Combining Rhythmic and Morphological ECG Features for Automatic Detection of Atrial Fibrillation: Local and Global Prediction Models

Atrial fibrillation (AF) is the most common type of heart arrhythmia. AF is highly associated with other cardiovascular diseases, such as heart failure, coronary artery disease and can lead to stroke. Unfortunately, in some cases people with atrial fibrillation have no explicit symptoms and are unaware of their condition until it is discovered during a physical examination. Thus, it is considered a priority to define highly accurate automatic approaches to detect such a pathology in the context of a massive screening. For this reason, in the recent years several approaches have been defined to automatically detect AF. These approaches are often based on machine learning techniques and—most of them—analyse the heart rhythm to make a prediction. Even if AF can be diagnosed by analysing the rhythm, the analysis of the morphology of a heart beat is also important. Indeed, during an AF events the P wave could be absent and fibrillation waves may appear in its place. This means that the presence of only arrhythmia could be not enough to detect an AF events. Based on the above consideration we have presented Morphythm, an approach that use machine learning to combine rhythm and morphological features to identify AF events. The results we achieved in an empirical evaluation seems promising. In this paper we present an extension of Morphythm, called Local Morphythm, aiming at further improving the detection accuracy of AF events. An empirical evaluation of Local Morphythm has shown significantly better results in the classification process with respect to Morphythm, particularly for what concerns the true positives and false negatives.