this model is due to its ability to diagnose several types of cardiovascular issues, this is driven by a single algorithm instead of the previous models which have specialized in detecting one pattern and required an engineered derivation of EG for that specific classification. This allows for a more efficient system as no pre-processing of the ECG data is needed. Validation in clinical settings is needed to ensure its accuracy for clinical use, but the model as seen in testing has potential to improve efficiency in ECG analysis.

Outcomes and Implications

The deep learning model developed in this study can classify a large range of different arrhythmias. In clinical use, it has the potential to increase accuracy, scalability, and efficiency of ECG analysis. The model proved to have a similar accuracy to cardiologist on the testing set of data it was used on. Since this model can assess for multiple types of pattern changes, it proves to be even more useful than previous models which have focused on the detection of only one pattern change.