Detecting Acute Ischemic Stroke from ECG Data - A Topological Approach
Acute ischemic stroke is one of the major causes of adult disability and mortality. Various studies have shown that patients with such stroke oftentimes experience autonomic imbalance which is reflected by a decreased heart rate variability. Hence, early detection of the stroke is made viable through the analysis of electrocardiogram (ECG) data. However, standard heart rate variability parameters are prone to human error, and they must be analyzed together with other physiological metrics such as respiratory rate. In this project, tools from both algebraic topology and machine learning are used in order to find a more reliable and robust way to differentiate patients with ischemic stroke through the ECG data. As tools from topology require a valid topological structure, the ECG data is first embedded into a Euclidean space as a point cloud with a Vietoris-Rips complex attached to it. Filtering the complex via the distance between points permits the use of persistent homology, a tool that tracks the evolution of homology over the filtration. Lastly, expressing the persistent homology in a vectorizable form so that machine learning models can be applied. Comparing to the traditional predictors, the model with topological predictors displays a vast increase in the accuracy of stroke patient classification.